IEEE Transactions on Power Systems, Vol. 9, No.

2, May 1994 957

A BOTTOM-UP APPROACH TO RESIDENTIAL LOAD MODELING

A. Capasso (*) IEEE Senior Member, W. Grattieri (**), R. Lamedica (*), A. Prudenzi (*) IEEE Member

(*)Electrical Engineering Dept., University of Rome "La Sapienza"

Via Eudossiana 18,00184 Rome, Italy
(**) ENEL-CREL (Italian Electricity Authority - Electrical Research Center)
Via A. Volta 1, Cologno Monzese, Milan, Italy

Abstract - A model of electric residential end-use is proposed for Over the last few years, Italy has given much attention to the appli-
establishing the load diagram of an area by a process of synthesis. cation of DSM policies in the residential sector. This has culminated
The model follows a "bottom-up" approach, allowing construction in recent legislation framed to permit residential customers to opt
of the relative load shape of the area, starting from knowledge of for a two-part tariff which provides for cheap power in off-peak
its most relevant socioeconomic and demographic characteristics, periods.
Unitary Energy Consumption (UEC) and the load profiles of indi- In order to arrive at a suitable method for evaluating the impact
vidual household appliances. Several probability functions have of DSM policies on residential customers, a simulation model
been introduced in order to cover the close relationship existing be- (ARGOS) has been developed following a "bottom-up'' ap-
tween the demand of residential customers and the psychological proach. Though such a methodology tends to be highly complex,
and behavioral factors typical of the household; the model makes it is extremely accurate, thus guaranteeing that the approach is
frequent use of the latter through a Montecarlo extraction process. adequately customer-oriented. Moreover, the implementation of a
The model has been applied for the simulation of a residential area validated customer response model, allowing to evaluate and predict
where field measurements of power demand had been made at 15- the impact of the different rate policies which may be implemented
minute intervals and a combined-mail survey had been conducted to within the framework of DSM programs, results more feasible.
investigate household energy usage. The good results obtained are attributable to the fact that the model
The paper reports the results of a comparison between recorded and allows reproduction of the load diagram of an end-use area by a
predicted load profiles. process of synthesis that permits differentiation between the behav-
ioral and engineering components of power consumption. The
synthesis involves aggregation of the load diagrams of each house
INTRODUCTION hold there, taking due account of the macroscopic socioeconomic
and demographic characteristics of the area concerned.
The development of models for ascertaining the load shape in end-
use areas is a cost-effective option for electricity utilities as it allows
reduction of load investigations to those needed for model tuning 1. OVERVIEW OF THE MODEL
and validation. This is especially true in the case of models which
also make provision for consideration of technological advances and Load shaping of residential end-users by a process of synthesis is a
behavioral evolution in residential energy usage. highly complex task, because household energy usage is intimately
Models provide extremely interesting possibilities where load s h a p linked to lifestyle-related psychological factors that are, of course,
ing is concerned, particularly as regards evaluation of the diverse extremely subjective and not easily defined with any degree of
Load Management (LM) options available to the utility within the precision. Nor does the definition of the standard behavior of the
framework of various Demand Side Management (DSM) programs, various types of customer through statistical correlations within the
as also indicated by Chan et al. [l]. framework of load-research fully resolve the problem, because it
Researchers operating in this sector have already produced several fails to consider the random variability of the demand. This aspect,
residential load-shaping models through the adoption of economet- however, is revealed by surveys covering a given type of family
ric-statistical, engineering-end-use or combined approaches [2,3,4], during the same day (cross-sectional analysis) and the same family
but there have been few attempts to broach the subject of the discre- over a period of several days (longitudinal analysis) [I]. The varia-
tionary nature of residential energy usage. bility factor, moreover, is inherent in analyses performed on end-
However Walker & Pokoski [5] have made a relevant contribution use areas of limited dimensions.
in this regard by introducing the concept of probability.functions In order to develop a model that takes account of all these require
such as "availability" and "proclivity" which are needed in order to ments, a load-shape synthesis approach based on the following two
assure correct defmition of psychological factors affecting energy levels has been adopted:
usage by residential customers.
Level 1 : aggregation of individual appliance demand so as to pro-
duce an individual household demand profile;
93 SM 498-6 PWRS A paper recommended and approved
by the IEEE Power System Engineering Committee of
Level 2: aggregation of the load shapes of various households so as
the IEEE Power Engineering Society for presentation
at the IEEE/PES 1993 Summer Meeting, Vancouver, B.C., to derive the end-use area load profile by synthesis.
Canada, July 18-22, 1993. Manuscript submitted Aug.
31, 1992; made available for printing May 7, 1993. In developing the model it was found essential to make provision for
consideration of the relevant psychological and behavioral aspects.
PRINTED IN USA This was achieved through logic rules, implemented in an automatic
procedure, thus permitting quantification of interrelations between
the various elements which constitute the essential basis of energy
usage by residential customers.

