1. Introduction
Climate change describes the long-term modification in Earth’s climate patterns, arising from a twisted interaction of natural factors with human activities. Natural factors contributing to climate change include differences in solar gain, volcanic eruptions, and greenhouse gas emissions from natural sources like oceans and vegetation. In this regard, the principal factor contributing to modern climate change is human activity. This includes deforestation, fossil fuel burning, industrial processes, and agriculture, all of which have significantly deteriorated the concentration of greenhouse gas in the atmosphere and cause global warming [
1,
2]. The European Environment Agency (EEA) projects a temperature increase of 2.5 to 4 °C across Europe by the period 2071–2100 [
3]. This issue leads to the emergence of urban heatwaves, driven by the interaction of global warming with the Urban Heat Island (UHI) phenomenon, which has become a critical issue for cities worldwide [
4,
5]. These heatwaves are driven by the dual effect of global warming and the localized heat buildup characteristics of UHI, making it a critical problem [
6]. For example, the 2003 summer heatwave in Europe had a devastating impact, resulting in over 35,000 deaths. Specifically, it claimed 14,729 lives in France [
7], 2139 in Wales and England [
8], near 2200 in the Netherlands [
9], and 7295 in Germany [
10]. During intense heatwaves, a building’s capability of adapting efficiently, absorbing, and mitigating the disruptive impacts of extremely high temperatures becomes crucial. Given the close relationship between humans and their living environment [
11,
12], this capability protects occupants from environmental heat stress and subsequently restores the building to its standard operational state. This crucial ability is commonly denoted as “resilience”, constituting a pivotal concept that encompasses a building’s capacity to adapt anticipate, and promptly recover from hazards and disruptive incidents [
13,
14]. Hence, within the realm of building design, in addition to energy conservation, buildings are also engineered to be thermally flexible. The resilience of buildings against specific types of shocks is evaluated and enhanced through policies implemented by building codes [
15,
16,
17]. In this regard, it is strongly advocated to optimize and augment the thermal efficacy of edifices to function as protective havens, safeguarding occupants during elevated ambient temperatures.
The term “optimization” is frequently used to describe the process or series of steps aimed at achieving the utmost perfection, functionality, or effectiveness in a design, system, or decision [
18]. In the conventional context of building performance simulation (BPS), “optimization” implies identifying the ideal solutions to address a problem, as this might be impractical given the inherent complexities of the issue [
19,
20] or limitations within the simulation process [
21]. Some researchers also use the term “optimization” to describe an improvement cycle utilizing computer-based simulations to obtain solutions that, while not the absolute best, are significantly improved [
19,
21,
22]. Alternatively, other authors have explored optimizing building performance through “sensitivity analysis” and “design of experiment” methods without relying on mathematical optimization [
23,
24]. Additional methods for building optimization include “brute-force search” [
25], and “expert-based” optimizations [
26]. Nonetheless, it is widely acknowledged among “simulation-based optimization” experts that this concept denotes a fully automated process relying solely on mathematical models and numerical simulations. In the conventional studies, this automation typically occurs through the integration of a building simulation platform with an optimization “engine” that incorporates some optimization methods and algorithms [
27].
More recently, the use of machine learning algorithms has emerged as a prominent and convenient method to optimize multiple objectives simultaneously, without the need for optimization ‘engines’ and relying solely on coding. This approach not only reduces operational complexity but also decreases calculation time [
28,
29,
30]. Machine learning models, such as XGBoost or Gaussian processes, serve as predictive tools that can approximate objective functions in optimization tasks. These surrogate models help accelerate the optimization process by providing fast and accurate predictions, reducing the need for computationally expensive simulations. By integrating these models with optimization algorithms, large solution spaces can be explored more efficiently than with traditional methods [
31]. Techniques like Bayesian optimization utilize these models to iteratively propose new solutions based on prior evaluations, balancing exploration and exploitation to converge on optimal solutions effectively. Additionally, data-driven optimization supports multi-objective scenarios by generating Pareto-optimal solutions, which is a method for selecting the optimal solution between various scenarios, offering a set of trade-off solutions among conflicting objectives [
32]. This is particularly useful in complex systems, where manual tuning would be infeasible. In summary, the application of machine learning in optimization facilitates a more streamlined, efficient, and versatile approach to solving complex, multi-objective problems, making it an invaluable tool in various fields such as engineering, finance, and operations research.
