Sustainable Building Optimization Model for Early-Stage Design

Abstract

Buildings represent the largest potential for carbon reduction worldwide. This highlights the need for a simulation and optimization method for energy management. The early design stage of buildings represents an important phase in which choices can be made to optimize design parameters. These parameters can focus on multiple areas, including energy and thermal comfort. This paper introduces the optimization of early-stage sustainable building design considering end-user energy consumption. It proposes an optimization model that integrates multiple layers, which consist of a parametric energy simulation, artificial neural network, and genetic algorithm. The proposed optimization model considers a single objective function to obtain the optimal design. The targeted goal is to obtain minimal energy consumption for residential buildings during the early design stages. Key design parameters of the building were identified for optimization and feasible ranges for them were obtained using genetic algorithms. Finally, the results of this paper include the identification of the optimal building design for the thermal comfort analysis and optimal energy performance. The model was applied to a case study in Egypt and the results showed that using the developed optimization model can lead to a 25% reduction in energy consumption.

1. Introduction

The number of new cities being built in Egypt has increased over the years. This has led to the construction of more buildings and residential compounds all over the country. The increase in household loads has been attributed to this expansion in new communities and residential compounds [1]. Additionally, the use of appliances and air conditioners during the summer has been on the rise. It is reported that the number of air conditioners has increased from 700,000 units in 2006 to 5.3 million units in 2022 [2,3]. According to [4], the buildings sector has the largest potential for and the lowest cost of carbon reduction. Since each building is unique, optimization tools are needed that could be applied to all building types. The energy performance of a building is dependent on several factors, including shape, form, design, components, aesthetics, the HVAC system, and energy consumption. Therefore, designing an energy efficient building is based on minimizing thermal loads, designing efficient HVAC systems, utilizing passive strategies, and using renewable energy [5].

Designers typically use parametric analyses to evaluate approaches for energy efficiency. The authors of [6] propose the use of an optimization-based design technique to overcome the shortcomings of the parametric method in energy efficiency calculations. According to [7], although there are several methods for building envelope optimization, many of them are evaluative and not prescriptive. Hence, this area needs to be further investigated. To achieve high-performance building designs, building energy simulation tools that can be coupled with optimization models are needed [1]. The authors of [8] proposed prediction and optimization models for thermal loads in office buildings. The authors of [9] proposed a modeling and optimization approach for HVAC systems. They were able to obtain optimal settings and reduce energy consumption while taking into consideration the indoor environment quality as well. The authors of [10] proposed an optimization model for residential building design in Kuwait. Therefore, this paper proposes a sustainable building optimization model that is integrated with a building energy simulation, with a specific focus on Egypt.

There are several methods for energy efficiency, including: passive design techniques, energy management, reducing HVAC energy, and using renewable energy [11]. In order to calculate the effect of these methods, simulation tools are needed to depict their use. Previous researchers have studied parametric optimization for the energy performance and daylight of school buildings [12], public buildings [13], and office towers [14]. The authors of [15] developed a parametric approach for green residential buildings at an early stage. They also conducted a correlation analysis between design parameters and energy use and found thermal performance to be the parameter with the highest influence. However, their model was specific to only one project [1].

Several challenges exist in performing early-stage building simulations, including uncertainty, changes in design, lack of information, large combinations of parameters, and time-consuming design creation [16]. However, early-stage building design studies have several benefits, including a reduction in the energy consumption by changing the building shape, orientation, or envelope [15]. Hence, a sustainable building optimization model would overcome the majority of these issues and increase the resulting benefits through the development of multiple combinations of parameters or design details. One benefit of using optimization models during the early design stage is that it can have a significant impact on the design even, though its accuracy is still low [17]. According to [18], early design models need to quickly evaluate design alternatives and be able to solve non-linear problems. There are several building performance tools available; however, their drawbacks include difficulty in use, slowness of processing, and simplified feedback [19]. The subject of building energy performance optimization has been tackled in different areas of built environments. Researchers have previously studied high-rise buildings [20,21,22,23,24,25], hospitals [26,27], heritage buildings [28,29], and educational buildings and offices [30].

This paper proposes an optimization method based on a parametric analysis and artificial neural network model in order to optimize residential building energy efficiency. This method will enable designers to calculate and optimize energy consumption in the early stages of building design, which will speed up decision-making regarding different designs. Additionally, this will support the interoperability between building modeling and energy simulation modeling.

