Tuomas Ruponen

Improving energy performance of buildings through exploitation

of available data

Rakennusten energiasuorituskyvyn parantaminen hyödyntämällä

saatavilla olevaa dataa

Master’s thesis
Espoo, October 3rd 2016

Supervisor: Assistant Professor Antti Peltokorpi D.Sc. (Tech.)

Advisor: Property Manager Jaana Ihalainen M.Sc. (Tech.)

 Aalto University, P.O. BOX 11000, 00076 AALTO
www.aalto.fi
Abstract of Master's thesis

Author Tuomas Ruponen

Title Improving energy performance of buildings through exploitation of available data
Degree programme Structural Engineering and Building Technology
Minor Construction Management and Economics Code IA3022
Thesis supervisor Assistant Professor Antti Peltokorpi D.Sc. (Tech.)
Thesis advisor Jaana Ihalainen M.Sc. (Tech.)
Date 03.10.2016 Number of pages 99+8 Language English

Abstract
The objective of this research is to find out how the energy performance of buildings can
be improved effectively by exploiting available data in the operations and maintenance
phase. In this research building automation systems, open data and Internet of Things
are studied as value generating technological solutions. The research process is based on
reviewing research literature, conducting interviews and analyzing measurement data
recorded by the building automation system of a case office building.

The study identified 11 initiatives to close existing energy performance gaps. These
initiatives belonged to categories of developing building services equipment control,
increasing the extent of available data, observing the state of user experience and
facilitating maintenance processes. Then the three most effective were chosen for a
feasibility study. This effectiveness of an initiative was judged by evaluating it in the
dimensions of expected benefits and challenge to implement in an indicative manner by
13 interviewees in four stakeholder groups. This simple evaluation method turned out to
serve its purpose well: Vague evaluation dimensions covered both quantitative and
qualitative aspects, while stakeholder groups had different perspectives on the initiatives.
Thus the method is recommended for similar problems, as long as only indicative results
are pursued.

Out of the 11 initiatives, the most effective ones were considered to be those that are
simple and do not require any installation work, or at the most the installation of
transmitters or sensors: 1) adaptive heating control, 2) user satisfaction measurement
systems, 3) energy performance monitoring systems and 4) selected equipment group
control interfaces. Feasibility studies suggested that adaptive heating control has the
potential to increase energy performance with negligible installation work, user
satisfaction measurement system would be sensible to pilot as a service, energy efficiency
monitoring in small-scale would be convenient to purchase as a service and selected group
control interfaces enable large savings with small trouble. The least effective initiatives
were considered to be the ones that are complex, risk user satisfaction or require the
integration of numerous systems.

Keywords energy performance, building automation, Internet of Things, open data

 Aalto-yliopisto, PL 11000, 00076 AALTO
www.aalto.fi
Diplomityön tiivistelmä

Tekijä Tuomas Ruponen

Työn nimi Rakennusten energiasuorituskyvyn parantaminen hyödyntämällä saatavilla
olevaa dataa
Koulutusohjelma Rakenne- ja rakennustuotantotekniikan koulutusohjelma
Sivuaine Rakentamistalous Koodi IA3022
Työn valvoja Apulaisprofessori, TkT Antti Peltokorpi
Työn ohjaaja DI Jaana Ihalainen
Päivämäärä 03.10.2016 Sivumäärä 99+8 Kieli englanti

Tiivistelmä
Tämän tutkimuksen tavoitteena on selvittää miten rakennusten energiasuorituskykyä
voidaan parantaa tehokkaasti hyödyntämällä saatavilla olevaa dataa käyttö- ja
ylläpitovaiheessa. Tutkimuksessa arvoa tuottavina teknologisina osaratkaisuina tutkitaan
rakennusautomaatiojärjestelmiä, avointa dataa ja Esineiden Internetiä. Tutkimus-
menetelminä käytetään kirjallisuustutkimusta, haastatteluja ja toimistorakennuksen
rakennusautomaatiojärjestelmän tuottaman mittaustiedon analysointia.

Tutkimus tuotti 11 energiasuorituskyvyn ongelmakohtien korjaamiseen tähtäävää

aloitetta. Näiden aloitteiden päämäärinä oli taloteknisten laitteiden ohjauksen
kehittäminen, käytettävissä olevan datan lisääminen, käyttäjätyytyväisyyden tarkkailu tai
ylläpidon prosessien helpottaminen. Aloitteista kolmelle tehokkaimmiksi arvioiduille
tehtiin tarkempi toteutettavuustutkimus. Tehokkuusarviointi perustui neljään
sidosryhmään jaetun 13 haastateltavan suuntaa-antavaan näkemykseen aloitteiden
toimeenpanon hyödyistä ja haasteista. Tämä yksinkertainen arviointitapa osoittautui
toimivaksi: Moniselitteiset arviointiulottuvuudet kattoivat sekä määrälliset että
laadulliset näkökulmat, kun taas eri sidosryhmät painottivat aloitteiden eri
ominaisuuksia. Näin ollen kyseinen arviointimenetelmä soveltuu samankaltaisiin
ongelmiin, kunhan tulosten suuntaa-antava taso on riittävä tutkimuksen tavoitteisiin
nähden.

Näistä 11 aloitteesta tehokkaimmiksi koettiin pääosin sellaiset, jotka ovat yksinkertaisia

eivätkä vaadi laajaa asennustyötä: 1) adaptiivinen lämmityksen säätö, 2)
käyttäjätyytyväisyyden mittausjärjestelmä, 3) energiatehokkuuden seurantajärjestelmä
ja 4) laitteiden ryhmähallintaan perustuvat käyttöliittymät. Toteutettavuustutkimusten
perusteella adaptiivinen lämmityksensäätö voisi parantaa energiasuorituskykyä pienellä
asennustyöllä, käyttäjätyytyväisyyden mittausjärjestelmää olisi järkevää aluksi kokeilla
palveluna, energiatehokkuuden seuranta pienessä mittakaavassa olisi kätevää ostaa
palveluna ja ryhmäohjaukseen perustuvat käyttöliittymät voisivat säästää huomattavasti
energiaa pienellä vaivalla. Tehottomimmiksi arvioidut aloitteet olivat monimutkaisia,
vaaransivat käyttäjätyytyväisyyden tai vaativat useiden järjestelmien yhteensovittamista.

Avainsanat energiatehokkuus, rakennusautomaatio, Esineiden Internet, avoin data

Acknowledgements
I express my gratitude to the number of people and organizations who contributed to this
Master’s thesis.

This thesis is a part of the project “Improving the life-cycle value of buildings: Digitalization
of performance-focused business” (DIGIBUILD). Within this framework the aim of the
thesis is to use a case study to develop directions to design processes and systems to improve
performance of buildings in the dimensions of user experience, use value, energy efficiency
and maintenance efficiency, with open data and Internet of things as central elements.

I thank my supervisor Assistant Professor Antti Peltokorpi D. Sc. (Tech.) for his guidance
during the research process. I am also grateful to my advisor Jaana Ihalainen M. Sc. (Tech.),
Teppo Salmikivi Lic. Sc. (Tech.), Aimo Hämäläinen M. Sc. (Tech.), Jouni Poikkeus M. Eng.
and Jussi Tuomola B. Eng. from Helsinki University for their counseling throughout this
multidisciplinary work, and to the CEO Sampsa Laine D. Sc. (Tech.) from DataRangers Oy
for allowing me to use their Louhin-platform for data analysis free of charge.

Similarly, I am thankful to all the people who went through the trouble to participate in the
interviews and shared their insight, but whose names cannot be listed here due to ethical
reasons.

Lastly, I appreciate Aalto University School of Science, Aalto University School of

Engineering, Helsinki University Center for Properties and Facilities, TA-Yhtymä Oy,
Granlund Oy and Fira Oy for committing to the DIGIBUILD project as sites of research,
and to Tekes, Aalto University Foundation, TA-Yhtymä Oy, University of Helsinki Funds,
Fira Oy and Granlund Oy for funding the project.

Helsinki, September 26th 2016

Tuomas Ruponen
Contents
Abstract
Tiivistelmä
Acknowledgements
Contents
Acronyms
Abbreviations
1 Introduction .................................................................................................................... 1
1.1 Research background .............................................................................................. 1
1.2 Research objectives ................................................................................................. 2
1.3 Research scope and structure .................................................................................. 3
2 Energy performance of buildings .................................................................................. 5
2.1 Overview of the performance approach in buildings .............................................. 5
2.2 Improving the energy performance of buildings ..................................................... 7
2.2.1 Energy performance of building services equipment ...................................... 7
2.2.2 Energy performance of equipment control ...................................................... 9
2.2.3 Quality of the indoor environment................................................................. 10
3 Development of building automation systems to improve energy performance of
buildings............................................................................................................................... 13
3.1 Overview on building automation systems ........................................................... 13
3.2 Traditional system architecture ............................................................................. 13
3.3 Development of system architecture ..................................................................... 14
3.3.1 Development in the context of Internet of Things paradigm ......................... 14
3.3.2 Development towards a service-oriented model of building automation ...... 15
3.4 Energy performance with available data ............................................................... 15
3.4.1 Traditionally collected data ........................................................................... 15
3.4.2 Deployment of Internet of Things related technology ................................... 15
3.4.3 Open data ....................................................................................................... 16
3.5 Development in building automation control systems .......................................... 16
3.5.1 Overview of building automation control ...................................................... 16
3.5.2 Conventional control...................................................................................... 16
3.5.3 Occupancy detection methods ....................................................................... 17
3.5.4 Model-based predictive control ..................................................................... 19
3.5.5 Artificial intelligence based control ............................................................... 20
3.6 Summary of the theoretical background ............................................................... 22
4 Research method .......................................................................................................... 23
4.1 Overview of the research method ......................................................................... 23
4.2 Description of the case study ................................................................................ 23
4.2.1 Case organization ........................................................................................... 23
4.2.2 Case building automation system .................................................................. 23
4.2.3 Case building ................................................................................................. 24
4.3 Research process, data and analysis ...................................................................... 24
4.3.1 Research process and data collection............................................................. 24
4.3.2 Data analysis tools ......................................................................................... 26
4.3.3 Limitations in the research method ................................................................ 27
5 Results .......................................................................................................................... 29
5.1 Identified energy performance gaps ...................................................................... 29
5.1.1 Current approach to managing energy performance...................................... 29
 5.1.2 Existing energy performance gaps ................................................................. 30
5.1.3 Summary of identified energy performance gaps .......................................... 31
5.2 Overview and limitations of available data ........................................................... 31
5.3 Evaluation of current control systems and their energy performance ................... 35
5.3.1 Heating system control................................................................................... 35
5.3.2 Cooling system control .................................................................................. 40
5.3.3 Ventilation system control ............................................................................. 41
5.3.4 Lighting system control ................................................................................. 43
5.3.5 Summary of all energy performance gaps ..................................................... 45
5.4 Identified solutions to close the energy performance gaps ................................... 46
5.4.1 Installing electronic radiator thermostats with smart heating functions ........ 46
5.4.2 Adaptive outdoor temperature compensation for heating .............................. 48
5.4.3 Room setback temperatures outside office hours .......................................... 50
5.4.4 Optimizing heating with weather forecast data.............................................. 54
5.4.5 Intelligent lighting and blind control with daylight harvesting...................... 57
5.4.6 Improving the manageability of large amounts of sensor data and actuators 61
5.4.7 Enabling communication between building automation and room reservation
systems to reduce energy consumption and manual input ........................................... 64
5.4.8 Satisfying the dynamic demands of users with artificial intelligence based
control of equipment .................................................................................................... 66
5.4.9 Utilizing context-aware alarm thresholds in fault detection .......................... 68
5.4.10 Introducing mobile applications to collect sensing data on the indoor
environment from building users ................................................................................. 70
5.4.11 Optimizing pressure differences with ventilation .......................................... 73
5.4.12 Summary of the initiatives ............................................................................. 74
5.5 Feasibility study of selected improvements .......................................................... 76
5.5.1 Selection rationale .......................................................................................... 76
5.5.2 Measuring user satisfaction ............................................................................ 78
5.5.3 Group control of actuators ............................................................................. 79
5.5.4 Energy efficiency monitoring ........................................................................ 83
5.5.5 Adaptive outdoor temperature compensation for heating .............................. 84
6 Conclusion ................................................................................................................... 87
6.1 Discussion of the findings ..................................................................................... 87
6.2 Theoretical contribution ........................................................................................ 88
6.3 Practical implications ............................................................................................ 88
6.4 Research limitations .............................................................................................. 89
6.5 Further research ..................................................................................................... 90
References ............................................................................................................................ 91
Appendices
Acronyms
P [kW] heating power of radiator network
Pexhaust [kW] electric power of an exhaust fan
Psupply [kW] electric power of a supply fan
SFP [kW/(m3/s)] specific fan power
T11 [°C] exhaust air inlet
T21 [°C] supply air inlet
T22 [°C] supply air outlet
c [kJ/(kg °C)] specific heat capacity of water
qmax [m3/s] the maximum of supply and exhaust air flows
m
[kg/h] water flow through a heat exchanger
Δt
ΔT [°C] district heating water cooling
η [-] heat recovery efficiency ratio
ω [%] heat exchanger valve position
Abbreviations
AI Artificial Intelligence
ANN Artificial Neural Network
ASHRAE American Society of Heating, Refrigerating, and Air-Conditioning Engineers
BAS Building Automation System
BEMS Building Energy Management System
CBD Central Business District
CPS Cyber-Physical System
DDC Direct Digital Control
EU European Union
FDD Fault Detection and Diagnostics
FK Foreign Key
FMI Finnish Meteorological Institute
GA Genetic Algorithm
GHG Greenhouse Gas
IAQ Indoor Air Quality
IEQ Indoor Environmental Quality
MBPC Model-based Predictive Control
PBB Performance-Based Building
PK Primary Key
RFID Radio-Frequency Identification
SFP Specific Fan Power
WSN Wireless Sensor Networks
1 Introduction
1.1 Research background
International action on climate change is increasing, and at the same time buildings are
responsible for over 30% of the global energy consumption (IEA, 2015). This has led to
emerging pressure from policymakers to increase the sector’s energy efficiency. However
as residential buildings account for over 40% of the building sector’s carbon dioxide (CO2)
emissions (WBCSD, 2009), encouraging collective action is difficult. It is a typical “tragedy
of the commons” (Lu, et al., 2010), as the financial incentive for individual homeowners to
improve energy efficiency is negligible in comparison to the global motivation. For an
average household in the United States for example, a 20-30% reduction in heating,
ventilation and air-conditioning (HVAC) energy consumption would correspond to savings
of around $15 per month, while the national savings would be $15 billion annually with 1.12
billion tons less pollutants in the atmosphere (Lu, et al., 2010). Yet, this is not the case with
large property owners. Saving 10% in the heating expenses of a 6,000m2 office building in
Helsinki could translate to savings of 2,000-3,000€ annually, which would be enough to
justify small scale investments. Thus, there is a stronger alignment of interests with large
property owners and policymakers, rather than with individual homeowners and
policymakers, which reduces the policy resistance of improving the energy performance of
buildings.

However, it seems that self-interest has not been expected to be a sufficient driving force in
the European Union (EU) that “… prefers to perceive itself as the frontrunner of
environmental policy” (Korhola, 2014). This determination has materialized as the 2020
climate & energy package, which in turn led to the 2010 Energy Performance of Buildings
Directive. This directive contains the requirement that all new buildings must be nearly zero
energy buildings by the end of 2020 (European Commission, 2016a). However, in the wake
of the Paris Agreement of December 2015 it has been observed that the current policy will
not be enough to meet the 2030 reduction goals for greenhouse gas (GHG) emissions, so
member states are given national reduction targets (European Commission, 2016b). For
Finland, the current Proposal for an Effort Sharing Regulation 2021-2030 includes a 2030
GHG reduction target of 39% (compared to 2005) for sectors outside Emission Trading
System such as transport, buildings and agriculture (European Commission, 2016c). Based
on the Climate Change Act 609/2015 of Finland it is evident that this target would reflect
the national building codes (Finlex, 2015). Thus, property owners can expect more binding
regulation driving towards increased energy performance.

Nonetheless, the energy performance of buildings cannot be prioritized over the health of
building users. It is essential that building automation systems (BAS) control the building
services equipment so that the criteria set for indoor environmental quality (IEQ) are met
while making sure that the equipment are used energy efficiently to meet that goal. The
philosophy behind this aspect of energy performance is demand-based control: Resources
are used sparingly to meet the fluctuating demand. To enable computerized demand-based
control, the demand must be estimated based on available data. This data is traditionally
collected with a number of sensors deployed throughout the building and equipment, but can
also include new technologies and open data. The energy performance of buildings can also
be improved by ensuring that equipment are operating as intended. Undetected faulty
equipment or suboptimal control can incur significant amounts of wasted energy over time.

1
To avoid this, building automation systems continuously process sensor data and generate
alarms whenever undesired phenomena is detected. Sensor data can also be used to monitor
and benchmark key figures such as specific resource consumptions of equipment.

The ability of the BAS to promote energy performance is always restricted, though. The
capabilities of existing building services equipment and the realities of the construction
market dictate to which extent it is feasible to strive towards a high level of performance that
is based on computerization. If the ventilation system is incapable of purging rooms
individually with different air flows, there is no way that a computer program could
transform such a ventilation system into a highly demand-controlled one. Also, as a
procurement, the BAS has a short life cycle in comparison with building services equipment
in general, which means that all the automation-based investments leading to savings should
have a short payback period. Moreover, large property owners can be cautious about giving
up their ability to invite tenders from different providers: Any building automation solution
that would unite the provider and customer to the foreseeable future is undesirable.

Research literature has presented many solutions that utilize computerization and available
data in order to improve the energy performance of buildings. Some techniques presented in
those solutions have also made their way through commercialization into specialized
products or services. Yet, there seems to be a gap in the knowledge regarding the
effectiveness of those solutions, or they have been found ineffective, for their adoption is
not widespread at least in Finland. It could also be that some solutions have not even been
granted a serious consideration due to policy resistance. If two of these options were the
case, it would be unfortunate as the energy performance of buildings appears to be a global
concern, and without self-driven improvement the industry will be forced to meet tightening
policymakers’ goals. Hence, this thesis is about exploring solutions to improve the energy
performance of buildings in a low-threshold approach by exploiting available data, and
investigating the effectiveness of those solutions.

1.2 Research objectives

The purpose of this research is to increase understanding on solutions that improve the
energy performance of a building by exploiting available data. This objective is reformulated
into the following research question:
• How the energy performance of buildings can be improved effectively by exploiting
available data?

Before answering this question, the concepts energy performance of a building and available
data need to be explored, which is done later in the thesis. Here, effectively refers to
implications of a solution that suggest the largest benefits with the smallest trouble or costs
associated to implementation and use.

The research question can be further divided into the three sub-questions:
1. What are the energy performance gaps of buildings?
2. What kind of solutions that are based on the exploitation of available data can close
those gaps?
3. How effective are those solutions?

2
1.3 Research scope and structure
This thesis concentrates in detecting energy performance gaps in buildings, seeking solutions
to fill them by exploiting available data and evaluating the effectiveness of those solutions.
The research only includes factors of energy performance that can be enhanced with
available data in the operations and maintenance phase of a building. This means that energy-
efficient equipment control principles and methods to supervise and manage equipment so
that they are operating energy-efficiently are included in the research as long as available
data is being exploited. Such tasks are often executed with a BAS, for which those systems
are also in the focus of research. Moreover, energy performance of a building cannot be
improved in the operations and maintenance phase without considerations regarding indoor
environmental quality. Therefore solutions that facilitate this aspect are also included in the
research.

As the thesis is focused on the operations and maintenance phase of a building’s life cycle,
the energy performance aspects in the design of a building or its services equipment are
excluded from research. Out of building types, the focus is on office buildings due to the
high level of computerization and range of equipment. Furthermore, due to the
multidisciplinary field of this research and often lack of relevant expertise available, the
proposed performance-enhancing solutions are not described in high detail. Only the
principles of the solutions are introduced so that the potential effectiveness can be estimated.

The thesis continues with chapter two, where the energy performance of buildings is
discussed. Chapter three introduces briefly building automation systems in the context of the
Internet of Things paradigm and advanced controlling strategies prevailing in research
literature. In chapter four the research method and case study are presented. Chapter five
includes the results of the complete research process: identifying energy performance gaps,
suggesting potential solutions to close them, selecting the most effective solutions and
studying their feasibility. In chapter six the research process is concluded and further
research opportunities suggested.

3
4
2 Energy performance of buildings
2.1 Overview of the performance approach in buildings
The performance approach is “the practice of thinking and working in terms of ends rather
than means” and suggested to be an effective way to promote innovation (Gibson, 1982, p.
4; Sexton & Barrett, 2005). In this approach the requirements for a building’s performance
are articulated, but the means to achieve such performance are left for the suppliers to decide.
It is argued that the current prescriptive codes reduce the motivation of businesses to
innovate in the building sector. (Sexton & Barrett, 2005).

There are several technical challenges related to the performance approach. Firstly,
formulating the target performance of a building is regarded difficult. It would include
objectives, functional attributes and performance requirements that are not covered by
building codes. So there is a need to create a framework of requirements for clients’ use.
Secondly, the availability of tools to design, deliver or evaluate the performance of a given
solution differ. Some performance areas have tools and some do not. Thirdly, the depth of
knowledge is very uneven across the field of building technology. In some performance
areas the formulation of quantified requirements is possible, while in other areas only
qualitative requirements can be imposed. Finally, the interdependencies of performance
requirements may not be comprehended resulting in exclusionary requirements. (Foliente,
et al., 2005).

Furthermore, there are concerns regarding the situation of not relying on building codes. The
code-based methods often deliver a cost-effective and a reliable solution, in other words, the
performance of code-based methods is not thought to be a concern. If a transition to
performance-based buildings (PBB) is made, routine designs become very laborious due to
the burden of proof on the contractor about a building’s performance. (Sexton & Barrett,
2005). Yet, this very aspect can be regarded as the essence of PBB: Buildings can fulfil all
the energy related building codes and meet low-energy certification requirements while
performing poorly energy-wise. The energy consumption of Leadership in Energy and
Environmental Design -certified buildings deviates by average more than 25% from design
projections (Turner & Frankel, 2008, p. 32), while Finnish national building codes have not
been observed to cover pressure differences over the building envelope adequately
(Lindgren, 2012, p. 37; Katainen & Vähämaa, 2015, p. 90). To achieve a higher level of
energy performance, the ends must gain importance over means. Also with PBB being
widely accepted, maintenance contracts could specify the end result such as IEQ rather than
defining the operation and maintenance work content itself (Ihasalo, 2012). Still, a
prerequisite condition for PBB is that actors throughout the supply chain are able and
interested in innovating individually and collectively, which may not be the case (Sexton &
Barrett, 2005). In reality, sometimes innovation might mean the cannibalization of existing
business which leads to reluctance to invest in such (Henderson, 1993).

Generally, the performance of a building is a multidimensional concept. As such it can be

approached differently by each stakeholder (Ihasalo, 2012). From user’s viewpoint the main
aspects of building performance may regard the satisfactory state of indoor microclimate and
the functionality of space, whereas the owner may be more concerned with maintenance
costs and energy consumption. Therefore it is essential to find ways to improve the
performance of the building for all the actors rather than focusing on just one field of

5
performance. For example advocating energy performance in an inappropriate manner could
jeopardize the longevity of the tenancy agreement through degraded indoor environment.
Depending on perspective, performance can then be evaluated quantitatively or qualitatively.
Therefore it can be assessed by physical measurements, findings of an evaluator or surveys
(Ihasalo, 2012).

Out of the performance indicators listed by Lavy et al. (2010) in Table 2.1, energy
performance can be indicated by operating costs, utility costs, building maintenance costs,
maintenance efficiency, building physical condition, health and safety, indoor environmental
quality (IEQ) and resource consumption. More specifically, in this thesis the energy
performance of a building is approached from three perspectives: 1) energy performance of
building services equipment, 2) energy performance of equipment control and 3) quality of
the indoor environment. Building automation systems (BAS) are a useful tool in evaluating
performance in these dimensions, as they produce a large amount of continuous physical
measurements.

Table 2.1 Performance indicators listed by Lavy et al. (2010).

Financial Functional Physical
Survey-based indicators
indicators indicators indicators
• Operating costs • Productivity • Building • Customer/building
• Occupancy costs • Parking physical occupants’ satisfaction
• Utility costs • Space utilization condition with products or services
• Capital costs • Employee or • Property and • Community satisfaction
• Building occupant’s real estate and participation
maintenance costs turnover rate • Waste • Learning environment,
• Grounds-keeping • Mission and • Health and educational suitability, and
costs vision, and safety appropriateness of facility
• Custodial and Mission • Indoor for its function
janitorial costs Dependency environmental • Appearance
• Current Index quality
replacement value • Adequacy of • Accessibility for
• Deferred space disabled
maintenance, and • Resource
deferred consumption
maintenance • Security
backlog • Site and
• Capital renewal location
• Maintenance
efficiency
indicators
• Facility conditions
index
• Churn rate and
churn costs

6
2.2 Improving the energy performance of buildings
In this thesis, the energy performance of buildings is approached from three different
perspectives: 1) energy performance of building services equipment, 2) energy performance
of equipment control and 3) quality of the indoor environment.

2.2.1 Energy performance of building services equipment

The proper functioning of building services equipment, particularly the HVAC systems, is
an important energy performance criterion because it determines if a satisfactory indoor
environment is even possible to attain economically. The energy performance of equipment
can be ensured by monitoring key figures that depict whether the piece of equipment is
operating optimally, and by having fault detection functionalities in place that monitor the
processes within the equipment. These key figures can be target figures provided by
suppliers or derived performance metrics, while fault detection systems range from simple
threshold alarm functions to more complicated fault detection and diagnostics (FDD) tools.

Unfortunately “… most buildings do not work properly” (Wang, et al., 2013, p. 1382).
Sometimes even simple abnormalities of the building services equipment such as
simultaneous heating and cooling go long undetected and waste energy (Wang, et al., 2013).
To detect and reduce such abnormalities, reviewing and benchmarking key figures can be
an efficient approach. Comparing specific heating and cooling energy consumptions of
properties should reveal if one of them had unsynchronized heating and cooling control. To
extrapolate the idea of utilizing data to monitor performance, Ihasalo (2012) developed a
HVAC performance metrics system that relies heavily on data generated by the BAS. For
example in an air handling unit (AHU) the system measures the energy performance of time
schedule and heat recovery, and the operational performance that is constructed by the
system availability, pressure and temperature.

The metrics are based on performance targets that can include one or two target values. With
one target value the metric is the actual measurement divided by the target value and with
two target values it is the time when the measurements are between minimum and maximum
values divided by the total measurement time. The energy performance of the time schedule
of an AHU is based on comparing the schedule with an optimal one, while the performance
of heat recovery is based on comparing the measured heat recovery efficiency ratio to a
target value provided by the supplier. This measured efficiency ratio can be calculated with
Expression 1. (Ihasalo, 2012, p. 121).

T22 - T21
η= (1)
T11 - T21
where

η = heat recovery efficiency ratio

T22 = supply air outlet (°C)
T21 = supply air inlet (°C)
T11 = exhaust air inlet (°C).

The operational performance metric is determined by the product of availability, pressure

and temperature as shown in Expression 2 (Ihasalo, 2012, p. 124). Again an example of an
AHU is used. Availability of the system is calculated by dividing the actual running time

7
with the total duration of the time schedule. The pressure factor, that depicts whether the
system is supplying the right amount of air into the building, is calculated by dividing the
time period during which the pressure is within 5% of the pressure set point with the total
measurement time. Finally, the temperature factor is calculated by dividing the time period
during which the temperature is within 0.5 °C of the temperature set point with the total
measurement time. (Ihasalo, 2012).

HVAC system metric = Availability × Pressure × Temperature (2)

Another useful figure to derive metrics from could be the specific fan power (SFP) that
represents the electric efficiency of a ventilation system, an AHU or a fan. According to the
National Building Code of Finland, the SFP of forced supply and exhaust ventilation systems
must not exceed 2.0kW/(m3/s), which is the total electric power consumed by all of the fans,
frequency converters and power control devices of the building divided by the design supply
or extract air flow, whichever is the largest (Finnish Ministry of the Environment, 2012b).
The SFP figure for an AHU is calculated according to Expression 3, and for a fan it is the
electric power divided by the air flow (Muhli, 2012, p. 25). The target SFP figures are
provided by the supplier.

Psupply + Pexhaust
SFP = (3)
qmax
where

SFP = specific fan power kW/(m3/s)

Psupply = electric power of the supply fan (kW)
Pexhaust = electric power of the exhaust fan (kW)
qmax = the maximum of the supply and exhaust air flows (m3/s).

