Development of a Machine Learning-Based Prediction Model for Phase Change Material Integrated Buildings with Natural Ventilation

Abstract

The building sector contributes to more than one-third of global energy-related emissions, witnessing a surge in cooling demand. The integration of phase change materials (PCMs) in buildings with natural ventilation (NV) shows a promising passive design strategy for reducing cooling energy consumption (CEC). However, evaluating this coupling performance remains limitedly explored with no surrogate model available to support early design stages. This research addresses this challenge by developing a machine learning (ML) framework that incorporates large-scale parametric simulations and feature generation and utilizing a multialgorithm prediction modelling to estimate CEC in PCM-integrated buildings with NV. Using nine residential building models, eight BSh cities, and seven commercial PCMs, this framework combines extensive building energy simulations capturing 15 key variables including NV operation, PCM and building geometric characterizations, and climatic conditions. Over 2.1 million monthly datapoints generated across fifteen key variables through Latin Hypercube Sampling. 8 ML algorithms, including neural network, ensemble, and gradient-boosting methods, then interpreted using partial dependence. The optimal model (XGBoost) demonstrated high prediction accuracy (R² > 0.99). Additionally, the research evaluated the performance of PCM+NV on both cooling energy reduction and thermal comfort improvement. The findings revealed that optimal PCM melting temperature lies between 28-31°C, saving up to 34% in CEC and 48% improvement in thermal comfort. In sum, the findings endorse the effectiveness of PCM-integrated buildings with NV strategies, and the robust ML model adequately forecasting CEC.