Sustainability 16 00667
Sustainability 16 00667
Sustainability 16 00667
Article
Research on Optimized Design of Rural Housing in Cold
Regions Based on Parametrization and Machine Learning
Minghui Sun, Yibing Xue * and Lei Wang
School of Architecture and Urban Planning, Shandong Jianzhu University, Jinan 250100, China
* Correspondence: [email protected]
Abstract: With the rapid development of urbanization, the emergence of more self-built buildings in
the countryside has brought about energy waste problems and decreased comfort. Achieving the
low-carbon goal and improving the quality of the human living environment through architectural
and planning means have become vital issues. In this study, from a parametric perspective, model
building and performance simulation are carried out using Rhino and Grasshopper, and a multi-
objective optimization method and a neural network model are used as the theoretical basis to
train the prediction model after data collection and processing. The model validation of R2 = 0.988
and MSE = 0.0148 indicates that the model can accurately reflect the program’s performance. By
establishing a rapid prediction model for the performance of rural residential buildings, decision-
makers can perform performance predictions under various parameter combinations at the early
design stage, facilitating the screening of building types with high energy consumption and costs. The
method can improve the efficiency of decision-making at the early stage of design, help save decision-
making costs by screening high-energy-consuming building types, improve the living conditions
of residents, reduce carbon emissions, and contribute to the sustainable development of residential
building renewal design in rural areas.
lifestyle and between the architectural appearance and the aesthetic norms [4]. On the
one hand, this cannot meet the residents’ demand for the use of residential space, and
on the other hand, it also makes the residents increasingly dependent on the indoor air-
conditioning system. Therefore, it is crucial to accelerate the scientific development of rural
residential architecture by scientific renewal and the construction of rural residences.
retrofit of buildings is to find the best combination of energy-saving measures among many
strategies. At the same time, energy-saving retrofit measures interact, meaning that a partic-
ular measure may significantly affect another measure or even achieve the opposite effect.
Thus, simply and linearly adding up the effects of individual measures is not adequately
accurate; hence, the decision-maker will face a multi-objective optimization problem, which
describes multiple competing objective functions to evaluate feasible alternatives [18].
With the continuous development of computer technology, an increasing number of
scholars have investigated the application of multi-objective optimization algorithms in
building energy efficiency. Some researchers focus on studying the optimization problem
regarding computational and building design parameters. For example, Gao Yuan et al. [19]
combined the SPEA-2 optimization algorithm in Rhino-Grasshopper with Energy Plus to
carry out multi-objective optimization design for multiple parameters such as orientation,
depth, window-to-wall ratio, etc., for country houses. Vukadinović et al. [20] used Design
Builder software (2014) to optimize the window-to-wall ratio of passive houses and glass
type, and other parameters were optimized and iterated using Latin hypercube sampling
to determine the heating and cooling energy consumption of the two design scenarios.
In addition, numerous studies on parameters such as building envelope and equipment
systems exist. For example, Fabrizio et al. [21] optimized thermal performance parameters
such as wall thermal inertness, insulation thickness, and external window transmittance
for a Mediterranean residential building envelope with the objectives of primary energy
consumption and thermal discomfort hours. Carlucci et al. [22] combined the GenOpt
optimization engine with the Energy Plus simulation engine to perform a multi-objective
optimization of a net-zero-energy Southern Italian residence for the multi-objective opti-
mization of thermal discomfort in winter and summer and visual discomfort from glare
and the amount of daylight.
paper investigates the impact of architectural design parameters and envelope structure
parameters on the residence’s energy efficiency, comfort, and economic viability. It proposes
a method for the rapid prediction of building performance. By integrating MOO (multi-
objective optimization) and ML (machine learning), this method establishes a predictive
model for forecasting multiple objectives. It helps architects predict the performance of
rural homes during retrofits and upgrades, provides decision support in the early stages
of building design, and reduces decision costs. By enhancing the economic benefits of
energy-efficient design as well as the living experience of residents, the goal of sustainable
development is achieved in both economic and social dimensions.
Figure
Figure 1.
1. Overall
Overallworkflow.
workflow.
Figure2.2. Aerial
Figure Aerial view
view of
of the
thelayout
layoutof
ofDongquan
DongquanVillage.
Village.