On the other hand, fault detection systems are an important part of ensuring energy-efficient
operation of equipment. Those systems reduce the energy, water consumption and
maintenance costs and improve the quality of the indoor environment and safety of users.
The key to all this is the early detection of faults, which makes it possible to plan
maintenance work beforehand and schedule it so that the inconvenience to users is
minimized. (Hyvärinen & Kärki, 1996). Traditionally, fault detection in building automation
systems is based on alarm limits on control variables. Such a system monitors for example
the supply water temperature of an air heater, and if an upper threshold value is exceeded an
alarm on the event is generated. When deciding upon the alarm limits there inevitably is a
trade off with probability of false alarms and fast detection of irregularities. FDD tools on
the other hand are more versatile. These tools collect data from a number of control variables
and analyze it in order to detect and diagnose the reasons for abnormalities. By using data
more extensively FDD tools can detect problems that simple threshold checks cannot.
(Ihasalo, 2012).

The main methods in automatic FDD are top-down and bottom-up approaches, model-based
methods, knowledge-based methods and process history –based methods. Top-down and
bottom-up describe the direction of reasoning. (Ihasalo, 2012). A top-down reasoning
process could start from elevated energy consumption of the building, lead to large amounts

8
of energy consumed by the heating system and end with the detection of a problem in the
heat recovery unit. Model-based methods rely on modelling mathematically the physical
processes involved and comparing predicted measurements to the actual ones, while
knowledge-based methods rely on expert if-then rules (Ihasalo, 2012). Obviously the
application of these two methods is laborious. Process history –based methods use large
amounts of historical data and categorize it to resemble different faults and normal operation
(Ihasalo, 2012). The challenges related to the adoption of FDD systems include small
incentives due to minor benefits with sometimes significant installation time and cost,
inadequate sensor infrastructure and robustness of the tool (Ihasalo, 2012).

2.2.2 Energy performance of equipment control

Automatic and energy-efficient equipment control is an important factor of energy
performance: Even though people are willing to conserve energy due to environmental and
financial motives (Kim, et al., 2008), changing long-term behavior towards energy-saving
can be difficult (Marchiori & Han, 2010). As a fundamental element in energy efficient
control of artificial lighting, heating, ventilation and air-conditioning, user activity and
behavior have long been used (Nguyen & Aiello, 2013). When a space is not occupied, the
cooling set point can be elevated, heating set point lowered and lighting level set to minimum
to conserve energy (Chen, et al., 2009). Another interesting approach into energy
conservation is prioritizing passive cooling, heating and illuminance whenever possible by
little energy consuming methods such as adjusting motorized shades and remotely opening
windows (Kolokotsa, et al., 2009). Both methodologies fall into the category of demand-
based control that is based on the principle that “for any given system, the energy use is
minimized if the supply exactly matches the demand” (Gruber, 2012, p. 2). For example in
artificial lighting control this could mean that in addition to occupancy, the controller takes
into account the amount of available daylight (constant luminance control), or that the
amount of fresh air supplied is controlled by occupancy and the CO2 concentration of indoor
air.

Yet, the objective of energy performance inevitably risks user comfort through degraded
IEQ when occupancy detection is inaccurate. The often used motion-detection sensors are
poor at detecting occupancy (Lu, et al., 2010): It is not unusual in areas where motion-
controlled lighting is adopted that the workers find themselves waving their hands in the air
in order to get the lighting switched back on. Therefore, researchers have developed more
accurate methods in detecting user activity based on sensor data. These methods are
presented in Chapter 3.5.3.

Simulation-based energy savings potential of user activity recognition based solutions can
however be misleading. In a review carried out by Nguyen & Aiello (2013) the energy saving
potential of solutions for HVAC systems, artificial lighting and plug appliances was charted.
It was observable that the potential savings from actual experiments ended up being much
lower than in simulations. Thus, real-world testbeds should be used to confirm saving
potential of initiatives, as simulations are often based on a significant amount of idealization.

The challenges in demand-controlling HVAC systems mostly relate to the delay of

conditioning and maintaining a decent indoor air quality (IAQ). The estimated energy
savings through occupancy-based control of HVAC systems range from 10% to 40%
(Nguyen & Aiello, 2013). What makes saving energy without loss of comfort problematic
with the heating and cooling functions is that the conditioning of a room must commence

9
before it is occupied (Erickson, et al., 2009). In research, this aspect is managed with various
ways for different building types, sensor networks and activity recognition methods.
Essentially, all the methods attempt to overcome the indeterminism of human behavior. On
the other hand, the ventilation of buildings is important to manage well in order to reduce
the energy load caused by it while maintaining a decent level of IAQ. By reducing the
ventilation rate, however, there is a risk of introducing a so-called sick building syndrome to
the users if the IAQ was to degrade. Demand-controlling ventilation systems are an efficient
way to manage the pollutant concentration of indoor air energy-efficiently. (Dounis &
Caraiscos, 2009). In such systems the ventilation rate is adjusted in real-time to keep the CO2
concentration acceptable.

The concepts in efficient lighting control are occupancy-detection, controlled spaces and
daylight harvesting. The potential savings achievable with the related solutions range then
from 33% to 58% (Singhvi, et al., 2005; Delaney, et al., 2009). The concept of controlled
spaces means that a large area can be partitioned into smaller, independent sections that can
be managed separately, whereas daylight harvesting is simply optimizing the artificial
lighting level with respect to the amount of natural light (Delaney, et al., 2009, p. 62). With
innovative sensing algorithms, the energy consumption of the sensor networks can also be
managed efficiently: Singhvi et al. (2005) developed an active sensing algorithm that enables
the prediction of the daylight intensity throughout the day with just one observation per
sensor. Overall, visual comfort is easier to satisfy than thermal comfort due to the immediate
effect of the appliance on the surroundings. However, a challenge with lighting appliances,
especially fluorescent lighting systems is considering the adverse effects of continuous state
transitions. It is not a good idea to switch the lights on or off every time users enter and leave
a room, as this would result in a shortened lifetime of the lighting equipment. The breakeven
with the user absence period is estimated to be around 5-10 minutes. (Harris & Cahill, 2005).
Another challenging area is the detection of glare in daylight harvesting systems (Kolokotsa,
2007).

Research with quantified energy savings potential regarding plug loads is not as extensive
as with the other study areas. A prototype system developed by Marchiori & Han (2010) that
controls irregularly used appliances based on occupancy, suggested energy savings between
7.1-14.6%. There are also second order effects introduced by running and idle appliances as
they generate heat which in turn stresses the ventilation system (Harris & Cahill, 2005). Yet,
what differentiates the management of plug loads from HVAC and lighting systems is the
fact that appliances such as TV, microwave or printer may not be required to be on stand-by
even if a room is occupied. Also a PC might experience a lot of idle periods during its
operation. Therefore effective power management requires information on the user behavior
in addition to his or her location (Harris & Cahill, 2005). When the power management
system is unable to correctly predict the user’s behavior it is possible that energy savings
fade due to costs related to the state transitions of devices. These costs include the extra
energy consumed during start-up and the reduced lifetime of the device (Harris & Cahill,
2005).

2.2.3 Quality of the indoor environment

Energy performance cannot be observed separately from IEQ. A number of standards are
available that present guidelines for evaluating IEQ with physical measurements. Generally,
indoor environments can be assessed by three parameters: thermal comfort, indoor air quality
and indoor visual comfort. However, the American Society of Heating, Refrigerating, and

10
Air-Conditioning Engineers (ASHRAE) presents guidelines to measure IEQ even more
comprehensively. ASHRAE guidelines cover measuring energy, water, IEQ, thermal
comfort, indoor air quality, lighting and acoustics performance of commercial buildings.
These guidelines are then divided into three protocol levels: basic, intermediate and
advanced according to the accuracy of the metrics. (Hunn, et al., 2012).

On the other hand, in Finland three categories for indoor environment are used: S1, S2 and
S3. S3 is satisfactory, S2 good and S1 a very good indoor environment. These categories are
similar by description to the ones presented in SFS-EN 15251. In S1 indoor environment the
operative temperature should be kept within 0.5 °C of the target value for 95% of the time
the building is occupied. Also maximum and minimum indoor temperature limitations are
imposed for different seasons, and the temperature must be adjustable by 1.5 °C by building
users. The general principle is similar in S2 class, however a deviation of 1.5 °C is allowed
for 90% of the time of occupancy. (Ihasalo, 2012). The classification also includes criteria
for indoor air quality, acoustics and lighting that are specified for example in RT 07-10946
(Rakennustieto, 2008).

Thermal comfort can also be assessed differently from Finnish guidelines. It is influenced
by temperature, humidity, indoor air velocity radiant temperature, metabolic rate and
insulation of the clothing. (Kolokotsa, 2007). With quantified information on these
parameters, a Predicted Mean Vote (PMV) index is calculated. This index is also introduced
in EN ISO 7730:2005 (ISO, 2005; Oancea & Caluianu, 2013). This thermal environment
can also be categorized into one of four categories of SFS-EN 15251 according to the
variance of PMV (SFS, 2007). Indoor air quality is controlled by reducing pollutants or by
the ventilation system. The principal pollutants in indoor air are CO2 and volatile organic
compounds. Indoor visual comfort is determined by illumination, brightness, contrast, glare
and psycho-physiological aspects such as quantity, distribution and quality of light.
(Kolokotsa, 2007).

11
12
3 Development of building automation systems to
improve energy performance of buildings
3.1 Overview on building automation systems
Building automation systems (BAS) have a large influence on the energy performance of
buildings by automating energy-saving functions such as switching lights off or adjusting
ventilation rates. A BAS is a computerized control system that monitors, controls and
manages building services such as HVAC systems, lighting systems and electrical systems
(Shengwei, 2009). The main functions of BAS include switching equipment automatically
on and off for example based on scheduling, observing and optimizing the operation of
building services, collecting data on environmental conditions, providing information on
energy consumption, managing electrical loads and enabling remote control of systems.
These functions result in improved energy efficiency, substantial amounts of data on
building performance and the possibility to centralize facility management operations.
(Månsson & McIntyre, 1997). Thus, BAS is a central platform for improving energy
performance with data.

There are many names for this kind of a system that essentially controls equipment in
buildings: BAS, building energy management system (BEMS), energy management control
system, building management system and facility management system (Ihasalo, 2012). The
different expressions highlight different perspectives on the systems: For example BEMS
communicate and use information at a larger scale than other systems to apply more
developed algorithms in order to optimize the energy use (Månsson & McIntyre, 1997). One
can also emphasize the management of user comfort along with energy performance by
labelling the system as a building energy and comfort management system.

3.2 Traditional system architecture

The classic architecture of BAS does not strongly support decentralized execution of smart
functions that are based on exploiting data collected elsewhere in the system. According to
Soucek & Loy (2007) building automation systems traditionally have three levels in their
hierarchy: field level, automation level and management level. The interaction with physical
processes occurs at field level: Sensor data is collected by direct digital control (DDC)
stations, which also send data to actuators. Actuators can be connected to DDC stations
directly or through a bus communication system (fieldbus), which connects sensors to
actuators. Examples of common fieldbus technologies are LonWorks, Konnex and BACnet.
At automation level, data from field level is used to form logical connections and control
loops. Server stations and building controllers implement automatic control sequences to
operate the actuators appropriately. The server stations can also prepare field-level data for
storing. The automation level entities are connected to the field level via the automation
network, which can also use the LonWorks, KNX and BACnet technologies. At management
level, servers monitor the overall functioning of the BAS. This task includes storing
historical data for trending services, producing reports, observing system operation,
alarming on malfunctions and enabling parameter-based operation such as scheduling. The
management network uses Internet Protocol or other high-bitrate media. (Soucek & Loy,
2007). The hierarchical model of BAS is presented in Figure 3.1.

13
 Figure 3.1 The hierarchical model of BAS. Based on Soucek & Loy (2007, p. 82).

3.3 Development of system architecture

3.3.1 Development in the context of Internet of Things paradigm
BAS are the solution pursuing the value drivers of Internet of Things (IoT) in buildings
context. Also the development of BAS seems to aim towards similar goals to the IoT
paradigm. IoT is about the ability to connect at any time and from anywhere to anything
(Atzori, et al., 2010). Conceptually IoT consists of three aspects on smart objects: Anything
is identified, anything communicates and anything interacts (Miorandi, et al., 2012). The
value drivers of IoT are automatic proximity and sensor triggering, automatic product
security and extensive user feedback. These factors support the development of applications
belonging to the categories of information and analysis or automation and control.
Information and analysis applications apply to tracking, situational awareness and sensor-
driven decision analytics whereas automation and control concerns process optimization and
optimized resource consumption. (Coetzee & Eksteen, 2011).

Both of the application categories of information and analysis and automation and control
are highly relevant in improving the energy performance of built environment and are mainly
covered by building automation systems. However, the conceptual aspects on smart objects
are not met in a traditional model. Even though all the control points are identified, not
everything can communicate or interact. Therefore, it may be more appropriate to involve
the concept of a Cyber-Physical System (CPS) rather than IoT with a traditional BAS. CPS
integrates computation, communication and control with emphasis on closed-loop
information exchange and feedback while IoT highlights networking and interconnectivity
of all things (Ma, 2011). However, the next described development of the BAS architecture
towards a service-oriented network model seems to aim towards such.

14
3.3.2 Development towards a service-oriented model of building
automation
In order to allow for more complicated building services in building automation, more data
needs to be made available to subsystems of building controllers. Also the aspiration towards
developing reliable systems calls for more decentralized services, which are an effective way
to reduce single points of failure. Therefore the former layers and boundaries of the
hierarchical model are not appropriate in a service-oriented model of BAS. (Soucek & Loy,
2007).

Soucek & Loy (2007) present four aspects in this integration: flat network model, service
distribution, protocol convergence and new protocols to integrate BAS with IT systems. In
a flat network model the same network domain may include automation, management and
the fieldbus level. This means that special fieldbus media are tunnelled through control
network/IP routers to IP-based transport, thus bringing the I/O points directly to management
level. By gaining access to data point services of field devices from management-level
systems, the execution of those services can be done closer to the field. Thus the services are
executed in smaller autonomous units rather than in a building controller. Those data
services include browse, data point, trending, alarming, scheduling and security. Conversely,
the area of protocol convergence is focused on enabling communication between systems in
different control network technologies. Traditionally such has been achieved with gateways,
but the matter can also be managed by including other network protocol stacks in devices.
The BAS integration with IT systems on the other hand is driven by demand for Web-based
interfaces. Setting up a Web service interface requires servers, and to avoid a single point of
failure it is suggested that there could be small Web servers distributed in the field level in
room controllers or control network/IP routers. (Soucek & Loy, 2007).

3.4 Energy performance with available data

3.4.1 Traditionally collected data
Building automation systems need large amounts of data to monitor and operate processes
of HVAC systems. Sensors measuring temperature, pressure and humidity are necessary as
parts of control loops that produce actuator output for each HVAC process state. These
actuators can be for example valves in heat exchangers and pipework, pumps with frequency
converters and fans. To monitor a process, it is essential that the BAS also makes the control
signal of each actuator available for analysis together with physical measurements.

3.4.2 Deployment of Internet of Things related technology

To increase the energy performance of buildings beyond properly operating equipment,
more contextual information is needed. This information can be acquired with IoT related
technologies that enable identification and sensing. For example, lighting can be switched
off when sensors do not detect user activity and user-preferences can be retrieved with tag
identification procedures. The deployment of these technologies in existing buildings has
been greatly facilitated by the introduction of low-cost wireless sensing technologies.
Without the need for additional wiring for each device the installation costs decrease, and
wide deployment is more likely to be feasible (Agarwal, et al., 2010).

Central technologies of IoT include radio-frequency identification (RFID) and wireless

sensor networks (WSN). RFID technology is based on tags and readers. Every tag possesses

15
a unique identifier which is transmitted to the reader when a reader sends a query for tags in
the environment. The tag can be very small and it often resembles an adhesive sticker.
Passive tags do not have their own power source whereas active ones do, which obviously
affects the radio signal coverage. Passive tags get their power from the query signal through
induction. All in all, RFID technology makes it easy to track objects without seeing them.
(Atzori, et al., 2010).

On the other hand, sensors are more suitable when context-awareness is required. With
wireless technology it is easy to reconfigure and extend sensor networks, and position the
sensors in places where it previously has not been possible because of the aesthetic,
conservatory or safety issues related to the cabling (Reinisch, et al., 2007). Challenges
associated with wireless technology include supplying power, managing with the
communication range, dealing with problems in data exchange and ensuring the security of
information (Ihasalo, 2012). When choosing the type sensors of a wireless sensor network,
important factors to consider are practicalities such as cost, size and ease of use, data quality,
ethical issues, data type and level of intrusiveness (Keeling, et al., 2013).

A hybrid approach to these two technologies is a passive RFID sensor network, where the
sensing and computation devices are in fact RFID tags while the readers are data sinks and
the power supply of the network (Atzori, et al., 2010).

3.4.3 Open data

To accumulate data that could be used for control purposes, open data programs deliver a
cost-effective opportunity. Open data is available for the use of everyone and may freely be
used to improve the energy performance of buildings. The definition of open data and
content is that it “…can be freely used, modified, and shared by anyone for any purpose”
(Open Definition, 2016). Generally, the aim of the open data movement is to make data,
especially publicly acquired, obtainable via Web in electronic form (Gurstein, 2011).

3.5 Development in building automation control systems

3.5.1 Overview of building automation control
The task of a control system is to decide the appropriate output for an actuator to meet the
predefined objectives based on its input data. When a control task is simple also a more
straight-forward control method is appropriate. On the other hand, whenever there is a non-
linear process to be optimized and an abundance of sensor data to be analyzed, also a more
sophisticated control method may be required.

The controlling task can be divided to sub-tasks that all are performed differently. For
example the conventional control methods introduced next are mainly associated with
process control: The output is adjusted based on an input signal that can be the error from a
target value. Other control methods introduced below relate in part to process control, but
also to increasing contextual awareness based on sensor data. The task for context-based
control can then be to judge the demand for process control.

3.5.2 Conventional control

Here, proportional, integral and derivative (PID) and adaptive control are regarded as
conventional, as those are common in a BAS. In a proportional control system, the input is

16
the deviation from the target value: The error between indoor temperature and the
temperature set point, for example. The main idea of proportional control is to set a high
output when the error is large, and a low output when the error is small. Proportional control
cannot completely eliminate offset without instability, however. This offset is the steady-
state deviation from the target value. (Shengwei, 2009).

To tackle this issue it is possible to add integral action into the control system. The principle
of integral control is illustrated in Figure 3.2. With such action, the control system reacts to
an accumulating steady-state error by altering its output until the error has been eliminated,
given that the working condition and disturbances remain the same during this process which
is not often the case. The weakness of integral control is its slow responsiveness to sudden
and large errors. (Shengwei, 2009). These two first terms of PID control (proportional and
integral) are the most widely used combination in building automation control.

To enable fast responsiveness, derivative action can be introduced in the control loop.
Derivative control takes into account the rate of change of error in its response, while it has
no effect on steady errors. (Shengwei, 2009). This last term of PID control is less common
in building automation control, as it requires fast responsiveness from the controlled process
itself. However, it can be used for example in the control of service water.

Figure 3.2 Principle of integral control. Based on Shengwei (2009, p. 121).

An adaptive controller can be regarded as “a controller with adjustable parameters and a

mechanism for adjusting the parameters” (Shengwei, 2009, p. 132). Typical adaptive control
applications are auto-tuning, gain scheduling, self-tuning regulator, model-reference
adaptive system and stochastic adaptive control. Only auto-tuning is briefly introduced to
shed light on the principles of adaptive control. The main idea of relay auto-tuning is to
create a system that is able to adjust the parameters of a PID controller every time tuning is
demanded. When the tuning is demanded by operating a switch, relay feedback is enabled
and the PID controller is disconnected from the process. After the relay feedback stabilizes
the system, the PID parameters are computed and the controller resumes normal operation.
(Shengwei, 2009).

3.5.3 Occupancy detection methods

Occupancy detection methods range from motion-detection to probabilistic models. In
occupancy detection the most common method is to use a passive infrared (PIR) sensor that
is cheap, consumes little energy, produces data that is easy to interpret and does not enable
privacy violations. However, PIR sensors have their limitations: When a PIR sensor does

17
not detect movement, it is difficult to deduce based on PIR sensor data alone whether the
space is unoccupied or the user is remaining still. Therefore many innovative and
complementary sensing ways are developed.

As a very straight-forward approach Agarwal et al. (2010) installed a magnetic reed switch
door sensor. By doing so, it became much easier to isolate different occupancy transition
scenarios with the door opening and closing, thus improving accuracy of occupancy
detection. Still, there were some scenarios that the system could not infer correctly. Other
method introduced by Hagras et al. (2004) is to use pressure pad sensors on furniture to
detect people sitting, or installing software that informs the control system about user activity
on computers.

A camera in comparison to a PIR sensor produces more comprehensive information on the

state of a space. Erickson et al. (2009) installed a wireless camera sensor network that uses
an algorithm to classify every pixel of a frame into background, object or shadow. Then the
object pixels are merged to a blob that is tracked throughout the frame. The main goal of the
study was to quantify the number of occupants in rooms as an input for HVAC system
control. (Erickson, et al., 2009). With the use of cameras, however, ethical issues and a
feeling of intrusiveness must be considered. Clear principles for data collection, accessibility
and usage should then be defined and applied to avoid conflicts.

Radio-frequency identification (RFID) technology can also be used to accurately locate

people in buildings (Keeling, et al., 2013). The use of such technology requires the occupants
to keep tags with them in order to get the rooms they occupy conditioned, which means that
the system causes inconvenience for example when the tag is forgotten or when visitors do
not possess any. A strong advantage of RFID-based occupancy detection is that the users
and thus their preferences can be identified and stored in a database, enabling tailored service
(Chen, et al., 2009).

As a comprehensive occupancy detection method, Dong & Andrews (2009) present a

procedure for constructing an occupancy model from sensor event patterns. They
experimented with a conference room that was equipped with acoustics, illumination,
motion, CO2, temperature and relative humidity sensors. These sensors would obviously
produce a lot of data during their deployment. First step of the model development was to
use a data mining technique called Episode Discovery, which filters the most notable and
frequent event sequences. Some of these sequences can then be identified as meaningful
events, such as someone entering or leaving the room. Next, all the sequences are treated as
states in a semi-Markov model, so that they form a network of transition probabilities
between states, with each state having an expected duration. (Dong & Andrews, 2009). The
network is illustrated in Figure 3.3. In the figure X~(a) is the expected duration a of a state
and the number below a transition is the probability of the transition. The letter sequences
depict sensor event patterns: G denotes users entering the room, A, B and H notate different
sensor sequences that indicate a user staying in the room, D and I indicate users leaving the
room and F means that the HVAC is running.

18
 Figure 3.3 Semi-Markov model of discovered patterns (Dong & Andrews, 2009, p. 1448).

There is great strength in such an approach, as it tackles issues that purely motion-detection
based systems cannot. This detection method reduces the likelihood of the system being
fooled into deducing that the room is empty when it is not, or reacting to events that would
not require conditioning of the space. For example the system would not start conditioning
the space when someone entered the room just to make a phone call, whereas it would do so
when a meeting commenced. (Dong & Andrews, 2009).

3.5.4 Model-based predictive control

Under a range of idealized assumptions that their study was based on, Gyalistras et al. (2010)
found predictive control automation systems especially promising when it comes to saving
energy. Even though the saving potential is only a theoretical estimation, the authors are
confident that their assumptions are on the conservative side. The importance of prediction
seemed to increase the higher the solar heat gains were on the building in their several
thousand simulations. (Gyalistras, et al., 2010).

Model predictive control or model based predictive control (MBPC) is an algorithm that is
used to control dynamic systems by predicting changes in dependent variables. The goal of
the algorithm is minimizing a so-called cost function for a given time horizon. The cost
function depicts the interrelatedness of system inputs and overall objectives in terms of
acceptable boundaries. A strong advantage of the algorithm is using a receding horizon
strategy: The calculations to predict the future trajectory of the system are based on the
current state of the system which keeps shifting forward.

Freire et al. (2008) conducted simulations with MBPC strategy, where indoor temperature
and relative humidity were used to determine the appropriate output for a HVAC device.
The challenge was to optimize energy consumption without violating the thermal comfort
sensation of occupants with respect to a psychometric chart or predicted mean vote index.
In their simulations they were able to reduce energy consumption in varying weather without
violating thermal comfort boundaries once.

Similar benefits were observed by Kolokotsa et al. (2009). In addition to temperature and
relative humidity, they included observing CO2 concentration and illuminance in their
predictive control model. The study environment was a 125m2 laboratory, where they taught
the system the relationships between the current state of the space and environmental factors
such as outdoor temperature, window opening or air-conditioning operating. As a result, the
system was able to closely predict the development of a given set of environmental
parameters. Energy-optimization was performed by a minimization of a weighted function
that resembles the electricity consumption of control measures. Over 90% of users assessed
the overall comfort good or very good. (Kolokotsa, et al., 2009).

19
3.5.5 Artificial intelligence based control
Artificial intelligence (AI) techniques are useful when the objective is to integrate aspects of
learning, reasoning and optimizing into control. There are several artificial intelligence (AI)
techniques, but only artificial neural networks, fuzzy logics and genetic algorithms are
introduced here. Describing the AI aspect of information processing to this extent is
considered adequate, as the techniques focus on separate areas of intelligence and they are
extensively used in the related research. In human-intelligence terms it can be said that
artificial neural networks relate to learning, fuzzy logics relates to capacity of reasoning with
uncertainties and genetic algorithms relate to improving performance based on previous
experience (Oancea & Caluianu, 2013).

Artificial neural network (ANN) is a technique that tries to simulate the learning process of
human brain. ANN models the input-output relationships of a system based on history data
and exploits this ability to predict the outcomes of new combinations of inputs (Kalogirou,
2006). An application of an ANN could be to use it to predict the optimum moment to start
heating a building after the night setback in order to have the indoor temperature at a
comfortable level by 8 o’clock in the morning. Some of the input parameters in this case
could be the outside air temperature and wind speed, while the output of the ANN will be
the indoor temperature. (Oancea & Caluianu, 2013). Moreover, ANN is a so-called black
box model, meaning that it does not need any information about the system it is learning
(Kalogirou, 2006). In other words, ANN is able to learn causalities between given inputs and
outputs within a system without any other knowledge about the system. An illustration of an
ANN structure is presented in Figure 3.4.

Figure 3.4 An artificial neural network. Based on Oancea & Caluianu (2013, p. 99).

The neural network has three layers of neurons: an input layer, a hidden layer and an output
layer. A neuron is a basic element of the network that processes its input with summation
and activation functions to produce its output, which is the input of the neurons in the next
layer. Additionally, each input and output have a corresponding neuron. The amount of
hidden layers and neurons in them is problem-specific. Based on training data the network
then learns the relevant relationships between neurons and determines the corresponding
weights of every neuron’s output. The most common learning algorithm is called the Back-
Propagation algorithm, which reduces the error of the network by varying the weights along

20
its gradient until an acceptable tolerance is reached. (Kalogirou, 2006; Oancea & Caluianu,
2013).

In order to solve problems with ANN one must select suitable learning rate, number of
hidden neurons and activation functions. What makes ANN a strong technique is that it is
fault tolerant, robust and immune to input noise. (Kalogirou, 2006). Still, there are some
challenges related to the technique: The learning process of ANN needs reliable training data
and the self-learning algorithms are limited to their experience (Kolokotsa, et al., 2009).
Therefore the system cannot take intelligent action if it has not experienced the situation
before.

Fuzzy logics is a powerful approach when managing systems with humanistic aspects
(Kolokotsa, 2007). It is a way to get the reasoning process of computers closer to the way
human brain does it (Oancea & Caluianu, 2013). Fuzzy logics requires some computational
tricks to illustrate the ambiguity of thinking, but as a great advantage the inference logic is
very easy to understand and evaluate. Fuzzy logic consists of fuzzy rules and fuzzy sets.
Fuzzy sets are the input of the reasoning process, which is in the form if-then as fuzzy rules.
Every item of a fuzzy set belongs to that set to a degree, and the rules take those partially
true facts and determine to what degree they are true (Oancea & Caluianu, 2013). This degree
of membership is determined with membership functions. These functions are chosen based
on their perceived fit on the fuzzy set in question. An example of fuzzy membership
functions for categories of thermal environment is illustrated in Figure 3.5. In the figure the
horizontal axis (thermal sensation) corresponds to PMV index values. With such
membership functions, thermal sensation index of value +2 belongs to the fuzzy set of hot
with a degree of 0.4 and to the fuzzy set of warm with a degree of 1 (Oancea & Caluianu,
2013). These degrees of memberships within fuzzy sets are the foundation of the fuzzy
reasoning process.