2. Background

2.1. Sustainable Building Optimization Design

A basic principle of sustainable building design is to reduce negative impacts on building construction environments while taking costs and other performance criteria into account. This principle was implemented in previous studies via four methods: empirical rules, simulation-based trial-and-error methods, knowledge-based methods, and optimization methods [31]. The optimization method aims at obtaining an optimal or near-optimal design solution that meets potentially conflicting design requirements [32]. For example, the aim might be to minimize the environmental impact while also minimizing development costs and respecting other predefined performance criteria. However, there are many difficulties in applying optimization to building design. These difficulties can be seen in terms of the following aspects:

• The design space is typically large.
• The interactions among parameters are complicated and there is more than one performance criterion to be considered.
• The evaluation process of the performance criteria of the design alternative is usually conducted using different commercial simulation programs.
• It is difficult coupling the simulation programs with an optimization model.

In order to reach a sustainable building design in the early design stages, it is important to perform an energy performance assessment first. Although this would be very effective, it is considered a complex and demanding process. Sustainable building design based on performance efficiency includes various choices that can satisfy specific criteria, including economic and environmental issues, such as minimizing energy consumption and enhancing environmental performance. To do so, certain criteria can be optimized, such as the building layout, geometry, position, ventilation, HVAC elements, etc. [33]. This would require a great amount of effort, time, and skill by designers to achieve an effective design. Several issues arise when conducting this optimization, including: (1) variations in the building form; (2) the efficient use of natural daylight; (3) the effects of the increased wall surface on the heat flow; (4) the window-to-wall ratio; (5) the dimensions of the windows; and (6) the amount of thermal mass. Since these issues provide a wide variety of alternatives in design, it is best to settle on some of the elements first and then conduct the optimization on a select number. This would reduce the variability and make the selection process easier and faster. Additionally, assessments of occupants’ thermal comfort and buildings’ energy performances are also needed in this step.

Recently, numerous methods have been suggested to address performance-based design issues, most of which have used optimization algorithms [34]. The authors of [35] proposed an optimization approach for new buildings and retrofits focusing on the building geometry and envelope. The authors of [36] proposed a multi-objective GA for energy efficiency and thermal comfort optimization. The authors of [37] proposed an optimization model that considers energy, economy, comfort, and sustainability. The authors of [8] proposed prediction and optimization models for thermal loads in office buildings to reduce the operation load. The authors of [38] proposed an energy optimization model for buildings in different areas of Iran combining EnergyPlus for the simulation and a multi-objective particle swarm optimization tool. They also conducted a single objective approach from three perspectives: cooling, heating, and lighting-based optimizations. The authors of [39] proposed an optimization model for thermal comfort and energy efficiency while considering 10 parameters. They found that the most sensitive parameters were the external wall’s insulation thickness, the window-to-wall ratio, the roof’s heat transmission coefficient, and the heat gain coefficient of the external wall. The authors of [40] proposed an optimization model to minimize energy and achieve a net-zero-energy building while targeting visual and thermal comfort. The authors of [41] proposed a combined simulation and optimization model focusing on energy consumption and thermal comfort. This resulted in dozens of potential designs with tradeoffs between thermal comfort and energy consumption. The authors of [42] proposed an optimization model to reduce energy consumption and insulation costs and maximize thermal comfort by focusing on the temperature of thermostats in residential buildings. The authors of [43] proposed an optimization model to select materials for an envelope design of a residential building. Wu et al. proposed an optimization model for the net-zero renovation of residential buildings and [44] proposed an optimization model for a residential building with a sunspace in Serbia. They studied structural and architectural parameters under two scenarios which included the energy required for heating and cooling as well as the discomfort hours. Therefore, following previous research, this paper uses the design optimization method to achieve sustainability during the early building design stages.

