Building energy prediction model with AI-based PSO-ANN approach integrating architectural and HVAC processes

Nowadays, due to the increasing energy demand, the importance of research on energy consumption estimation in buildings is increasing. This study aims to estimate and interpret energy consumption in buildings by integrating the processes of architecture and heating-ventilation-air conditioning (HVAC) components with an approach that combines environmental impact. For this purpose, the data were obtained from residential buildings in Nottingham and City of London, and the hybrid artificial neural network algorithm and Shapley Additive Explanations (SHAP) analyses were used as methods. To improve the performance of the Artificial Neural Network (ANN) model, a hybrid method was created with Particle Swarm Optimisation (PSO) (PSO-ANN). In order to ensure the interpretability of the model, both tree-based and neural network-based SHAP method were used. The outputs of this AI-supported method were compared and interpreted. The hybrid model developed in this study achieved an R-squared value of 0.99 in the test data sets in both data groups, which is a significant improvement compared to standard ANN models (0.95–0.97). In addition, when the training data is also included, the overall improvement rate is calculated as 3–5 %. According to the results of the analyses, CO₂ emission rate and total area are the factors that affect the energy consumption the most. In addition, the positive effects of heating control systems and lighting factors on energy efficiency are evaluated, and their importance is emphasized. These results provide an effective prediction model that combines building energy consumption and environmental performance and integrates the processes of architecture and HVAC components. © 2025 Elsevier Ltd