Figure 3.5 Fuzzy membership functions for categories of sensation of thermal environment. Based on
Oancea & Caluianu (2013, p. 97).

To clarify the concept of fuzzy logic, consider the following example. A cooler is equipped
with a temperature sensor, and the comfort temperature is considered to be 21°C. A
measured temperature of 28°C could then belong to a predefined fuzzy set of warm with a
membership degree of 0.5, to the set of hot with 0.8 and to the set of comfortable with a
degree of 0. The deviation from the temperature set point and changes in cooler output would
then also be given linguistic representations. A fuzzy rule could state that if the temperature
is hot and the deviation is growing then increase the cooler output. Now in this case the
cooler would add more output the warmer it is and the more the deviation is growing. The
strength of such logic is that the controlling rules can contain ambiguity and cover many
different situations (Paiho, et al., 2002).

21
Genetic algorithms (GA), which are a form of evolutionary computing, are good at
optimizing solutions. The algorithm mimics the selection process of nature by producing an
initial population of chromosomes (solutions) and evaluating their fitness. The best
candidates are selected as parents and their offspring is generated by recombination of their
parents’ qualities, and the process goes on and on until the required fitness is achieved. The
fitness of the chromosomes is assessed with a fitness function and an objective function.
Also a number of constraints can be imposed on the selection process. (Oancea & Caluianu,
2013).

Let us consider an example of optimization with genetic algorithms. In energy consumption

and thermal comfort management the objective function can be defined with the predicted
mean vote index and a formula connecting energy consumption with the HVAC operation.
Whenever the room temperature set point is changed, the mean radiant temperature
(radiators), air velocity and relative humidity (humidifiers and ventilation system) need to
be optimized with respect to the comfort index and energy consumption. This is when the
GA starts the evolutionary computation and evaluation of candidate solutions. Ultimately
the algorithm produces the optimum combination of system outputs to achieve the best level
of thermal comfort with the least energy consumption. (Oancea & Caluianu, 2013).

3.6 Summary of the theoretical background

Based on the theoretical background, the relationships of energy performance and available
data are illustrated in Figure 3.6 By reinforcing aspects of open data, BAS and IoT
technology, the ability of building service equipment to achieve a higher level of equipment
performance, control performance and indoor environmental quality is enhanced. This
improvement is not necessarily completely attributable to BAS, but also to a property
manager working with the BAS. Also instead of relying exclusively on sensor data to
evaluate IEQ, feedback could be collected from building users with IoT technology.

Figure 3.6 Relationships of energy performance and available data.

22
4 Research method
The objective of the empirical study is to provide practical evidence and understanding on
solutions that improve the energy performance of buildings by exploiting available data.
This is done by answering to the three research sub questions in a real-life context:
• What are the energy performance gaps of buildings?
• What kind of solutions that are based on the exploitation of available data can close
those gaps?
• How effective are those solutions?

4.1 Overview of the research method

The research method of this thesis is a case study on a property maintenance organization
and an office building in Helsinki, Finland. Case study was chosen as the research method
because it allows the research to take place within a real environment, which is convenient
for detecting existing energy performance gaps and methods to close them, while the theory-
building process itself verifies the results (Eisenhardt, 1989, p. 547). Still, it is important in
case studies to collect evidence from multiple sources to capture the richness of the research
environment and phenomena (Yin, 2009, p. 2). The overall approach was constructive:
Focus was not on what has been done to close energy performance gaps, but on what could
be done to close the current ones.

4.2 Description of the case study

4.2.1 Case organization
Helsinki University Center for Properties and Facilities is responsible for tasks related to
owning, developing and managing properties that are owned by Helsinki University. The
center also executes the university’s capital expenditure plan and construction projects. The
building portfolio ranges from warehouses and offices to chemical laboratories and medical
facilities. The turnover is 140 million euros and the amount of staff is 430. (University of
Helsinki, 2013; Helsinki University Intranet, 2014). The total of 300 buildings are distributed
over 33 regions with a clear majority in Helsinki. (Helsinki University Intranet, 2014). Out
of the 300 buildings, 30 are remotely managed from a control room in Helsinki.

Being responsible for property maintenance, the case organization is well-equipped to make
informed decisions regarding building services engineering in the construction phase. In
their decisions they emphasize usability, modifiability, adjustability and serviceability of
buildings throughout their lifecycle. Additionally, with such a large amount of properties,
there is an incentive to consider the energy efficiency of buildings. Therefore the case
organization is interested in automating energy conservative functions in building services
equipment and increasing energy awareness among building users.

4.2.2 Case building automation system

The case organization uses six different building automation system controller software.
These are Visonik and Desigo from Siemens, Trend, Deos OPENweb ControlPanel, Fidelix
and Regin. The one studied the most was OPENweb ControlPanel by Deos. It is a web-based
BACnet BMS software capable of integrating systems that use standardized interfaces,
therefore allowing for data exchange via SQL, OPC, M-bus and Modbus. OPENweb also
supports control of KNX systems for lighting and shading control. An Event Control Center

23
modular expansion was used for handling alarms and messages. Another modular expansion
that Deos provides is a Fidelio hotel booking system which makes Check-In, Check-Out and
Pre-Check-In available as BACnet objects. The software providers also promise that the
open architecture allows for individual expansions if any special application is required.
(Deos, 2016). Also Desigo, Fidelix and Regin systems were studied a little for benchmarking
purposes, however the less intuitive interfaces or half-completed sensor infrastructure
resulted in the decision to focus on Deos.

4.2.3 Case building

The building studied in this thesis is an office building in Helsinki central business district
(CBD). The original purpose of this thesis was to study both offices and laboratories that
were controlled in Deos software. Laboratories are very intensive in their energy
consumption in comparison to offices, which is mainly attributable to research needs: Large
volumes of air flow, strict temperature tolerances and energy intensive equipment such as
particle accelerators. Much of the potential savings can be attributed to making research
equipment more energy efficient, which did not quite fit well with the scope of the thesis.

Therefore the most attention is given to a recently renovated office building that is referred
to OfficeOne in this thesis. It has the most comprehensive sensor infrastructure of all the
buildings managed by Helsinki University Properties and Facilities. OfficeOne has eight
storeys and its gross internal area is 6100m2. It is connected to the district heating and cooling
network, has 16 air handling units and produces well over 600 measurement streams from
its sensor infrastructure. The later suggested improvements that include sensor data
utilization are mostly generic, so that they are applicable to both laboratories and office space
unless some research needs are violated.

4.3 Research process, data and analysis

4.3.1 Research process and data collection
The research is conducted in the following phases:
1. Identifying energy performance gaps. This phase is based on analyzing historical
sensor data available in the building automation system and semi-structured
interviews of key personnel in the case organization. The sensor data used in the
analysis was recorded in Deos OPENweb ControlPanel between 1st February 2016
and 26th September 2016. This time window was chosen because the final sensor
infrastructure of the later introduced case building OfficeOne was completed in the
end of January 2016. There were four key personnel that were interviewed: The
people responsible for property maintenance, building automation, building services
engineering and the case building. The themes explored in the interviews were
energy consumption of properties, sources of energy waste, adjustability of HVAC
systems, energy efficiency by building automation, energy management and the
stakeholders in energy saving activities. The interviews were recorded and
transcribed.
2. Seeking solutions to close the gaps. Existing ideas and solutions that eliminate or
reduce the identified inefficiency were sought from research literature and
commercial providers. If no existing solutions were found, such was suggested.
3. Evaluating solution initiatives. These ideas and solutions were introduced to
experts of property management, building automation, building services engineering
and property maintenance in separate workshops of various group sizes from one

24
 expert to four experts at a time. These experts belonged to the case organization or
its interest groups, and are listed in Table 4.1. They were asked to share thoughts and
evaluate the proposed solutions in the dimensions of challenge to implement and
expected benefits in the scale of 1-10. Such a method was chosen to roughly group
the suggestions into groups of low hanging fruits, continuous improvement,
strategical development projects and lemons according to an organizational change
initiative evaluation framework presented by Peltokorpi et al. (2008). The grouping
was used to support later decision making, and the complete evaluation results are
presented in Appendix 1. The fourfold evaluation table is presented in Figure 4.1:
The plotted dots resemble the averages of evaluations per respondent group, while
the ellipse axis depicts the standard deviation of evaluation in both dimensions. If the
respondent was unable to evaluate challenge or benefits, it was plotted as 5 but not
taken into account in the average or standard deviation calculation. Descriptions of
the four groups are as follows (Peltokorpi, et al., 2008):
• “Low hanging fruits are changes that lead to assured results with only minor
effort.”
• “Continuous improvement means an ongoing series of minor intervention
which creates steady, but marginal, growth.”
• “Contrary to continuous improvement, strategic development projects require
the managers’ attention. The project’s results are breakthrough improvements
…”
• “Lemons are projects when assessed in detail prove to be impossible or too
labor intensive to implement for the expected results, and thus must be
dismissed.”
4. Studying the feasibility of selected improvements. The project steering group
chose three most effective solutions for further examination to provide estimates on
their monetary costs and benefits. This selection was based partly on the evaluation
results and partly on their own judgment.

Table 4.1 Experts evaluating proposed improvements.

Expertise area Background Total
amount
Building automation Contractor (2) and designer (2) 4
Building services Designer (3) 3
engineering
Property management Manager (3) 3
Property maintenance Plumber (3) 3

25
 Figure 4.1 The fourfold table for evaluation results.

4.3.2 Data analysis tools

The data analysis tools and software used in the process are presented in Table 4.2.

Table 4.2 Tools and software used in data analysis

More
Purpose Software/Tool name information
from:
Provider: Deos
Exporting sensor data. (Phases 1 and 4 of
OPENweb ControlPanel (Deos, 2016)
research.)
Provider: DataRangers
Creating self-organizing maps and a user
(DataRangers,
survey. (Phases 1 and 2 of research.) Louhin
2016a)
Building analyses and managing large
(DataRangers,
amounts of sensor data. (Phases 1 and 2 of DataMiner
2016b)
research.)
Provider: Microsoft
Data clustering. (Phase 2 of research.) SQL Server Data Tools for
(MSDN, 2016a)
SQL Server 2014
Excel add-in for data clustering. (Phase 2 of SQL Server Data Mining
(MSDN, 2016b)
research.) Add-Ins for Office
Visualizing data and fitting trend lines
Excel (MSDN, 2016c)
(Phases 1 and 4 of research.)
Creating reports that are based on
dimensional data modelling. (Phase 2 of Power BI Desktop (Microsoft, 2016)
research.)

26
4.3.3 Limitations in the research method
The research was limited by the existing sensor infrastructure of the case building,
unfinished tuning of building services equipment and the subjective evaluation method of
initiatives aimed to close energy performance gaps. Due to unavailability of relevant sensor
data, some naïve assumptions were made regarding causalities and processes to estimate the
significance of energy consumption sources. These assumptions are always stated in the text.

Additionally, the HVAC equipment were partly in tuning phase due to recent renovation in
the case building OfficeOne. This means that some of the systems operated temporarily in a
way that may cause higher energy consumption by purpose and thus mislead the energy
efficiency analysis, however these sources of energy inefficiency were acknowledged in the
text whenever identified. OfficeOne was still chosen to be the case building due to its most
extensive sensor measurements and a user-friendly interface.

Grouping the proposed improvements according to their attractiveness was thought to be

convenient to base on subjective expert knowledge, because test results in other parts of the
world, simulation results or product brochures were not considered adequate sources of
information, nor the wide scope supported real-world testing. Although, the evaluation
meetings of proposed improvements were heterogeneous. The number of participants varied
and a meeting is never identical to another. Hence there is a social aspect to the evaluation
process and unavoidable misunderstandings may lead to distorted and incomparable
evaluation grades. Anyway, same information content on the idea of an improvement,
expected benefits and identified challenges was provided and a short discussion followed to
specify unclear aspects of the subject if necessary.

27
28
5 Results
5.1 Identified energy performance gaps
5.1.1 Current approach to managing energy performance
The case organization deals with buildings from construction to maintenance, which is why
they emphasize usability, modifiability, adjustability and serviceability of buildings
throughout their lifecycle. This manifests as distributed systems, loose sizing of equipment
and light integration of hardware into the building to facilitate replacements and upgrades.
The distributed system architecture can be accomplished with appropriate air handling unit
zoning with variable air volume systems and allocation of cooling units, whereas loose sizing
of equipment can mean for example reserve capacity in heating and cooling power of
equipment or in ventilation system components and ducts. It was suggested that in the future
the philosophy of distributed systems would lead to integrated panel heating and cooling
solutions to better meet area specific indoor conditions and allow for comprehensive
recycling of return water heat content.

The energy saving goals of the case organization are in part externally motivated. There are
long-term energy efficiency and sustainability objectives imposed by the management and
obligations regarding energy auditing based on the law on energy efficiency 1429/2014 of
Finland (Finlex, 2014). In the past, energy has mostly been saved by basic and effective
things rather than complicated control systems: Making sure that the time-scheduling of
equipment operation times is appropriate, heat recovery of exhaust air is functioning and
occupancy-based lighting control is working. If automation-based solutions are introduced,
they cannot be prioritized over the controllability of systems. The lack of excess funds
however is an impediment to upgrading the energy performance of the case properties, which
is why most investments are timed to coincide with building modernizations.

It should also be made sure that the energy costs remain within budget in the short-term.
Ensuring this requires continuous consumption monitoring. If an increase in energy
consumption is detected by an energy manager company with whom the case organization
has a contract, the underlying causes are determined and action is taken with the
organization’s own resources if necessary. These resources extend all the way to own
maintenance-men that replace actuators in the field. There are however too often problems
associated with poor tuning of control loops of building systems, and the reason for this is
suggested to be the instalment schedules in building automation contracts that leave no
financial incentive for the contractor to finish the work properly. The contractors are mainly
involved with the case organization through maintenance contracts that concern server and
sub-distribution board aspects of building automation systems. Anyhow, there is reluctance
to move towards solutions provided by energy service companies, because all the necessary
special expertise is already in-house. Energy service professionals are however considered
appropriate for occasional consulting purposes.

29
5.1.2 Existing energy performance gaps
The interviewees identified multiple needs for improvement with regard to building
automation system functionalities, energy consumption metering and systems integration to
be met in the future. None of the improvements should include extensive electrical work or
large amounts of equipment replacements, however. When it comes to building automation,
there are goals of reducing energy consumption when the building is not in use and
eliminating overheating of spaces. Currently no system exists that contains the opening hours
and holiday periods of the property portfolio, on which the operating times programming of
equipment could be based on. With a large number of properties it is impossible to keep up
with the programming task, as every piece of equipment needs to be programmed separately.
It was suggested that a system that contains this information and automatically programs the
time schedules for equipment should be developed, an interface between building
automation and room reservation systems should be created or at least that the equipment
could be controlled in purposeful groups. Group control in general was thought to facilitate
tasks as supply water temperature programming adjustments according to seasons, supply
air temperature control and air handling unit operating times. Falling behind in doing these
tasks manually was considered to lead to energy waste.

A few issues were brought up concerning the overheating of spaces: Forgotten radiator
heating compensation curve adjustments, poor adjustability of radiator heating, lack of
protection from heat loads and the fact that weather forecasts are not utilized in any way.
With the older building automation systems that represent a vast majority, supply water
temperature of radiators is defined with a two point compensation curve. These curves tend
to get shifted upwards during cold winters to supply warmer water into the system, but
unfortunately it often is forgotten to make the shift back after the cold period is over. Large
benefits were thought to be expected should those system adjustments be automated reliably.
On the other hand, most of the case buildings have a single radiator network which impedes
area specific heating optimization. If the cellar and façades had their own networks, heating
could be optimized better with respect to heat loads caused by the sun for example.
Moreover, Finland was thought to be behind in protecting buildings from solar radiation
according to an interviewee. Automatically operating motorized sun blinds could reduce heat
loads when they are not desired and utilize them when needed. Additionally, the fact that the
heating system does not use weather forecast data can lead suboptimal performance when
facing a very cold night or a sharp temperature rise.

The case organization is aiming towards a role in energy leadership in the future to increase
energy awareness among building users. Energy consumption measurement infrastructure
and reporting are central aspects of this goal. Currently energy consumption is reported
inconsistently across buildings and much thought is given to the matters of what to measure
and how should the information be managed. Identified needs however are user friendliness,
information content that serves maintenance, automated and detailed energy monitoring and
reporting capabilities that enable drilling through to the underlying equipment operation, and
monetary feedback to motivate short response times of the staff. The needs of user
friendliness and maintenance serving information content highlight the fact that the system
would be used by staff with various backgrounds and tasks. Only the relevant information
should then be produced for each user.

There were two more energy performance gaps, but those were based on observation rather
than interviews: User complaints that were received in personal e-mails inboxes and single

30
value threshold alarms. User complaints are a problematic area as the indoor environment is
not likely to ever satisfy everybody. In the case building users in the same room feel cold
and hot. With more knowledge on the relative numbers of satisfied and unsatisfied occupants
the interventions through the BAS could be better informed. Also the feedback channel could
be more efficient: Gathering feedback could be systematic and the satisfaction rate could be
made visible to the whole maintenance organization. Conversely, with single value threshold
alarms there is inevitably a trade-off with the amount of false positives and early detection
of faults. If the detection accuracy was better, the amount of false positives could be reduced.

5.1.3 Summary of identified energy performance gaps

The identified energy performance gaps are listed in Table 5.1. Next, measurement data
available through the BAS is used to verify identified energy performance gaps and detect
new ones.

Table 5.1 Summary of identified energy performance gaps.

Performance
Description
area
• Poor area-specific adjustability of heating
• Reactive essence of heating control
Heating
• Need for shifting the compensation curves to adapt the heating power
to seasonal needs
Cooling • Lack of protection from heat loads
• Manual input intensive processes
• Inability to control actuators in groups
• Inefficient channels for feedback from building users
Processes • Lack of knowledge on the relative numbers of satisfied and unsatisfied
building users
• Inaccurate single-value threshold alarms in BAS
• Inconsistent energy consumption reporting

5.2 Overview and limitations of available data

There is a large amount of measurement data generated and recorded by the building
automation system of OfficeOne, but in some cases the constraints in sensor infrastructure
make it difficult to evaluate the energy performance of equipment control. The data is
collected in Deos OPENweb ControlPanel building automation system controller software
that monitors the state of spaces and equipment extensively. The most common
measurement point types are listed in Table 5.2. These measurement types include over 600
measurement streams which are gathered and recorded in 25 sub-distribution boards. The
measurements can be downloaded as a comma separated value file in which the first column
states the date, the second states the time and the remaining columns state the measurement
values for a given sensor.

31
Table 5.2 Available sensor data in OfficeOne.
Parameter/actuator name Parameter units or alternative Recorded Actuator
states (X) (X)
ROOM
Temperature °C X
Temperature set point °C
CO2 concentration ppm X
CO2 concentration set point ppm
Motion detection state Detected/Not detected
Chilled beam valve opening 0% - 100% X
Supply air damper state Minimum/Normal/Augmented X X
air flow
Exhaust air damper state Minimum/Normal/Augmented X X
air flow
FLOOR
Plug load of the northern or southern half kW X
of the floor
Lighting load of the northern or southern kW X
half of the floor
AIR HANDLING UNITS
Energy consumption per a switchgear kWh X
Inlet air temperature °C X
Pressure difference over the air inlet filter Pa X
Air temperature after heat recovery °C X
Return water temperature in the heating °C X
radiator circuit
Pump state in the heating radiator circuit Running/Not running X
Valve opening in the heating radiator 0% - 100% X X
circuit
Valve opening in the cooling coil circuit 0% - 100% X X
Electronically commutated supply air fan 0% - 100% X X
Inlet pressure Pa X
Inlet pressure set point Pa X
Supply air temperature °C X
Supply air humidity 0% - 100% X
Exhaust air temperature °C X
Terminal pressure Pa X
Terminal pressure set point Pa X
Pressure difference over exhaust air filter Pa X
Pressure difference over heat recovery Pa X
Heat recovery disc 0% - 100% X X
Outlet air temperature °C X
Electronically commutated exhaust air fan 0% - 100% X X
Bypass damper 0% - 100% X X
HEATING SYSTEM
Energy consumption MWh X
District heating water usage m3 X
Inbound district heating water °C X
temperature
Outbound district heating water °C X
temperature
Outdoor temperature °C X

32
Parameter/actuator name Parameter units or alternative Recorded Actuator
states (X) (X)
Service water heating
Service water usage m3 X
Heat exchanger valve of the warm service 0% - 100% X X
water system
Service water temperature °C X
Circulated warm service water pump state Running/Not running X
Radiator system
Heat exchanger valve of the radiator 0% - 100% X X
system
Radiator system supply water °C X
temperature
Radiator system supply water °C X
temperature set point
Radiator system pressure Pa X
Radiator system pump Running/Not running X
Radiator system return water °C X
temperature
Floor heating system
Heat exchanger valve of the floor heating 0% - 100% X X
system
Floor heating system supply water °C X
temperature
Floor heating system supply water °C X
temperature set point
Floor heating system pressure Pa X
Floor heating system pump Running/Not running X
Floor heating system return water °C X
temperature
Air heater
Heat exchanger valve of the air heater 0% - 100% X X
network
Air heater supply water temperature °C X
Air heater supply water temperature set °C X
point
Air heater system pressure Pa X
Air heater system pump Running/Not running X
Air heater return water temperature °C X
COOLING SYSTEM
Energy consumption MWh X
District cooling water usage X
Inbound district cooling water °C X
temperature
Outbound district cooling water °C X
temperature
Cooling convector system
Heat exchanger valve of the convector 0% - 100% X X
system
Convector system pump Running/Not running X
Convector system supply water °C X
Convector system supply water set point °C X
Convector system pressure Pa X
Convector system return water °C X

33
 Parameter/actuator name Parameter units or alternative Recorded Actuator
states (X) (X)
Air cooler
Heat exchanger valve of the circuit 0% - 100% X X
formed by air cooler and chilled beam
system
Air cooler supply water temperature °C X
Air cooler supply water temperature set °C X
point
Air cooler system pressure Pa X
Air cooler circuit pump Running/Not running X
Air cooler return water temperature °C X
Chilled beams
Valve of the beam circuit 0% - 100% X X
Chilled beam system pump Running/Not running X
Chilled beam system supply water °C X
temperature
Chilled beam system supply water °C X
temperature set point
Chilled beam system pressure Pa X
Chilled beam system return water °C X
temperature

The controller software also collects data regarding the state of separate systems such as
street lighting, special outlet fans and fire safety systems, and manages events and alarms
based on threshold values and rules.

There are some issues with the produced data that pose challenges to its utilization:
• Bulk measurements. Energy consumption measurements often represent more than
one separate system: The electricity consumption of 16 air handling units is
measured in bulks in 4 switchgears which results in noise in the data depending on
the operating schedules. For example in the occupied zone the consumption may be
cyclical whereas in toilets it is not. Additionally, the heating and cooling energy
consumption metering covers multiple separate systems. This complicates system-
specific energy consumption analysis.
• Limitations in hardware. Not all systems can be controlled digitally. Mechanical
radiator thermostats are self-operated unit controllers that cannot be connected to
building automation. Nonetheless, the radiator valve position for each room is vital
information when assessing the effect of the heating system on room temperature.
• Missing target values. Room temperature and carbon dioxide content of air set
points are not recorded. This means that it is impossible to know what conditions
the chilled beams and supply air dampers were trying to achieve in historical
analysis.
• Locally controlled systems. Lighting systems are not connected to building
automation with the exception of corridor lighting and outdoor lights. The
functionality of the lighting system cannot therefore be assessed remotely. Also the
information produced through lighting control regarding daylight availability and
lighting state are not available.

34
5.3 Evaluation of current control systems and their energy
performance
5.3.1 Heating system control
Heating is controlled with the BAS. The source of heat in OfficeOne is the water in the
district heating network. The heat of this water is used in four systems: Radiator network,
air heater, service water heating and floor heating. The required temperature of supply water
flowing into the radiator network is determined by outdoor temperature with a compensation
curve that is presented in Figure 5.1, where the vertical axis corresponds to supply water
temperature and the horizontal axis corresponds to outdoor temperature. To reach the
appropriate supply water temperature with the heat content of district heating water, a heat
exchanger valve is controlled in the radiator network. The control target of the supply water
temperature of air heating and floor heating is also defined with an outdoor temperature
compensation curve, whereas warm service water temperature is being kept constant
regardless of the outdoor temperature.

The amount of supply water used for the actual heating is adjusted with separate valves. The
air handling units have their own valves that heat the intake air if heat recovery is not able
to warm it up sufficiently. Radiators also have their own valves that adjust the water flow
and thus the heat that is transferred. This valve is controlled by a mechanical thermostat that
can only be adjusted manually within a small range. The purpose of mechanical thermostats
is to reduce the heating power of radiators when free heat is available.

Figure 5.1 The outdoor compensation curve of the radiator heating system.

OfficeOne is heated constantly to keep the rooms in uniform temperature regardless of the
cyclical occupancy patterns, however around half of the air handling units (AHU) are
supplying a minimum air flow until midnight and then turned off for the rest of the night.
The reason why all AHUs are not on minimum air flow is that the building was recently
renovated and the night time ventilation is used to reduce smells. As a reminder, the heating
system energy consumption metering covers radiator heating, air heater, floor heating and
service water heating. During winter the energy consumption is quite static around the clock,
though there are some night time spikes as observable in Figure 5.2. Based on Figure 5.3 it
can be suggested that those spikes are caused by some air handling units running and
radiators heating more during lowering outdoor temperatures (the temperature dropped
towards the 4th and 6th of February). The midnight drops may have something to do with the

35
respective drops in AHU operation that can be seen later in Figure 5.11. Still, it is suspicious
how the consumption profile stays rather static with such dynamic overall operation.

Figure 5.2 Hourly energy consumption of the heating system during winter. The consumption profile
stays quite static with occasional night time spikes.

Figure 5.3 Heating system heat exchanger valve opening trends as two hour averages to reduce
oscillation of the data. The consumption spikes seem to be explained by increasing demand for
radiator and air heating.

During warm spring days heating energy is mainly consumed outside office hours, as can
be seen in Figure 5.4. This is explained by heating system valve opening trends in Figure
5.5 as radiators and the ventilation system do not require as much energy during the
daytime. Also the almost zero energy consumption during warm days suggests that service
water heating and floor heating are negligible. However, the frequent and sharp drops from
110kWh to 0kWh in hourly heating energy consumption occurring within the span of two
hours can be a performance concern. The thermal mass of a building can cause undesirable
situations: Cold structures may cause thermal discomfort in the building after heating is
lowered due to rising outdoor temperatures, and warm structures may induce inertia after
cooling is commenced. Therefore information on future heat loads may help in anticipating
situations when cooling systems are battling the thermal mass, or heating systems are
fighting cold structures with reduced output. The mechanical radiator thermostats do
handle the situation reactively so that radiator heating is stopped after a certain room

36
temperature is reached, but with predictive capabilities free heat could be utilized to save
purchased energy.

Figure 5.4 Hourly energy consumption of the heating system during spring. There are occasional
daytime drops.

Figure 5.5 Heating system heat exchanger valve opening trends as two hour averages to reduce
oscillation in data. The daytime drops seem to be caused by reduced radiator and air heating.

The effect of heat loads on overheating the building, indicated by the chilled beam network
valve opening in the cool distribution system, was explored by studying the correlation
between outdoor temperature and chilled beam network valve, plug loads and chilled beam
network valve, and produced solar energy and chilled beam network valve. The data sets
were recorded between 12th April and 12th August 2016 between 10:00 and 15:00 on
working days to assume a relatively stable occupancy and thus utilization of chilled beams.
Chilled beam valve signals were aggregated into hourly averages. Plug loads were
assumed to correlate with the amount of users, and the electricity produced in the solar
panels was assumed to correlate with the solar radiation heating the building that is quite
evenly exposed to sunlight from the East, South and the West.

Based on the coefficients of determination it seems that the most valuable information (out
of the three observed sets) for predictive control would be the future outdoor temperature,
which is quite trivial. This correlation is seen in Figure 5.6. Plug loads and solar radiation
did not correlate with cooling demand at all. The correlations are compiled in Table 5.3.

