Comparative study of hybrid machine learning models to predict the energy consumption of buildings enhanced with PCM: A case study in Morocco

Abstract

Predicting the thermal performance of buildings is a key research area in the context of improving energy efficiency and reducing environmental impacts. Several approaches have been developed to model and predict thermal performance. Among these approaches, machine learning techniques are distinguished by their ability to exploit large amounts of data and model complex systems, but their effectiveness remains to be demonstrated in different contexts. This work therefore explores the application of hybrid machine learning models. Six different models, including ANN-LR, ANN-RR, ANN-RF, ANN-GB, ANN-DT, and ANN-ELM were evaluated and compared to the standalone model (ANN) based on the statistical metrics. Using the TRNSYS tool, the dynamic simulation of a building enhanced with PCM into the roof was performed to generate the data. The findings proved the effectiveness of the hybrid machine learning techniques, with ANN-LR and ANN-GB emerging as the most reliable hybrid approaches for accurate prediction, showcasing their robustness and suitability for complex prediction tasks, while ANN-RR model proved to be the least effective. Furthermore, the performance of the models varied considerably depending on the target, with total energy consumption appearing more complex and challenging for prediction.