1.1. The Literature Review
Extensive research has focused on enhancing building performance across diverse regions. Through the integration of several input factors and advanced strategies, these publications have shown valuable outcomes in decreasing energy consumption and enhancing indoor thermal conditions. Zhong et al. investigated the impact of various behavioral models of window openings on indoor environments in Hong Kong, revealing that static temperature threshold models can lead to discrepancies in predicting window use and indoor parameters [
33]. Liu et al. [
34] proposed a method to optimize variable air volume (VAV) terminal zone operations in buildings with “mixed-mode” ventilation, suggesting thermostat control adjustments to encourage window opening during favorable conditions. Similarly, Liu et al. [
35] demonstrated that regulating window operations can achieve energy savings of 3–16%. Moreover, Zhang et al. [
36] investigated the effects of window design on energy use in office buildings, finding that the south window-to-wall ratio (WWR) and glazing type had the most significant impact. Liu et al. [
37] also discuss the use of operable windows in modern office buildings in North America, highlighting their benefits and potential hazards related to using extra energy. The study analyzes data on windows and the use of thermostat in two buildings with mixed-mode ventilation in Ottawa, Canada, and uses various models in order to predict the likelihood of these activities and identify triggers. Based on the findings, the paper presented preliminary recommendations to enhance terminal device arranging in mixed-mode ventilation buildings in cold environments, including applying thermostat set point setbacks to encourage energy-saving window use.
Recent studies have combined data science with traditional methods to optimize building performance. For instance, Wu et al. [
38] conducted research on improving thermal comfort, energy usage, and daylighting in a residential building using XGBoost machine learning algorithm. The study focused on 11 objectives and proposed an optimization strategy for combining the “Bayesian method” and the “non-dominated sorting genetic algorithm II (NSGA-II)”. According to the results, the BO-XGBoost model could provide excellent predictive performance, and achieve multi-objective optimization effectively. Additionally, Bre et al. [
39] devised an innovative approach for multi-objective optimization in building performance area. They paired the multi-objective NSGA-II with artificial neural network (ANN) meta-models, trained using data from EnergyPlus software simulations. The results showed that the method can reduce the required number of physics-based simulations by almost 75% while sustaining the precision. Moreover, Chen.et al. established an innovative multi-objective optimization outline that merges building information modeling (BIM) with advanced machine learning methods like least squares support vector machine (LSSVM) and NSGAII. Their study, validated by investigating a teaching building as the case study, demonstrated that the LSSVM-NSGA-II approach is highly effective, achieving a 10.6% decrease in energy usage and a 32.2% progress in thermal efficiency [
40]. Guo et al. [
41] focused on optimizing the performance and economy of village houses in cold regions of China by artificial neural networks (ANN) and multi-objective optimization techniques such as NSGA-III. The research recommends increasing the population size and iteration count for NSGA-III to achieve better results, suggesting that a population size of 50 or more and at least 200 iterations yield optimal performance in the optimization process. Shen et al. [
42] presented a novel approach that integrates a combined model with a multi-objective optimization algorithm. Results showed a significant improvement in prediction accuracy, achieving a 9.19% increase in the R
2 value and substantial reductions in error metrics. By incorporating SHAP values, the study also provides valuable insights into the influence of various architectural design parameters on building performance. Similarly, Seraj et al. [
43] highlighted the significant energy consumption of residential buildings in the UK and the urgent need for retrofitting to meet efficiency targets. They developed a data-driven AI model to predict energy performance based on building features and retrofit scenarios using the Energy Performance Certificate (EPC) dataset, finding that the artificial neural network model is the most accurate, with an R
2 of 0.82 and a mean percentage error of 11.9%. However, inconsistencies were noted across models for different retrofit scenarios, suggesting that AI tools can enhance traditional physics-based models, improving the retrofitting process in residential buildings. Furthermore, Tariq et al. [