In order
In order to to improve
improve thethe efficiency
efficiency and and accuracy
accuracy of of the
the field
field survey,
survey, the
the village
village layout
layout
andthe
and theform
formofof building
building layout
layout were
were firstfirst analyzed
analyzed basedbased
on theon the current
current status ofstatus of the
the village
village
plan. In plan. In thehouses,
the village villagethehouses, the main
main house mainlyhouse
has mainly has the
the primary primary
function function
of living as the of
livingfunctional
main as the main functional
space; space; itsform
its architectural architectural form
is a regular is a regular
rectangle; rectangle; the com-
the compartments and
partments
inverted and inverted
houses mainly havehouses mainly
storage andhave storage
other and functional
auxiliary other auxiliary functional
spaces. spaces.
By integrating
By integrating
the the existing
existing village village
layout plans andlayout plans and
conducting conducting
actual surveyingactual surveying
in Dongquan in Dong-
Village, we
quan Village,
collected we collected
dimensional dimensional
and structural dataand
forstructural
various typesdataofforbuilding
variouscourtyards.
types of building
After
courtyards.
assessing theAfter assessing
research data of the research
183 data of
farmhouses in 183 farmhouses
the village, in thethat
we know village, we know
the houses in
that the houses in the village are mainly categorized into “—”, “L”, “U”, and “□” shapes.
the village are mainly categorized into “—”, “L”, “U”, and “ □ ” shapes. The four types of
The four
houses intypes of houses
the village in the“L”,
are “—”, village
“U”,are “—”,
and “□”,“L”,
as “U”,
shown andin“□”, as shown
Figure 3. The in Figureof
number 3.
The numberand
“U”-shaped of “U”-shaped
“Mouth”-shaped and “Mouth”-shaped dwellings
dwellings is the highest, is the highest,
accounting accounting
for 36.1% and 34.4%,for
36.1% and 34.4%,
respectively, followed respectively,
by “L”-shaped followed by “L”-shaped
dwellings dwellings atdwellings
at 23%. The “—”-shaped 23%. Theare “—”-
the
shaped dwellings are the least, only 6.6%. Therefore, the objectives of this study are mainly
least, only 6.6%. Therefore, the objectives of this study are mainly the following:
the following:
(1) Type 2: “L”-shaped courtyard house.
(2) Type 3: “U”-shaped courtyard house.
(3) Type 4: “□”-shaped courtyard house.
courtyards. After assessing the research data of 183 farmhouses in the village, we know
that the houses in the village are mainly categorized into “—”, “L”, “U”, and “□” shapes.
The four types of houses in the village are “—”, “L”, “U”, and “□”, as shown in Figure 3.
The number of “U”-shaped and “Mouth”-shaped dwellings is the highest, accounting for
Sustainability 2024, 16, 667
36.1% and 34.4%, respectively, followed by “L”-shaped dwellings at 23%. The 6“—”- of 19
shaped dwellings are the least, only 6.6%. Therefore, the objectives of this study are mainly
the following:
(1) Type 2: “L”-shaped courtyard house.
(1) Type 2: “L”-shaped courtyard house.
(2) Type 3: “U”-shaped courtyard house.
(2) Type 3: “U”-shaped courtyard house.
(3) Type 4: “□”-shaped courtyard house.
(3) Type 4: “□”-shaped courtyard house.
Figure 3.
Figure 3. Schematic
Schematic diagram
diagram of
of rural
rural residential
residential building
building forms.
forms.
An analysis
An analysis ofof survey
surveydata
datahas
hasrevealed
revealedthat
thatthethescale
scaleofofcourtyards
courtyards is is primarily
primarily de-
deter-
termined
mined by by
thethe dimensions
dimensions of the
of the mainmain house
house andandthe the
wing wing rooms,
rooms, which,
which, together
together withwith
the
the surrounding walls, form the courtyard. The variation in width primarily
surrounding walls, form the courtyard. The variation in width primarily reflects differences reflects dif-
ferences
in in the functional
the internal internal functional layout.
layout. With theWith
rise the rise in economic
in economic levels
levels and and increasing
increasing usage
usage demands,
demands, enhancing
enhancing the building’s
the building’s widthwidth to increase
to increase its area
its floor floorallows
area allows for addi-
for additional
tional functionalities,
functionalities, such assuch
the as the provision
provision of garages
of garages or the or the layout
layout of primary
of primary and sec-
and secondary
ondary bedrooms.