Optimization models are an important tool used in the decision-making process to select the optimal solution among a set of solutions. Any optimization model has three main components, which are: objective function, variables, and constraints. Optimization models can be categorized based on the objective function and the nature of their constraints. The objective function is the mathematical function to be optimized [33]. Among the available methods for optimization are genetic algorithms (GAs), which are commonly used to reach an optimal solution over many generations. GA optimization starts searching by randomly sampling within an optimization solution space, after which it uses stochastic operators to lead the optimization process based on the preset objective function values. In GAs, the entire solutions’ group is labelled as a population while a specific solution to the optimization problem is labelled as an individual. Whenever a new population of individuals is created, it is referred to as a generation. Each individual is represented by a binary string, reals, integers, characters, or a list of roles referred to as a chromosome. These chromosomes code the parameters chosen which correspond to an individual. The fitness of an individual is linked to the value of the objective function at that point [45]. There are three basic genetic operators that control the evolution of the solutions’ generations of the optimization problem. These are reproduction, crossover, and mutation. The probability of a given solution being chosen for reproduction is directly proportional to the fitness of that solution. Crossover implies that parts of two randomly chosen chromosomes are exchanged in order to create a new individual. Finally, mutation involves randomly varying an allele in a solution to search for new options in the solution space.

2.2. Previous Research on Design Optimization Techniques

Multiple models exist for building energy prediction, which can be classified into forward models and data-driven models [46]. Forward models rely on inputs from the built environment and require a lot of information and time. Data-driven models, on the other hand, are based on mathematical models and can rely on tools, such as machine learning algorithms [46]. A large dataset is needed to train and optimize the models for accurate predictions.

Several methods have been proposed by previous researchers for the prediction of energy consumption. The authors of [47] used machine learning to predict energy consumption and applied it to a case study in Malaysia. The authors of [46] used deep learning and other machine learning techniques for predicting energy consumption. The authors of [6] proposed a method to assess a cost-effective design while maximizing its energy efficiency and minimizing the life cycle costs for residential buildings in Tunisia.

The process of finding the best values for the variables of a particular problem to maximize/minimize an objective function is called optimization. Optimization problems exist in different fields of studies. Optimization can be classified into single objective, multi-objective, and weighted sum [4]. This depends on the number of objectives the user wishes to optimize. Complicated design problems may require automated ways of optimization, including genetic or evolutionary algorithms [1]. These would be used to produce variations in designs, understand the tradeoffs, and graphically evaluate the variations.

Two main methods exist for optimization: mathematical and stochastic. The mathematical optimization method relies on gradient-based information of the involved functions to reach the optimal solution. However, this method suffers from local optima entrapment, which means that the algorithm assumes that the local solution it reaches is the global solution. Hence, it fails to obtain the global optimum. These methods can also be ineffective for problems with unknown or computationally expensive derivations [48].

The second algorithm is stochastic optimization, which relies on random operators that allow it to avoid local optima. The stochastic method starts the optimization process by generating a set of random solutions for a given problem. In contrast to mathematical optimization methods, stochastic methods do not need to calculate the gradient of a solution. The evaluation of solutions is performed based on the calculated objective function(s). The most popular algorithms are the genetic algorithm (GA), particle swarm optimization (PSO), ant colony optimization (ACO), and differential evolution (DE). Both classes of nature-inspired algorithms are similar in that solutions are improved until an end criterion is satisfied. In GAs, solutions are considered as individuals and the parameters of the solutions take the place of their genes. The survival of the fittest individuals is the main inspiration for this algorithm, in which the best individuals tend to participate more in improving poor solutions [49].

Many algorithms have been developed and used for the energy design optimization of buildings. According to [50], the algorithm must be selected based on the nature of the optimization problem to be solved, since no algorithm can be the best in all performance indices. Several researchers have used GAs for building simulations. The authors of [51] used GA to optimize 11 parameters and minimize the life cycle cost of a residential building. The authors of [52] used GA for minimizing annual energy use. The authors of [53] used GA to optimize building designs in the Mediterranean area. The authors of [54] used GA to optimize building forms based on site-specific building forms. Additionally, previous research has compared several optimization techniques and found GA to have the best performance [51,55,56]. The authors of [57] used GA to create an energy calculation model with an optimization program. However, GAs are observed to have several limitations, including their difficult and time-consuming optimization process [17].

3. Research Motivation

Several studies have highlighted the importance of integrating energy prediction with optimization methods for building design optimization [10]. The performative computational architecture field bases its designs on an iterative method that considers form finding, performance evaluation, and optimization. Advanced optimization applications/plugins exist in Grasshopper, such as Galapagos, Goat, Silvereye, Opossum, Dodo, and Nelder Mead. However, these optimization tools are more suitable for a specific case study, not a generalized framework. There are other advanced applications that overcome these drawbacks; however, they require that an architect create the entire building model along with energy simulation settings to get the optimal design for a specific case study. This is time-consuming and requires a lot of effort. To overcome this issue, the authors propose an integrated optimization model, which encompasses modeling, a simulation, and energy predictions based on the inputted information for the weather data and materials. The model focuses on an early-stage design with an application in Egypt.

