AI-enhanced automation of building energy optimization using a hybrid stacked model and genetic algorithms: experiments with seven machine learning techniques and a deep neural network

Energy efficiency is a key concern of architectural design since real estate consumers demand lower energy costs. In addition, Governments are increasingly committed to prioritizing energy efficiency initiatives and implementing energy measures to address climate change and its impacts. Recent advances in artificial intelligence (AI) may allow for precise measurement and optimization of building energy saving, which is yet to be explored. This paper aims to examine the integration of AI in automating the prediction and optimization of energy performance in residential buildings. Unlike the common practices, the scenario development and building optimization tend to be enhanced by using AI in the present paper, using a Python-based script alongside the Non-dominated Sorting Genetic Algorithm (NSGA-II) through EnergyPlus simulations. This approach balances multiple objectives, such as energy consumption and thermal comfort, to streamline the identification of optimal building configurations. Seven machine learning (ML) models, including Linear Regression (LR), Decision Trees (DT), Random Forest Regressor (RFR), Gradient Boosting Machines (GBM), Support Vector Regressor (SVR), K-Nearest Neighbors (KNN), and Extreme Gradient Boosting (XGB), and a deep Feedforward Neural Network (FNN) are developed and assessed in predicting three key performance metrics: Energy Use Intensity (EUI), Predicted Percentage Dissatisfied (PPD), and Heating Load. A hybrid stacked model, combining FNN with XGB, using GBM meta learner, emerged as the top performer, achieving an impressive Coefficient of Determination (R²) of 0.99 and Mean Absolute Percentage Error (MAPE) of 0.02 across all targets. This model was trained and validated using simulation data from selected areas of London. It was further evaluated on unseen data from diverse UK cities without retraining, confirming its predictive power across varying climatic conditions. Feature importance analysis revealed that occupant behaviour and infiltration play the most significant role in energy performance, surpassing structural and building envelope characteristics. This paper's contributions lie in the workflow designed for automating selected tasks, the AI-driven optimization framework, and the robust hybrid modeling approach, offering novel tools for energy-efficient building design and retrofitting. The findings are particularly valuable for architects, urban planners, and policymakers seeking scalable, data-driven solutions to reduce energy consumption while maintaining thermal comfort.