44] show that energy consumption in educational facilities significantly affects sustainable development, contributing to approximately 37% of carbon dioxide emissions. This research explores AI models, including decision trees, K-nearest neighbors, gradient boosting, and LSTM, to predict energy use in schools. Findings indicate that school size and air conditioning capacity are major factors influencing energy consumption, while the type of school has a weaker correlation. The decision tree model achieves an average prediction error of 3.58%, whereas gradient boosting and LSTM effectively manage data variability. This research underscores the potential of sustainable educational buildings to promote energy efficiency and environmental stewardship, serving as interactive learning environments for students. lastly, Golafshani et al. [
45] investigates energy consumption dynamics in buildings through data-driven modeling with machine learning algorithms, analyzing 66,800 energy records from office buildings in 40 cities. Utilizing algorithms like random forest, extra trees, gradient boosting (GBR), and eXtreme gradient boosting (XGBR), the study found GBR and XGBR models to have a mean absolute percentage error under 1.2% and a coefficient of determination above 0.99. It emphasizes the significant impact of building shape and temperature on energy consumption and employs the grey wolf optimizer to identify optimal design parameters, laying a foundation for energy-efficient buildings in diverse climates.
1.2. Research Gaps and Objectives
Current building design optimization methods face several limitations, particularly in terms of computational efficiency, adaptability to future climate scenarios, and the ability to address dynamic comfort metrics. Traditional simulation-based methods like EnergyPlus, though effective in detailed modeling, require extensive computational resources and iterative simulations that make them impractical for quick applications. Additionally, while optimization techniques such as genetic algorithms (GA) have shown promise, their integration with advanced simulation tools remains underdeveloped, limiting the potential for a more efficient, combined approach.
Although some studies have explored machine learning techniques, there is still a significant gap in their combined usage with GA to optimize building energy performance and occupant comfort, particularly in terms of multi-criteria such as total energy (TE), indoor overheating degree (IOD), and predicted percentage dissatisfied (PPD). Furthermore, the influence of future climate scenarios on building energy optimization is often overlooked, which is essential for ensuring long-term sustainability and resilience. Lastly, the optimization of building management systems, especially concerning dynamic control strategies for heating, cooling, and ventilation based on real-time environmental conditions, remains underexplored.
To address the mentioned gaps, the primary objective of this research is to develop a comprehensive and computationally efficient optimization framework that integrates Bayesian optimization and XGBoost algorithms with the non-dominated sorting genetic algorithm II (NSGA-II). This framework aims to optimize building design and operation, specifically focusing on key performance indicators such as total energy (TE), indoor overheating degree (IOD), and predicted percentage dissatisfied (PPD). The study will also explore multi-objective optimization under different climate scenarios—historical (2020), mid-future (2050), and future (2080)—to ensure building designs are resilient and sustainable. Additionally, the research seeks to optimize building management systems by incorporating dynamic control strategies to enhance both energy efficiency and occupant comfort.
4. Discussion
4.1. Analysis of the Influential Factors in Different Environmental and Climatic Contexts
The investigation of the influence of various design parameters on TE, PPD, and IOD across different periods reveals both consistencies and deviations from previous research. In analyzing the influence of various factors on TE over different periods, the WWR consistently emerges as the most significant variable. In the baseline scenario, WWR has a substantial negative impact on TE, indicating that the window-to-wall ratio is the most decisive variable that can be controlled to decrease the total energy of buildings. For instance, larger windows enhance daylight penetration and air circulation between the indoor and outdoor environments, subsequently reducing the energy required for air conditioning and lighting. This observation is consistent with prior studies emphasizing the importance of window management in energy efficiency strategies [
97]. Additionally, Tout is identified as the second most significant variable, also negatively affecting TE due to its impact on Tin, which increases energy demands. However, some researchers suggest that the negative effects of WWR can be mitigated with advanced glazing technologies and shading devices [
98]. In future scenarios, WWR remains significant, with a trend towards a more positive impact, indicating that advancements in building technologies and improved radiation control can enhance energy efficiency. These findings suggest that future building designs should incorporate innovative solutions to effectively manage the adverse effects of large window areas on energy consumption.