37
With conditional estimation methods the data on produced solar energy and plug loads
could prove useful too, though. This is illustrated by data sets presented in Figure 5.7 and
Figure 5.8. Within segments of the more defining variables the previously irrelevant
variables gain significance, which is suggested by the increased coefficients of
determination. This finding implies that techniques of predictive analytics may be able to
increase control performance.

Table 5.3 Correlations of heat loads and cooling demand.

Coefficient of
Description Conditions Trend line
determination
Outdoor
temperature vs. None y = 0.7912 x + 2.6109 R2 = 0.2317
Cooling demand
Plug loads vs.
None y = 0.4191 x + 10.405 R2 = 0.0566
Cooling demand
Solar radiation vs.
None y = 0.3148 x + 15.09 R2 = 0.0698
Cooling demand
Plug loads vs. Outdoor temperature above
y = 1.6263 x + 5.5926 R2 = 0.3536
Cooling demand 23°C
Outdoor temperature above
Solar radiation vs.
20°C and plug loads between y = 0.7285 x + 13.035 R2 = 0.3333
Cooling demand
16kWh and 19kWh

Figure 5.6 Outdoor temperature vs. cooling demand. Outdoor temperature was the most defining
variable.

38
 Figure 5.7 Plug loads vs. cooling demand when data is filtered with outdoor temperature. Plug loads
gained significance with a limited outdoor temperature interval.

Figure 5.8 Solar energy vs. cooling demand when the data has been filtered with outdoor temperature
and plug loads. Solar radiation then gained more significance.

Anyhow, these observations suggest that the largest energy savings could result from
demand-controlling ventilation and radiator heating, of which ventilation system is covered
in Chapter 5.3.3. Demand-controlling radiator heating would mean reducing heating while
users are not present and when free energy is available. Reducing heating while users are
not present is also known as using night setback temperatures, which means that the room
temperatures are lowered by a few degrees for the night. Decreasing heating when free
energy is available means that the controller calculates the forecasted heat loads and
estimates an appropriate level of heating output, respectively.

The energy performance of heating control is then considered satisfactory: Heating demand
is estimated based on outdoor temperature and overheating is eliminated with self-operated
unit controllers (mechanical radiator thermostats) in a reactive manner. This kind of control
is reliable and affordable. A problem however is that these mechanical unit controllers
cannot be adjusted remotely, which means that they cannot be synchronized with cooling
control either. The energy performance could be enhanced by adding predictive capabilities,
allocating heat output to periods of occupancy and enabling digital control.

39
5.3.2 Cooling system control
Cooling is controlled with the BAS. The principle of cooling is also heat exchange with
district network water however in reverse to heating. The cool water is used in three circuits:
Convectors, air coolers and chilled beams. The convectors and air cooler have a single set
point for the supply water temperature while chilled beam network has an interval of a few
degrees with a fixed raise to the current dew point temperature. The cooling demand is
determined by room temperature measurements and intake air temperature. The cooling
valve of convectors and chilled beams starts opening once a room temperature set point is
exceeded. Utility service rooms are equipped with convectors, whereas spaces in the
occupancy zone are equipped with chilled beams. The beams are active meaning that supply
air is distributed through them. The total chilled air supplied by chilled beams consists of
conditioned supply air by 30% and recycled indoor air by 70%. Conversely, AHUs start
cooling the intake air whenever the supply air set point is not reached with zero usage of
both the heat recovery and air heater.

The chilled beam operation is limited exclusively to the periods when there is a user present
in the room and the temperature set point is exceeded, which means that control is based on
demand. Likewise, the air cooler starts functioning only when the air is too warm. Therefore
there is little to do to save energy other than adjusting set points. The energy consumption
profile and valve openings of the heat exchangers in the cooling system are presented in
Figure 5.9 and Figure 5.10. It can be inferred that the energy consumption of the convector
system is negligible with its hourly energy consumption of approximately 1kWh.

The energy performance of cooling system control is therefore considered excellent. Cooling
energy is consumed exclusively during periods of demand: Chilled beams operate only when
users are present in an overheating room and air coolers operate only when the intake air is
too hot. The only sensible way to improve cooling performance is then to decrease cooling
demand by protecting spaces from heat loads.

Figure 5.9 Hourly energy consumption of the cooling system during a spring week. Consumption
occurs exclusively during periods of occupancy.

40
Figure 5.10 Cooling system valve opening trends as two hour averages to reduce oscillation in data. It
is apparent that the total consumption is defined by chilled beams and air coolers.

5.3.3 Ventilation system control

Ventilation is controlled with the BAS. The ventilation system consists of 16 air handling
units (AHU) that are mainly concerned with conditioning the intake air into supply air of
appropriate temperature and upholding a pressure in the supply and exhaust duct network.
The amount of air supplied and exhausted from the room is then controlled by dampers. A
typical AHU in OfficeOne consists of a supply damper, filters, a heat recovery unit, an air
heater, an air cooler, a supply fan, an exhaust fan and an exhaust damper. The AHUs operate
according to time schedules, and with less power during weekends. In OfficeOne the
operating schedule is from 6.30am to 11.59pm. The desired supply air temperature is defined
with a compensation curve with respect to the exhaust air temperature, and that temperature
set point is achieved with heat recovery that transfers heat from exhaust air to intake air if
necessary. If the effect of heat recovery is not sufficient, air heater valve is opened until the
desired air temperature is reached. If cooling is necessary then the air cooler valve is opened
respectively. The system also monitors that the supply air is not too humid.

The dampers have three states: minimum, normal and augmented air flow. The appropriate
state is defined by occupancy detection and the CO2 concentration of indoor air as follows:
A minimum air flow is maintained when the room is empty, a normal air flow is maintained
during occupancy and an augmented air flow is provided if an upper limit of CO2
concentration is reached.

As mentioned earlier, some AHUs operate around the clock due to smells from renovation.
Also some exhaust fans are running constantly in toilets and stairways. This causes noise in
some of the energy consumption data that obscures the cyclicity of the AHUs serving the
occupancy zone. The electricity consumption of the groups of AHUs in different switchgears
is presented in Figure 5.11. In the figure the spike on the early afternoon of Thu 12th May is
caused by a break exercise session in one area. It is important to note that this data does not
include the air heater and air cooler energy consumption.

41
 Figure 5.11 Electricity consumption of the ventilation system. The trends are largely obscure.

The cyclicity according to occupancy patterns is apparent in the green curve

6xAHUand1xExhaust. That group consists of three AHUs serving offices, two AHU serving
cafeterias and one AHU serving a stairway. The blue curve 1xAHU consists of a single AHU
serving a seminar room and utility room, which explains the irregular daytime spikes. The
purple curve 3xAHU is rather constant due to the needs imposed by renovation. It serves
various areas but mainly the lobby, meeting rooms in the cellar and toilets. The red curve
6xAHU6xExhaust is the one with the noisiest profile. It consists of three AHU serving office
areas, one AHU serving meeting rooms, one AHU serving toilets and one AHU serving a
stairway. The six exhaust fans serve utility rooms.

Yet, there is cyclicity hidden in the obscure trends too as observable in Figure 5.12. The
trends show the fan and related damper operation for office space, which is very cyclical.
The principles of ventilation system control can be seen from those trends as well: operating
schedules of the fan, less power on weekends, morning purges after the downtime and
occupancy detection based damper state. Note that the average value for damper state
consists of almost 30 dampers, so the augmented air flows in individual rooms have limited
effect on the average value.

Figure 5.12 Single AHU supply fan and related damper operation.

The energy performance of ventilation control is considered excellent as well. The electricity
consumption is mainly allocated to time periods of occupancy and heightened demand for

42
fresh air. Even though the control is by nature reactive it is not that much of a concern
because the inertia of the ventilation process is small in comparison to heating: Rising CO2
concentrations can be quickly and effectively managed by temporarily augmenting the
supplied air flow. The only concern in ventilation control is that AHU downtimes may have
adverse effects on the pressure differences over the building envelope.

5.3.4 Lighting system control

The lighting is controlled locally separately from the BAS, except for corridor and outdoor
lighting that are based on schedules and outdoor illuminance. Room lighting is controlled by
occupancy detection with a time-out period and a daylight harvesting setting. The artificial
lighting intensity does not adjust gradually, so there is a single daylight illuminance
threshold when the lights switch off. As data used in lighting control is unavailable through
the building automation system, only high level observations can be made. The electricity
consumption of the lighting system of OfficeOne versus the produced solar energy (assumed
to correlate with daylight availability) at the rooftop during a week is shown in Figure 5.13
and in larger scale in Figure 5.14. The data in the figure was collected between 1st February
and 1st July 2016 daily from 9:00 to 17:00 in order to focus on periods of occupancy.

Without knowledge on the operation permissions given by the lighting controller it cannot
be stated whether the reduced electricity consumption is attributable to user behavior or the
daylight harvesting function, whereas the efficiency of occupancy detection is evident
judging by the cyclicity of the consumption profile. The lighting controller’s operation
should be scrutinized to see to which extent daylight is really harvested. If the reduction of
around 5.6kWh in hourly electricity consumption can be attributed to a daylight harvesting
function, it is saving 3.8€ every sunny day with a price of electricity of 8.4c/kWh.

Figure 5.13 Hourly electricity consumption of the lighting system in comparison to hourly produced
solar energy during a week. The consumption is focused on periods of occupancy.

43
 Figure 5.14 Daylight availability vs. lighting electricity consumption. The consumption is reducing
towards increasing daylight availability.

The energy performance of lighting control is considered good. The electricity consumption
concentrates to periods of occupancy, however it is unclear how well the daylight harvesting
function is operating. Even if it is operating as intended by switching lights off when a
threshold is exceeded, the performance could be increased by constant luminance control:
Adjusting artificial lighting level gradually according to available daylight to keep a constant
luminance.

44
5.3.5 Summary of all energy performance gaps
All energy performance gaps identified in the research are listed in Table 5.4.

Table 5.4 Summary of all energy performance gaps.

Identified by Identified by
Description
interview observation
Heating
Poor area-specific adjustability of heating X X
Reactive essence of heating control X X
Heating control not being founded on the cyclical
X
demand
Need for shifting the compensation curves to adapt the
X
heating power to seasonal needs
No synchronization with cooling X
Cooling
Lack of protection from heat loads X
Ventilation
Potentially harmful implications on pressure
differences over the building envelope due to air X
handling unit downtimes
Lighting
Potential lack of daylight harvesting X
Processes
Manual input intensive processes X
Inability to control actuators in groups X
Inefficient channels for feedback from building users X
Lack of knowledge on the relative numbers of satisfied
X
and unsatisfied building users
Inaccurate single-value threshold alarms in BAS X
Inconsistent energy consumption reporting X X

45
5.4 Identified solutions to close the energy performance gaps
Eleven potential solutions were identified based on energy performance gaps covered in
interviews or observed in measurement data. These solution initiatives and their evaluations
are presented next.

5.4.1 Installing electronic radiator thermostats with smart heating

functions
Mechanical radiator thermostats are cheap and reliable, but have some drawbacks. The
temperature set points of mechanical radiator thermostats and chilled beams may differ from
one another. Radiators may be set to a higher room temperature than what the cooling system
is trying to achieve without any remote means observe or control the situation. Moreover,
mechanical thermostats measure room temperature right next to the radiator. This may cause
trouble as curtains or furniture right next to the radiator distort the temperature measurement.
If the measured temperature is much higher than the temperature in the occupancy zone, the
radiator valve may be closed even when the room temperature is lower than desired.

The situation can be improved by installing electronic radiator thermostats that could also
be connected to the building automation system via radio:
• The thermostats can be given an offset according to the distorted temperature
measurement, which means that the valve is operated as if the measured temperature
was higher or lower than the measured value next to the radiator.
• The radiator valve operation can be determined by a temperature measurement from
the occupancy zone.
• The radiator valve operation can also be linked to occupancy detection through
different modes: When nobody is around an economy mode is assumed, and when
occupancy is detected a comfort mode is resumed. This allows for a slight daytime
room-specific optimization of heating.
• Controlling logic can be applied regarding the synchronized operation of the chilled
beams and radiators to minimize the simultaneous cooling and heating while
preventing draft from windows.
• By enabling remote control over radiator valves the imperfections of valve functions
can be compensated (Salama, 2014, p. 52).
• Open windows are detected and valve is closed for a set period to minimize heat loss
(Salama, 2014).
• Valve operation can be optimized when using room temperature setbacks or
predictive heating control. For example during morning setback recoveries radiator
valves with mechanical thermostats tend to be completely open. This causes uneven
water flow in radiators and thus uneven heating if the network is out of balance. (TA,
2011, p. 9).
• The valve can be programmed to regularly close and open completely to prevent
calcification.

There are also some drawbacks related to the installation of electronic radiator thermostats:
• The hardware and installation cost a lot in large scale.
• Electronic thermostats require a power source, which means battery replacements
every one to three years or electrical wiring.

46
 • The cost of building automation programming work that depend on the amount of
logic integrated to the functioning of the valves.

The remote operation of radiator valves requires communication between the building
automation system and the electronic thermostat. This can be done wirelessly over radio,
which is supported by KNX, LON, Modbus and BACnet protocols (Ylitalo, 2012; Morris,
et al., 2012; Martocci, 2008). Some commercially available electronic thermostats are
operated by a central controller over radio mesh network technology, but essentially the
requirement for wireless communication is a radio transmitter connected to the thermostat.
If the radiator thermostat operation in OfficeOne is desired to be determined by a separate
room temperature sensor or chilled beam state, it is possible through building automation
where that information is available. This improvement would lead to a lot more data
generated into the building automation system (over 300 points of control). Therefore
considerations must be given to the graphical representation of this new data.

The evaluation results for this proposition are presented in Figure 5.15. Respondents with
building automation background mainly consider this proposition to pose minor difficulties,
however the estimated benefits vary a lot. Connecting the electronic thermostats to building
automation was considered easy, but the associated costs troubled. Some valued the
increased controllability of heating radiators high, while others did not see large benefits to
be expected from this. Respondents with building services engineering background mainly
saw this proposition as beneficial, however the assessed challenge of implementation
diverged. In general the ability to synchronize heating and cooling was considered valuable.
Respondents with property management background did not agree on the challenge but
generally the improved heating control was welcomed. Respondents in property
maintenance estimated the challenges of implementation to be medium, and the positive
influence to be large. The positive influence was mainly attributed to the ability to adjust the
heating power of radiators with regard to the actual room temperature rather than the
temperature measured right next to the radiator. The necessity of changing the batteries did
not trouble them.

Figure 5.15 Evaluation results of the initiative of installing electronic radiator thermostats. The ellipse
axis length corresponds to the standard deviation of answers, while the dot is the average.

47
5.4.2 Adaptive outdoor temperature compensation for heating
The vast majority of the buildings of the case organization have a radiator heating system
where supply water temperature set point is determined by a compensation curve that is a
straight line, which is a suboptimal heating control method. A straight line is not an ideal
shape for a compensation curve as the required heating power is not linearly dependent on
the outdoor temperature nor the heating power of radiators is linearly dependent on the
supply water temperature (Paiho, et al., 2002). Therefore the curve placement or steepness
often needs adjustment after users complain about thermal discomfort. A typical case is that
during winters users complain about coldness and the curve is shifted upwards. Yet, after
outdoor temperatures rise it is easy to forget to shift the curve back to its original position.
This results in the radiator supply water being often warmer than necessary.

The automatic adjustment of the compensation curve requires only programming effort if
room temperature measurements are available through the building automation system. An
example of such a fuzzy logic based program is presented by Paiho et al. (2002, p. 69). The
idea is to adjust the radiator supply water temperature according to outdoor temperature and
room temperatures. The domains of these three parameters are divided into fuzzy sets that
are used in control with fuzzy rules that are presented in Table 5.5. The configuring process
begins by moving the original compensation curve to coincide with the current outdoor
temperature and supply water temperature. Then the fuzzy reasoning is done twice: Late in
the evening and in the early hours. If the result of both reasoning processes is similar in
means of need and direction of change, a new point is recorded accordingly. Every point that
is recorded in the future overwrites the existing point that is the closest, and so the
compensation curve is updated by connecting the points. (Paiho, et al., 2002). This process
is illustrated in Figure 5.16. By constantly checking for the need to update the compensation
curve it is ensured that the supply water temperature is kept at an optimal level. The
defuzzification method of the program was not specified.

Table 5.5 Rule base for the fuzzy controller. Translated from Paiho et al. (2002, p. 69).
1. IF room temperature is “cold” AND if the outdoor temperature is “cold” THEN “raise” the supply
water temperature
2. IF room temperature is “cold” AND if the outdoor temperature is “mild” THEN “raise a lot” the
supply water temperature
3. IF room temperature is “hot” AND if the outdoor temperature is “cold” THEN “drop” the supply
water temperature
4. IF room temperature is “hot” AND if the outdoor temperature is “mild” THEN “drop a lot” the
supply water temperature
5. IF room temperature is “suitable” AND if the outdoor temperature is “cold” THEN “don’t
change” the supply water temperature
6. IF room temperature is “suitable” AND if the outdoor temperature is “mild” THEN “don’t
change” the supply water temperature

48
Figure 5.16 The principle of compensation curve modification process. Based on Paiho et al. (2002, p.
69).

The potential benefits of applying this program include smaller deviations from desired room
temperatures, reduced energy consumption and a decrease in the workload of maintenance
personnel. Naturally the extent to which heating can be optimized depends on the physical
radiator network, whether it is the same for the whole building or if it is façade-specific.
With façade-specific radiator networks other parameters can be utilized to a larger extent
such as solar radiation. This increased adjustability has been found to enable savings of 8%
in heating expenses (Laaksonen, 1995, p. 629). Anyhow, with adaptive heating control in a
20,000m3 residential building with an annual heating energy consumption of 600MWh and
a room-specific heating power of 184kW, Paiho et al. (2002, pp. 93-94) estimated the
following benefits if the average room temperatures decrease by 1°C and maintenance
workhours decrease by 5 hours a year:
• annual energy savings of 3.3% with a total heating energy consumption of 600MWh
(800€ with 40€/MWh),
• savings of 170€ due to reduced workhours and
• smaller deviations in room temperatures.

There are some risks and drawbacks associated to the implementation of the program, too.
First of all, the program cannot perform well if the measured room temperatures do not
represent the temperature in the occupancy zone. For example if the temperature sensors are
located inside a chilled beam problems are likely to arise. The decision on which room
temperature measurements to include and how is building-specific. Secondly, the program
requires constant monitoring by a property maintenance professional for a year to ensure its
proper functioning. For property managers it is more likely a matter of resources than
additional staff, however. Lastly, some precautionary measures must be taken to ensure the
appropriate functioning of the heating system when room temperature measurements are
temporarily distorted or unavailable. Such can include a default compensation curve when
an abnormal situation is detected.

49
The prerequisites for the implementation of this program are easily met. Radiator supply
water temperature control and outdoor temperature measurements are basic points of control
and available through building automation. Also the room temperature measurements are
often produced, but if that is not the case, retrofitting some wireless temperature sensors is
not too big of a task.

The evaluation results for this proposition are presented in Figure 5.17. Respondents with
building automation background considered this proposition to be highly beneficial with the
dominant opinion being that the implementation would be easy, too. The main arguments
were that efficient adjustment of heating could save a lot of energy, and often no added
infrastructure would be required, or at least it would be easy to install. Respondents with
building services engineering background mainly considered the application to be easy,
however the benefits were estimated mainly to be medium. The person that estimated the
challenge of implementation (and also the benefits) to be high was concerned that in Finland
during periods of large daily temperature changes the program’s performance may be
suboptimal. Respondents with property management background estimated both the
challenge and benefits to be medium, while respondents in property maintenance thought
uniformly that the positive influence is high but challenging to obtain. They thought that
successfully automating the shifts of compensation curves would save a lot of energy and
time, but choosing the right room temperatures for input would be difficult.

Figure 5.17 Evaluation results of the initiative of programming adaptive outdoor temperature
compensation for heating control.

5.4.3 Room setback temperatures outside office hours

Unlike with the ventilation system, cooling system and plug appliances, the energy
consumption of the radiator heating system of OfficeOne is not cyclical. Radiator heating
output is kept at a level determined by the outdoor temperature regardless of the time and
occupancy. By adopting setback temperature schedules for nights and weekends some
energy could potentially be saved by lowering room temperatures temporarily by 1-3°C.

There are many ways to use setback temperatures. Setback schedules and depths can be
given as fixed values based on safe side expert knowledge, or those can be the outputs of an
algorithm that calculates optimal setback depths and likely reheating times so that overall
energy consumption is minimized. Moreover, the existing equipment defines how the
setback can be implemented. With radiators that have mechanical thermostats setbacks can
only be applied by lowering the supply water temperature or reducing the water flow in the

50
network. If the system was equipped with remotely controllable electrical thermostats, the
setbacks could also be executed by controlling individual radiator valves. Anyhow, the
limitations of the equipment should be considered before implementing setback
temperatures. For example in the case organization’s buildings the heat exchanger valves of
the radiator network have usually been sized to meet the requirements of a cold winter, which
means that with temperatures from -5°C to +10°C the control is rather coarse and the system
is not precise in its adjustments.

Regardless of the method it is important to correctly determine the setback schedules to

minimize user inconvenience but also maximise savings. To facilitate this aspect, a self-
learning algorithm to learn and store reheating rates is presented next. The algorithm has
been developed by Paiho et al. (2002, p. 127) to control a building’s air supply based heating
system in relation to building occupancy patterns. It was applied in a low-energy office
building that consumes approximately 60% less energy than offices in average (Paiho, et al.,
2002, p. 116).

The algorithm is based on the idea that the setback recovery rate is dependent on outdoor
temperature alone. Then the time in minutes that it takes to raise the room temperature by
1°C is recorded for each outdoor temperature and used for control, as in the future this
information can be used to derive the required advance to start reheating. The initial values
for reheating rates are generated with test runs and the assumption of linear relationship
between outdoor temperature and reheating rates. Also a boundary condition is applied: The
reheating rates with indoor temperatures of 21°C and above are 0 min/°C. By using this
information, a straight line can be fitted that contains all the initial values. (Paiho, et al.,
2002). Yet, it may be wise to rule out setback temperatures during very cold periods when
the excess heating power of radiators is used up.

The learning process modifies the initially linear curve based on true setback recovery
durations. If the reheating duration is overestimated by let us say 20 minutes for a 2°C
setback depth, a new point is inserted corresponding to that outdoor temperature with the
initial reheating rate subtracted by 20 min / 2°C = 10 min/°C. On the other hand when the
estimated reheating duration of 200 minutes is underestimated by let us say 30 minutes and
the difference to the suitable room temperature is 0.6°C, then the new point corresponds to
that outdoor temperature and the initial reheating rate increased by 0.6°C * 200 min/°C =
120 min/°C. Then this new point is connected to the nearest points. For the first learning
event the point is connected to the initial curve’s end points. This learning process is
visualized in Figure 5.18. To impose restrictions to learning Paiho et al. (2002) used a 0.5°C
deviation limit for resulting indoor temperature to differentiate between a successful
reheating and a failed one, and the maximum allowed change to the initial reheating rate was
limited to 50% to reduce the effect of errors. The effect of errors could also be effectively
managed by the method of ordinary least squares. (Paiho, et al., 2002).

As the radiator heating system is more likely to experience long lags between control
measures and room temperatures compared to air supply based heating, forecasted outdoor
temperatures may become a necessary input to make better control decisions. Additionally
it may be appropriate to integrate learning of the cooling times versus outdoor temperatures
to enable better timing.

51
 Figure 5.18 The learning process. Based on Paiho et al. (2002, p. 132).

Simulations on room setback temperatures suggest a decent amount of savings. In his

Master’s thesis Salama (2014) simulated 3°C night setbacks in radiator heating. In
simulations that were based on a mathematical model of room thermodynamics in Python
language the expected energy savings were 4.3% with the air-conditioning switched on for
the night, and 22.5% with air-conditioning turned off for the night. The outdoor temperatures
corresponded to temperatures in March in Finland. If the setback depth and length were
optimized with regard to outdoor temperature the savings were 8% with average outdoor
temperature corresponding to Southern Finland’s annual average of 6.5°C. With an IDA
simulator of VTT Technical Research Centre of Finland the simulated 3°C night setbacks
resulted in energy savings of 12.6% between December and February in Finland. (Salama,
2014). To estimate the scale of the savings consider an office building in Helsinki with a
surface area of 6100m2. The consumed district heating energy during the past year was
roughly 600MWh. If half of it is assumed to go towards radiator heating, a let us say 10%
decrease in energy consumption would then equal to annual savings in money of 1200€ or
0.20€/m2 with the price of 40€/MWh for district heating energy.

There are some problems associated with the implementation of setback temperatures:
Setback recoveries can use up more energy than what was saved during the lower
temperature period, the desired temperature may not be reached by the time users arrive and
the feasibility of the method is situation-specific. In the reheating phase of a setback period
the required heating power exceeds the power required in a normal situation. This fact leads
to a building-specific break-even for the setback duration that determines whether any
energy is saved. Moreover, the change in the heat content of the radiator network water uses
up energy. For example in a 40,000m3 educational facility the radiator network has 3m3 of
water. In fixed 5°C setbacks of supply water temperature the simple action of reheating the
network water changes the water’s heat content every time by 17.5kWh, and annually by
4.4MWh if the heating season is assumed 250 days long. This amount of energy corresponds

52
to 1.1% of the total estimated radiator heating energy consumption of 400MWh in that
building if the losses during heat exchange are not considered.

If the algorithm fails in estimating the duration for the reheating process and rooms remain
chilly when users arrive, complaints can be expected. It is important to note that with
mechanical radiator thermostats all the radiator valves are often completely open during the
setback period. This leads to uneven water flow and so uneven heating in the setback
recovery phase if the network is out of balance. (TA, 2011). Therefore it is possible that
some rooms remain chilly longer than others causing thermal discomfort. If this problem
cannot be solved quickly with the current resources it is more likely that the whole setback
function is disabled rather than an external expert is summoned. Furthermore, an important
notion is that temperature sensors often measure air temperature instead of operative
temperature, on which the Finnish indoor environment categories S1-S3 and user’s thermal
comfort are based. Therefore successful implementation of setback temperatures requires
constant monitoring by a property maintenance professional and knowledge on the
building’s characteristics to minimize the amount of dissatisfaction within the users and
maximize the longevity of the setback temperature function.

The last major drawback of the implementation of setback temperatures is that its feasibility
is very context-specific. First of all, to even consider its implementation the building should
have periodical use patterns and the user’s operations should not be harmfully affected by
the temperature drop. Secondly, the function should adapt to the surrounding conditions.
With temperatures much below zero the building’s heating system is struggling to maintain
a constant room temperature and there is not enough heating power to raise the room
temperature rapidly. On the other hand, the setback depth should be optimized: Whenever
there is much excess heating power available and room temperature can be raised quickly,
the temperature drop can be deeper. Thirdly, it is often the case that the excess heating power
has been minimized in a district heating based radiator heating system. Up to 30%-50% of
the heating bill can consist of a payment that reserves a desired heating power capacity for
the ultimate heating situation (Laaksonen, 1995, p. 581), however by average it is around
16% (Göös, 2012). This means that there is inherently a trade-off with the savings achievable
by night setbacks and the costs of the required heating power.

The requirements of the implementation of setback temperatures on sensor and actuator

infrastructure depend on the method. If temperature drops are carried out based on expert
knowledge with mechanical radiator thermostats by lowering supply water temperatures, the
existing automation systems are sufficient with data on outdoor temperature and control over
supply water temperature. If setback lengths and depths are optimized, room temperatures
and possibly weather forecasts are needed. As mentioned in Chapter 5.4.2, room
temperatures are often available in the building automation system and if this is not the case,
retrofitting a few wireless sensors is not a major issue. Weather forecast data can be acquired
from commercial providers like Foreca or for free from Finnish Meteorological Institute, if
such is not done already. Finnish Meteorological Institute’s open data program is explained
in more detail in Chapter 5.4.4. Importing weather data requires constant queries from the
building automation system to a web feature service and partitioning the resulting file into
exploitable parameters. If control over every radiator’s valve position and temperature set
point is desired, electronic thermostats with wireless control from building automation
system is required as explained in Chapter 5.4.1.

53
The evaluation results for this proposition are presented in Figure 5.19. Respondents with
building automation background considered the implementation of setback temperatures
easy, however the expected benefits varied a lot. For some it was unclear how much of the
saved energy would be used up in the reheating phases of setback periods, and whether
anything would be saved at all. The programming part was not thought to be a problem,
though. Respondents with building services engineering background mainly thought the
proposition to be challenging, while the potential benefits were mainly estimated medium.
It was emphasized that the balance of radiator networks should be verified before applying
night setbacks. Respondents in property management did not see much benefits and
considered the temperature setbacks to be a challenging area. The fact that the available
excess heating power has been minimized does not strongly encourage the implementation
of this proposition. Respondents in property maintenance estimated the benefits to be large,
but also the challenge. They thought that with careful implementation setback temperatures
would be a viable way to save heating energy.