Sustainability 2024, 16, x FOR PEER REVIEW
bedrooms. Similarly,
Similarly, the differences
the differences in depth in also
depth also reflect
reflect variations
variations in the in the internal
7 of
internal 19
func-
functional
tional layoutlayout ofhouse.
of the the house. Buildings
Buildings withwith
lowerlower economic
economic status
status andand earlier
earlier construc-
construction
tion dates
dates typically
typically have have
moremore singular
singular functions,
functions, merely merely
catering catering to cooking
to cooking and needs,
and living living
needs,
resultingresulting
in in relatively
relatively more more
minor minor
overalloverall
depths. depths.
As As economic
economic
the era of house construction changes, the functions within houses gradually become levels
levels improve
improve and and
the
era of house construction changes, the functions within houses
more numerous and diverse, leading to an increase in the depth of the buildings. gradually become more
numerous and diverse, leading to an increase in the depth of the buildings.
To create the surveyed cohort types of the representatives of most rural residential
To create the surveyed cohort types of the representatives of most rural residential
buildings in the village, the range of values taken from six perspectives, “Principal House
buildings in the village, the range of values taken from six perspectives, “Principal House
Width” (PHW), “Principal House Depth” (PHD), “Wing House Width” (WHW), “Wing
Width” (PHW), “Principal House Depth” (PHD), “Wing House Width” (WHW), “Wing
House Depth” (WHD), “Length of the Courtyard” (LIC), and “Width of the Courtyard”
House Depth” (WHD), “Length of the Courtyard” (LIC), and “Width of the Courtyard”
(WIC), were used to understand the characteristics of the rural residential buildings. As
(WIC), were used to understand the characteristics of the rural residential buildings. As
shown in Figure 4, Type 2 PHW is mainly in the range of 10–15. Type 4 has a more even
shown in Figure 4, Type 2 PHW is mainly in the range of 10–15. Type 4 has a more even
yard size distribution than the other types. The performance of WHD for all three types
yard size distribution than the other types. The performance of WHD for all three types of
of compounds is more dispersed, indicating that this indicator is subject to more signifi-
compounds is more dispersed, indicating that this indicator is subject to more significant
cant variation depending on the type of compound. The above information on PHW,
variation depending on the type of compound. The above information on PHW, PHD,
PHD, WHW, WHD, LIC, and WIC for the different compound types was obtained by
WHW, WHD, LIC, and WIC for the different compound types was obtained by compiling
compiling the data
the data from from theresearch
the building buildingin research in the village.
the village.
Figure
Figure4.4.Range
Rangeofofvalues
valuesofofcritical
criticalvariables
variablesfor
fordifferent
differenttypes
typesofofbuildings.
buildings.
3.1.2. Parameterization
The parametric model created with Rhino and Ladybug Tools was able to simulate
the performance of the building based on the local climate and the site environment. HB
Energy is used to simulate the energy consumption of the building by using Ladybug from
https://www.ladybug.tools/epwmap/ (accessed on 23 October 2023) to download the
Jinan City weather file as the weather file for this simulation study.
The parameters for the energy simulation were set as follows: the simulation period was
the whole year, the heat transfer coefficient of the facade was taken as K = 0.959 (W·m−2 ·K−1),
the heat transfer coefficient of the roof was taken as K = 1.167 (W·m−2 ·K−1), and the type
of windows was taken as Type1. The personnel presence rate, lighting power, and electrical
equipment power are set according to national and local standards.
Table 1. VariableAnalysis
3.2. Correlation se ing and initial investment (definition of abbreviations: PHW = principal house
width; PHD = principal house depth; WHW (W, E) = wing house width (western, east); WHD = wing
This section discusses the architectural parameter variables and simulates the relevant
house depth; RS = roof slope; EW = eave width; HCW = height of the courtyard’s wall; BH = building
parameter variables of the main house and compartments, respectively. Subsequently, a
height; BO = building orientation; WT = window type; and EPS, XPS = insulation material type).
Pearson correlation analysis of the data was conducted using the “scipy.stats” library in
Building Parameters Range of Values Unit Initial Investment
PHW 10–20 m /
PHD 3–12 m /
7 3Low-E + 0.1Air + 3Low-E 0.656 2.22 240
Figure
Figure 7.