0885-8950/94/$04.00 0 1993 IEEE

Authorized licensed use limited to: MULTIMEDIA UNIVERSITY. Downloaded on August 29,2021 at 13:19:16 UTC from IEEE Xplore. Restrictions apply.
 958
The elementary units composing the residential customer model Extractions obtained by the Montecarlo method are introduced to
have been identified as: simulate random differences from the typical average daily behav-
iors assigned. In the proposed model the household load diagram
* appliance * household member thus ensues from each member activating one or more of the appli-
ances present in the home. Activation is considered to be deter-
Combination of these elementary units leads to the basic aggregation mined by heuristic rules which interrelate engineering and behavior-
referred to as "the household". al probability functions.
In other words, it is considered that an end-use area load shape is
determined essentially by interrelations between appliances and
members of the household. However such interrelations may pro- 2. MODEL STRUCTURE AND IMPLEMENTATION
duce different demand-profile results, depending on the geographic
area involved and the type of household concerned. In order to By means of the two-level synthesis procedure, the proposed model
reproduce the psychological factors which form the very basis of permits simulation of one of the probable load diagrams of a
domestic energy usage, the interrelations in question have been household to which specific socioeconomic and demographic char-
expressed as rules and constraints affecting certain functions which acteristics have been assigned. The resulting load profile is then
have been introduced in the model. These functions reflect the combined with those obtained from simulations on all the other
average tendency of members of the household to carry out various households in the area concerned. Each simulation step thus pro-
specific activities at home at given times of the day. vides an update of the demand as a running total. On a step-by-step
T h e functions introduced (generally of probabilistic basis, therefore, the results of the ongoing simulation for a house-
type) may be summarized as follows: hold having specific characteristics are also known in terms of
power, consumption and the contribution of each specific appliance
Behavioral functions to the demand profile.
A schematic representation of the model structure and implementa-
- Availability at home of each member of the household; a histo- tion (ARGOS procedure) is shown in Fig. l.
gram representing percentage values is assigned to the various
time intervals which make up the day. The time reference of this
function refers to such critical events as going out to school/work,
coming home, meal-times and bed-times.
- Home-activities;the various activities involving the use of electric I I
energy by members are grouped under four main headings:
housework, personal-hygiene, cooking and leisure. The per- Build main characteristics Build the oppliances set used
centage probability of their distribution during the day is given.
for the simulated day assign each appliance working/
- Appliance usage; appliances most frequently found in a household
(loop for m members of
are assigned a percentage distribution of usage during the day. the household)
- Proclivityfor home-activities; each member is assigned his or her
percentage availability to carry out each of the above home-
activities.
I
- Human resources; each member is assigned a batch of resources Build the household doily
such as eyes, ears and hands which are expressed symbolically in
terms of percentage. The ensuing values are useful for quantifying
the possibility of each member using more than one appliance at
Aggregate household load
the same time (the function is introduced to exclude - for example profiles to build load shape
- the simultaneous use of a flat-iron and a hair-drier, while permit- of end-use area
ting the simultaneous use of the latter and a radio or TV). (loop for n households)

- Appliance ownership; each type of household is assigned a param-

eter indicative of a given set of appliances available. This parame- Fig. 1 -Schematic flow-chart of model implementation (ARGOS).
ter depends on certain features that are characteristic of the type
of household, such as assumed income, number of members, 2.1 Specit%data-fde
socioeconomic characteristics of the relevant end-use area, etc..
All the data needed for characterization of the elementary units
As an alternative to the histograms the functions availability at home (namely appliances and members of the household) are contained in
and home activities of the various types of household members can a specific data-file constructed by means of a batch procedure, or an
be represented by normal type probability distributions. In this case automatic preprocessing procedure. The file must contain the fol-
functions are described in terms of time instants of home "in/outgo- lowing parameters regarding appliances and household members:
ing" (or "startlend" activity) and of duration of the corresponding
availability at home (or activity) (see also Para. 4). appliance
. recognition label which connects appliance usage to the pertinent
Engineering functions home-activity (cooking, housework, personal-hygiene, leisure)
.'mode of operation, which may b e completely automatic
- Appliance mode of operation: (e.g.refrigerator, freezer) or call for the use of human resources.
.Cycle or activation time As regards this last mode of operation, the procedure further
.Power demand (depending on the type, size or model) distinguishes between activation and usage (e.g. hair-drier) on the
*Averageannual consumption one hand, and activation only (e.g. dish-washer, clothes-washer)
- Household demand limit (fmed by contract, in Italy) on the other
- Technological penetration; saturation of appliances embodying . power demand limits (minimum, maximum)
any technological innovation. . workinglduty cycle typical duration