Figure 5.19 Evaluation results of the initiative of using room setback temperatures.

5.4.4 Optimizing heating with weather forecast data

During spring and autumn the changes in outdoor temperature can be large between the day
and the night, which may lead to suboptimal heating control: Compensating the supply water
temperature with outdoor temperature in real time is poor at taking into account the thermal
inertia of the building. For example if outdoor temperatures suddenly rise in the spring after
a long and cold period, the heating power is cut down instantly and users may consequently
experience thermal discomfort due to the still cold structures. On the other hand in energy
saving perspective, free energy is unharnessed when the building is heated during the early
morning only to stop heating completely and start cooling a few hours later.

The solutions to utilize predicted heat loads vary. A simple approach is to use an average of
outdoor temperatures of a certain time frame to adapt to fluctuating conditions. The
challenge then is to determine that appropriate time frame for every building. Another type
of a solution is based on Roger Taesler’s (2007) concept of equivalent temperature.
Equivalent temperature is a transformed outdoor temperature that can be used to control the
supply water temperature directly so that expected heat gains and heat losses are taken into
account to maximize the use of free energy and minimize the use of purchased energy.
(Taesler, 2007). It is essentially a mathematical model derived from a building’s heat balance
that relies on technical building characteristics, site characteristics and weather forecast
parameters. A commercially available product called eGain forecasting™ uses equivalent

54
temperature in heating control (eGain, 2016). A more decentralized solution is the control
of every radiator equipped with an electronic thermostat individually. Salama (2014)
developed a neural network based solution that is used to control the room temperature set
point based on weather forecast. The idea is to lower the room temperature in advance so
that the changing outdoor temperature elevates it to the desired level. The neural network is
found by a genetic algorithm that runs simulations. The inputs of the neural network are 11
different second time derivatives of outdoor temperature and the output is the deviation with
the room temperature set point. In his Python language –based simulations of a thermal
model of a room the achieved energy savings were 9% with the average outdoor temperature
corresponding to Southern Finland’s annual average of 6.5°C. The simulations were a part
of the product development of the smart thermostat of Fourdeg. (Salama, 2014; Fourdeg,
2016). To estimate the scale of the savings consider an office building in Helsinki with a
surface area of 6100m2. The consumed district heating energy past year was roughly
600MWh. If half of it is assumed to go towards radiator heating, a 9% decrease in energy
consumption would then equal to annual savings in money of 1080€ or 0.18€/m2 with the
price of 40€/MWh for district heating energy.

The main risks and drawbacks associated to continuous model-based predictive heating
control are related to user dissatisfaction, difficulties in problem diagnosis and costs of
installation. Similarly to the use of setback temperatures, user dissatisfaction leads quickly
to the disconnection of the smart predictive control system, and to the situation when money
has been spent for nothing. Additionally, the case organization’s property managers are
reluctant towards the installation of systems where problem diagnosis can only be done by
external consultants. Whenever the parameters of a mathematical model are wrong, the
maintenance personnel will face major difficulties trying to diagnose the problem, which
also leads to the previously mentioned situation if the service provider is not
accommodating. Anyhow, all of the predictive methods require constant supervision by a
property maintenance professional for at least a year to ensure appropriate operation, which
absorbs resources of the maintenance organization. Drawbacks of the installation of
electronic thermostats have been discussed earlier in Chapter 5.4.1.

Again the required information and control infrastructure depends on the amount of
operations included, though weather forecasts must be imported to the system in all of the
cases. Those forecasts can be received from a number of providers, however the Finnish
Meteorological Institute (FMI) provides them for free as a part of their open data program.
By querying the web feature service of FMI one has continuous access to weather data.
These queries can be made single point and parameter-specifically. It however is necessary
to register for the service and obtain a user-specific API key that is used in the construction
of a query. There are also limitations imposed to the number of requests made by a single
API key in different time windows, however these limits are reasonable. (Finnish
Meteorological Institute, 2014). The following findings were made based on FMI weather
forecasts and subsequent observation data:
• By average the forecasted outdoor temperature was 2.0°C lower than the observation.
• By average the forecasted cloud cover exceeded the observation by 2.4 units on a
cloudiness scale of 0-8.
• By average the forecasted wind speed exceeded the observation by 1.5 m/s.
• By average the forecasted air humidity was 5.4% higher than the observation.

55
The findings were based on 888 hourly forecasts between 20.5.2016 5:00:00 - 20.7.2016
7:00 with scopes up to 24 hours. Observations were not available coordinate-specifically, so
the values are weather station (suburb) specific. The forecast coordinates belong to that
suburb though. Also cloud cover observations were limited to a scale of 0-8 whereas
forecasts were on a scale of 0-100 (requiring conversion to 0-8).

The web service queries can be made with a web browser that opens a XML file containing
time and parameter stamped forecast values. For example a query to obtain the temperature,
total cloud cover, wind speed, wind direction, dew point and accumulated solar radiation
forecasts for a randomly chosen point in Åland Island would be of form:

http://data.fmi.fi/fmi-apikey/<API-
key>/wfs?request=getFeature&storedquery_id=fmi::forecast::hirlam::surface::point::timevalue
pair&latlon=<Latitude and longitude, for example:
“60.2,20”>&parameters=temperature,totalcloudcover,
windspeedms,dewpoint,winddirection,radiationglobalaccumulation

After the forecast provider has been chosen, it is only a matter of creating the software
interface to import the data into building automation. The rest of the information and control
related challenges with regard to supply water temperature, outdoor temperature, room
temperatures and electronic radiator thermostats have been discussed earlier in Chapters
5.4.1 and 5.4.2. Still it should be emphasized that the room temperature measurements need
to be representative of the conditions in the occupancy zone. Neglecting this and operative
temperature considerations could lead to user complaints and disconnection of the whole
function. Operative temperature can be well below the room temperature for example when
the windows are large, and heating power is too low to eliminate the cold draft.

The evaluation results for this proposition are presented in Figure 5.20. Respondents with
building automation background considered the challenges in weather data utilization
primarily minor. Weather data is available from a number of sources and importing it is
straight-forward. The expected gains differed, though. The usage of weather data directly in
radiator heating control raised suspicion, however more confidence was placed on the
benefits in special heating control such as snow melting. According to a respondent these
savings can be significant when the warm weather of the next day is used to melt the snow
instead of electricity. Respondents with building services engineering background
considered weather data utilization to be easy as well, with mainly large benefits to be
expected. After all it is about harvesting free energy. Respondents in property management
estimated the benefits to be medium, but thought that predictive heating optimization would
be difficult without compromising thermal comfort. Respondents in property maintenance
regarded this field as both challenging and rewarding. They thought that at best this
improvement would help in keeping the indoor thermal conditions more stable with dynamic
weather.

56
 Figure 5.20 Evaluation results of the initiative of utilizing weather forecast data in heating control.

5.4.5 Intelligent lighting and blind control with daylight harvesting

Occupancy detection is utilized extensively in lighting control, however available daylight
is not. The disregard of users can easily undermine the attempts to harvest the daylight, too.
When the sun is shining strong and low the blinds are often closed by the user, but after the
sun has risen the blinds remain closed and lights on even if the daylight would be enough to
illuminate the space. It is impossible to know currently how well the daylight harvesting
function is working in OfficeOne based on Chapter 5.3.4, but it can be suggested that
gradually adjusting artificial lighting could make more use of the available daylight than a
simple on-off switch.

To make the most of the available daylight and to automate energy conservative behavior it
would be necessary to integrate blinds and lighting control. The idea is to let daylight into
the room whenever there is no risk of glare, and correspondingly adjust the artificial lighting
level. Doing so would obviously require motorized blinds. Moreover, the controller requires
information on the available outdoor illuminance, which can be provided by illuminance
sensors. The sensors can be placed in rooms or on the outside wall to detect the space-
specific daylight intensity. Then with information on solar position the likelihood of glare
can be inferred. Next the outlines of two solutions are presented.

Two solutions developed around the latitude of Finland shared a number of similarities.
Paiho et al. (2002, p. 211) developed a fuzzy control system that is based on vertical surface
daylight intensity, sun altitude and the azimuth as inputs, and blind angle together with
artificial lighting level as outputs. The fuzzy rules are essentially based on the notion that
glare is likely when the sun’s altitude is low and the vertical surface illuminance level is over
20klx. (Paiho, et al., 2002). The glare is then eliminated by adjusting the blind angle. A
similar system was developed and tested by Bülow-Hübe (2007). The idea is to keep the slat
angle equal or greater than the cut-off angle of the present effective solar altitude when the
vertical surface illuminance threshold of 20klx is exceeded. The effective solar altitude takes
into account the latitude and orientation of the building, and the cut-off angle is the angle
that prevents direct sunlight. In the case of negative cut-off angles the blinds remain
horizontal to eliminate a direct view of the sky. On the other hand, with a long period of little
available daylight the blinds will be raised to maximize daylight intake (Bülow-Hübe, 2007).
Some cut-off angles for effective solar altitudes are presented in Figure 5.21.

57
 Figure 5.21 Cut-off angles for effective solar altitudes. Based on Bülow-Hübe (2007, p. 16).

The solar altitude angles for each time of day change constantly in the latitudes of Sweden
and Finland. This makes the blind angle control more complicated than in countries closer
to the equator. Bülow-Hübe (2007) solved this problem by using three different slat angles
during the day that are listed for each month. Solar position can also be calculated with an
algorithm. In one presented by Reda & Andreas (2004) the solar azimuth and zenith angles
can be determined with an accuracy of around ±0.0003° and validity from the year -2000 to
6000. The required parameters for the solar position calculation are date, coordinated
universal time, the fraction of a second that is added to the coordinated universal time to
adjust for Earth’s irregular rotational rate. The predicted values for these fractions of a
second are reported weekly for example by the United States Naval Observatory. Other
necessary parameters are longitude, latitude, elevation and annual local pressure and
temperature. The coordinated universal time is based on universal time, also known as
Greenwich civil time, and is kept within 0.9 seconds of universal time by one second steps.
These steps are reported based on observation. Also the slope and azimuth rotation for the
observation surface are required to determine the angles of incidence from the solar angles.
Note that the solar azimuth angle is measured from north towards the east, and the solar
zenith angle is measured from up towards the horizon in this application as depicted in Figure
5.22. (Reda & Andreas, 2004).

An important objective is to avoid inconvenience to the users due to fluctuating control. For
example consider the situation when a cloud moves in front of the sun that causes glare:
Before the event the blinds were closed and lighting level was high, and after the cloud
blocks some of the sunlight, blind angle is lowered and lighting level decreased. Soon
afterwards as the cloud moves away, the blinds are closed and lighting level increased again.
To avoid such annoyance there should be a sufficient delay of around 15-30 minutes
integrated to the action of opening the blinds, however the action of closing the blinds should
be swift. The delay length could also depend on the amount of fluctuation in the measured
daylight illuminance. (Paiho, et al., 2002).

58
 Figure 5.22 The two solar angles: the zenith and the azimuth.

The benefits of integrated blind and lighting control include reduced electricity consumption
of artificial lighting and reduced cooling energy consumption. These benefits can be
attributed to decreasing heat loads and visual discomfort with blinds while maximizing the
use of available daylight. In their simulations Paiho et al. (2002, p. 224) achieved average
lighting energy savings of 44.3% in comparison to the situation when lights are constantly
on at their nominal output during working hours. Therefore these savings are not comparable
to occupancy detection based lighting control situation. Similar results were observed by a
real test bed in Sweden with average annual savings of approximately 50%. The savings
were however really seasonal due to available daylight: 77% in May and 5% in November,
respectively. Also if working desks are located far from the windows the expected savings
reduce to around 25%. (Bülow-Hübe, 2007). When estimating the scale of these savings,
consider an office building in Helsinki with a surface area of 6100m2. The average amount
of electricity used for lighting is 400kWh daily which in money corresponds to 33€ daily
and about 12 152€ annually with the price of electricity of 8.4c/kWh. The potential savings
are then annually 6076€ or 1.0€/m2 in comparison to the situation when lights are constantly
on at nominal output during working hours.

The drawbacks of integrated blind and lighting control include the investment and lifecycle
costs of motorized blinds, and the annoyance of users especially if the system is not working
properly. Motorizing the blinds costs and even though the blinds can be wirelessly
controlled, they need a power supply and thus electrical wiring. This adds to the installation
cost. The tuning process of the system on the other hand can be annoying for users. Even the
fluctuating artificial lighting level can be a distracting element, but when blind movement
and angle adjustment is added the whole system risks becoming an inconvenience. Anyhow,
as there is little experience of motorized blinds in Finland some policy resistance is only
natural.

Lighting systems are challenging in the information and control infrastructure sense. In the
case organization’s buildings lighting control is often implemented locally with DALI
protocol. This means that the lights are not connected to building automation, and all the
programming must be done on the spot. Many building automation control software support
the inclusion of systems operating on DALI protocol, but doing so requires expertise in that
field from the building automation contractor. In OfficeOne for example, corridor lights and
lights in the outdoor areas that are operating based on time schedules are connected to
OPENweb ControlPanel, but room lighting is left out. It is perhaps because of the specialized

59
nature of the lighting field: There are lighting management solutions such as Helvar Imagine
(2016) commercially available that can be integrated to building automation, and systems
such as Houm (2016) that are completely cloud-based and do not emphasize integration with
building automation. Anyway, the decision on whether to integrate these systems or not is
not straight forward. Integration always requires maintenance when systems update or are
replaced, and if both of the systems enable remote problem diagnosis and system supervision
separately, there may not be enough grounds to integrate them. The decision will
nevertheless determine whether the sensor data used in lighting control will be available in
the control of other devices through building automation.

Other information availability and control infrastructure problems relate to illuminance

inference and blind control. Installing illuminance sensors is not a major task, however if
daylight availability estimation is based on a vertical surface illuminance measurement and
solar position calculations, some astronomical parameters need to be imported to the
program. The shading devices on the other hand can be controlled with building automation
for example by installing commercially available controlling solutions such as Somfy
Animeo (2016) product family that supports IP, KNX and LON protocols.

The evaluation results for this proposition are presented in Figure 5.23. The respondents with
building automation background did not have similar opinions on neither the challenge nor
the benefits of a daylight harvesting blinds and lighting control system. Daylight harvesting
as a lighting control principle was generally seen as a desirable goal, whereas including
motorized blinds was not. Respondents with building services engineering background
considered the proposition to be challenging with varying estimated benefits. Respondents
with property management background estimated diverging benefits, and the challenge of
implementation was considered above medium. This can be attributed to previous
experience: Motorized blinds were installed in a building and the users were not satisfied
because of inconvenient blind movement. Respondents in property maintenance regarded
this field as both challenging and rewarding. They thought that currently the available
daylight is not utilized sufficiently.

Figure 5.23 Evaluation results of the initiative of daylight harvesting with motorized blinds and
constant luminance lighting control.

60
5.4.6 Improving the manageability of large amounts of sensor data and
actuators
Large amounts of sensor data nor actuators can be managed well without case-specific
definitions. Forming a control group of snow melting heaters of all the buildings in the same
suburb would require programming effort, or querying the efficiency of all the heat recovery
units would require manual gathering of all the relevant measurements. There are no
hierarchical system- and location-specific attributes for actuators or sensor data that would
ease the work of the maintenance organization. Note that OfficeOne alone produces
continuously over 600 different measurements.

On the other hand, allowing property managers to create refined data streams and views
based on their needs would facilitate their supervisory task and benchmarking. These refined
data streams could include SFP figures, heat recovery efficiencies and other HVAC system
metrics. If the software allows the creation of user’s own views, these figures can be laid out
on the same page across different buildings. Attributes would be of great help in the process
of key figure definition, as all the relevant measurements could be easily referred to as their
attributes rather than sensor network nodes. Moreover, if better alarm event management is
desired, the attribute-based data modelling philosophy should be extended there. Also the
ability to define triggers and data fields in alarm events such as Boolean alarm criteria, traffic
lights to visualize status, response time calculation and cumulative costs of maintenance
negligence the alarm event management could produce more useful figures to an active
property manager. The matter of alarm events with attributes is not considered here further,
however.

The basic idea is to add attributes to sensors and actuators that describe its location and the
measurement or actuator type. Moreover, these attributes should be structured hierarchically
when possible. An example of this procedure with room temperature measurements is
presented in Table 5.6. By utilizing the properties of relational databases, repetitive attributes
can be input only once with primary and foreign key relationships. In the tables below, the
ID column higher in the hierarchy is the primary key (PK) to which the tables lower in the
hierarchy refer to with their case-specific foreign key (FK) column.

61
Table 5.6 Normalization of sensor data.
BEFORE NORMALIZATION
Point-wise measurements
Date Time Room 101 Room 102 Room 201 Room 202
Temperature Temperature Temperature Temperature
(°C) (°C) (°C) (°C)
1.4.2016 07:00:00 20 21 20.3 20.7
1.4.2016 07:05:00 20.5 21.3 20.6 21.1
… … … … … …
AFTER NORMALIZATION
Floor table
ID (PK 1) Floor Name
1 1st floor
2 2nd floor
… …
Room table
ID (PK 2) Room Name Floor ID (FK 1)
1 101 1
2 102 1
3 201 2
4 202 2
… … …
System table
ID (PK 3) System name
1 Room measurements
2 Radiator heating system
… …

Measurement type table

ID (PK4) Measurement type Measurement unit System ID (FK 3)
1 Temperature °C 1
2 Indoor air CO2 concentration ppm 1
3 Illuminance lux 1
4 Supply water temperature °C 2
… … … …
Fact table
Date and Time Measurement Room ID (FK 2) Measurement ID (FK 4)
1.4.2016 07:00:00 20 1 1
1.4.2016 07:00:00 21 2 1
1.4.2016 07:00:00 20.3 3 1
1.4.2016 07:00:00 20.7 4 1
… … … …

The example above is meant only for introductory purposes, as in reality the database design
task is much more complex. Such a database structure would face problems with
measurements that are not limited to a single room, but are distributed across several areas
such as in an AHU. Therefore it is necessary to develop a more flexible database structure
so that areas can be defined. It may also be a good idea to allow for relationship definition
between systems and locations in the database structure, even if that is not utilized
immediately in applications. This matter is further illustrated in Appendix 2.

The benefits of attributes in measurement data and actuator control are indirect. These
attribute-based features would allow for faster routine operations such as switching seasonal
heaters on and off, user friendlier measurement data management by data slicing
functionalities and easier monitoring through refined information and benchmarking. These
features would not save energy alone, but only through active harnessing of a property
manager. A lot of energy is wasted because of failures in simple things that go long

62
unnoticed, such as forgetting those seasonal heaters on or by having the heating power
unnecessarily high. With a better data model and user interface the property manager could
easily define the key figures that reflect his concerns, and drill from abnormalities through
to the underlying problems.

By focusing in attributes in application development the hardware gradually loses its

importance in programming sense. All references in applications can be made through
attributes rather than network addresses. Also, adding or replacing sensors is only a matter
of adding rows in tables rather than a data structure affecting operation. Extrapolating the
database thinking of sensor data and attribute-based controls leads to many cross-platform
possibilities as every measurement and actuator could be linked to a location in a building
information model for example. Acquiring attribute-based control capabilities can be
problematic as it either is a fundamental element of a building automation system or then it
is not. The development of such features can be expensive, and so it would require sufficient
demand in the market.

When it comes to adding attributes to sensor data the problems are easier to overcome.
Measurement data can be imported from a building automation system to an external server
that then writes a database with every piece of measurement data containing attributes.
Anyhow, software development and database management is not free. Properties produce
huge amounts of data that require processing capabilities, data transfer, server rooms with
cooling and database maintenance.

The evaluation results for this proposition are presented in Figure 5.24. Respondents with
building automation background considered the challenges of implementation in improving
the manageability of large amounts of sensor data and actuators to be medium with medium
to high benefits to be expected. Some of the controller software already support user-defined
attributes to an extent. Group control of some previously defined actuator groups was
generally considered desirable with a large amount of properties to manage, but everyone
did not see the benefits in an interface enabling arbitrary control groups. The estimated
challenge and benefits diverged dramatically among respondents with building services
engineering background. Two of them thought that time could be saved a lot making sensor
data and actuator management more efficient, whereas one person did not see these benefits.
This person is not involved with a similar stock of properties every day, though. Respondents
in property management regarded the proposal to be somewhat of a low hanging fruit. In
their perspective the natural direction of controller software development should include a
data model to better manage with the huge amounts of data. Also benchmarking views were
thought to be desirable for energy management purposes. Respondents in property
maintenance were frustrated with the current snow melting control. Currently it is based on
outdoor temperature limits so that the heaters are on during most of the heating season.
Stricter limits have not served their purpose as the heaters have then been off when they have
been needed. The property maintenance respondents therefore thought it best to have a single
switch to stop snow melting altogether within a sensible area when there clearly is no need,
and thought that this would save a significant amount of energy. Also the ability to define
time schedules of equipment in groups was desired.

63
 Figure 5.24 Evaluation results of the initiative of improving the manageability of large amounts of
sensor data and actuators.

5.4.7 Enabling communication between building automation and room

reservation systems to reduce energy consumption and manual
input
Energy is wasted when empty rooms are being conditioned. With multi-user office space in
the case properties this is not much of an issue because the rooms are rarely completely
empty during office hours. Yet, with irregularly used rooms such as lecture halls or meeting
rooms the energy inefficiency can become more of a concern. In some of the case properties’
lecture halls, air handling unit operation times are set by hand according to teaching
timetables. Air supply can also be based on a fixed schedule or occupancy detection. None
of these solutions alone is optimal: Manual input is laborious, fixed schedules disregard
empty rooms and occupancy detection requires time-out periods to compensate for
inaccuracies in detection. On the other hand, there is a discrepancy between meeting room
bookings and true usage due to last minute cancellations and over-estimated time needs.
Empty rooms that are free for use may then appear to be reserved when looked up through
the system.

A solution is to enable communication between the user, room reservation system and
building automation system. A system that does this has been developed by Padmanabh et
al. (2009). It monitors the state of meeting room with wireless sensors and compares it to the
actual bookings. The system consists of an application server, database server, SMS server
and a sensor cloud. Their system operates as a proxy between Microsoft Outlook booking
system and the user. To use the system, the user is required to give their phone number. If a
user is trying to book a room that is currently unavailable for the desired time, an option
becomes available to be put on a waiting list that can be room or zone specific. The system
keeps track of chances made in the bookings, and if a meeting is cancelled and there is
someone in the waiting list, they are alerted through SMS and e-mail to book the room. If
60 minutes have passed without any bookings from the person first in the waiting list, the
next person is given a chance. Additionally, if a room has been booked and five minutes
have passed since the meeting start time without any activity detected by sensors, the
meeting organizer receives a SMS telling them to cancel their booking. Once the booking is
cancelled, the waiting list gets serviced. The system also alerts the meeting organizer if they
have left lighting or air conditioning on after a meeting. (Padmanabh, et al., 2009).

64
Implemented with the case buildings the system would be slightly different from the one
presented above as observable in Figure 5.25. The application would require an interface
with building automation to gain access to sensor data, whereas the previously introduced
system had its own sensors. Moreover, if it is desired that the application can directly control
equipment, some security concerns may have to be tackled. Still, essentially the system
would be very similar to the one presented by Padmanabh et al. (2009), except that lecture
hall reservations are given as an input rather than booked through the system. It would
monitor the reservations of meeting rooms in Microsoft Outlook and also the reservations of
lecture halls. Based on this information supply air unit operation schedules and room
temperature set points are imposed with sufficient advance to the event starting time. If
occupancy sensors do not detect any activity within 5 minutes of starting time, the lecture or
meeting organizer is asked with a SMS if anything will take place at all in the room. If not,
the conditioning will be stopped. If the user answers yes or no answer is received at all,
conditioning of the room continues. The system can be notified about the premature end of
an event by a SMS, and if no movement is sensed for 15 minutes the organizer is asked
through a SMS if the event has ended. It is noteworthy that the system does not make a
decision to degrade the indoor environment’s quality unless a confirmation is made with a
SMS.

Figure 5.25 Architecture of the proposed system cluster.

The benefits of the application are reduced energy consumption, increased room utilization
and reduced manual input. Padmanabh et al. (2009) tested their application during 60 days
in one of the buildings of their company in India. The company has more than 100,000
employees and approximately 2,700 meeting rooms in total. They were able to elevate the
utilization of the meeting rooms from 67% to 90%. For their working hours this equals to
about 2 extra hours per day in a conference room. If the utility is extrapolated to their whole
stock of meeting rooms, the improved utilization corresponds to a utility of 583 new
conference rooms. With their alerting mechanisms on air conditioning or lighting being
accidentally left on they were able to save 13% of the electricity consumption.

The application communicating with reservation systems, Outlook and building automation
systems will be difficult to upkeep due to many integration interfaces. Problems may arise
every time any of these systems updates, thus incurring maintenance costs and downtime.
The situation when some of the systems is replaced is completely another issue. Another
drawbacks are that majority of the energy savings are achievable simply by occupancy-based
control of air supply and lighting, and the building automation system can produce reports
on room occupancy if the discrepancy between bookings and true usage is suspected to be

65
high. These undesired habits can then be mended with organizational policy. The feasibility
of this application however rises together with the number of meeting rooms and the demand
for them. The main technical challenges relate to enabling the communication between
software, and if the application is allowed to control any equipment, the security of the
whole. The integration would require a lot of coordination between experts of different
software to come up with a properly functioning entity. The readiness of software to integrate
with anything and the interests of the organization to co-operate can vary a lot adding to the
difficulties. Nonetheless, occupancy sensor data of the meeting rooms needs to be available
in the building automation system.

The evaluation results for this proposition are presented in Figure 5.26. Some respondents
in property management, building services engineering and building automation that have
been involved in software development projects saw the integration of a number of systems
as both too expensive and laborious when much of the benefits can be gained simply by
occupancy detection. The other respondents generally considered the challenge of
implementation to be lower for this proposition. Respondents in property maintenance and
building automation who evaluated the benefits high justified this by the present amount of
daily manual work that could be eliminated, while the others generally estimated the benefits
to be smaller.

Figure 5.26 Evaluation results of the initiative of utilizing room reservation data on equipment control
and managing the discrepancy between bookings and true occupancy.

5.4.8 Satisfying the dynamic demands of users with artificial

intelligence based control of equipment
Building users are not always able to adjust indoor conditions towards their liking which can
be a factor that leads to complaints. These complaints can then cause a large amount of work
for property managers who try to fix the situation, as every change in room conditions could
lead to new complaints from other users. If the users were given the ability to tell a computer
what conditions they want and where, and the computer could satisfy these demands to a
reasonable extent, this work would be reduced greatly.

Artificial intelligence (AI) can be a useful framework to tackle these kind of issues because
of capabilities in learning, reasoning and optimizing. If predictive capabilities are desired in
control, a possible approach is to create a decision unit that uses measurement data to analyze
situations and determine interventions to meet previously defined objectives. These
objectives can be anything that is measured or possible to derive from measurements, such

66
as room temperature, carbon dioxide content of air, deviations from target values,
illuminance and energy consumption. In other words, the decision unit could perform any
optimization tasks, even non-linear ones, if inputs and outputs are measured. The building
users could then tell the decision unit what they want, and the unit will control the equipment
accordingly within appropriate boundaries. The decision unit could basically be a
functionality of a room controller, however the heavy processing tasks such as machine
learning could be done on a separate server.

There are two essential functionalities of the decision unit: predictive capabilities and
efficient optimization. Examples of these are presented next. Predictive capabilities can be
gained with data mining models such as neural networks, whereas optimization can be done
with a genetic algorithm where fitness of an individual can be weighed to favor energy
efficiency or comfort. The data mining model would require extensive training data
including every possible control situation in every possible context to achieve a satisfactory
level of predictive accuracy, because the optimization technique will only be as good as the
predictive model due to its usage in the solution optimization. Including energy consumption
optimization in every control action by a weight that the user has approved and giving users
more control could result in energy savings and reduced complaints. Also the less time is
used in dealing with complaints the more resources are available for other tasks.