7. Simulation
Simulation results
results of
of courtyard
courtyard parameters.
parameters.
The line
The line graph
graphshows
showsthat
thatBO, EW,
BO, andand
EW, HCW werewere
HCW significantly correlated
significantly with UTCI
correlated with
and not significantly correlated with TLI and COST.
UTCI and not significantly correlated with TLI and COST.
The line graph shows that BO, EW, and HCW were significantly correlated with
Sustainability 2024, 16, 667 UTCI and not significantly correlated with TLI and COST. 10 of 19
Figure
Figure 8.
8. Simulation
Simulation results
results of
of enclosure structure parameters.
enclosure structure parameters.
As shown
As shown inin the
the line
line graphs,
graphs, EPS
EPS and
and XPS
XPS have
have aa significant
significant correlation
correlation with
with TLI
TLI and
and
COST but
COST but no
no significant
significant correlation
correlation with
with UTCI.
UTCI. WT
WT has
has no
no significant
significant correlation
correlation with
with all
all
three but has a varying degree of influence on TLI and
three but has a varying degree of influence on TLI and COST.COST.
the mean. There may be slight deviations in the evaluation results, but they have little
effect on the trend of the optimization results. Then, the variables of building parameters,
as well as the initial investment in Tables 3 and 4, are input into the gene port to optimize
Type2, Type 3, and Type 4, respectively. The number of iterations is 40 generations, the
subset of each generation is 50, and the number of populations is 2000. Table 3 shows the
genetic algorithm parameters.
Figure 99shows
Figure showsthethesolution
solutionset
setofofthe
the multi-objective
multi-objective optimization
optimization as aasscatter
a sca plot
er plot
on
on the Wallacei coordinate system. TLI, Cost, and UTCI correspond to the x, y,
the Wallacei coordinate system. TLI, Cost, and UTCI correspond to the x, y, and z axes. and z axes.
The
The image
PCP PCP image in Figure
in Figure 10 the
10 shows shows the analysis
analysis of the optimization
of the optimization process toprocess
achieve to achieve
minimum
minimum
TLI, TLI, COST,
minimum minimumand COST,
maximum and UTCI.
maximum UTCI.
Figure 9.
Figure 9. Distribution
Distribution of
of Pareto
Pareto frontier
frontier solutions
solutions for
for three
three types
types of
of RRB.
RRB.
Figure 10. Distribution of parallel coordinate plot for three types of RRB.
By analyzing the results of the multi-objective optimization and the standard devia-
Sustainability 2024, 16, 667 12 of 19
Figure 9. Distribution of Pareto frontier solutions for three types of RRB.
Figure
Figure10.
10.Distribution
Distributionof
ofparallel
parallelcoordinate
coordinateplot
plotfor
forthree
threetypes
typesof
ofRRB.
RRB.
Byanalyzing
By analyzing the
the results
results of
of the
themulti-objective
multi-objectiveoptimization
optimizationandandthe standard
the standard deviation
devia-
plots of the process, the overall evolutionary direction of the algorithm tends
tion plots of the process, the overall evolutionary direction of the algorithm tends to be stable,
to be
and all of them converge to the optimal trend. Among the 194 Pareto frontier solutions of
stable, and all of them converge to the optimal trend. Among the 194 Pareto frontier solu-
Type 2, the objective range of COST is from 8000 to 28,000, the objective range of TLI is from
tions of Type 2, the objective range of COST is from 8000 to 28,000, the objective range of
70 kWh/m2 to 148 kWh/m2 , and the mean value of UTCI is between 16.7 ◦ C and 17.3 ◦ C;
TLI is from 70 kWh/m2 to 148 kWh/m2, and the mean value of UTCI is between 16.7 °C
among the 229 Pareto frontier solutions of Type 3, the objective range of cost is from 9000 to
and 17.3 °C; among the 229 Pareto frontier solutions of Type2 3, the objective range of cost
32,000, the target range of TLI is 70 kWh/m2 to 160 kWh/m , and the mean value of UTCI
is from 9000 to 32,000, the target range of TLI is 70 kWh/m 2 to 160 kWh/m2, and the mean
is between 16.2 ◦ C and 17.2 ◦ C. For the 326 Pareto frontier solutions of Type 4, the target
value of UTCI is between 16.2 °C and 17.2 °C. For the 326 Pareto frontier solutions of Type
range of cost is 13,000 to 48,000, the target range of TLI is 63 kWh/m2 to 138 kWh/m2 , and
4, the target range of cost is 13,000 to 48,000,◦the target◦range of TLI is 63 kWh/m2 to 138
the mean value of UTCI ranges between 16 C and 17 C. The average value of the
Sustainability 2024, 16, x FOR PEER REVIEW
UTCI
13 value
of 19
kWh/m2, and the mean value of UTCI ranges between 16 °C and 17 °C. The average
is between 16 ◦ C and 17 ◦ C. The average value of the UTCI is between 16 ◦ C and 17 ◦ C
of the UTCI is between 16 °C and 17 °C. The average value of the UTCI is between 16 °C
(Figure 11).