Therefore, the main contribution of this work is to provide a speedy and accurate model for achieving optimal design solutions in the early design stage. Furthermore, the current study explores the capabilities of ANN and genetic algorithms to improve the optimal performance of the design of residential buildings while considering energy consumption. This will aid architects and decision-makers in their decisions to minimize energy usage swiftly and accurately during early design stages.

4. Methodology

This paper presents a novel optimization model that aims to obtain the minimal energy consumption for residential buildings during the early design stages. The model is based on a single objective function to obtain the optimal design. According to [58], a building design optimization analysis consists of four steps: (1) identifying the design variables; (2) choosing a simulation tool and creating the building; (3) choosing an objective function; and (4) choosing an optimization algorithm. These four steps were integrated into the three phases of this study. Phase 1 consisted of the creation of the energy simulation (discussed in Section 4.1), phase 2 consisted of the creation and validation of the ANN model (discussed in Section 4.2), and the final part, phase 3, consisted of the novel GA optimization model (discussed in Section 4.3). Figure 1 shows the optimization model workflow from the creation of the energy simulation to the development of the ANN model and its validation to the final genetic algorithm. In order to discuss the proposed optimization model, a previous step must be first introduced, which includes the previously proposed parametric model and energy simulation analysis. These were introduced by the authors in [59,60].

4.1. Parametric Model and Energy Simulation

This section briefly discusses the entire framework, which includes the previously created models in order to introduce the new optimization model. The framework consists of three phases: (1) constructing a parametric model with an energy simulation engine [60]; (2) developing an artificial neural network (ANN) model for energy prediction [59]; and (3) developing an optimization model that is integrated with the ANN model.

Parametric Model with an Energy Simulation

The parametric model’s workflow consists of four phases:

(a)

The choice of construction materials. This includes materials for the building envelope, including the walls, roof, slab on grade, windows, and glazing;

(b)

Building geometry modeling. This includes modeling the building’s length, width, height, and orientation using Grasshopper;

(c)

Energy simulations. Various combinations of parameters were tested in order to generate different scenarios and determine their corresponding range for energy consumption. These simulations were performed in EnergyPlus;

(d)

The generation of the database. Twelve scenarios were used with 1000 simulations for each one, for a total of 12,000 generated simulations. These simulations were stored, including the inputs and outputs of each one.

This model takes into consideration different design parameters as well as climate and thermal comfort to generate the annual energy consumption in a building. It considers energy consumption due to heating, lighting, cooling, and equipment. The generated scenarios form a database which could be used for predicting the energy consumption of buildings.

4.2. Artificial Neural Network Model for Energy Prediction

The ANN model aimed to predict the energy usage of buildings emanating from heating, cooling, lighting, and other plug-ins. It combines parametric modeling with simulation tools to generate a database for energy consumption. Although there are several machine learning algorithms, ANNs have been widely used in the area of energy analysis and prediction. For example, [61] reported the superiority of ANN models in the prediction of energy consumption over other methods. Hence, ANN was used in this study for its effectiveness and superiority. The network used in this study was a supervised learning, feed-forward neural network with a back-propagation algorithm [59]. The networks were trained using the Levenberg–Marquardt back-propagation algorithm, which is appropriate for modeling non-linear functions, such as those in this study [62]. It was developed over three phases: the creation of the initial model, the optimization of it, and the evaluation of its performance. A database of residential buildings was used to develop this model [63] and a user-friendly interface was also developed to enable quick and easy energy predictions.

4.3. GA’s Optimization Model

This section discusses the proposed novel optimization model. This represents the final section of the study proposed by [60] in the overall study methodology. This proposed model aims to support architects and designers in their quest to achieve optimal energy performance in building design. The optimization model is carried out with an objective of minimizing the building’s energy consumption, as shown in Equation (1). The main focus of the optimization process is on the conceptual design phase due to its significance in determining the energy usage performance as well as thermal indoor comfort.

Min. ∮ Annual predicted Energy Use Intensity: pEUI

(1)

The previously proposed ANN model is used to calculate the simulation parameters. These parameters are then optimized in the optimization model to evaluate the design space options. In the optimization model, a building is limited to a massing box with known building dimensions (length, depth, and height). The defined variables for this study are the wall type, SOG type, roof type, building orientation, window-to-wall ratio for each façade (north, south, east, and west), Glass U-value, Glass SHGC, Glass VT, cooling set point, and heating set point. In this paper, GA is used to select the optimal building design based on minimal energy usage.