Authorized licensed use limited to: MULTIMEDIA UNIVERSITY. Downloaded on August 29,2021 at 13:19:16 UTC from IEEE Xplore. Restrictions apply.
 959
. usage profile (not essential because if not explicitly assigned it is that concerning the activation, by the i-th member, of an appliance
replaced by the hourly profile of the corresponding homeactivity) whose operation, characterized by a power requirement and a lapse
. unitary energy consumption (UEC) of time, is compatible with the residual power and energy. A more
* penetration detailed mathematical description of the model is in the Appendix.
. probability of daily usage Fig.2 illustrates schematically the two main phases through which
* human resources engaged in usage (eyes, ears and hands as per the model builds up the load diagram of an individual household.
centages) The first, concerning the search for time intervals in which there is
* deferrability (possibility of shifting the usage in time) at least one person at home, starts from an average seasonal profile;
* working/duty cycle of certain appliances forming the base load of in order to derive the daily profile of the simulation, the procedure
the household's global demand diagram (see 8 2.2) executes a Montecarlo extraction phase on the times of entering and
leaving home, having considered a probability distribution within
household member the range of time intervals covering a reasonable variability toler-
. type (adult male, adult female, boy, girl) ance (graph AT=f(t) of Fig.2). The intersection of the profile of the
. income (high, medium, low) presumed on basis of job presence of members at home (by way of example, F i g 2 indicates
* availability at home profile that for kitchen activities, Sj= f(t)) identifies the time intervals
* proclivity for home-activities (assignment of a percentage of the during which the i-th member could activate one of the appliances
total availability daily at home) associated with the activity in question.
The latter hourly profile is multiplied by a coefficient (Y. This takes
In order to aggregate the elementary units in households, it is also account of the capacity of the member to perform the relevant activ-
necessary to assign the following data which serve to characterize ity and also of whether that activity can be performed, namely, that
the different households present in the area simulated: the person concerned is not doing something else during the same
period of time. The last profile indicated in the first row of Pig.2,
i. number and type of household members Tij=f(t), is obtained in this way.
ii. number of such households in the area Regarding appliance selection, the procedure extracts from a list
iii. average electricity consumption of household which covers the average range of appliances present in the house
iv. appliance ownership hold simulated, taking account of the daily-use probability assigned
to each of these. The range reflects the data concerning appliance
The data for Parameters i and ii can be assigned either by a deter- penetration as a function of the socioeconomic conditions derived
ministic procedure or on the basis of demographic statistics. from ISTAT Consumption Tables (see paragraph 3).
The procedure extracts the duration and power requirement of the
2.2 Composition of household load diagram activated appliance's working-cycle, or its duty-cycle in the case of
an automatic cycle (e.g. dish-washer or clothes-washer).
The ARGOS procedure reconstructs a probable load diagram of a After having ascertained the energy necessary for the appliance to
household with given characteristics, through the aggregation of the operate, the procedure indicates the activation thereof on the T, pro-
load diagrams of the individual appliances utilized during the day. file, where possible.
The procedure is designed to attain two distinct objectives. In the Fig.2 example the only position possible on the final P(t)
The first is that of identifying those time intervals when the follow- diagram is between 18:OO and 22:30 hours; a Montecarlo extraction

1; ,
ing conditions exist contemporaneously: at least one member of the procedure is employed to select the time of activation.
household at home, and the possibility and wish of that member to In the hour considered for the simulation, the allocation follows
activate one of the household appliances. The second objective is general criteria, favoring the percentually-predominant activity.

~ ; ' ; p ~ ~ ~ p:Ai-th
levoi ~
household member from
*'
h~; ,o h
; member i ~the ~simulated
for ~ , ~:day~; ~t ;$mj-th:
doily
o hprofile ; o f ; c r u ; ; u hhome-activity j - i
averoge typological
profile c3'
' 0 4 8 12 16 20 24h 0 4 8 12 16 20 24h0 4 8 12 16 20 24h

PM1
Extraction of k-th Pllrnlt
appliance related to Presidual
j-th home-activity Ek=energy consumed Pappl'ance
from appliances by k-th appliance 0
doto set working cycle
4h
"L"J
d=durotion of usage
PW1 t
Plirnit
Presidual

Pdernand

Fig. 2 -Simplified schematic of appliance load diagram allocation procedure for construction of household load diagram for simulated day.
Schematic refers the case of i-th member of household using electric oven.