Still, these positive aspects also embody a negative side, and there are many downsides in
the implementation of AI based control. First of all, gathering the training data can be a very
long task especially with systems that have long delays between inputs and outputs such as
radiators. Also the number of systems adds to the training time as every state of system
operation in every context needs to be trained. Historical data from normal operation is
unlikely to be suitable for training, especially if the control targets have always been the
same. It is virtually impossible to completely train a predictive model, which leads to the
problem of unreliable predictions in situations that have not been experienced before. On the
other hand, giving users control over the room’s conditions may cause larger problems with
the thermal functionality of the area as a whole. Having separate areas of a large room trying
to achieve completely different thermal conditions leads to a high energy consumption. Also
the amount of statistical intelligence in control makes it extremely difficult to diagnose the
reason for upcoming problems, and an external consultant is likely to be needed every time
the system is not working properly. Perhaps the largest problem of all is that once any of the
building systems controlled by AI is replaced, the model needs to be trained again.

A predictive analytics model will require all relevant contextual information, control signals
and outcomes as direct measurements or derived values. For example in OfficeOne such
control would require electronic thermostat valves in heating and as much recorded data as
possible. The required data depends on the application, but could include occupancy
detection, radiator valve positions, chilled beam valve positions, lighting output, available
daylight and illuminance at every workplace. As described in Chapter 5.4.5 the act of
integrating lighting control in building automation is a challenge itself. Also to tackle the
problem of bulk energy consumption measurements devices would need to be run separately
in the training phase, or this information would need to be derived based on some other data.

The evaluation results for this proposition are presented in Figure 5.27. Respondents with
building automation background considered the challenge of implementation of AI based
control to be high with varying benefits to be expected. The weight of expected benefits was

67
below medium, indicating a slight orientation towards a lemon proposition. This can largely
be attributed to the reduced manageability of systems: Among these respondents there was
a desire to be able to know exactly what each actuator is doing and why. Two respondents
with building services engineering background expected large benefits with challenging
implementation. The reasoning behind the benefits was the potential ability to save energy
whenever users permit it, but accomplishing this would need significant tuning, dividing
space appropriately for control and sufficient resources for the unit’s upkeep. The outlier of
small benefits and challenge considered AI more appropriate in smaller nonlinear
optimization tasks, such as choosing an ideal starting time for heating setback recovery.
Respondents in property management expected medium benefits with medium to high
challenge, while respondents in property maintenance expected both medium benefits and
challenge. There was no discussion regarding these evaluations.

Figure 5.27 Evaluation results of the initiative of using artificial intelligence in equipment control.

5.4.9 Utilizing context-aware alarm thresholds in fault detection

In the case study it was observed that most of the alarm events on HVAC processes are based
on a single value threshold. With single value thresholds there is a trade off with early
detection of faults and false alarms: If thresholds are very strict, faults can be detected early
with the cost of many false alarms and if the thresholds are not strict, faults may not be
detected before the system stops working completely. Some reduction in false alarms could
be achieved if the fault detection tool could learn to distinguish between normal operation
and a fault.

Such functionality can be realized by statistical models that group sensor measurements
automatically into different situations. This idea is visualized with self-organizing maps in
Figure 5.28 where each cell in a field corresponds to the same situation between fields. For
example the top right corner of every field is the same situation, and the usual parameter
value is shown on the bar to the right of every field. The statistical model has information
on the typical value and the usual deviation range of every parameter for every situation.
Therefore the distance from a situation that is considered normal is not measured in one
dimension as it is done currently, but in as many dimensions as there are measurements.
With a group of measurements as an input, the outputs of for example Microsoft Clustering
are the nearest cluster, distance from the nearest cluster, probability that the measurements
belong to a cluster or that these measurements are included in the model at all. These outputs
can be used to detect faults after normal operation has been shown to the model. A clustering
model was used to detect a manually inserted heat recovery malfunction in Appendix 3.

68
With more accurate fault detection the upcoming alarm events would be taken more
seriously, as maintenance personnel would experience that by default an alarm is likely to
require action (although this attitude can be reinforced by prioritizing fault events too). This
can incur major benefits to the maintenance organization as every fault has implications on
user inconvenience, maintenance costs and energy consumption if not detected in time. The
statistical models could also provide alarm-specific feedback on the usual energy
consumption of that situation versus normal operation to encourage early involvement. The
model cannot produce anything that is not measured, however.

The drawbacks of statistical models are that they need extensive and reliable training data,
or at least supervised learning where an expert specifies normal operation to the model while
it is observing the measurements. Regardless of the means of the training, there is need for
expertise in statistical tools and building systems during the tuning process of the model,
which is costly. Also no matter how well the training is done, eventually the model will face
a situation that has not been trained, and then its performance will be poor. Therefore if some
accuracy in fault detection is gained, it comes with the price of dependence on consultants
that help in tuning the model.

Figure 5.28 Self-organizing map of air handling unit sensor data.

The evaluation results for this proposition are presented in Figure 5.29. Respondents with
building automation background mainly regarded the challenge of implementation of
context-aware alarm thresholds to be medium and the benefits to be small, with a slight
overall orientation towards a lemon proposition again. It was mostly difficult to see what
there was to gain by eliminating some false alarms, and the tuning process of the tool was
considered laborious. It was suspected that the tool would not be able to capture all the states
of normal operation, which could lead to even more false alarms than currently. The
evaluations of respondents with building services engineering and property management
background diverged without much discussion on the reasons for the grades. They would

69
not have to deal with alarm events in their work. Respondents in property maintenance,
however, considered the ability to differentiate between regular abnormal situations and true
faults desirable. They regarded the challenge of implementation from medium to high.

Figure 5.29 Evaluation results of the initiative of utilizing context-aware alarm thresholds.

5.4.10 Introducing mobile applications to collect sensing data on

the indoor environment from building users
There is currently a large amount of inefficiency within the process of submitting user
feedback that is consequently used as basis for BAS interventions. User complaints are
currently received through e-mail. The messages do not always contain accurate information
regarding the location of the problem which leads to more messages and delays with the
problem handling, nor does it on the amount of satisfied people in the same room.
Additionally, e-mail inboxes of different people in the maintenance organization are not the
best place to store user feedback, as the feedback is not visible to anyone else and it may be
difficult to look up old feedback. On the other hand, user satisfaction metrics are hard to
derive from separate messages.

Today the majority of people carry smart phones, which are a convenient platform for
collecting feedback. There are hardware and applications commercially available that enable
gathering location-specific user input from mobile phones to a server in the cloud. For
example Bluetooth low energy proximity transmitters broadcast a device-specific ID which
may correspond to a particular location in the building by a definition in the server, and this
identification by the server will trigger a push-notification that invites the user to give
feedback through an application in the phone. An example of a survey created with the
Louhin platform of DataRangers (2016b) that supports those functionalities is shown below
in Figure 5.30. The process of giving feedback should be fast and easy, yet produce all the
necessary information and allow users to give text-form feedback if necessary. If the user is
satisfied, ticking yes and submitting ends the survey, but if something is wrong then the
survey expands just enough to drill through to the exact problem. After the survey has ended,
the user could be informed on the current satisfaction rate and room condition measurements
from that location.

The feedback data could also be used in maintenance processes. By importing this location-
specific feedback to the building automation control software and visualizing it for example
in a floor plan or room view the maintenance personnel can observe both the room conditions

70
and user feedback at the same time. With time stamps of feedback events it is possible to
compare conditions to the trend in user satisfaction to diagnose problems. After set points
have been adjusted, the submitter of the feedback can be notified that action has been taken
and another survey can be sent later to see if the situation has improved. On the other hand,
this feedback module should enable forwarding messages with location information to
maintenance-men should the problem require a visit. Anyhow, it is vital to notify the sender
of the feedback on interventions to ensure their commitment on giving feedback.

Figure 5.30 An example of an occupant survey created with Louhin platform.

This method of collecting user feedback prevents the system from becoming a channel for
complaining only, as users are invited to give feedback according to preprogramming.
Therefore the feedback should represent the room conditions better than in the situation
when feedback is given solely because something is wrong. The programming can also
include smart functions such as verifying surveys when abnormal feedback is given.

The benefits of collecting sensing data from users regularly include reduction of waste in
user complaint handling, more comprehensive data on user satisfaction and charting effects
of maintenance interventions. Such benefits are hard to estimate in currency, however by
attaining a better understanding on user satisfaction some energy could be saved by avoiding
unneccessary interventions and reducing excess output. For example the philosophy of the
WarmEnough concept developed by Granlund and Nomenal is to increase comfort and
productivity by optimizing heating based on collected location-specific feedback data
(Dooley, 2016). The results could then occasionally indicate an opportunity to reduce
heating output. A screenshot of the WarmEnough application is presented in Figure 5.31.

71
 Figure 5.31 A snapshot of the location-specific temperature feedback view (Dooley, 2016).

The challenges and drawbacks associated with this proposition are the fluctuating interests
of building users to give feedback, battery replacements of bluetooth transmitters and
integrity of the feedback data. Especially keeping up the interest of building users seems to
be difficult: In an interview it was mentioned that a feedback system was piloted in the past,
and within a span of a few months the users stopped giving feedback completely. Therefore
it is imperative to ensure the commitment of users by fortifying their perception on the
feedback leading to action whenever there are reasonable grounds for that. Complementarily
by making the process of giving feedback more fun, some users might become more eager
to participate. For example Ken Dooley (2016) in Granlund is experimenting with PlaySign
how user experience can be improved with playable surveys. A snapshot of such a survey is
presented in Figure 5.32. There are no results published from this experimentation.

Figure 5.32 A snapshot of a playable user feedback survey (Dooley, 2016).

If bluetooth or any other transmitters are used in the infrastructure, battery replacements are
required every few years. Conversely, the sensation of thermal comfort can be affected by a
number of psychological factors in addition to the actual room temperature, which is why
the integrity of feedback data should be ensured by verification procedures when abnormal
input is received. For example when the temperature is 22°C and a user is feeling very cold,
the feedback should be verified by sending the survey to other users in the room or the next
user entering the room.

When feedback data is collected to a server of a separate service provider from the building
automation contractor, a software interface is required to enable a feedback module within
the controller software. As discussed earlier, such integration requires upkeep in the case of
software updates and replacements. Another option is to use a completely separate feedback
platform, which imports room condition data from separate sensors or building automation.

The evaluation results for this proposition are presented in Figure 5.33. The majority of
respondents with building automation background considered the proposal to be a low
hanging fruit as the necessary technology is available and the produced data could prove

72
valuable whenever adjusting set points. Respondents with building services engineering
background estimated the challenges of implementation to be medium with mainly high
gains to be expected, essentially for the same reasons as above. Respondents in property
management thought the benefits to be above medium with diverging opinions on the
associated difficulties. There was varying optimism regarding the interests of users to
participate in the surveys, but the produced data on user satisfaction was considered valuable.
Respondents in property maintenance evaluated the gains to be high, but also the challenges
of implementation. After all the suggested transmitter infrastructure needs maintenance, and
the users must be committed to giving feedback. The awkwardness of dealing with user
complaints was highlighted particularly by property maintenance staff: It is not uncommon
that a person feeling cold has someone sitting next to them that is feeling hot. The situation
cannot therefore be resolved successfully if both are complaining. Not adjusting set points
due to a single complaint could be better justified if everyone else was happy with the room
conditions.

Figure 5.33 Evaluation results of the initiative of collecting sensing data from building users.

5.4.11 Optimizing pressure differences with ventilation

Current saving activities in ventilation may induce adverse effects on IAQ. Air supply is
often minimized for the nights for saving purposes, but some exhaust fans for example in
stairways, kitchens and toilets keep running. These actions have effects on the pressure
differences of the building that are hard to evaluate without separate studies. In Finland the
indoor air of the building should be in a slight vacuum of 0-10Pa in comparison to the
outdoor air (Seppänen, 2010, p. 18). Slight vacuum is considered best, as positive pressure
would cause long term moisture stress on the building envelope and high vacuum would risk
the indoor air quality through suction of impurities from the holes and cracks of the
structures. The vacuum should therefore not exceed 30Pa (Finnish Ministry of the
Environment, 2012a, p. 19). Field study results on pressure differences can be distorted in
many ways, as the pressure difference is influenced by ventilation system, wind, fluctuating
outdoor temperatures and user activities of for example opening doors and windows.
Therefore conclusions regarding the pressure differences of a building should be based on
long term monitoring. (Katainen & Vähämaa, 2015).

The solution is to install sensors measuring the pressure difference between indoor air and
outdoor air. As pressure differences not only depend on air supplied and exhausted but also
on wind speed and temperature differences, some refined information should be supplied in
order to detect potentially harmful time periods from the oscillating data. The refined data

73
could simply be hourly averages on pressure differences, wind speed and temperature
difference between outdoor and indoor air. The visual representation of pressure difference
data should include schematics of areas of different air handling unit.

By optimizing ventilation system operation schedules the energy consumption could rise to
compensate for the exhaust fans running around the clock, but in proportion the risks on
indoor air quality can be better managed. The drawbacks are the large amount of sensor
installation work and the necessity of expert resources to manage the ventilation system
effectively based on the produced information. It is noteworthy that the air-tightness of the
building must be sufficient, and the ventilation system must allow for space-specific purging
if the measurement data was to be utilized. In other cases the measurement data could only
indicate underlying reasons of problems without much means to resolve them. Therefore the
issue should be considered already in the design phase of a construction project, for example
by allocating separate air handling units for the toilets, kitchens and stairways.

The evaluation results for this proposition are presented in Figure 5.34. This proposal was
included in the initiative portfolio later on by the project steering group, so the number of
respondents is smaller (N=9). There was a single evaluation by a person with building
automation background. He estimated the benefits to be high and the challenge to be small,
because the provided information might help in indoor air quality problem diagnosis simply
with sensor data. The two respondents with building services engineering background
considered the produced data beneficial, but the sometimes poor adjustability of the
ventilation system could mean that the identified problem would not be possible to resolve
effectively. The three respondents with property management background estimated high
benefits for the same reason as above with diverging estimated challenge. The required
amount of sensors and drilled holes was a concern. The three respondents with property
maintenance background considered both the benefits and challenge to be high for the same
reason.

Figure 5.34 Evaluation results of the initiative of collecting long-term sensor data on pressure
differences over building envelope.

5.4.12 Summary of the initiatives

The initiatives to close the energy performance gaps are summarized in Table 5.7. In the last
column the distribution of evaluation result categories among the stakeholder groups is
shown. If an evaluation result happened to be exactly at the boundary of a category, then
half a stakeholder group was appointed to both categories, respectively.

74
Table 5.7 Summary of the initiatives.
Initiative Benefits Challenges Category
1. Installing • Better area-specific • Installation costs • Strategic
electronic radiator adjustability of heating • Power supply development
thermostats with project (2.5/4)
smart heating • Low hanging
functions fruit (1.5/4)
2. Adaptive outdoor • Automatic adjustments of • Choosing the appropriate • Low hanging
temperature heating power to meet room temperature as the fruit (3/4)
compensation for seasonal needs adaptive variable • Strategic
heating • Requires supervision to development
ensure proper functioning project (1/4)
3. Room setback • Allocating heating power • Lack of excess heating • Strategic
temperatures to better meet the cyclical power during cold periods development
outside office hours demand • Uneven reheating with project (2/4)
unbalanced radiator • Low hanging
networks fruit (1/4)
• Elevated energy • Lemon (1/4)
consumption during
reheating
• Risking thermal comfort
4. Optimizing • Predictive control of • Difficult problem diagnosis • Low hanging
heating with heating • Cost of installation with fruit (2/4)
weather forecast electronic thermostats • Strategic
data • Risking thermal comfort development
project (2/4)
5. Intelligent lighting • Daylight harvesting • Cost of installation • Strategic
and blind control • Protection from heat loads • Risking visual comfort development
with daylight • Power supply project (3.5/4)
harvesting • Low hanging
fruit (0.5/4)
6. Improving the • Ability to control actuators • Actuator control requires • Low hanging
manageability of in groups that are based commitment of BAS fruit (2.5/4)
large amounts of on their attributes software providers • Strategic
sensor data and • Ability to create key • Database development development
actuators figures with references to project for sensor data project (1.5/4)
attributes • Database maintenance
7. Enabling • Automatic equipment • Many interfaces between • Strategic
communication operation scheduling software development
between building based on room bookings project (2/4)
automation and • Room booking • Lemon (2/4)
room reservation management to improve
systems to reduce utilization
energy consumption
and manual input
8. Satisfying the • AI satisfies users’ demands • Acquiring training data • Strategic
dynamic demands of while optimizing energy • Need of new training after development
users with artificial consumption replacements project (2/4)
intelligence based • Need of better adjustability • Low hanging
control of of heating fruit (1/4)
equipment • Difficulties in problem • Lemon (1/4)
diagnosis

75
 Initiative Benefits Challenges Category
9. Utilizing context- • Alarm thresholds adjust • Need of training data • Strategic
aware alarm for each situation • Need of supervision and development
thresholds in fault tuning project (2.5/4)
detection • Lemon (1.5/4)
10. Introducing • Comprehensive • Committing users to give • Low hanging
mobile applications knowledge on the relative feedback fruit (2.5/4)
to collect sensing numbers of satisfied users • Costs of procuring and/or • Strategic
data on the indoor to unsatisfied ones licensing development
environment from • Feedback data available • Battery replacements for project (1.5/4)
building users for anyone in the transmitters
maintenance organization
to review
11. Optimizing • Comprehensive • Cost of installation • Strategic
pressure differences knowledge on the long- • Required expertise to act development
with ventilation term behavior of pressure on the produced data project (3/4)
differences over building • The ventilation system may • Low hanging
envelope to help diagnose not enable energy efficient fruit (1/4)
potential IAQ problems resolving of identified
issues

5.5 Feasibility study of selected improvements

5.5.1 Selection rationale
The project steering group familiarized with the evaluation results and selected three of the
most promising and different initiatives for further evaluation. This means that information
on costs of implementation and potential monetary benefits were sought. Moreover, the ends
were emphasized over the means according to performance philosophy, so some of the
feasibility studies included different options to find the most cost efficient way to close
performance gaps.

The selection process was based partly on the evaluation results and partly on the steering
group’s judgment. The initiatives were scored by weighing the benefits by 70% and
challenges by 30% according to Expression 4, after which they were ranked. Influence was
weighed over challenge in the scoring due to larger uncertainty in the latter dimension. Then
the highest scoring and most heterogeneous initiatives were chosen for further study. The
ranking is shown in Table 5.8. The ranks are numbered from I to XI with the top three
highlighted green, bottom three red and the rest yellow, respectively.

Score = 0.7 × Influence + (1 - 0.7) × (10 - Challenge) (4)

76
Table 5.8 Score-based ranking of proposals by respondent background.
Proposal/Background Building Building Property Property
automation services manage maintena All
engineering ment nce
2. Adaptive outdoor
temperature
III VI IV VII I
compensation for
heating
10. Collecting sensing
II V III IX II
data from users
6. Improving the
manageability of
large amounts of IV IV I V III
sensor data and
actuators
11. Optimizing
I II V IV IV
pressure differences
1. Electronic radiator
VI III II III V
thermostats
4. Utilizing weather
forecast data in VIII I VII VIII VI
heating control
5. Daylight harvesting
with blinds and VII X VIII I VII
lighting control
3. Room setback
V VII X X VIII
temperatures
7. Communication
between room
IX XI XI II IX
reservation systems
and BAS
9. Context-aware
X IX IX VI X
alarm thresholds
8. Artificial
intelligence based XI VIII VI XI XI
control of equipment

As a result, 2. Adaptive outdoor temperature compensation for heating, 6. Improving the

manageability of large amounts of sensor data and actuators and 9. Collecting sensing data
from users were chosen for further study. 9. Collecting sensing data from users was
reformulated into measuring user satisfaction to better capture the identified value driver.
Moreover, for the same reason the proposition 6. Improving the manageability of large
amounts of sensor data and actuators was split in two: Group control of actuators and
energy efficiency monitoring.

77
5.5.2 Measuring user satisfaction
The project steering group considered the main benefit of collecting sensing data from users
with mobile applications to be the information on user satisfaction, which is an intrinsic
value as it fosters the longevity of tenancy agreements. Therefore in addition to the solution
described in Chapter 5.4.10 some services that provide user satisfaction measurements and
information on indoor air quality are studied.

Two courses for implementation were identified: One is to simply purchase everything as a
service, while the other is to design the infrastructure and user interfaces yourself and acquire
a license for data collection and handling. The latter option leaves more freedom in
generating functions that support maintenance operations, however. Another benefit, if user
dissatisfaction could be reduced with the designed functions, would be the decrease in the
workload related to handling complaints. A person in the case organization estimated that
complaints from a very problematic building cause 2-3 hours of work every week, which
corresponds to roughly 7,800€ of resources annually (salary costs of 60€/h). Next the costs
of options are considered: Providers A and B are service providers, whereas Provider C is a
license and consultation provider.

The service provided by Provider A is piloted in one of the buildings of the case organization.
The piloted service includes indoor air quality measurements visualized to users on screens:
Relative humidity of air, temperature, total volatile organic compound content of air and
CO2 content of air. There was an option to include particulate matter (PM1, PM2.5 and PM10)
and noise sensors for an additional fee. The tablets also let users to submit feedback. All the
measurements and feedback results can then be reviewed on a web page. The pricing is based
on fixed annual costs for sensors, gateways, tablets and routers with installation included,
while the license fee is surface area based. The total cost of the piloted service is 11.4€/m2
annually with an area of 450m2. Particulate matter and noise measurements would have
added 4.7€/m2 each. For an office building with a room area of 5,000m2 the service fee
would then be 57,000€ annually.

Provider B offers a service that measures user satisfaction. Feedback is given with a browser
(computer or mobile phone) and with separate button pairs that can measure for example
overall satisfaction, thermal comfort or indoor air quality. The feedback is not location-
specific within the building unless the submitter provides it in text-form. The running costs
are based on the amount of users and buttons. The running costs would be 2,520€ annually
with a service allowing over 300 users and including seven button pairs per building. The
initial costs are 400€ per measured entity (for example a building). The feedback results can
be reviewed on a webpage.

Provider C offers a cloud environment to conduct mobile surveys and report the results. The
environment supports a low energy Bluetooth transmitter infrastructure and has capabilities
to easily export data to, and import data from other servers. This option seems fit if sensing
data is to be utilized in controller software and maintenance operations. Based on an estimate
given by the service provider, the fixed costs of the cloud environment are approximately
10,000€ annually, while the one-time consultation costs for a first-time setup are estimated
to be another 10,000€. The required in-house resources for setup would then be 6,000€ per
building (derived from the consultation costs above with 60€/h). Based on a case building
floor plan the transmitters would cost 0.22€/m2 (around 20€ each), while the recurring
battery replacement costs every three years would be 0.03€/m2 annually (1 hour per floor

78
every three years à 40€/h). So for a single 5,000m2 office building the costs would be
27 100€ first year and 10 150€ annually thenceforth. A comparison of the costs between
suppliers is summarized in Table 5.9 over five buildings and a five year period.

Table 5.9 Benefits and costs of user satisfaction measurement systems

Costs
Present
Description First Following
value
year years
Provider A
• Information on user satisfaction 228,000€ 228,000€ -987,240€
• Potential to reduce complaints by visualizing
room conditions
Provider B
• Information on user satisfaction 14,600€ 12,600€ -56,462€
Provider C
• Information on user satisfaction 56,250€ 10,750€ -89,864€
• Potential to reduce complaints by informing on
room conditions
• Potential to utilize the system in maintenance
operations

The options over five 5,000m2 buildings and a five year period. The interest rate for present value estimation
was chosen to be 5%. The resultant discount factor for periodical payments is 4.33.

It seems that the most economical option is to use the service provided by Provider B, if data
on user satisfaction is the only thing that is desired. Also due to the lowest initial costs, it
may be a good idea to determine the interest of users to give feedback with the service of
Provider B. However, if two-way feedback and area-specific satisfaction data are going to
be required, then those features can be included with an approximated additional 5-year
contribution of 30,000€.

5.5.3 Group control of actuators

The benefits of group control are largely attributable to facilitating routine processes. On
one hand time is saved and on the other hand energy is saved because some adjustments
cannot simply be made currently quickly enough due to the large amount of manual work
involved. It is however not realistic to assume that a building automation controller would
support attribute-based group control introduced in Chapter 5.4.6 without sufficient demand
in the market, so the feasibility of group control is studied by examples of predefined control
groups. Two examples are considered as sources of benefits: AHU time schedule
programming and snow melting control.

Facilitating AHU time schedule programming is a simple example of saving time. Let us
consider an upcoming holiday period in the university, and to save energy the air handling
units are programmed to operate on half power and minimum air flow except for the utility
services rooms. As starting information let us assume 30 buildings, eight AHUs per building
of which two serve spaces with fixed ventilation demand and five minutes of programming
per AHU (includes looking up the closing period and occasionally the AHU operating area;
the time estimate is based on observing this work in the control room briefly). In the first
scenario every AHU is programmed separately which leads to 15 hours of work. This work

79
is to be done before and after the holiday period, and occasionally during the holiday should
some unexpected use of the building occur. Still, let us assume 30 hours of work in total. In
the second scenario all the AHUs in a building can be controlled in groups to holiday mode
and back with a single click according to an interface presented in Figure 5.35. In the same
interface it is possible to edit the holiday period time schedules and check the operating
areas. This would essentially save all the 30 hours work once the settings have been defined.
With a salary of 40€/h that equals to 1,200€ annually. A contractor that was asked about the
costs of developing such a feature told that the idea could be implemented in a different way
to the one presented here (changing time schedules), and a cost estimate cannot be provided
without a thorough study.

Figure 5.35 AHU group control interface.

The other example is the control of electric snow melting in gutters and spouts, which has
long been a concern for the case organization. The purpose of snow melting cables is to keep
the gutters and water spouts clear of ice and snow in metal-sheeted gable roofs. The ideal
situation would be that the heaters are only on when there is snow on the roof and the snow
is melting or it is raining. The heaters also require a time-out period to let the rainwater
system dry. On some of the roofs there are sensors detecting snow, rain and humidity, but
problems may arise when snow is melting in other parts of the roof undetected by the sensors.
If the gutters get clogged by ice, the water that has melted starts to dam up, and the risk of
moisture damage grows. Another risk is that the water starts to overflow the gutters instead
of flowing through the water spouts, which easily results in icicles that need to be cleared in
the CBD area. Also because AHUs are often located on the top floor there is a significant
amount of heat radiating from underneath the roof. With a thick layer of snow on the roof
and the lowest layer melting, the pressure can cause the water to leak through the metal sheet
seams. Based on the experience in the case organization, even the smallest water leakage
costs 5,000-10,000€ to treat while a larger water damage costs 10,000-100,000€ to repair.
Therefore the reliability of the controlling method is extremely important.

80
However, on the vast majority of the roofs the heaters are controlled by the outdoor
temperature so that heating is on when the temperature is between -5°C and +5°C. Subzero
temperatures have been included because snow can melt with temperatures below zero when
it is sunny or with buildings that have a lot of heat leakage through the roof. Obviously
purely outdoor temperature based control is not optimal, and next the wasted electricity is
being estimated.

Between 1st September 2015 and 1st September 2016 the heating criteria was satisfied for
2964 hours according to measurement data collected by an outdoor temperature sensors in
Helsinki CBD. This corresponds to 34% of the whole year. Based on data collected from
technical drawings in Table 5.10 snow melting power is roughly 10kW per 1,000m2 flat
projected roof area. It is important to note that in some buildings the heating cables are self-
adjusting the heating power, which reduces the electricity consumption when there is no
need for full output. Still, the nominal output is used in the following calculations.

Table 5.10 Snow melting power and projected roof areas for four buildings in Helsinki CBD.
Building Snow melting power [kW] Projected roof area [m2] Ratio [W/m2]
A 9.6 1035 9.3
B 9.2 946 9.7
C 5.6 434 12.9
D 5.6 540 10.3

Let us consider an imaginary roof with 700m2 projected roof area and 7kW of snow melting
power. If the heaters are on with full power for 2964 hours, 20.7MWh of electricity is
consumed. This equals to a bill of 1743€ annually with the price of electricity of 84€/MWh.
If that roof area is the average among the 30 buildings that are remotely managed by the case
organization in Helsinki CBD, the total bill is then 52,285€.

The savings potential can be estimated in two stages: Firstly by eliminating the days without
snow and rain, and secondly by eliminating the days when there is snow or it rains, but
melting is not needed. The estimates are based on an outdoor temperature sensor in Helsinki
CBD and FMI observation data on daily snow depth, cumulative 24 hour precipitation, daily
minimum temperature and daily maximum temperature between 1st September 2015 and 1st
September 2016. This weather data can be downloaded for Helsinki CBD once an API-key
is provided 1.