Grasshopper is used to perform the design simulations, which are stored in the database. These data are then used for training and validation using an artificial neural network. The proposed ANN model is automatically linked to the GA for calculating the energy consumption for each generated solution without needing simulation tools. The GA elements for the design optimization model consist of four steps: the determination of the gene representation; determination of the initial population size; determination of the ranked fitness; and determination of the GA operators and their probabilities, as discussed below.

4.3.1. Gene Representation

In GAs, a coding scheme is needed to encode the variables of the problem into a gene. The variables are usually coded into a fixed-length string of binary representations or real values. This study selected the encoded method with real values. Figure 2 shows an example of the solution representation to a given GA optimization problem. Each solution (alternative design) is represented by a string of elements with a length equal to the number of variables (16 variables), with each element (chromosome) in the solution string representing the optimization variable.

4.3.2. Population Size

One of the advantages of GA is that it can search numerous points in the search space in parallel. The parallel search size is called the population size and is equal to the number of genes. The population size selection is one of the most important aspects of the design of GAs. Currently, most GAs work with a constant population size (N), which is determined as a user-controlled optimization input parameter. The population size selection needs to be determined using experiments and is considered as a dependent problem that affects the GA’s performance and processing time significantly. A larger population size (in the hundreds) would enable the exploration of a larger search space and increase the probability of obtaining a global optimal solution. However, it could slow the convergence process and increase the processing time. A smaller population size, however, would have inaccurate solutions [64].

4.3.3. Fitness Function

The fitness function is fundamentally the objective function and is used to evaluate the solutions (energy consumption) based on the objective function. The GA process starts with a randomized population of solutions. As the energy consumption decreases, a higher fitness is assigned to an individual. The fitness value of each chromosome (optimal solution) is calculated using Equation (2). A good individual is one that has a high fitness value. In this research, each individual was ranked based on its relative merit, which is calculated as the individual’s fitness divided by the total fitness of the whole population [65].

fitness (i) = 1/pEUI

(2)

4.3.4. Genetic Operators

• Reproduction

GA is based on generating an initial population that reproduces into generations in order to reach the best solution. In addition, the number of generations needed to reach convergence are problem-dependent and have to be determined experimentally. The following steps implement the reproduction operator: (1) scaling the raw fitness scores; (2) identifying the better string (selection operator) with the higher fitness value; (3) duplicating the better chromosome (optimal solution); and (4) eliminating the worst solution with the lower fitness value.

The most significant issue for the operator section is the selection pressure; this is a way of ensuring the “survival of the fittest” strings. Many methods are used to implement this operator but proportional selection is the most common reproduction operator. The probability of an individual being selected is calculated using Equation (3).

$$Psurvival \left(i\right)=\frac{fitness \left(i\right)}{{{\displaystyle \sum }}_{j=1}^{Pop size}fitness \left(i\right)}$$

(3)

This type of selection is performed using the roulette wheel mechanism n times. A random number, r, in the range of 0 to 1 is generated each time, in which the individual with the highest fitness value has a higher chance of being selected.

• Crossover

This operator is used to select the best solution then pass it on to the mating pool and discard the worst solution. Therefore, crossover is used to recombine the selected solutions (parent genes) to generate new solutions (offspring/children). Crossover can be implemented in different ways. In this study, each of the two parent genes is randomly selected with the exchange of information between the two genes performed to create a child gene with more than one cross point. Figure 3 shows an illustrative example of the multi-point crossover operation. This method, known as the N-point of the crossover operator, is presented, in which one copy of each parent solution (“A” and “B”) along with one offspring are present after the crossover operation [64]. However, not all the genes in the population will be subjected to the crossover operation; this is controlled by the crossover probability (P_c), which is experimentally determined.

Mutation is another operation that disturbs the chromosome values in the confidence of finding a best solution and avoids the local optimum solutions by randomly and suddenly altering the gene values. Mutation is important because it can ensure the discovery of more solutions in the solution space. Moreover, mutation prevents premature convergence by increasing the population diversity [66]. Similar to crossover, the mutation process is controlled by probability (P_m). For this study, a number “r” is randomly generated in the range of 0 to 1. If r is smaller than (P_m), the mutation takes place and generates a new gene randomly. The crossover rate normally ranges from 0.6 to 0.9 and the mutation rate ranges from 0.05 to 0.2; hence, the values used for experimentation were within these ranges.