Authorized licensed use limited to: MULTIMEDIA UNIVERSITY. Downloaded on August 29,2021 at 13:19:16 UTC from IEEE Xplore. Restrictions apply.
 960
Activation probabilities are assigned by criteria that limit the aver- TABLE I: Geographical breakdown of household size (5%).
age daily energy within tolerance bands around the values indicated
by national statistics. Number of Northerr C e n t r a l 3out h e r n
members Italy Italy Italy
2.3 Composition of load diagram of entire simulated area
1 7.5
The procedure described in 2.2 is repeated for all the households in 2 16.3
the end-use area. 3 27.8 26.4
The model provides a wide range of computational and intermediate 4 33.7 28.2 34.0
processing options for simulation of an entire end-use area. Simula- 5 11.1 17.3 20.8
tion results are reproduced in the form of output tables or diagrams 2 6 4.4
referring to:

- daily load diagrams calculated at preassigned intervals (e.g. fif- TABLE 11: Geographical breakdown of male proclivity for home-
teen minutes) of: activities such as housework and cooking (5%).
-entire area (average per user demand)
-individual household
-household types Italy Italy Italy
-individual appliance (its contribution both to the individual
household load profile, and to the load profile of the entire area) Often
-individual home-activity (its contribution to the load profile of
the entire area). Never 27.8 26.4 16.0
- energy consumption of individual appliances, individual home-
activities, individual and global household types
- partial and global energy consumption expenses evaluated on the diverse activities performed during the daily life of a household,
basis of the various applicable tariffs. thus providing all the necessary data concerning psychological and
behavioral factors affecting energy usage at home.
By way of explanation, Table 11 reports data on the participation of
3. DATA SOURCES AND DEVELOPMENT an employed male in housework and cooking. On the basis of simi-
lar data has been possible to outline average profiles on the proclivi-
The data needed to simulate the load shape of a residential area can ty to engage in various home-activities.
be broken down into the following classes: As things stand at the moment, the procedure still do not permit
breakdown of the data on the basis of calendar time
Demographic and socioeconomic data (weekday/weekend, wintedsummer).
. Data concerning the lifestyle of the persons concerned and their ISTAT has been running its Household Consumption Survey con-
electricity-usage habits tinuously for some years now. This provides the main characteris-
. Engineering data on operation of the relevant household appli- tics on the national-user's electric energy consumption and, more
ances. especially, the relationship of this with socioeconomic, demographic
and regional conditions [I.
These data have been utilized to establish the "Behavioral functions"
and "Engineering functions'' already mentioned.

Demonrauhic and socioeconomic data

These data are available from statistical studies conducted within the
framework of nationwide demographic and social surveys made Income c l a s s High Middle Low
periodically by organizations such as ISTAT (Italian Central Insti-
tute of Statistics) and CENSIS (Social Investment Studies Center). Northern Electric [$I 8 7 5
They permit very detailed characterization of social and demograph- Italy Total [ $ I 737 637 486
ic aspects. Regarding the latter, for example, Table I reports the
data on the breakdown of families by size in north, central and Central Electric [ $ ] 10 7 6
southern Italy. Italy Total [ $ ] 639 573 322

Customers' lifestvle and habits Southern Electric [ $ ] 9 7 6

The two main sources of such data are: Italy Total [ $ I 557 452 312
Survey of Use of Time in Italy
+hrvey of Household Consumption
Table 111 gives electricity consumption, with breakdown by pre-
The Time Use Survey was conducted by ISTAT within the context sumed income and geographic area.
of a Multipurpose Survey covering a sample of about 40000 persons
between June 1988 and May 1989. The persons concerned were Ennineerinp data on auvliances
asked to keep a diary listing all their activities during a given day, These data stem mainly from experimental investigations made by
and indicating in each case: a) starting and finishing times; b) brief constructors and organizations such as ENEL (Italian Electricity
description of the activity; c) the place concerned; d) the person(s) Authority), ENEA (National Institute for New Technologies,
present during the activity; e) other activities performed at the same Energy and Environment) and CESI (Italian Experimental Electric
time, if any [6]. Center) [8]. The electrical aspects of the workinglduty cycle of the
Such surveys, of course, permit description and quantification of the commonest household appliances have been characterized on the