and 17 °C (Figure 11).
Figure
Figure11.
11.Box
Boxdiagram
diagramofofPareto
Paretofrontier
frontiersolution
solutiondistribution.
distribution.
The
Theabove
abovedata
datawere
wereexported
exportedfrom
fromGrasshopper
GrasshoppertotoExcel
Excelvia
viathe
theTT
TTToolbox
Toolboxtool
toolfor
for
the
thefollowing
followingmachine
machinelearning
learningmodel
modeltraining.
training.
3.4.Predictive
3.4. PredictiveModel
ModelConstruction
Construction
Thisstudy
This studyemploys
employsthe thefeedforward
feedforwardneural
neuralnetwork
networkalgorithm
algorithmand andcommonly
commonlyused used
machine learning libraries for model construction. Scikit-learn is the premier Python library
machine learning libraries for model construction. Scikit-learn is the premier Python li-
for machine learning, constructed upon foundational data science packages like numpy
brary for machine learning, constructed upon foundational data science packages like
and matplotlib. It encompasses various machine learning functionalities, including data
numpy and matplotlib. It encompasses various machine learning functionalities, includ-
sampling, preprocessing, model validation, classification, and regression [29]. In this study,
ing data sampling, preprocessing, model validation, classification, and regression [29]. In
a regression prediction model is employed. Neural networks, which are artificial neural
this study, a regression prediction model is employed. Neural networks, which are artifi-
networks inspired by biological neural systems, are applicable in various machine learning
cial neural networks inspired by biological neural systems, are applicable in various ma-
applications, including regression prediction. Furthermore, this method is noted for its
chine learning applications, including regression prediction. Furthermore, this method is
comprehensive efficiency and accuracy as a machine learning technique.
noted for its comprehensive efficiency and accuracy as a machine learning technique.
Neural network fitting for regression prediction is a technique for making predictions
Neural network fi ing for regression prediction is a technique for making predictions
on a given dataset by training a neural network. In a neural network, nodes are organized
on a given dataset by training a neural network. In a neural network, nodes are organized
into a hierarchy. The input layer receives raw data, the intermediate layer processes
into
the ainput
hierarchy. The extracts
data and input layer receives
features, andrawthedata, thelayer
output intermediate
generates layer processesEach
predictions. the
input data and extracts features, and the output layer generates predictions.
node has an associated weight and bias, and these values can be trained and adjusted by Each node
has an associated weight
backpropagation and to
algorithms bias, and these
minimize valuesbetween
the error can be trained and adjusted
the predicted and actualby results.
back-
propagation algorithms to minimize the error between the predicted and actual
The fitting process is usually divided into two phases: training and testing. In the training results.
The fi ing
phase, theprocess
datasetisisusually
divided divided into twoand
into training phases: training
validation andThe
sets. testing. In theset
training training
is then
phase,
used tothetrain
dataset
the is divided
neural into training
network to findand
the validation sets. The
optimal weights andtraining
biasessetforisminimizing
then used
to train the neural network to find the optimal weights and biases for minimizing the pre-
diction error. During training, optimization algorithms such as stochastic gradient descent
can help optimize the weights and biases of the neural network.
the prediction error. During training, optimization algorithms such as stochastic gradient
descent can help optimize the weights and biases of the neural network.
Figure
Figure 12.
12. Structure
Structure diagram
diagram of
of MLP
MLP neural
neural network.
network.