• Optimization Model Implementation

As previously mentioned, the process of the design optimization of buildings while considering energy efficiency requires both simulations and ANN tools to be used for evaluating objective functions during the optimization process. Before starting the optimization, a user usually needs to define the optimization problem’s parameters. It must be noted that GA’s performance can be sensitive to its control parameters. The change in the crossover or mutation probability rate would have an effect on the performance of the algorithms. To identify the proper control parameters’ values, the GA performance would need to be assessed for various sets of parameters. The main control parameters that were applied in the optimization model and their values are presented in

Table 1. GA control parameters sets.

Control Parameter	Values
Population size	[100,150,200]
Crossover probability	[0.8,0.9]
Mutation probability	[0.1,0.2]
No. of Generations/Cycle	[25,50,100]

.

The integration between the GA and ANN is modeled and coded using Visual Basic (VB). The algorithm code is based on the idea of intelligently maximizing the call for an objective function (called the ANN model) before proceeding to the optimization process as well as during the optimization process.

A user interface application is developed using VB to facilitate the process of data entry. Figure 4 represents the proposed GA pseudo code for the random creation of an initial population, evaluation of each solution (alternative design) via the ANN model, determination of the best solution (minimum energy) and worst solution (maximum energy), and iterative performance of mutation and crossover until the termination criterion is reached. Upon starting the proposed VB model, the user is asked to enter the input data that start up the optimization using NeuroSolutions. The input data include: the building length, building depth, building height, population size, number of generations/cycles, and crossover probability (P_c), as shown in Figure 5.

In order to detect the most appropriate parameters, different sets of GA control parameters are generated. This is implemented on a sample building to be optimized with different parameter values. A massing box represents a residential building in the conceptual design phase with one thermal zone with the following dimensions: the building length = 15.0 m, building depth = 20.0 m, and building height = 12.0 m.

In this study, the Cairo airport weather file is selected, which represents the surrounding model’s environmental conditions. The control parameter sets that are used for the experiment are presented in

Table 2. Optimization model parameter sets.

Population Size	Crossover Rate	Generation Number
	0.7	25
100	0.7	50
	0.7	100
	0.9	25
	0.9	50
	0.9	100
	0.7	25
150	0.7	50
	0.7	100
	0.9	25
	0.9	50
	0.9	100
	0.7	25
200	0.7	50
	0.7	100
	0.9	25
	0.9	50
	0.9	100

. Three population sizes are chosen (100, 150, 200) and the crossover rate is either 0.7 or 0.9 while the generation number ranges from 25 to 100.

Due to the stochastic nature of the GA, each run is expected to produce different results for the same set of parameters. As such, 10 runs are performed and the best solution which represents the minimal energy consumption for each parameter set is presented in

Table 3. Values of objective function for the best solution.

Index	Parameter Sets	Best Solution (Watts)
1	[100,0.7,25]	21,044.80
2	[100,0.7,50]	20,105.18
3	[100,0.7,100]	20,325.30
4	[100,0.9,25]	19,827.70
5	[100,0.9,50]	20,249.90
6	[100,0.9,100]	20,192.84
7	[150,0.7,25]	19,832.80
8	[150,0.7,50]	19,962.80
9	[150,0.7,100]	19,767.65
10	[150,0.9,25]	20,334.80
11	[150,0.9,50]	20,047.10
12	[150,0.9,100]	19,708.00
13	[200,0.7,25]	20,489.46
14	[200,0.7,50]	19,913.17
15	[200,0.7,100]	20,080.30
16	[200,0.9,25]	19,888.20
17	[200,0.9,50]	19,600.60
18	[200,0.9,100]	20,106.80

.

The results presented in Table 3 show that the lowest minimum energy consumption is achieved by index no. 17 (19,600.60 watts) as compared to the other parameter sets, which means that the recommended GA parameters for the population size, crossover rate, mutation rate, and generation number are 200, 0.9, 0.1, and 50, respectively. In this study, the GA process stops when convergence occurs if the objective does not improve in ten consecutive cycles. Figure 6 presents the experiments with different population sizes (100, 150, and 200) and numbers of generations for different evolutionary cycles (25, 50, and 100) with a crossover probability of 0.7 and 0.9, which illustrates the whole process and the effect of varying the population size. The figure shows that results can be improved if a larger population size is used.