The idea is to compartmentalize the runtime of the snow melting heaters into different states
that are either waste or not, and have durations based on the weather and sensor data. This
concept is illustrated in Figure 5.36. In the figure observation data on daily snow depth,
cumulative 24 hour precipitation, daily minimum temperature and daily maximum
temperature are used to compartmentalize the snow melting runtime (determined with an
outdoor temperature sensor) into states with durations. These states are ultimately labeled
into waste of electricity or not waste of electricity.

1
http://data.fmi.fi/fmi-apikey/<APIkey>/wfs?request=getFeature&storedquery_id=fmi::observations
::weather::daily::timevaluepair&fmisid=100971&parameters=snow,rrday,tmin,tmax&starttime=2015-09-
01T00:00:00Z&endtime=2016-09-01T00:00:00Z

81
 Figure 5.36 Snow melting waste estimation method.

In the first stage the duration is calculated when temperature is between -5°C and +5°C but
there is no snow and it has not rained. According to Figure 5.36 this duration is 842 hours.
It was suggested by the technical managers that there should be one switch for all of the
snow melting heaters within the same suburb, so that the heaters could be turned off
whenever there is clearly no need for heating. When the switch is turned back on, the
building specific controller operation is resumed. By eliminating the duration without snow
or rain in this way, 5.89MWh of electricity could be saved per building (assuming 7kW
output). A contractor that was asked about the costs of creating such a switch told that it is
quite easy and does not require much work as long as the heaters are connected to BAS,
however no cost estimate was provided.

The second stage is to eliminate the time periods when there is snow, it has not rained and
the temperatures are well below zero (but above -5°C) so that snow is unlikely to melt (in
total 49 hours) and when there is no snow, it has rained but the temperatures are not freezing
(in total 769 hours). The total duration meeting this criteria is then 818 hours according to
Figure 5.36. This corresponds to additional savings of 5.73MWh per building. To eliminate
this runtime, one must rely on automation, and a number of sensors would need to be
installed on the roofs with different orientations. The cost of procuring a controller and the
sensors (snow, temperature and humidity) for a building with roof area of 700m2 is estimated
to be 5,000€ based on a previous contract. The estimate is subject to uncertainty as it is based
on a lump-sum from memory from which the portion of heating cables has been removed
based on the cost of another contract. Yet, the cost is not far-fetched as a controller costs
1,000€ (Taloon, 2016a), four temperature and humidity sensors 800€ (Taloon, 2016b), and
a power supply 170€ (Talotuote, 2016), adding up to 1,970€ without installation costs. The
prices are not wholesale prices, though.

82
Another option is to install self-adjusting cables that halve the heating output whenever the
surroundings are dry or rainwater is unlikely to freeze (approximated as 842 hours plus 769
hours plus 49 hours). The costs related to this option per building are estimated below:
• 200m of self-adjusting cable à 8€/m (Finnparttia, 2016), and
• cost of installation of 1,650€ (based on a historical contract).

The savings potential of different actions is then compiled in Table 5.11 assuming 30
buildings with average roof area of 700m2, heating power of 7kW and price of electricity of
84€/MWh.

Table 5.11 Compiled snow melting savings potential versus costs for 30 buildings.
Savings
Description Cost
Energy Money Percent
Option 1
Creating a single switch for all the
176.8MWh 14,853€ 28.4% -
snow melting cables within a suburb
Option 2
Automating snow melting
348.6MWh 29,282€ 56.0% 150,000€
successfully
Option 3
Installing self-adjusting melting
174.3MWh 14,641€ 28.0% 97,500€
cables

Group control is clearly a feasible tool to take into use immediately, as large benefits are
possible to reap with a minimal cost. Moreover, the savings potential of advanced snow
melting controllers should be verified with appropriate consumption metering. If the savings
estimate above can be validated, it would be sensible to install such controllers at least
whenever replacing the old snow melting systems.

5.5.4 Energy efficiency monitoring

The main value driver in improving the manageability of sensor data is the ability to observe
large amounts of data of interest across different buildings in a single view. This data of
interest can be specific consumptions of properties, specific powers of equipment or other
key figures such as heat recovery efficiencies. This value could be captured by a software
and database development project that covers the principles presented in Chapter 5.4.6 and
Appendix 2, but the costs of such a project were not possible to estimate accurately within
the limits of this thesis. The costs depend highly on naming policies of control points
throughout the building automation systems, and the capabilities of the systems to export
and import data, though. Instead two service providers who offer web-based energy
management tools are presented as options. The options are compiled in Table 5.12.

Provider A offers a web-based energy management tool that is used globally. Assuming that
all the measurements they need are specified in a .csv file at the customer side, the pricing
is as follows: 3€ a month per measurement and a 2,000€ initial cost per building for basic
reporting. If 50 resource consumption and key performance indicator measurements were
included, the total annual cost would then be 3,800€ for the first year and 1,800€ the
following years. If a lot more measurements were included, the price per measurement would
start to decrease towards 2€ a month per measurement, while the initial costs would rise to
accommodate for the more in-depth reporting.

83
Provider B offers a web-based performance monitoring tool that converts sensor data,
control signals and alarms into performance metrics. The performance of a building is then
visualized on various dashboards. The initial costs are 2,200€ and annual costs 2,400€ when
procuring the tool for an average office building with around 10 AHUs and 150 room
measurements. If the BAS is a Visonik, Desigo or Deos the initial costs may elevate slightly.

Table 5.12 Options for an energy management tool.

Costs
Description
First year Following years
Provider A
• Visualizations and reports are tailored to meet user’s 3,800€ per 1,800€ per
needs building building
Provider B
• Sensor data is converted into performance metrics
2,200€ per 2,400€ per
• The performance of a building is visualized on various
building building
dashboards

Due to the amount of ongoing costs it may be more feasible to develop one’s own system if
all the buildings were monitored. However, if the biggest consumers of the building stock
were the only ones monitored, it could be more sensible to purchase the system as a service
instead of developing one’s own.

5.5.5 Adaptive outdoor temperature compensation for heating

The energy performance gap being closed by adaptive outdoor temperature compensation is
automating the changes in heating power to meet the demands of different seasons while
maintaining thermal comfort. It was not possible to test such a program on the case building,
so the potential benefits from adaptive outdoor temperature compensation are estimated
based on simulated situations of excessive heat content in radiator supply water. In other
words, radiator supply water temperature is raised temporarily and the implications on
energy consumption are studied. It is important to note that the capability of the presented
program to close this gap has not been verified, so the following is estimating the gap itself.

There is an option in OfficeOne to use 5°C shifts in radiator supply water temperature during
cold periods. These shifts were then made for testing purposes during weekends of the mild
period when nobody was present in the building (to avoid inconvenience), and the energy
consumption was compared to a normal situation. Unfortunately this thesis took place
mainly outside the heating season, so it was not possible to gather extensive test data for
analysis. Some preliminary results could be obtained, however.

The shift periods were 26th August 15:00 – 28th August 16:30, 2nd September 15:00 – 4th
September 15:00 and 9th September 15:00 – 11th September 15:00 (year 2016) while the
complete data set was recorded from 1st February 00:00 to 11th September 23:59 (year 2016).
Because of bulk energy consumption measurements and around the clock running AHUs it
was impossible to differentiate between air heater and radiator energy consumption, so the
estimates were based on thermodynamic calculations. First, the specific heat capacity of
water is converted to watt-hours per kilogram degrees Celsius according to Expression 5.
Then the water flow through the heat exchanger is estimated based on average hourly valve

84
position and the design flow of the exchanger (0.49dm3/s) according to Expression 6. This
number is subject to uncertainty due to hysteresis. Finally, radiator heating power is
calculated as the heat divided by the time interval of one hour according to Expression 7.
That number is also subject to uncertainty due to losses of heat exchange, mixing outbound
district heating water and the fact that district heating water is used in preheating service
water after heating the radiator water. The cooling of district heating water was calculated
based on hourly averages of temperatures, with the outbound district heating water
temperature being a mix of the return district water temperatures of all the four heat
exchangers. Additionally, the portion that was used to preheat service water was impossible
to estimate because the temperature of district heating water was not measured after the
radiator network heat exchanger. However, this uncertainty slightly loses importance
because of the fact that it is the difference in heating power that is being analyzed.

kJ kWh
c = 4.19 = 4.19 × 0.000278
kg × °C kg × °C (5)

where
c = specific heat capacity of water (kWh/kg°C)

m ω kg
≈ × 1764 (6)
Δt 100% h
where
m
Δt
= water flow through the heat exchanger (kg/h)
ω = heat exchanger valve position (%)

m
P≈c× × ΔT (7)
Δt
where
P = total heating power of the radiator network (kW)
ΔT = district heating water cooling (°C)

The resulting data on radiator heating power in a normal situation and in a shift situation is
presented in Figure 5.37. The data suggests that heating power is higher in a shift situation
with outdoor temperatures ranging between 17-21°C.

85
 Figure 5.37 Radiator heating power with normal and elevated supply water temperature.

At the design temperature of -26°C the trend line of the normal situation would suggest a
heating power of 210kW, while technical drawings state a design power of 144kW. Thus,
the validity of these calculations is very unclear before data near design temperatures have
been acquired. Yet, some preliminary results can be obtained for outdoor temperatures
between 17-20°C. The average difference of radiator heating power with normal and
elevated supply water temperature is 7.67kW while the duration within that temperature
interval is 1360 hours, respectively. This indicates an annual energy performance gap of
10.4MWh, if 5°C warmer water was supplied throughout the year. More durations are
compiled in Table 5.13 to facilitate the possible utilization of further tests.

However, due to these experimentations occurring during weekends, this estimation method
does not take into account how much of the elevated heating power can be attributed to the
absence of heat loads caused by user activity. If only weekends were observed, the heating
power was still 6.64kW more by average in the elevated supply water temperature scenario.
This aspect contributes also to the margin of error, but does not reverse the findings.

Table 5.13 A sheet to estimate the total benefits with further study.
Interval
]-∞,-5[ [-5,+0[ [+0,+8[ [+8,+17[ [+17,+21[ [+21,+∞[
[°C]
Duration 569h 556h 3123h 2365h 1360h 789h
No excess Negligible
Hourly Requires Requires Requires
power 7.32kW water flow
gap assumed
tests tests tests
assumed
Annual 9.96MWh
gap (398€)

Based on the preliminary results it can be suggested that if a control program of adaptive
outdoor temperature compensation is able to eliminate or narrow this gap without any
installation work, its implementation could be beneficial. Unfortunately no cost estimates
for the installation of adaptive heating control program were obtained, even though two
contractors were asked.

86
6 Conclusion
6.1 Discussion of the findings
The objective of this thesis was to investigate how the energy performance of buildings could
be improved effectively by exploiting available data. This objective was subsequently
formulated into three research sub-questions:
1. What are the energy performance gaps of buildings?
2. What kind of solutions that are based on the exploitation of available data can close
those gaps?
3. How effective are those solutions?

The research process was based on reviewing research literature, conducting interviews and
analyzing measurement data recorded by the building automation system of a case office
building. The research only included factors of energy performance that could be enhanced
with available data in the operations and maintenance phase of a building’s lifecycle. This
means that energy-efficient equipment control principles, and methods to supervise and
manage the energy performance of equipment were included in research. Additionally, being
the main tool for computerized intervention, building automation systems were within the
scope of the thesis. Also solutions that facilitate the management of satisfactory indoor
environmental quality were included.

The identified energy performance gaps of buildings related to 1) inadequate heating

adjustability and control, 2) suboptimal utilization of free energies, 3) uncharted implications
of existing saving measures, 4) insufficient knowledge on the satisfaction of building users
and 5) inefficiency of processes. The initiatives to close these gaps were based on acquiring
new equipment to enhance adjustability, installing new sensors and transmitters to collect
more data, programming new control methods or developing new information systems. Out
of these initiatives the most effective were considered to be the ones that are simple and do
not require any installation work, or at the most, the installation of transmitters or sensors.
The least effective initiatives were considered to be the ones that are complex, risk user
satisfaction or require integration of numerous systems.

To increase feasibility, some compromises were made with regard to the originally proposed
top three initiatives: Instead of studying the costs and benefits of those three initiatives as
such, four more commercially realistic entities and options within them were studied. Those
four entities were measuring user satisfaction, group control of actuators, energy efficiency
monitoring and adaptive outdoor temperature compensation for heating. Based on the
feasibility studies it is suggested that adaptive heating control has the potential to increase
energy performance with negligible installation work, user satisfaction measurement system
would be sensible to pilot as a service, energy efficiency monitoring in small scale would be
convenient to purchase as a service and selected group control interfaces would enable large
savings with small trouble.

87
6.2 Theoretical contribution
The results of this research produce feedback from an end-user organization (and its
stakeholders) to the research community and support some findings made in earlier research.

The findings increase understanding on the perceived effectiveness of energy performance

enhancing initiatives from the perspectives of professionals in the fields of building
automation, building services engineering, property management and property maintenance.
Even with such a small sample size, the perceived challenges and benefits can be utilized to
clarify the value propositions of solutions that rely on the same technologies. By harnessing
the identified value drivers while minimizing the impact of drawbacks, the policy resistance
regarding the adoption of those solutions is likely to decrease.

There are also results that support findings and statements made in earlier research:
• The findings of this research verified that adaptive outdoor temperature compensation
would reduce the workload of maintenance, therefore confirming the suggested benefits
by Paiho et al. (2002, p. 94). Moreover, their control method was considered highly
effective.
• Forming half of the initiative 6. Improving manageability of large amounts of sensor data
and actuators that ranked third among all the initiatives in effectiveness, energy efficiency
monitoring was considered very beneficial. This supports the notion of Ihasalo (2012, pp.
2, 137) that there is a need for tools that monitor performance. Moreover, based on a semi-
structured interview the monitoring dashboards should be tailored to user-specific needs.
• The limitations of available data confirmed that training data required by artificial neural
networks may not be available as suggested by Kolokotsa et al. (2009, p. 1850). The
historical data available through the BAS is deterministic due to control objectives and
limited due to sensor infrastructure: For example the resultant room temperature with
different actions and contexts cannot be taught with historical data because control always
has had the same goal, and not all inputs are measured.
• Based on the evaluations it was observed that the initiative 9. Utilizing context-aware
alarm thresholds in fault detection was not perceived beneficial enough versus the
development costs. This finding supports the statement of Ihasalo (2012, p. 42) that FDD
tools are not widespread because of little quantified information about the benefits with
sometimes large costs of installation. The only stakeholder group that saw clear benefits
in improving fault detection was property maintenance.

6.3 Practical implications

The results of this thesis have implications regarding the design guidelines of heating
systems, ventilation systems and lighting systems, and the requirements for building
automation systems: Decisions made then by building owners, property managers, building
automation contractors and building services engineering designers may have a huge impact
on energy performance and the applicability of the presented initiatives later on. These
implications are listed below:
• In heating system design, the exposure of the building to heat loads such as sunlight
should be considered based on window size, façade type and immediate protective
environment. If it is likely that the heat loads are going to be large at a façade at a
time during the heating season, then a separate radiator network for each façade could
improve heating control performance significantly. Also allocating a separate
network for spaces with constant heating demand would serve the same purpose.

88
 • In ventilation system design, the division of operating areas and decentralization of
ventilation is essential to attain a good level of energy performance. To an extent,
areas with constant ventilation demand and areas serving different functions or
occupants should have their own AHUs to increase adjustability.
• In lighting system design, the availability of natural light in working areas should be
scrutinized based on window area and floor layout. If there is large availability of
natural light due to large windows and working areas mainly near the windows,
constant luminance control with motorized blinds may introduce large savings of
electricity.
• In requirements for a building automation system, attention should be given to the
naming policy of control points and capabilities to export and import data, if a
database project for sensor measurements was being considered. Such a database
project would face problems if control points were not named in a standardized and
unique way, or if importing data from the BAS was difficult. If the BAS had
interfaces requiring only credential, link and refresh interval specification from the
database server, this aspect would be greatly facilitated. Additionally, snow melting
heaters should be connected to building automation.

6.4 Research limitations

The findings made from this research are subject to limitations of the case study. The
limitations of the case study were caused by the occasional unavailability of relevant sensor
data, unfinished tuning of building services equipment, the subjective evaluation method of
initiatives aimed to close energy performance gaps and the small sample of evaluators
(N=13). Consequently, let us recall the three research sub-questions:
1. What are the energy performance gaps of buildings?
2. What kind of solutions that are based on the exploitation of available data can close
those gaps?
3. How effective are those solutions?

The case study limited the answers to the first research question especially with respect to
lighting control performance due to unavailability of cause level data. All the findings in this
area were made based on effect level data. In other aspects of energy performance, there was
cause level data available supported occasionally by statements made in the semi-structured
interviews. Unfinished tuning did not limit the research, as it was possible to isolate
equipment still in tuning phase from analysis.

Answers to the second research question were limited by the lack of experience of the author
about the field of research. The effect of this limitation was reduced by a literature review
before commencing the research, multidisciplinary steering within the case organization and
information provided by experts during the evaluation interviews.

Answers to the third research question were subject to limitations caused by the small
amount of experts that evaluated the initiatives and occasionally the unavailability of
relevant expertise. Not many of the respondents were familiar with artificial intelligence or
database development, which reduced the input or its degree of certainty especially in the
challenge to implement dimension. Therefore, the results occasionally answer a question:
How effective are those solutions perceived to be? Still, the effect of the small sample size
was reduced by interviewing people from different organizations.

89
6.5 Further research
Identified needs for further research relate on one hand to clarifying the value proposition of
some of the potential solutions and on the other hand to monitoring the influence of the
selected solutions should they be implemented.

There was vagueness in the value proposition of installing electronic thermostats and
motorized blinds. Even though the installation of electronic radiator thermostats was
considered beneficial, the quantitative influence of smart heating functions on the energy
performance of the heating (and cooling) system remained unclear. By applying the solution
to a part of a building, deploying the radiator pipes with sensors measuring water flow and
recording chilled beam operation (if any) the outcome could be compared quantitatively to
the situation with mechanical thermostats. Similarly with daylight harvesting, the referred
test results did not state the benefit of including motorized blinds in constant luminance
controlled lighting system. Constant luminance control is much simpler to implement
without motorizing blinds, thus the value proposition of including them should be
scrutinized.

The capability of the selected solutions to improve energy performance in real-world

testbeds is also an interesting study area. Especially the commitment of building users to
giving feedback and the ability to utilize it in BAS interventions, and the capability of
adaptive heating control to reduce excess output are intriguing areas of study.

90
References
Agarwal, Y. & Balaji, B. & Gupta, R. & Lyles, J. & Wei, M. & Weng, T. 2010.
Occupancy-driven energy management for smart building automation. BuildSys '10
Proceedings of the 2nd ACM Workshop on Embedded Sensing Systems for Energy-
Efficiency in Building. [Online]. PP. 1-6. [Cited 5.2.2016]. DOI:
10.1145/1878431.1878433. ISBN: 978-1-4503-0458-0.

Atzori, L. & Iera, A. & Morabito, G. 2010. The Internet of things: A survey. Computer
networks. [Online]. Vol. 54:15. PP. 2787-2805. [Cited 12.4.2016]. ISSN: 1389-1286. DOI:
10.1016/j.comnet.2010.05.010.

Rakennustieto. 2008. RT 07-10946 Sisäilmastoluokitus 2008 [in Finnish]. [Online]. 22

pages. [Cited 8.7.2016]. Available at:
https://www.rakennustieto.fi/kortistot/rt/kortit/10946.

Bülow-Hübe, H. 2007. BELOK - Solar Shading and Daylight Redirection. Lund, Sweden.
Energy and Building Design, Architecture and Built Environment, Lund University. 69
pages. [Cited 29.6.2016]. ISSN: 1651-8128. ISBN: 978-91-85147-20-5.

Chen, H. & Chou, P. & Duri, S. & Lei, H. & Reason, J. 2009. The design and
implementation of a smart building control system. ICEBE '09 Proceedings of the 2009
IEEE International Conference on e-Business Engineering. [Online] PP. 255-262. [Cited
12.2.2016]. ISBN: 978-0-7695-3842-6. DOI: 10.1109/ICEBE.2009.42255.

Coetzee, L. & Eksteen, J. 2011. The Internet of Things-promise for the future? An
introduction. IST-Africa Conference Proceedings. [Online]. PP. 1-9. [Cited 31.3.2016].
ISBN: 978-1-905824-26-7.

DataRangers. 2016a. We help you to create value with your people. [Online].
Available at: http://www.datarangers.fi. [Accessed 6.7.2016].

DataRangers. 2016b. Louhin - Change the way you collect and use data. [Online].
Available at: http://www.louhin.com/?lang=en. [Accessed 4.7.2016].

Delaney, D. T. & O'Hare, G. M. P. & Ruzzelli, A. G. 2009. Evaluation of energy-

efficiency in lighting systems using sensor networks. BuildSys '09 Proceedings of the First
ACM Workshop on Embedded Sensing Systems. [Online]. PP. 61–66. [Cited 22.2.2016].
ISBN: 978-1-60558-824-7. DOI: 10.1145/1810279.1810293.

Deos, 2016. OPENweb Brochure. [Online]. Available at: http://www.deos-

ag.com/en/downloads/. [Accessed 7.6.2016].

Dong, B. & Andrews, B. 2009. Sensor-based occupancy behavioral pattern recognition for
energy and comfort management in intelligent buildings. Proceedings of building
simulation. [Online]. PP. 1444-1451. [Cited 5.2.2016]. Available at:
http://www.ibpsa.org/proceedings/BS2009/BS09_1444_1451.pdf.

91
Dooley, K. 2016. Building a Culture of Experimentation. Energistä rakentamista blog of
Granlund. [Online]. Available at: https://energistarakentamista.com/2016/06/29/building-
a-culture-of-experimentation-kokeilukulttuuri/. [Cited 6.7.2016].

Dounis, A. I. & Caraiscos, C. 2009. Advanced control systems engineering for energy and
comfort management in a building environment—A review. Renewable and Sustainable
Energy Reviews. [Online]. Vol. 13:6. PP. 1246-1261. [Cited 5.2.2016]. DOI:
10.1016/j.rser.2008.09.015.

eGain. 2016. eGain Company presentation 2016. [Online]. Available at:

http://egain.se/en/about-us/press/. [Accessed 29.6.2016].

Eisenhardt, K. M. 1989. Building theories from case study research. Academy of

management review. [Online]. Vol. 14:4. PP. 532-550. [Cited 18.9.2016]. DOI:
10.1016/j.rser.2008.09.015. ISSN: 0363-7425. DOI: 10.5465/AMR.1989.4308385.

Erickson, V. L. & Lin, Y. & Kamthe, A. & Brahme, R. & Surana, A. & Cerpa, A. & Sohn,
M. D. & Narayanan, S. 2009. Energy efficient building environment control strategies
using real-time occupancy measurements. BuildSys '09 Proceedings of the First ACM
Workshop on Embedded Sensing Systems for Energy-Efficiency in Buildings. [Online]. PP.
19-24. [Cited 12.2.2016]. ISBN: 978-1-60558-824-7. DOI: 10.1145/1810279.1810284.

European Commission. 2016a. Buildings. [Online]. Available at:

https://ec.europa.eu/energy/en/topics/energy-efficiency/buildings. [Accessed 23.8.2016].

European Commission. 2016b. Proposal for a REGULATION OF THE EUROPEAN

PARLIAMENT AND OF THE COUNCIL on binding annual greenhouse gas emission
reductions by Member States from 2021 to 2030 for a resilient Energy Union and to meet
commitments under the Paris Agreement and amending Reg. [Online]. Available at:
http://ec.europa.eu/clima/policies/effort/proposal/documentation_en.htm. [Accessed
23.8.2016].

European Commission. 2016c. Proposal for an Effort Sharing Regulation 2021-2030.

[Online]. Available at: http://ec.europa.eu/clima/policies/effort/proposal/index_en.htm.
[Accessed 23.8.2016].

Finlex. 2014. Law on Energy Efficiency 1429/2014 [in Finnish]. [Online]. Available at:
http://www.finlex.fi/fi/laki/alkup/2014/20141429. [Accessed 23.8.2016].

Finlex. 2015. Climate Change Act 609/2015 English. [Online]. Available at:
http://www.finlex.fi/en/laki/kaannokset/2015/20150609. [Accessed 23.8.2016].

Finnish Meteorological Institute. 2014. Open data manual. [Online]. Available at:
http://en.ilmatieteenlaitos.fi/open-data-manual. [Accessed 13.5.2016].

Finnish Ministry of the Environment. 2012a. D2 Indoor climate and ventilation of

buildings [in Finnish]. [Online]. Available at: http://www.finlex.fi/data/normit/37187-D2-
2012_Suomi.pdf. [Accessed 4.7.2016].

92
Finnish Ministry of the Environment. 2012b. D3 Energy management in buildings [in
Finnish]. [Online]. Available at: http://www.finlex.fi/data/normit/37188-D3-
2012_Suomi.pdf. [Accessed 28.7.2016].

Finnparttia. 2016. LK IST23 itsesäätyvä lämpökaapeli [in Finnish]. [Online]. Available at:
http://www.finnparttia.fi/epages/finnparttia.sf/fi_FI/?ObjectPath=/Shops/2014102905/Prod
ucts/LKIST23. [Accessed 1.9.2016].

Foliente, G. et al., 2005. Performance Based Building. R&D ROADMAP, PeBBu Final
Report, EUR, 21988. [Online]. 13 pages. [Cited 14.3.2016]. Available at:
http://www.construction-
innovation.info/images/pdfs/conference_cd_2004/conference/Papers/Refereed%20Papers/
146%20Foliente.pdf.

Fourdeg. 2016. Fourdeg - Energy Efficiency in Buildings and District Heating Network.
[Online]. Available at: http://www.fourdeg.com/. [Accessed 29.6.2016].

Freire, R. Z. & Oliveira, G. H. C. & Mendes, N. 2008. Predictive controllers for thermal
comfort optimization and energy savings. Energy and Buildings. [Online]. Vol. 40:7. PP.
1353-1365. [Cited 5.2.2016]. ISSN: 0378-7788. DOI: 10.1016/j.enbuild.2007.12.007.

Gibson, E. J. 1982. Working with the performance approach in building. CIB Report,
Publication 64. [Online]. PP. 1-30. [Cited 14.4.2016]. Available at:
https://www.irbnet.de/daten/iconda/CIB_DC23969.pdf.

Gruber, M. 2012. Demand-based control of indoor climate in office buildings. Design of

local and central feed-forward control system for high comfort, low energy use and low
peak power. Gothenburg, Sweden. Chalmers University of Technology, Department of
Energy and Environment, Building Services Engineering, Licentiate's thesis. [Online]. 172
pages. [Cited 9.9.2016]. ISSN: 1652-6007. Available at:
http://publications.lib.chalmers.se/records/fulltext/164192/164192.pdf.

Gurstein, M. 2011. Open data: Empowering the empowered or effective data use for
everyone?. First Monday. [Online]. Vol. 16:2. [Cited 10.9.2016]. ISSN: 1396-0466.
Available at: http://firstmonday.org/ojs/index.php/fm/article/view/3316/2764. DOI:
10.5210/fm.v16i2.3316.

Gyalistras, D. & Gwerder, M. & Oldewurtel, F. & Jones, C. N. & Morari, M. & Lehmann,
B. & Wirth, K. & Stauch, V. 2010. Analysis of energy savings potentials for integrated
room automation. Clima-RHEVA World Congress (No. EPFL-CONF-169732). [Online]. 8
pages. [Cited 5.2.2016]. Available at:
https://infoscience.epfl.ch/record/169732/files/Gyal_10_Proc-Clima2010.pdf.

Göös, T. 2012. Investointilaskentamenetelmien soveltuvuus lämmitysenergiansäästöön

kohdistuvassa investoinnissa [in Finnish]. 93 pages. [Cited 29.6.2016]. Tampere, Finland.
Tampere University of Technology, Master's Thesis. Available at:
http://static.ecome.fi/upload/3829/Diplomity%C3%B6_G%C3%B6%C3%B6s_Tommi.pdf
.

93
Hagras, H. & Callaghan, V. & Colley, M. & Clarke, G. & Pounds-Cornish, A. & Duman,
H. 2004. Creating an ambient-intelligence environment using embedded agents. IEEE
Intelligent Systems. [Online]. Vol. 19:6. PP. 12–20. [Cited 22.2.2016]. ISSN: 1541-1672.
DOI: 10.1109/MIS.2004.61.

Harris, C. & Cahill, V. 2005. Exploiting user behaviour for context-aware power
management. WiMob'2005 IEEE International Conference on Wireless And Mobile
Computing, Networking And Communication. [Online]. Vol. 4. PP. 122–130. [Cited
22.2.2016]. ISBN: 0-7803-9181-0. DOI: 10.1109/WIMOB.2005.1512959.