In
In MLP,
MLP, the
the data
data are
are input
input as
as aa matrix,
matrix, and
and neurons
neurons areare activated
activated in
in the
the implicit
implicit layer
layer
for
for weight
weightresponse.
response. TheThemodel
modelcancanconsider
considerthetheeffect
effectof
ofeach
eachperformance
performancefeature
featureduring
during
backpropagation.
backpropagation. The relationship between
The relationship betweenthe theeffects
effectsofofthethethree
three performance
performance parame-
parameters
ters impacts the fi ing process. The data in this paper are fi ed to a neural network
impacts the fitting process. The data in this paper are fitted to a neural network using
using an
an array of 6000 observations with 14 features as predictor variables and observations with
array of 6000 observations with 14 features as predictor variables and observations with
three features
three featuresas asresponse
responsevariables.
variables.The The “Batch
“Batch Size”
Size” was wassetset to 128,
to 128, andand the the model’s
model’s re-
regression performance was evaluated using the “MSE” (Mean Squared
gression performance was evaluated using the “MSE” (Mean Squared Error) loss function Error) loss function
and R2
and R2 (R-Square)
(R-Square) metrics.
metrics.
The dataset is divided into
The dataset is divided into aatraining
trainingsetsetfor
fortraining
trainingthethemachine
machinelearning
learningmodel
modelandand
a test set for validating the model. In this study, 70% of the data was randomly
a test set for validating the model. In this study, 70% of the data was randomly selected as selected as
the training set, 20% as the test set, and 10% as the validation
the training set, 20% as the test set, and 10% as the validation set. set.
3.4.3. Model
3.4.3. Model Evaluation
Evaluation
The training
The training regression
regressionwaswasconducted
conductedbyby modeling
modeling thethe
neural network
neural networkthrough Py-
through
thon forfor
Python a total of 300
a total rounds
of 300 roundsandandconverged
converged after the the
after 100th round.
100th Cross-validation
round. Cross-validationwas
usedused
was to select andand
to select evaluate
evaluatesimulations,
simulations,culminating
culminatingininchoosing
choosing the best-performing
the best-performing
model. The
model. The specific
specific model effect is
model effect is shown
shown in in Figures
Figures 1313 and
and 14.
14. From
From thethe fitting
fi ing effect
effect
graph, the
graph, the model
modelhas
hasR2R2== 0.988
0.988 for
for the
the training
trainingset,
set,R2
R2== 0.986
0.986 for
for the
the testing
testing set,
set,R2
R2== 0.985
0.985
for the validation set, and R2 = 0.988 for all data. The prediction model converged in all
for the validation set, and R2 = 0.988 for all data. The prediction model converged in all
three parts,
three parts, thus
thus avoiding
avoiding the
the issue
issue ofof overfitting.
overfi ing.
Figure 13.
Figure 13. Training
Training and
and validation
validation loss
loss over
over time.
time.
3.4.4.Code
3.4.4. CodeEncapsulation
Encapsulation
Aftervalidating
After validatingthe
thefeedforward
feedforwardneural
neuralnetwork
networkprediction
predictionmodel
modelconstructed
constructedbased
based
onthe
on themulti-objective
multi-objective dataset
dataset of of rural
rural residential
residential buildings
buildings in cold
in cold regions,
regions, the system
the system code
code was encapsulated to enhance its usefulness for designers. The software was built
was encapsulated to enhance its usefulness for designers. The software was built with
with Tkinter,
Tkinter, a standard
a standard GUI for
GUI library library for Python,
Python, and mainlyandpredicted
mainly predicted the three
the three target target
variables.
The interface of the software is shown in Figure 15.
variables. The interface of the software is shown in Figure 15.
Figure15.
Figure 15.Software
Softwareuser
userinterface.
interface.
4.4.Results
Resultsand
andDiscussion
Discussion
After the model training is completed, we save the trained model. By plotting the
comparison line graphs of the actual values versus the predicted values for the three
types of prediction targets in the test and validation sets(Figure 16), the predictive model
demonstrates a relatively good level of prediction for all three targets. Notably, the model’s
performance in predicting COST is better than the other two indicators.
Then, to test the saved model’s accuracy, the saved regression prediction model is
imported, and three groups of variables are randomly selected as cases in each type of RRB
for prediction. The differences between the prediction results and the actual performance
levels are compared. The case parameter variables are shown in Table 4. A comparison of
actual and predicted target values is shown in Table 5.