Authorized licensed use limited to: MULTIMEDIA UNIVERSITY. Downloaded on August 29,2021 at 13:19:16 UTC from IEEE Xplore. Restrictions apply.
 961
basis of the results, which also indicated the main technological The program can access this set both during automatic compilation
innovations designed to favor energy-saving. of the specific data-file and during the construction of the load
Some of the data utilized are as follows: diagrams for the individual households and the entire area.
- UEC of each appliance grouped by sue, brand, etc. The data needed for application of the model to specific, limited
- Typical setting-dependent load diagrams for each appliance end-use areas so that they can be characterized in order to optimize
- Average annual consumption of each appliance class (dish- existing plants or to plan plant resources in developing areas can be
washer, refrigerator, etc.) supplemented by field surveys (census of household energy-use
- Technological characteristics and innovations habits in the area), and records of hourly load diagrams.
The model has been applied to a residential customer area including
The residential electric load in Italy has some peculiar characteris- 95 households grouped in four buildings in the outskirts of Milan, in
tics such as: order to run certain checks on the validity of the simulation results
a) a fmed contractual demand limit for customers (3 or 6 kw); for a limited end-use area.
b) very low levels of penetration ( < 2-3 W ) of weather-dependent Recordings have been made of the global demand profile measured
appliances such as air-conditioners and ventilators; every quarter of an hour on a daily basis and over a one-month
c) home heating provided almost entirely by natural gas or other period during the winter season. A combined-mail survey, based on
fossil fuels. a questionnaire to which 70% of the households replied, has also
The model still does not include any data concerning residential been conducted in the same area in order to investigate the demo-
buildings or the average daily temperature profile. graphic characteristics of households as well as the main aspects of
Moreover, no thermal models have yet been implemented for elec- energy usage at home.
tric water heaters and refrigerators, whose duty-cycle is also partial- Probabilistic distributions gathered from answers to the question-
ly influenced by climatic conditions. While awaiting the validation naire, together with those resulting from regional statistical surveys
of available thermal models through field tests, which are under and load-research have been fed into the model in the form of
way, reference has been made to average load diagrams obtained behavioral and engineering probability functions. Fig. 3 illustrates
from measurements of a large sample of such appliances produced the probabilistic profiles of the different home-activities, as deduced
by the main manufacturers. from elaboration of customers' answers. Further details of the
The "bottom-up'' approach used by the model provides a large de- survey have been given by Capasso et al. [9].
gree of flexibility when inserting in the data set appliances which
ercent (%)
incorporate new manufacturing technologies and also new consump- 1 oc 100
tion models as determined by load research.

4. CASE STUDY 8C

The specific data-file forming the basis of the simulation of a resi-

dential end-use area can be compiled specially for the study of an 60
area where the data listed in Q 2.1 are available, or they can be
obtained automatically, with assistance from the program user.
In the latter case the user must assign certain macroscopic character-
40
istics, such as:
* day to be simulated

. geographic location of area

. number of households simulated 20
. presumed income band.
The socioeconomic conditions of the area are included in these
macroscopic characteristics because they affect the consumers' life
styles, of course, and hence determine different presences at home 0 4 8 12 16 20 24h
-B- tiousowork -+ cooklng
and different consumption patterns. They also serve to assign labels
0personal hygiene leisure time
to the various prepared lists of appliances. The lists themselves
indicate the appliances possessed by the various households in Fig. 3 -Home-activities profiles of surveyed households
presumed income band.
The macroscopic characteristics assigned by the user permit the The performance of the model for the prediction of recorded load
execution of a preprocessing procedure to select certain sets of data profiles has been evaluated by comparison of simulated and meas-
relating to the desired simulation. They thus allow production of the ured power demand on a daily, weekly or monthly time basis.
specific data-fde. Fig. 4 illustrates the variability range of recorded and simulated
To facilitate the search for the residential load and the analysis daily load profiles obtained for each of the 22 working days within
thereof, the model is also implemented complete with certain utility the period examined.
procedures contained in an extraction and aggregation rules set. Fig. 5 compares recorded and simulated average load profiles on a
As the instructions representing these rules are in a meta-prograrn- monthly and weekly time basis.
rning simplified language, the user has no difficulty in rapidly The following Normalized Variation Factor (NVF) can be used to
modihing the heuristic rules which go to make up the electric load produce a numerical evaluation which determines the extent to
profile. Such modifications can be extremely useful if the imple- which the simulations refled the recorded data:
mented behavioral and engineering probability functions have to be
changed, following the acquisition of a new correlation law result- Ci@redi - reci)2
ing from Load Research and econometric studies or application of NVF = (3)
the model to new typological and functional realities. n.reco2