Helsinki University Intranet. 2014. University Facilities. [Online]. Available at:

https://flamma.helsinki.fi/portal/home/login. [Accessed 5.7.2016].

Helvar. 2016. Helvar Solutions. [Online]. Available at: http://www.helvar.com/solutions.

[Accessed 30.6.2016].

Henderson, R. 1993. Underinvestment and incompetence as responses to radical

innovation: Evidence from the photolithographic alignment equipment industry. The
RAND Journal of Economics. Vol. 24:2. PP. 248-270. [Cited 14.4.2016]. Available at:
http://www.jstor.org/stable/2555761.

Houm. 2016. Houm - High-end lighting control for everyone. [Online]. Available at:
http://houm.io/en/. [Accessed 30.6.2016].

Hunn, B. D. & Haberl, J. S. & Davies, H. & Owens, B. 2012. Measuring Commercial
Building Performance. ASHRAE Journal, July 2012. [Online]. Vol. 54:7. PP. 48-59. [Cited
29.4.2016]. Available at:
http://search.proquest.com/docview/1034969948?accountid=11365.

Hyvärinen, J. & Kärki, S. 1996. Building optimization and fault diagnosis source book.
Espoo, Finland. Technical Research Centre of Finland, VTT Building Technology. 324
pages. [Cited 29.4.2016]. ISBN: 952-5004-10-4. Available at:
http://www.ecbcs.org/docs/annex_25_source_book.pdf.

IEA. 2015. Building Energy Performance Metrics. [Online]. Available at:

http://www.iea.org/publications/freepublications/publication/BuildingEnergyPerformance
Metrics.pdf. [Accessed 23.8.2016].

Ihasalo, H. 2012. Transforming building automation data into building performance

metrics-design, implementation and evaluation of use of a performance monitoring and
management system. Helsinki, Finland. Aalto University publication series DOCTORAL
DISSERTATIONS 26/2012. 188 pages. [Cited 24.4.2016]. ISSN: 1799-4934 (printed).
ISSN: 1799-4942 (electronic). ISBN: 978-952-60-4539-9 (printed). ISBN: 978-952-60-
4540-5 (electronic). Available at: https://aaltodoc.aalto.fi/handle/123456789/4828.

ISO. 2005. EN ISO 7730:2005 Ergonomics of the Thermal Environment - Analytical

Determination and Interpretation of Thermal Comfort using Calculation of the PMV and
PPD Indices and Local Thermal Comfort Criteria. [Online]. Geneva, Switzerland.

94
International Organisation for Standardisation. 52 pages. [Cited 22.4.2016]. Available at:
http://www.iso.org/iso/catalogue_detail.htm?csnumber=39155.

Kalogirou, S. A. 2006. Artificial neural networks in energy applications in buildings.

International Journal of Low Carbon Technologies. [Online]. Vol. 1:3. PP. 201-216.
[Cited 22.2.2016]. DOI: 10.1093/ijlct/1.3.201.

Katainen, V. & Vähämaa, K. 2015. Paine-erojen pitkäaikainen seuranta ja painesuhteiden

vaihtelu rakennuksissa [in Finnish]. [Online]. University of Eastern Finland, Bachelor's
thesis. 95 pages. [Cited 15.6.2016]. Available at:
https://www2.uef.fi/documents/976466/2568699/VahamaaKatainen_virallinen2015.pdf/3a
3baf31-412e-4716-9552-f86d075fb276.

Keeling, T. & Clements-Croome, D. J. & Luck, R. & Pointer, P. 2013. Wireless sensors for
monitoring people and their close environment. In: Intelligent Buildings - Design,
Management and Operation. 2nd ed. s.l. ICE Publishing. PP. 107-118. [Cited 21.2.2016].
ISBN: 978-0-7277-5734-0. DOI: 10.1680/ib.57340.107.
Kim, Y. & Charbiwala, Z. & Singhania, A. & Schmid, T. & Srivastava, M. B. 2008.
SPOTLIGHT: Personal natural resource consumption profiler. Proceedings of the 5th
Workshop on Hot Topics in Embedded Networked Sensors. [Online]. 5 pages. [Cited
22.2.2016]. Available at: http://nslab.ee.ntu.edu.tw/NetworkSeminar/papers/spot.pdf.

Kolokotsa, D. 2007. Artificial intelligence in buildings: a review on the application of

fuzzy logic. Advances in Building Energy Research. [Online]. Vol. 1:1. PP. 29-54. [Cited
22.2.2016]. ISSN: 1751-2549 (printed). ISSN: 1756-2201 (electronic). DOI:
10.1080/17512549.2007.9687268.

Kolokotsa, D. & Pouliezos, A. & Stavrakakis, G. & Lazos, C. 2009. Predictive control
techniques for energy and indoor environmental quality management in buildings.
Buildings and Environment. [Online]. Vol. 44:9. PP. 1850-1863. [Cited 5.2.2016]. DOI:
10.1016/j.buildenv.2008.12.007

Korhola, E.-R. 2014. The rise and fall of the Kyoto Protocol : climate change as a political
process. Helsinki, Finland. University of Helsinki, Faculty of Biological and
Environmental Sciences, Department of Environmental Sciences, Doctoral dissertation.
321 pages. [Cited 23.8.2016]. ISBN: 978-951-51-0233-1 (printed). ISBN: 978-951-51-
0234-8 (electronic). Available at: http://urn.fi/URN:ISBN:978-951-51-0234-8.

Laaksonen, E. 1995. Isännöitsijän käsikirja [in Finnish]. 10th edition. Helsinki, Finland.
Kiinteistöalan kustannus. 1006 pages. ISBN: 951-9378-90-1.

Lavy, S. & Garcia, J. A. & Dixit, M. K. 2010. Establishment of KPIs for facility
performance measurement: review of literature. Facilities. [Online]. Vol. 28:9. PP. 440-
464. [Cited 24.8.2016]. ISSN: 0263-2772. DOI: 10.1108/02632771011057189.

Lindgren, H. 2012. Koulurakennuksen painesuhteiden mittaus liitettynä ilmavirtojen

säätöön [in Finnish]. [Online]. 38 pages. [Cited 15.6.2016]. Mikkeli University of Applied
Science, Bachelor's thesis. Available at: http://urn.fi/URN:NBN:fi:amk-201203123266.

95
Lu, J. & Sookoor, T. & Srinivasan, V. & Gao, G. & Holben, B. & Stankovic, J. & Field, E.
& Whitehouse, K. 2010. The smart thermostat: using occupancy sensors to save energy in
homes. SenSys '10 Proceedings of the 8th ACM Conference on Embedded Networked
Sensor Systems. [Online]. PP. 211-224. [Cited 12.2.2016]. ISBN: 978-1-4503-0344-6.
DOI: 10.1145/1869983.1870005.

Ma, H.-D. 2011. Internet of things: Objectives and scientific challenges. Journal of
Computer Science and Technology. [Online]. Vol. 26:6. PP. 919-924. [Cited 31.3.2016].
DOI: 10.1007/s11390-011-1189-5.

Marchiori, A. & Han, Q. 2010. Distributed wireless control for building energy
management. BuildSys '10 Proceedings of the 2nd ACM Workshop on Embedded Sensing
Systems for Energy-Efficiency in Building. [Online]. PP. 37-42. [Cited 22.2.2016]. ISBN:
978-1-4503-0458-0. DOI: 10.1145/1878431.1878441.

Martocci, J. P. 2008. BACnet® Unplugged: ZigBee® and BACnet Connect. ASHRAE

Journal. [Online]. Vol. 50:6. PP. 42-46. [Cited 28.6.2016]. Available at:
http://bookstore.ashrae.biz/journal/download.php?file=Martocci.pdf.

Microsoft. 2016. Power BI - Bring your data to life. [Online]. Available at:
https://powerbi.microsoft.com/en-us/. [Accessed 22.7.2016].

Miorandi, D. & Sicari, S. & Pellegrini, F. D. & Chlamtac, I. 2012. Internet of things:
Vision, applications and research challenges. Ad Hoc Networks. [Online]. Vol. 10:7. PP.
1497-1516. [Cited 31.3.2016]. DOI: 10.1016/j.adhoc.2012.02.016.

Morris, T. & Vaughn, R. & Dandass, Y. 2012. A retrofit network intrusion detection
system for MODBUS RTU and ASCII industrial control systems. 2012 45th Hawaii
International Conference on System Sciences. [Online]. PP. 2338-2345. [Cited 28.6.2016].
ISBN: 978-0-7695-4525-7. DOI: 10.1109/HICSS.2012.78.

MSDN. 2016a. Download SQL Server Data Tools (SSDT). [Online]. Available at:
https://msdn.microsoft.com/en-us/library/mt204009.aspx. [Accessed 6.7.2016].

MSDN. 2016b. SQL Server Data Mining Add-Ins for Office. [Online]. Available at:
https://msdn.microsoft.com/en-us/library/dn282373.aspx. [Accessed 6.7.2016].

MSDN. 2016c. Excel. [Online]. Available at: https://msdn.microsoft.com/en-

us/library/office/fp179694.aspx. [Accessed 13.9.2016].

Muhli, J. 2012. Ilmanvaihtokoneen puhaltimen uusinta energiatehokkaasti [in Finnish].

Metropolia University of Applied Sciences, Bachelor's thesis. [Online]. 34 pages. [Cited
28.7.2016]. Available at: http://urn.fi/URN:NBN:fi:amk-201205117596.

Månsson, L.-G. & McIntyre, D. 1997. Technical Synthesis Report. A Summary of IEA
Annexes 16 & 17 Building Energy Management Systems. International Energy Agency,
Energy Conservation in Buildings and Community Systems. [Online]. PP. 1-41. [Cited
28.4.2016]. Available at: http://www.ecbcs.org/docs/Annex%2016-
17%20TSR%20web.pdf.

96
Nguyen, T. A. & Aiello, M. 2013. Energy intelligent buildings based on user activity: A
survey. Energy and Buildings. [Online]. Vol. 56. PP. 244-257. [Cited 5.2.2016]. ISSN:
0378-7788. DOI: 10.1016/j.enbuild.2012.09.005.
Oancea, C. & Caluianu, S. 2013. Designing intelligent buildings for people's well-being
using an artificial intelligence approach. In: Intelligent Buildings - Design, Management
and Operation. 2nd ed. s.l. ICE Publishing. PP. 89-105. [Cited 21.2.2016]. ISBN: 978-0-
7277-5734-0. DOI: 10.1680/ib.57340.089.

Open Definition. 2016. The Open Definition. [Online]. Available at:

http://www.opendefinition.org/. [Accessed 10.9.2016].

Padmanabh, K. & Malikarjuna, V. A. & Sen, S. & Katru, S. P. & Kumar, A. &
Pawankumar, C. S. & Vuppala, S. K. & Paul, S. 2009. iSense: a wireless sensor network
based conference room management system. BuildSys'09 Proceedings of the First ACM
Workshop on Embedded Sensing System for Energy-Efficiency in Buildings. [Online]. PP.
37-42. [Cited 21.2.2016]. ISBN: 978-1-60558-824-7. DOI: 10.1145/1810279.1810288.

Paiho, S. & Karjalainen, S. & Alanne, K. & Norvasuo, M. & Eriksson, M. & Pöyhönen, S.
& Kaartinen, J. & Lehtovaara, J. 2002. Rakennusten uudet säätö- ja
energianhallintaratkaisut [in Finnish]. Espoo, Finland. VTT Research notes 2134. 279
pages. [Cited 9.5.2016]. ISSN: 1235-0605 (printed). ISSN: 1455-0865 (electronic). ISBN:
951-38-5951-7 (printed). ISBN: 951-38-5952-5 (electronic).

Peltokorpi, A. & Alho, A. & Kujala, J. & Aitamurto, J. & Parvinen, P. 2008. Stakeholder
approach for evaluating organizational change projects. International Journal of Health
Care Quality Assurance. [Online]. Vol. 21:5. PP. 418-434. [Cited 25.5.2016]. ISSN: 0952-
6862. DOI: 10.1108/09526860810890413.

Reda, I. & Andreas, A. 2004. Solar position algorithm for solar radiation applications.
Solar energy. [Online]. Vol. 76:5. PP. 577-589. [Cited 25.5.2016]. ISSN: 0038-092X.
DOI: 10.1016/j.solener.2003.12.003.

Reinisch, C. & Kastner, W. & Neugschwandtner, G. & Granzer, W. 2007. Wireless

technologies in home and building automation. Industrial Informatics, 5th IEEE
International Conference. [Online]. Vol. 1. [Cited 28.4.2016]. ISSN: 1935-4576 (printed).
ISSN: 2378-363X (electronic). ISBN: 978-1-4244-0850-4 (printed). ISBN: 978-1-4244-
0851-1 (electronic). DOI: 10.1109/INDIN.2007.4384737.

Salama, M. 2014. Älykäs lämmönsäädin [in Finnish]. Espoo, Finland. Aalto University
School of Electrical Engineering, Master's thesis. [Online]. 63 pages. [Cited 21.6.2016].
Available at: http://urn.fi/URN:NBN:fi:aalto-201411042992.

Seppänen, K. 2010. Painesuhteet rakennuksen ulkovaipan yli [in Finnish]. Kuopio,

Finland. University of Eastern Finland, Aducate Reports and Books 9/2010, Bachelor's
thesis. [Online]. 41 pages. [Cited 15.6.2016]. ISSN: 1798-9116. ISBN: 978-952-61-0055-5
(printed). ISBN: 978-952-61-0056-2 (electronic). Available at:
http://epublications.uef.fi/pub/urn_isbn_978-952-61-0056-2/urn_isbn_978-952-61-0056-
2.pdf.

97
Sexton, M. & Barrett, P. 2005. Performance-based building and innovation: balancing
client and industry needs. Building Research & Information. [Online]. Vol. 33:2. PP. 142-
148. [Cited 14.3.2016]. ISSN: 0961-3218 (printed). ISSN: 1466-4321 (electronic). DOI:
10.1080/0961321042000323789.

SFS. 2007. SFS-EN 15251:2007 Indoor environmental input parameters for design and
assessment of energy performance of buildings addressing indoor air quality, thermal
environment, lighting and acoustics. [Online]. Helsinki, Finland. Suomen
Standardisoimisliitto SFS. 52 pages. [Cited 29.4.2016]. Available at:
https://sales.sfs.fi/en/index/tuotteet/SFS/CEN/ID2/1/41945.html.stx.

Shengwei, W. 2009. Intelligent buildings and building automation. 1st ed. s.l. Taylor &
Francis. [Online]. 248 pages. [Cited 21.2.2016]. ISBN: 978-0-415-47570-9 (printed).
ISBN: 978-0- 203-89081-3 (electronic).

Singhvi, V. & Krause, A. & Guestrin, C. Garrett Jr., J. H. & Matthews, H. S. 2005.
Intelligent light control using sensor networks. SenSys '05 Proceedings of the 3rd
international conference on Embedded networked sensor systems. [Online]. PP. 218–229.
[Cited 22.2.2016]. ISBN: 1-59593-054-X. DOI: 10.1145/1098918.1098942.

Somfy. 2016. Somfy Solutions - Building Control. [Online]. Available at:

http://www.somfy-
architecture.com/index.cfm?page=/buildings/home/solutions_tertiaires&action=navi&catid
=1&t=1. [Accessed 30.6.2016].

Soucek, S. & Loy, D. 2007. Vertical integration in building automation systems. Industrial
Informatics, 5th IEEE International Conference. [Online]. Vol: 1. PP. 81-86. [Cited
28.4.2016]. ISSN: 1935-4576 (print). ISSN: 2378-363X (electronic). ISBN: 978-1-4244-
0850-4 (printed). ISBN: 978-1-4244-0851-1 (electronic). DOI:
10.1109/INDIN.2007.4384735.

TA. 2011. Patteriverkoston säätäminen [in Finnish]. s.l. s.n. [Online]. 68 pages. [Cited
9.6.2016]. Available at: http://vantalvi.fi/wp-content/uploads/2014/05/TA-
K%C3%84SIKIRJA-Patteriverkoston-S%C3%A4%C3%A4t%C3%A4minen-2011.pdf.

Taesler, R. 2007. Method of controlling the heating of a building at a desired level. Patent
No. WO 2007061357 A1. Available at:
https://worldwide.espacenet.com/publicationDetails/biblio?CC=WO&NR=2007061357A1
&KC=A1&FT=D&ND=1&date=20070531&DB=&locale=.

Taloon 2016a. Sulanapitotermostaatti Devireg 850 keskusyksikkö [in Finnish]. [Online].

Available at: http://www.taloon.com/sulanapitotermostaatti-devireg-850-keskusyksikko/S-
3531184/dp?openGroup=10386. [Accessed 1.9.2016].

Taloon 2016b. Kattoanturi Devireg 850 IP67 [in Finnish]. [Online]. Available at:
http://www.taloon.com/kattoanturi-devireg-850-ip67/S-3531183/dp?openGroup=10387.
[Accessed 1.9.2016].

98
Talotuote 2016. Virtalähde DEVI DC-PSU 24V/24W (D850/WEB-HOME) [in Finnish].
[Online]. Available at:
http://www.talotuote.fi/epages/talotuote.sf/fi_FI/?ObjectPath=/Shops/2013112707/Product
s/3531191. [Accessed 1.9.2016].

Turner, C. & Frankel, M. 2008. Energy performance of LEED for new construction
buildings. New Buildings Institute. [Online]. PP. 1-42. [Cited 9.9.2016]. Available at:
http://www.usgbc.org/Docs/Archive/General/Docs3930.pdf.

University of Helsinki. 2013. Helsinki University Center for Properties and Facilities [in
Finnish]. [Online]. Available at: http://www.helsinki.fi/teknos/. [Accessed 22.4.2016].

Wang, L. & Greenberg, S. & Fiegel, J. & Rubalcava, A. & Earni, S. & Pang, X. & Yin, R.
& Woodworth, S. & Hernandez-Maldonado, J. 2013. Monitoring-based HVAC
commissioning of an existing office building for energy efficiency. Applied energy.
[Online]. Vol. 102. PP. 1382-1390. [Cited 5.2.2016]. ISSN: 0306-2619. DOI:
10.1016/j.apenergy.2012.09.005.

WBCSD. 2009. Transforming the Market: Energy Efficiency in Buildings. [Online].

Available at: http://www.wbcsd.org/transformingthemarketeeb.aspx. [Accessed
23.8.2016].

Vine, E. 2005. An international survey of the energy service company (ESCO) industry.
Energy Policy. [Online]. Vol. 33:5. PP. 691-704. [Cited 15.4.2016]. ISSN: 0301-4215.
DOI: 10.1016/j.enpol.2003.09.014.

Yin, R. K. 2009. Case study research: Design and methods. 4th edition. 240 pages. [Cited
18.9.2016]. Sage publications. ISBN: 978-1-4129-6099 (printed).

Ylitalo, J. 2012. Rakennusautomaation väylät ja integraatio [in Finnish]. Metropolia

University of Applied Sciences, Bachelor's thesis. [Online]. 32 pages. [Cited 20.6.2016].
Available at: http://urn.fi/URN:NBN:fi:amk-2012121219255.

99
Appendices

Appendix 1. Evaluation results. 2 pages.

Appendix 2. Sensor data modelling. 4 pages.
Appendix 3. Data clustering. 2 pages.
 Appendix 1 (1/2)

Appendix 1

Evaluation results
The interviewees are listed in Table 0.1 and evaluation results in Table 0.2, respectively.

Table 0.1 Interviewees.

Interviewee Interview Supplementary Expertise Background
date interview for
proposal 11
I 27.5.2016 - Building automation Contractor
II 30.5.2016 - Building services Designer
engineering
III 6.6.2016 12.8.2016 Building services Manager
engineering
IV 6.6.2016 12.8.2016 Property Manager
management
V 6.6.2016 12.8.2016 Property Manager
management
VI 6.6.2016 12.8.2016 Building services Designer
engineering
VII 15.6.2016 - Building automation Designer
VIII 15.6.2016 - Building automation Designer
IX 13.7.2016 - Property maintenance Plumber
X 13.7.2016 - Property maintenance Plumber
XI 13.7.2016 - Property maintenance Plumber
XII 2.8.2016 - Building automation Contractor
XIII 17.8.2016 - Property Manager
management
Appendix 1 (2/2)

Table 0.2 Evaluation results.

Interviewee / Proposal I II III IV V VI VII VIII IX X XI XII XIII
Influence 10 7 9 8 5 7 2 4 8 8 8 9 10
1
Challenge 7 8 6 8 4 3 2 3 7 4 5 2 3
Influence 10 5 9 6 6 5 8 8 9 9 8 9 8
2
Challenge 8 1 7 4 6 2 3 2 9 8 8 2 4
Influence 10 4 8 4 4 5 2 3 7 7 7 9 5
3
Challenge 1 1 8 7 7 7 2 4 7 8 9 3 5
Influence 10 3 10 7 4 8 2 2 8 8 8 9 null
4
Challenge 7 1 2 8 6 4 2 3 7 7 8 1 6
Influence 10 5 8 3 4 3 2 5 9 9 9 10 9
5
Challenge 1 8 7 8 6 8 8 7 8 5 9 4 5
Influence 10 2 10 8 8 10 4 6 8 8 9 10 10
6
Challenge 6 2 8 3 7 null 4 5 7 5 9 3 3
Influence 10 5 5 5 4 2 2 6 9 9 9 9 null
7
Challenge 5 4 8 7 5 7 8 6 8 5 9 4 9
Influence null 1 8 6 6 9 2 3 5 6 5 7 7
8
Challenge null 2 9 9 5 9 9 8 4 4 5 8 9
Influence null 2 8 4 6 null 3 2 7 8 8 8 6
9
Challenge null 2 8 8 5 7 null 8 5 5 8 6 8
Influence 10 4 8 6 7 10 8 8 8 7 8 8 6
10
Challenge 1 6 5 3 6 4 2 3 8 7 8 6 2
Influence null null 8 6 7 9 null null 8 9 9 9 7
11
Challenge null null 9 2 9 4 null null 8 6 7 2 8
 Appendix 2 (1/4)

Appendix 2

Sensor data modelling

Below there are three figures visualizing the functionalities of attribute-based modelling of
measurement data and one figure demonstrating an interactive user view. The visualizations
were made with Microsoft Excel and Microsoft PowerBI.

In Figure 0.1 the multidimensional data cube has not been sliced in any dimension. On top
of the page there are functions that enable slicing the cube:
• All fact data (both room and equipment measurements) can be sliced in two
dimensions: Time and location.
• Room measurements can be sliced further in the dimension: Measurement type.
• Equipment measurements can be sliced further in two dimensions: Equipment and
Measurement type.
• It is also the dimensions that can be sliced with one another: Choosing a physical
location removes all the irrelevant equipment and vice versa.

In Figure 0.2 the cube has been sliced in the location dimension by choosing the room
H312B. This removes all the measurements that are not related to that room: For example
air handling unit 204 can be chosen but not 203. Also plug loads and lighting electricity
consumption measurements are only available for that portion of the floor.

In Figure 0.3 the cube has been sliced in the location dimension by choosing the room H123.
The data is further sliced with measurement types.

The dimensions can be hierarchized in many different ways but the idea is that the data
model itself contains information on what measurements are relevant in which physical
location, so there is no need to go through drawings when analyzing measurement data.
Appendix 2 (2/4)

Figure 0.1 Unsliced sensor data view. All fact data (both room and equipment measurements) can be
sliced in two dimensions: Time and location. It is also the dimensions that can be sliced with one
another: Choosing a physical location removes all the irrelevant equipment and vice versa.
 Appendix 2 (3/4)

Figure 0.2 Sensor data sliced with room 312B. The data model includes the information that it is the
air handling unit 204 serving that area.
Appendix 2 (4/4)

Figure 0.3 Sensor data sliced with the room H123. The data model includes the information that it is
the air handling unit 203 serving that area.
 Appendix 3 (1/2)

Appendix 3

Data Clustering

A Microsoft Clustering model was created with SQL Server Data Tools for SQL Server 2014
and SQL Server Data Mining Add-In for Office to group sensor data sets into clusters. The
model was trained with air handling unit measurement data between 1st January 2016 and
30th March 2016 that consists of

• Month,
• Weekday,
• Hour,
• Room air CO2 concentration,
• Exhaust pressure,
• Exhaust pressure set point,
• Exhaust fan,
• Inlet pressure,
• Inlet pressure set point,
• Inlet fan,
• Heat recovery,
• Supply air temperature,
• Supply air temperature set point,
• Air heater valve,
• Air heater return water,
• Air cooler valve,
• Air humidity and
• Intake air temperature.

The resulting clusters were named according to their favored occurrence times and outdoor
temperatures.

Then the model was given data from 11th April to produce four outputs: cluster, cluster
distance, cluster probability and model inclusion likelihood. There was an inserted
malfunction at 6.50am: The heat recovery disc stopped operating while the air heater was
operating. That row is bolded and underlined in Table 0.3.

At 6.50am the model inclusion likelihood drops to 0.08 indicating that the combination of
operating air heater and non-operating heat recovery disc is abnormal, which it is.
Appendix 3 (2/2)

Table 0.3 Air handling unit measurements in clusters.

Supply Air Intake Cluste Inclusio
Heat Air Heat Air r Cluster n
Room Recove Temperatu er Temperatu Dista Probabili Likeliho
Time CO2 ry re Valve re Cluster nce ty od
11.4.16 Temperate
4:40 430,7 0,0 19,4 0,4 10,3 Night 0,00 1,00 1,00
11.4.16 Temperate
4:45 427,5 0,0 19,4 0,7 10,3 Night 0,00 1,00 1,00
11.4.16 Temperate
4:50 430,1 0,0 19,4 0,0 10,2 Night 0,00 1,00 1,00
11.4.16 Temperate
4:55 430,4 0,0 19,4 1,1 10,3 Night 0,00 1,00 1,00
11.4.16 Temperate
5:00 428,6 0,0 19,4 1,9 10,3 Night 0,00 1,00 1,00
11.4.16 Temperate
5:05 433,3 0,0 19,4 3,2 10,3 Night 0,00 1,00 1,00
11.4.16 Temperate
5:10 431,2 0,0 19,4 8,6 10,3 Night 0,00 1,00 1,00
11.4.16 Temperate
5:15 431,5 0,0 19,4 16,3 10,3 Night 0,00 1,00 1,00
11.4.16 Temperate
5:20 427,0 0,0 19,4 22,9 10,3 Night 0,00 1,00 1,00
11.4.16 Temperate
5:25 418,7 0,0 19,4 8,6 10,2 Night 0,00 1,00 1,00
11.4.16 Temperate
5:30 418,7 0,0 19,4 0,1 10,2 Night 0,00 1,00 1,00
11.4.16 Temperate
5:35 431,2 0,0 19,4 0,1 10,2 Night 0,00 1,00 1,00
11.4.16 Temperate
5:40 430,1 0,0 19,4 0,0 10,2 Night 0,00 1,00 1,00
11.4.16 Temperate
5:45 421,3 0,0 19,4 0,5 10,1 Night 0,00 1,00 1,00
11.4.16 Temperate
5:50 426,9 0,0 19,4 0,5 10,2 Night 0,00 1,00 1,00
11.4.16 Temperate
5:55 425,5 0,0 19,4 0,9 10,2 Night 0,00 1,00 1,00
11.4.16 Temperate
6:00 429,1 0,0 19,4 1,9 10,1 Night 0,00 1,00 1,00
11.4.16
6:05 430,8 100,0 19,7 21,9 7,1 Early hours 0,00 1,00 1,00
11.4.16
6:10 423,4 100,0 20,0 21,1 5,8 Early hours 0,00 1,00 1,00
11.4.16
6:15 426,5 100,0 19,7 22,5 5,4 Early hours 0,00 1,00 1,00
11.4.16
6:20 429,6 100,0 19,6 24,6 5,1 Early hours 0,00 1,00 1,00
11.4.16
6:25 424,4 100,0 19,6 26,7 4,9 Early hours 0,00 1,00 1,00
11.4.16
6:30 424,4 100,0 19,7 28,6 4,9 Early hours 0,00 1,00 1,00
11.4.16
6:35 427,0 100,0 18,5 27,4 4,9 Early hours 0,00 1,00 0,65
11.4.16
6:40 424,1 100,0 17,9 36,5 4,8 Early hours 0,00 1,00 0,86
11.4.16
6:45 435,4 100,0 18,3 41,9 4,8 Spring evening 0,02 0,98 1,00
11.4.16
6:50 437,9 0,0 18,6 44,8 4,7 Spring evening 0,00 1,00 0,08
11.4.16
6:55 430,6 100,0 18,9 45,6 4,6 Early hours 0,08 0,92 0,51