As shown in Table 5, the model shows a good prediction performance for both TLI
and UTCI, and the prediction for COST sometimes produces large fluctuations. However,
overall, the model predictions are still accurate, and the prediction accuracy should be
higher if more samples are provided in the multi-objective optimization stage, and more
case data are generated.
Figure
Figure 16.
16. Comparison
Comparison of
of the
the true
true value
value of
of the
the forecasted target with
forecasted target with the
the forecasted
forecasted value.
value.
Author Contributions: Conceptualization, Y.X. and M.S.; Methodology, Y.X. and M.S.; Software, M.S.;
Validation, M.S.; Formal analysis, M.S.; Investigation, M.S. and L.W.; Resources, M.S.; Writing—original
draft, M.S.; Writing—review and editing, M.S.; Visualization, M.S.; Supervision, Y.X.; Project administra-
tion, Y.X. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: At this time, the data and source code for the machine learning models
produced in this study are not available as they are required for the subsequent phase of the research.
Conflicts of Interest: The authors declare that they have no known competing financial interests or
personal relationships that could have appeared to influence the work reported in this paper.
Nomenclature
RRB rural residential buildings EC energy consumption
IPCC intergovernmental panel on climate change CES cost of the enclosure structure
MOO multi-objective optimization RF roof slope
PHW principal house width BO building orientation
PHD principal house depth HCW height of the courtyard’s wall
WHW wing house width EPS EPS thickness
WHD wing house depth XPS XPS thickness
BH building height WD window type
EW eave width TLI total loads intensity
ML machine learning UTCI Universal Thermal Climate Index
WIC width of the courtyard LIC “length of the courtyard”
Sustainability 2024, 16, 667 18 of 19
References
1. Merlet, Y.; Rouchier, S.; Jay, A.; Cellier, N.; Woloszyn, M. Integration of phasing on multi-objective optimization of building stock
energy retrofit. Energy Build. 2022, 257, 111776. [CrossRef]
2. China Building Energy Efficiency Association Building Energy Consumption and Carbon Emission Data Specialized Committee.
Efficiency, Research Report on Building Energy Consumption in China; Chinese Construction Industry Chubns: Beijing, China, 2022.
Available online: https://www.cabee.org/ (accessed on 23 June 2023).
3. Building Energy Efficiency Research Center, Tsinghua University. China Building Energy Efficiency Development Research Report
2023; China Building Industry Press: Beijing, China, 2023.
4. Wu, Y. Research on Low Carbon Adaptive Construction Methods and Practices of Rural Community Spatial Form. Ph.D. Thesis,
Zhejiang University, Zhejiang, China, 2016.
5. He, H.; Xiong, K.; Li, Y. Status and outlook of the quality of concrete residential projects in villages and towns in China. Concrete
2010, 247, 129–131.
6. Liang, L.; Wu, W.L.; Lal, R.; Guo, Y.B. Structural change and carbon emission of rural household energy consumption in Huantai,
northern China. Renew. Sustain. Energy Rev. 2013, 28, 767–776. [CrossRef]
7. Wang, R.; Jiang, Z. Energy consumption in China’s rural areas: A study based on the village energy survey. Clean. Prod. 2017, 143,
452–461. [CrossRef]
8. Fan, J.L.; Yu, H.; Wei, Y.M. Residential energy-related carbon emissions in urban and rural China during 1996–2012: From the
perspective of five end-use activities. Energy Build. 2015, 96, 201–209. [CrossRef]
9. Xu, J.; Gao, W.J.; Huo, X. Analysis on energy consumption of rural building based on survey in northern China. Energy Sustain.
Dev. 2018, 47, 34–38.
10. He, B.-J.; Yang, L.; Ye, M. Buildings energy efficiency in China rural areas: Situation, drawbacks, challenges, corresponding
measures and policies. Sustain. Cities Soc. 2014, 11, 7–15. [CrossRef]
11. Jermyn, D.; Richman, R. A process for developing deep energy retrofit strategies for single-family housing typologies: Three
Toronto case studies. Energy Build. 2016, 116, 522–534. [CrossRef]
12. Zhang, Y.; Yang, L. Comparative study on the indoor thermal environment of traditional and modern residential houses in
northern Jiangsu Province. Archit. Technol. 2017, 48, 707–710.