Authorized licensed use limited to: MULTIMEDIA UNIVERSITY. Downloaded on August 29,2021 at 13:19:16 UTC from IEEE Xplore. Restrictions apply.
 962
where: The results obtained for the maximum calculated values of weekly
predi = predicted demand at i-th time step and monthly averages respectively are:
reci = recorded demand at i-th time step
rec,, = average recorded demand NVF = 0.0124, NVF = 0.0099 (6)
n number of time steps in a day
XlW xi04 The calculated values reveal the "time averaging" effect of the
simulations, thus accounting for the steady reduction in the NVF
i 71-
value calculated on a daily, weekly and monthly basis, as also
reported by Walker & Pokoski [5].
6 6

-
g 4
5
-.-
g 4
5

1m w Male-employed €3
Availability at home
3 3 I m ~ , H a u s e - w i f a'(present and awake
2 2

1 1
clothes-washer
dish-washer
:0 5 lo 15 20 ' tv
a) time (hour) b) time (hour) liohting
Fig. 4 -Daily load profiles of 22 winter working days (month) as miscellaneous (leisure)
obtained from: a) recordings; b) simulation. miscellaneous (cooking)

refrigerator (B)
refrigerator (BQ

Fig. 6 -Predicted daily demand profile for a household.

[U1
700
I I I I I

0' 5 10 15 20
a) time (hour) b) time (hour)

Fig. 5 -Average load profiles on: a) weekly and b) monthly time

basis. Comparison between (-) recorded and (--) predicted
per household values.

Several NVF numerical evaluations have been made by assuming

that the recordings and the simulations fit sufficiently well, provided
that differences between predicted and recorded values fall within
the variation range of the load profiles recorded on different days.
Pairs of different days of recordings were compared in order to 1
calculate the NVF value. The resulting variability range of the 11

coefficient was: Fig. 7 -Predicted average (per household) daily load profile of the
simulated area and contribution of appliances related to
0.0801 2 NVF 2 0.0220 (4) housework (clothes-washer, dish-washer, miscellaneous).

The same evaluation was also made by comparing day-to-day varia- The possibilities guaranteed by the model to indicate the individual
tion of predicted versus recorded values. The result was as follows: contributions to the simulated-area load diagram, listed in 8 2.3,
permit precise estimations of indices such as "coincidence factor"
0.0527 2 NVF 2 0.0094 (5) and "load factor", which are particularly useful for analyses of
system peak load and DSM studies [9,10].
It is evident from comparison of (4) and (5) that the model provides Fig. 6, for instance, illustrates the results of the daily load profile
a good approximation of the daily recorded load diagram. The simulation of a household with certain characteristics. It shows:
highest value of the variability range for the daily simulations indi- - breakdown of time spent at home by each member of the family,
cated in (3,appears to fall well within the variability range of the derived from the extraction stage of 8 2.1;
recordings referred to in (4). - list of appliances being activated during the day;
NVF values have also been calculated in order to compare record- - total daily load shape.
ings and simulations on a weekly (5 working days) and monthly (22
working days) basis. Fig. 7 illustrates the contribution to the average global load diagram