Regarding PPD, the base case scenario identifies Tout and Tin as the most significant variables, demonstrating the predominantly negative and positive impacts, respectively. This emphasizes the critical role of thermal regulation in maintaining occupant convenience, as larger temperature differences between indoor and outdoor environments contribute to human discomfort [
99]. These results are consistent with studies emphasizing the need for efficient thermal insulation and ventilation systems to reduce occupant dissatisfaction [
100]. Although WWR and opening area also play important roles, their effects on PPD are secondary to the direct thermal exchanges between indoor and outdoor environments. In mid-future and future scenarios, the pattern remains similar, with a strong negative influence of Tout on PPD. This persistent trend indicates the necessity for improved building envelopes and adaptive thermal comfort strategies to maintain occupant satisfaction. Some negative viewpoints from the literature suggest that without significant advancements in insulation and climate control technologies, achieving satisfactory PPD levels will remain challenging in extreme climates [
101].
IOD is significantly influenced by the WWR and Tin, both of which exert a considerable positive effect. Larger windows increase solar heat gain, raising indoor temperatures and exacerbating overheating, particularly in warm climates. This relationship is supported by previous research emphasizing the thermal load implications of extensive glazing [
102]. Additionally, Tout plays a crucial role on understanding the broader impact of external environmental factors on indoor thermal conditions. The variability in Tout directly affects the amount of heat transfer between the outside and inside of buildings, thereby influencing indoor comfort levels and energy demands. This outcome underscores the significance of pondering local climate patterns and seasonal variations when designing and managing indoor environments to mitigate overheating risks. Looking ahead, the persistent influence of WWR underscores the ongoing challenge of managing solar heat gain through effective window design and shading strategies. Despite advancements in architectural design and materials, such as dynamic shading systems and low solar heat gain glazing, the enduring positive impact of WWR on IOD necessitates continual innovation in building practices to ensure optimal indoor thermal comfort. Addressing these challenges comprehensively is essential for adapting to climate change and promoting sustainable indoor environments.
Ultimately, it can be concluded that climate change is decreasing the sensitivity range of sensors for both outside and inside temperatures, as they control the number of openings and efficiency of operations. As a result, sensors may be used less frequently in the future for controlling indoor conditions. These findings have the potential to impact the building envelopes, facilitating the production of resilient and adaptable structures capable of efficiently responding to the shifting situations. Furthermore, they might influence building programs and guidelines, which could result in several instructions that promote the use of building envelope design principles. Additionally, the findings of this study could serve as foundational material for advanced research and the expansion of innovative control systems and technologies. Consequently, there will be a particular emphasis on smart construction mechanisms for dynamically adapting to the residents’ desires in real time. The outcomes of the sensitivity analysis offer a thorough comprehension of influencing factors, modification of control policies, optimization of design parameters and development of an operative strategies. Such perceptions help improve indoor circumstances and resident’s convenience, reduce energy consumption, and finally decrease the indoor overheating rate.