5. Case Study

The proposed model was applied to a case study representing an actual construction project in Egypt. It was applied to explore the tradeoff between the energy consumption and the thermal comfort of occupants in a typical residential building. The case study is a single-family residential building located in New Cairo (Figure 7).

At the project design stage, the design team needs to determine the most appropriate design that achieves the minimum energy consumption among alternative designs.

Table 4. Characteristics of the case study building.

Location	New Cairo
Number of floors	2 stories
Floor area	252 m2
Building Length	14 m
Building Depth	18 m
Floor-to-floor height	3 m
Number of floors	3
Exterior wall area	380 m2
Orientation	45°
North façade (WWR)	20%
South façade (WWR)	10%
East façade (WWR)	15%
West façade (WWR)	15%
Wall Type	Wall type 1
Roof Type	Roof type 1
SOG Type	SOG type 1
Glass Type	Single glass 6 mm
Glass U-value	5.67 w/m2k
Glass SHGC	1.0
Glass VT	0.7
Heating Set point	22 °C
Cooling Set Point	22 °C

summarizes the set of information that represents the basic characteristics of the building envelope and thermal energy simulation parameters. These characteristics are the baseline design options for the building. For example, the building’s baseline orientation is 45°. There are many possible combinations of building design options, which can make the selection process complicated. Additionally, a large number of combinations require a long computing time and the interoperability between the modeling tool, simulation tool, and optimization engine can be a complex issue. Hence, the design team can face difficulties when considering the tradeoffs between designs to calculate the energy consumption. The proposed model can aid the design team in their choice and streamline the process.

The simulation for the baseline design option was performed using EnergyPlus. The model was applied to the project and the energy consumption for the baseline design option was 35,623.07 watts. After entering the user information in the optimization model, the optimization was performed to reach the design that achieved the minimum energy consumption, as listed in

Table 5. The optimum design for the case study.

#	Variables	Value
1	Wall Type	2
2	SOG Type	1
3	Roof Type	1
4	Orientation (°)	48
5	South windows to wall ratio (%)	16%
6	East windows to wall ratio (%)	13%
7	North windows to wall ratio (%)	18%
8	West windows to wall ratio (%)	11%
9	Glass U-value	0.6
10	SHGC	0.5
11	VT	0.5
12	Cooling set point	22
13	Heating set point	20
	Energy Consumption	26,421.12 watts

. The wall, SOG, and roof types had limitations in the simulation process when considering the building dimensions as constraints [60]. As a result, the most appropriate design that achieved the minimum energy usage included the increase in the envelope walls and roof thickness. In addition to that, the WWR had a great influence, in which the best ratio was found in the north and south, while avoiding the huge amount of heat gained from the western and eastern facades. The resulting percentage of energy savings was 25.8% when using the developed optimization algorithms. Project design teams would value this process and outcome because it allows designers to select the best design accurately and quickly while overcoming the interoperability issue.

6. Discussion

This paper presented an optimization model for energy consumption to facilitate early-stage building design. The model was based on the integration of a simulation tool and a genetic algorithm optimization engine using Middleware, which is an ANN. The model has a user-friendly interface that does not require any technical experience to aid decision-makers in their choice of an optimal building design. In order to select the optimization parameters, experiments were conducted with different population sizes (100, 150, and 200) and numbers of generations for different evolutionary cycles (25, 50, and 100) with a crossover probability of 0.7 and 0.9. The recommended optimization parameters were 200 for the population size, 50 for the generations of cycles, and a 0.9 crossover probability. These parameters were then used in a case study to reach the optimal design of a residential building in New Cairo. The optimized design was reached within 22 generations (1100 generated genes). These design examples show the ability of the proposed model to reach an optimal or near-optimal solution. It should be noted that the solutions of these cases took about two to three minutes while running on a 300 MHz Pentium PC. Designers can then obtain valuable information on multiple design scenarios and make effective and optimal design choices. The proposed optimization model is part of a holistic workflow that aims to reach an optimal design using one platform to simplify the user experience and streamline the process of choosing the optimal design. The optimization result (26421.12 watts) was compared with that obtained from the simulation for the generated design without performing the optimization (35,623.07 watts) which shows an energy saving of about 28%. This shows the capability of the optimization model to significantly save on energy consumption.