Authorized licensed use limited to: MULTIMEDIA UNIVERSITY. Downloaded on August 29,2021 at 13:19:16 UTC from IEEE Xplore. Restrictions apply.
 963
on a wintertime weekday of the demand stemming from all the applications, in view of a future likely increase of penetration in the
appliances connected with housework, as predicted by simulation. Italian residential sector; b) determination - through field measure-
Starting from the observed distribution given by the histograms, ments and experimental tests - of a model of customer response to
(see for instance home-activity diagrams of Fig. 3) a check was Utility price signals to be-introduced in order to evaluate the impact
made on the fitting possibilities offered by normal probabilistic of various Load Management policies (Time of Day, Time of Use
distribution. tariffs).
Mean values and standard deviations (expressed in terms of time)
were evaluated on the basis of the frequency distributions surveyed,
in order to identify characteristic parameters of the corresponding APPENDIX:Mathematical description of the model
probability density function.
A good fitting of the statistical data with normal distribution func- List of svmbols
tions was not always found to be possible. However such a distribu-
tion was systematically adopted, because the greatest differences j: home activity index;
were detected in the case of distributions with small numbers of i: household member index;
samples. Research is scheduled to explore the application of differ- 1: human resource index (1= 1t3)
ent standard probability functions to non-normal distributions. Til: resource of the i-th member;
For instance, Fig. 8 illustrates the comparison between the availabil- pij: proclivity of i-th member for j-th home activity;
ity at home frequency distribution of a student (histogram), as ob- Ai: availability at home daily profile of i-th member (average for
tained from the answers to the questionnaire of the survey, and each member type;
those simulated assigning in input Gaussian probability functions A;: availability at home daily schedule of the i-th member for the
having several standard deviation values. simulated day;
Sj: j-th home activity daily profile;
Pmt: contractual power demand limit of individual household;
k: mdex of appliance activated during day simulated;
P-,:minimum and maximum power demanded by the k-th appli-
ance;
d-,:miniium and maximum duration of dutylworking cycle of
the k-th appliance;
d,: duration of use of the k-th appliance;
Dg usage daily profile of the k-th appliance.
Fig.8 -Availability at home diagrams. Comparison of probabilistic The daily load diagram of the user is simulated without taking
functions and frequency distribution surveyed (histograms). account of the chronological sequence during the day of the various
household-memberlapplianceinteractions, but instead locating the
use of each appliance in the different time intervals when the condi-
5.CONCLUSIONS tions necessary for its usage occur (see also paragraph 2.2).
By extraction on the relative average probabilistic profile (Ai), each
A model based on a "bottom-up'' approach is proposed for establish- member is assigned an hourly histogram which is representative of
ing the load profile of a residential area by a process of synthesis. hidher availability at home on the specific day simulated (A*i).
The following characteristics have to be provided in order to Starting from the hourly probability profile of the various activities
achieve accurate simulation of the parameters considered fundamen- (Sj), the following is determined for each member:
tal for residential load research:
- Synthesis procedure involving aggregation on a two-level basis: a) Tij = (A: fl Sj).ailj
aggregation of the contribution made by individual appliances to where:
the household demand profile; b) aggregation of the various rele- cyilj = ri 'p..
vant household profiles, so as to arrive at the demand shape of the
entire area; In this manner various daily profiles are determined, one for each
- Introduction of "behavioral" and "engineering" probability func- resource and for each activity (see Fig. 2). The human resources
tions in order to reproduce the psychological factors affecting voluntarily made available by the entire household are thus:
residential demand;
- Adoption of a Montecarlo extraction process applied to average Tlj = CiTij (A.3)
behavior and characteristics represented by the above functions,
so as to take account of the highly random variability of the Each appliance k is ascribed a load diagram by probability extrac-
demand. tion of the installed power and the duration of operation, within the
The proposed model has been used to simulate a residential area, average characteristic ranges (P-, ,,P d-, d-). The total
where load recordings and a survey targeted on each individual energy absorbed during the day is:
household were carried out simultaneously. The results are encour-
aging. E, = P;d,
Because of the very flexible nature of the model stemming from its
bottom-up approach and structure, repeated changes can be made to The time profile of the resources necessary for using the k-th appli-
the behavioral and engineering functions as required, in order to ance activated is:
reproduce end-use variations resulting from technological innova-
tions affecting appliances and the socioeconomic evolution of IU = riledk
households.
Aspects of the model needing further development are: a) imple- Indicating with:
mentation of weather-dependent appliances models, such as HVAC Emt = Pmt.d,

Authorized licensed use limited to: MULTIMEDIA UNIVERSITY. Downloaded on August 29,2021 at 13:19:16 UTC from IEEE Xplore. Restrictions apply.
 964
if: computer simulation of an electric utility load shape, IEEE
E, < Ewt Transactions on Power Apparatus and Systems, Vol. PAS-100.n.
4, January 1981.
the load cycle and hence the energy of the k-th appliance can be [4] EPRI, Combining engineering and statistical approaches to
located in a time interval of the simulated day. The residual energy estimate end-use load shapes, %l. 2: Methodology and results,
available for usage of the remaining appliances is: Report EA-4310, Electric Power Research Institute, Oct. 1985.
[5] C.F. Walker, J.L. Pokoski, Residential load shape modeling
Em,, = E,,-E, based on customer behavior, IEEE Transactions on Power
Apparatus and Systems, Vol. PAS-104. n. 7, 1985.
hence, the time intervals (S,) of possible location of the successive [6] ISTAT, Time Use Survey, Notiziario n. 10, 1990 (in Italian).
appliances are limited to those for which: [A ISTAT, Household consumption: year 1987, vol. 14, 1989 (in
Italian).
[8] ENEA, Domestic appliances energetic consumptions, vols. 7-
10, 1992 (in Italian).
Some privileged time bands are identified for using the appliance k. [9] A. Capasso, W. Grattieri, A. Invernizzi, F. Insinga, R.
This is done on the basis of the usage daily profile (D,,if Lamedica, A. Prudenzi, Wlidation tests and applications of a
assigned), or on the basis of the daily profile of the activity j with model for Demand Side Management studies m residential bad
which the appliance service is connected. The profile of the re- areas, CIRED 1993, Birmingham (UK), May 17-21, 1993.
sources made available within the whole household for the k-th [10]EPlU, Selected statktical methorls for analysis of load research
appliance activation is then: data, Report EA-3467, Electric Power Research Institute, May
1984.
F~~= ( c ~ T ~n~D,
)
or: (A. 10)
F,~= (ciT ~ n~ sj) BIOGRAPHIES