4.2. Reliability Analysis for Design Purposes
To assess the reliability of the models, sensitivity analysis was employed, a method that evaluates how variations in input parameters affect the outcomes. Through this process, critical design parameters, such as the window-to-wall ratio (WWR) and indoor temperatures (Tin) were identified as significantly influencing energy performance metrics. By systematically adjusting these parameters and observing the resulting changes in TE, PPD, and IOD, the most influential factors that architects and engineers should prioritize during the design phase were pinpointed. In addition, simulation results were juxtaposed with empirical data from similar case studies to ensure that the models accurately represented real-world conditions. This comparison not only highlighted potential discrepancies but also allowed for model calibration, thereby enhancing predictive accuracy. Given the complexity of buildings as systems with numerous interdependent variables, ensuring that the models reflect this complexity is paramount for their practical application. Moreover, current analysis indicates that the implications of reliability extend beyond individual design decisions. They have broader ramifications for regulatory frameworks and building codes, which increasingly emphasize performance-based metrics. As the building industry shifts towards more sustainable practices, the demand for reliable predictive models will grow, providing a framework for evaluating compliance with energy efficiency standards and occupant comfort requirements.
4.3. Potential Limitations and Suggestions
It is crucial to acknowledge the limitations of the current research. Primarily, the reliance on simulation implements and algorithms might restrict its applicability to real-world scenarios. The precision of the simulations may be influenced by the accuracy of the input values and the hypotheses included in the frameworks. Additionally, the use of a simplified “shoebox” model, while effective for initial validation of the optimization framework, may not fully represent the complexities of real-world building designs. Although this model is commonly used in energy analysis, future studies should explore more complex and realistic models to better assess the framework’s applicability in practical scenarios. Furthermore, the case scenario is hypothetical, aimed at comparing various control strategies and their outcomes in the envelope design. The comparative investigation was limited to a particular climate, which may further restrict the applicability of the results to other climatic conditions.
To mitigate these constraints, future studies could implement extensive fieldwork to validate the parameters, modeling applications, and results derived from previous studies. Improving the quality of the input data by utilizing real-time data collection systems, such as building management systems (BMS), and incorporating more precise occupant behavior models will enhance the representativeness of the results. Fieldwork-based data collection, including monitoring energy use and indoor environmental quality, can provide a better understanding of actual building performance under varying conditions. This type of research can yield significant insights to identify the most suitable control strategies that accurately reflect user behavior concerning building openings. Furthermore, investigating possible variations in residents’ activities within real buildings is essential to gain a comprehensive understanding of user behaviors and preferences.
5. Conclusions
This study deeply explores the intricate interplay of environmental and climatic factors on building performance, occupant comfort, and indoor overheating risk. The findings underscore the profound impact of variables such as WWR, external and internal temperatures (Tout and Tin), and their complex relationships with TE, PPD, and IOD. Climate change exacerbates these challenges, narrowing the margin of error for building sensors that regulate environmental conditions, potentially diminishing their efficacy over time. This critical issue necessitates a reevaluation of building envelope strategies to enhance resilience and adaptability against increasingly volatile weather patterns. As climate conditions continue to evolve, the increased external temperatures (Tout) exacerbate energy consumption and discomfort levels, revealing that traditional metrics like the predicted mean vote (PMV) fail to fully capture the dynamic nature of thermal discomfort.
The introduction of IOD in this study provides a more comprehensive and precise measure by quantifying the frequency and intensity of overheating events, proving to be a more reliable metric for thermal comfort assessment.
Additionally, this research highlights that increasing WWR and indoor temperatures contribute to elevated IOD values, signaling the need for careful management of building envelopes and HVAC systems, especially under future climate scenarios.
The integration of IOD with energy efficiency metrics, such as energy use intensity (EUI), offers a more holistic approach to assessing building performance, optimizing energy efficiency, thermal comfort, and occupant well-being.
The implications extend beyond theoretical insights, urging policy makers to reconsider building codes and regulations that prioritize adaptive design principles. Furthermore, this study advocates for advancing smart building technologies capable of dynamically adjusting indoor environments in response to real-time data, thereby optimizing both energy efficiency and occupant well-being. The application of advanced machine learning techniques can further refine simulation models, ensuring their robustness and applicability across varying building types and climates. This research provides a critical foundation for more resilient and energy-efficient building designs, advancing sustainable practices to mitigate the long-term impacts of climate change.