13. Asadi, E.; Da Silva, M.G.; Antunes, C.H.; Dias, L. Multi-objective optimization for building retrofit strategies: A model and an
application. Energy Build. 2012, 44, 81–87. [CrossRef]
14. Liu, Y.; Huang, Z.; Wang, Y. Energy-efficient design of a farmhouse based on seasonal differences in indoor and outdoor
multi-space thermal environment quality and user space utilization. Build. Sci. 2022, 38, 216–222.
15. Al-Masri, N.; Abu-Hijleh, B. Courtyard housing in midrise buildings: An environmental assessment in hot-arid climate. Renew.
Sustain. Energy Rev. 2012, 16, 1892–1898. [CrossRef]
16. Taleghani, M. Outdoor thermal comfort by different heat mitigation strategies—A review. Renew. Sustain. Energy Rev. 2018, 81,
2011–2018. [CrossRef]
17. Zamani, Z.; Heidari, S.; Hanachi, P. Reviewing the thermal and microclimatic function of courtyards. Renew. Sustain. Energy Rev.
2018, 93, 580–595. [CrossRef]
18. Tavakolan, M.; Mostafazadeh, F.; Eirdmousa, S.J.; Safari, A.; Mirzaei, K. Parallel computing simulation-based multi-objective
optimization framework for economic analysis of building energy retrofit: A case study in Iran. J. Build. Eng. 2022, 45, 103485.
[CrossRef]
19. Gao, Y.; Hu, K.; Yue, X.; Yuan, J. Multi—Objective optimization design of form parameters for energy saving and thermal comfort
in northern country houses. Overseas Chin. Univ. Sci. (Nat. Sci. Ed.) 2021, 42, 619–627.
20. Vukadinović, A.; Radosavljević, J.; Ðord̄ević, A.; Protić, M.; Petrović, N. Multi—Objective optimization of energy performance for
a detached residential building with a sunspace using the NSGA-II genetic algorithm. Sol. Energy 2021, 224, 1426–1444. [CrossRef]
21. Ascione, F.; Bianco, N.; De Masi, R.F.; Mauro, G.M.; Vanoli, G.P. Design of the building envelope: A novel multi-objective
approach for the optimization of energy performance and thermal comfort. Sustainability 2015, 7, 10809–10836. [CrossRef]
22. Carlucci, S.; Cattarin, G.; Causone, F.; Pagliano, L. Multi—Objective optimization of a nearly zero-energy building based on
thermal and visual discomfort minimization using a non-dominated sorting genetic algorithm (NSGA-II). Energy Build. 2015, 104,
378–394. [CrossRef]
23. Yan, H.; Yan, K.; Ji, G. Optimization and prediction in the early design stage of office buildings using genetic and XG Boost
algorithms. Build. Environ. 2022, 218, 109081. [CrossRef]
24. Zekić-Sušac, M.; Has, A.; Knežević, M. Predicting energy cost of public buildings by artificial neural networks, CART, and
random forest. Neurocomputing 2021, 439, 223–233. [CrossRef]
25. Guo, J.; Zhou, J.; Li, M.; Lu, S. Based on ANN and many-objective optimization to improve the performance and economy of
village houses in Chinese cold regions. J. Build. Perform. Simul. 2023, 16, 526–536. [CrossRef]
26. Liu, C.L.; Tseng, C.J.; Huang, T.H.; Yang, J.S.; Huang, K.B. A multi-task learning model for building electrical load prediction.
Energy Build. 2023, 278, 112601. [CrossRef]
27. Lagios, K.; Niemasz, J.; Reinhart, C.F. Animated building performance simulation (ABPS)–linking Rhinoceros/Grasshopper with
Radiance/Daysim. Proc. SimBuild 2010, 4, 321–327.
Sustainability 2024, 16, 667 19 of 19
28. Tian, Y.; Huang, Q. Multi-objective optimization design method for building performance and its application—An example of
genetic algorithm. New Archit. 2021, 198, 84–89.
29. Jiang, J. Research on the Optimization of High-Density Neighborhood Morphology Design Based on Genetic Algorithm. Master’s
Thesis, Shenzhen University, Shenzhen, China, 2020.
30. Nelli, F. Machine Learning with sci-kit-learn. In Python Data Analytics; Apress: Berkeley, CA, USA, 2018.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.