If it ensues that: Alfonso Cauasso (M.72, SM.'92) was born in Padua, Italy, on
Ilj C F,j. (A.ll) January 24, 1945. He graduated in electrical engineering from the
University of Rome, Rome, Italy, in 1968. He then joined the
the daily time bands utilisable for time allocation of E, are identified Faculty of Engineering, University of Rome "La Sapienza", and
through the following time profile: was appointed Assistant Professor in Electric Power Systems in
1969, Associate Professor in 1973 and Full Professor of Electric
H, = s, n F~~n rlj (A. 12) Power Systems in 1980.
His main interests are in computer applications to Electric Power
A Monte Carlo extraction on the probability profile Hljidentifies the Systems and in Railway Electrification.
definitive time allocation. Prof. Capasso is a member of the AEI (Italian Electrical Associa-
The procedure continues with the updating of the profiles utilized tion), the CEI Committee 110 (EMC), the Italian Committee of
until the last extracted appliance is located. CIRED and CIFI (Italian Railway Engineers Assoc.).
The model considers the refrigerator and electric water-heater (the
latter in continuous-operation mode) as base loads in the house- Walter Grattieri was born in Romano di Lombardia, Italy, on
hold's global load diagram. Hence the working cycle of these appli- November 9, 1954. He graduated in Electrical Engineering from the
ances is broken down into two distinct contributions: Polytechnic of Milan in 1978. He joined the ENEL Electrical
Research Center, in Milan, in 1980. In 1990 he was a fellow of the
Pb = C, + BC, (A. 13) Advanced Study Program at the MIT's Center for Advanced Engi-
neering Study.
C, takes account of the power consumption resulting from automat- His main interests are in the field of Load Forecasting, Distribution
ic, thermostatically-controlled cyclical activation which occurs Planning and Demand Side Management.
throughout the day to maintain the temperature set for the appliance, He is a member of the IEA working group for the creation of an
while BC, takes account of power consumption resulting from use Implementing Agreement on Technologies and Programmes for
of the appliance by a member of the household (opening the refrig- Demand Side Management.
erator, or turning on a hot-water faucet).
The load diagram of each of the k appliances activated is indicated Regina Lamedica was born in Foggia, Italy, on December 6. 1952.
by P, and the total load diagram of the household by P(t). so: She graduated in Electrical Engineering (Hons.) from the University
of Bologna, Italy, in 1976. From 1976 to 1986 she was Assistant
P(t) = E, P, + E, P, (A. 14) Professor in Electric Power Systems at the University of Rome "La
Sapienza",Italy and since 1987 has been Associate Professor in
Electric Power Systems for Transportation.
REFERENCES Her main interests are in computer applications to Electrified
Transportation Systems and to Power Systems Analysis.
[l] M.L. Chan, E.N. Marsh, J.Y. Yoon, G.B.Ackerman, N . She is a member of CIFI (Italian Railway Engineers Association).
Stoughton, Simulation-based load synthesis methodology for
evaluating load-management programs, IEEE Transactions on Alberto Prudenzi (M.'91) was born in Rome, Italy, on December
Power Apparatus and Systems, Vol. PAS-100, n. 4, April 1981. 10, 1961. He graduated in Electrical Engineering from the Universi-
[2] J. Broehl, An end-use approach to demand forecasting, IEEE ty of Rome in 1987. Since 1992 he has been Assistant Professor in
Transactions on Power Apparatus and Systems, Vol. PAS-100, Electric Power Systems.
n.6. June 1981. His main interests are in computer applications in Power Systems.
[3] C.W. Gellings, R.W. Taylor, Electric load curve synthesis - A He is a member of AEI (Italian Electrical Association).

Authorized licensed use limited to: MULTIMEDIA UNIVERSITY. Downloaded on August 29,2021 at 13:19:16 UTC from IEEE Xplore. Restrictions apply.