Practically, there is no perfect approach that can be applied to all research fields and there is no computational algorithm that fits all problems. Artificial intelligence techniques have strongly gained researchers’ interest due to their wide use in several domains, including online learning [67], scheduling [68,69,70], multi-objective optimization [71], manufacturing, business intelligence, transportation [72], and healthcare [73]. This interest has led to the proposal of a plethora of approaches by researchers and practitioners in the area of optimization.

As previously mentioned, this study focused on using genetic algorithms to find the global solution for an optimization problem with incomplete/imperfect information. This was due to its reported capability to work with random variables, objective functions, and constraints. However, several advanced optimization algorithms have recently been proposed in the literature to solve other complex engineering problems. These algorithms can also be applied to the energy optimization problem and their accuracy can be compared with that of the proposed GA model. For example, Ref. [71] proposed a novel hybrid multi-objective evolutionary optimization model that can be used to analyze the tradeoffs between conflicting objectives. This can be of use in the energy domain if multiple objectives are studied, such as energy minimization and the optimal cost during the building’s life cycle. The authors of [73] applied two metaheuristics, NSGA-II (Non-dominated Sorting Genetic Algorithm II) and MOPSO (Multi-Objective Particle Swarm Optimization) and reported the superiority of the first method. These methods can be used to facilitate the decision-making process for selecting energy efficiency parameters. The authors of [68] proposed a novel heuristic algorithm that can improve the quality of the obtained solutions after performing crossover and mutation operations. This can be used to enhance the GA operations and the practical considerations of a building’s energy performance design. The authors of [73] proposed a novel memetic algorithm that can adjust the mutation rates based on a deterministic parameter strategy. Similarly, Ref. [74] used a memetic algorithm to minimize the total energy consumption, aiming at obtaining a good initial population through a chromosome-encoding method and a local search operator. The authors of [69] proposed a new mixed-integer linear programming mathematical model that can change the parameters throughout the search process. All these new techniques can be applied to the energy consumption optimization problem in residential buildings.

7. Recommendations for Future Work

Our recommendations for future work include applying a multi-objective perspective to the optimization issue. This is due to the fact that building performance optimization consists of tradeoffs between various objectives [17]. Multi-objective optimization can also be conducted while considering both energy and costs to minimize energy consumption along with life cycle costs. Optimization can be conducted at multiple stages; it can start at the preliminary design stage and be conducted again as the design progresses into the detailed stage [58]. Other optimization tools can be explored to identify solutions since building simulations can be non-linear [58]. Design optimization can also be studied at a larger scale, for neighborhoods or residential compounds, which would be beneficial to occupants. The number of design energy parameters can also be increased to include more efficiency measures depending on the user’s needs or the design team’s objectives. Daylight analyses can also be added to the study along with energy consumption to obtain a full picture of the energy performance of buildings. The proposed tool can also be compared to other available tools and sensitivity analyses conducted on them. Additionally, more experiments can be conducted to assess the efficiency of the various algorithms, such as ant colony optimizations, tabu searches, variable neighborhood searches, and simulated annealing. Future studies can also develop a hybrid algorithm and take into consideration more factors, such as building exergy. The analysis can be performed at a more macro scale to not only look at separate buildings, but also consider blocks of buildings or even cities. Finally, new heuristics can be developed to ensure population diversity, chromosome representations, and parameter updates [69].

8. Conclusions

This paper introduced an optimization model for a building design during the conceptual design stage in terms of energy consumption. The optimization model was created by developing a VB application to facilitate the optimization process and allow flexible data entry and efficient calculations. The developed model used a genetic algorithm integrated with an ANN model to facilitate the objective function evaluation process without needing to perform any energy simulations. The main tenets of the proposed model were that it: (1) was able to assess various design alternatives; (2) overcame the interoperability problem by integrating the simulation and optimization tools; and (3) suggested optimization parameters for residential buildings in the early design stage. The effectiveness of different GA control parameter sets was examined to find solutions to a building optimization problem. Sets of parameters were tested and the results highlight that the GA’s efficiency is sensitive to the control parameter values, which are the population size, probability rate of crossover, and number of generations per each cycle. However, the population size was the control parameter that had the highest effect. The proposed model was applied to a case study and resulted in a 25% reduction in the energy consumption. This highlights the importance and effectiveness of studying energy consumption in the early stages of building design.