Machine learning prediction of heating and cooling loads based on Athenian residential buildings’ simulation dataset

Abstract

Energy is a critical utility and infrastructure of today that powers development in society. With increasing demand, particularly for heating and cooling, there are increasingly more worries over energy shortages and pollution. Accurate prediction of heating and cooling energy usage is the most important method to achieve the highest level of energy efficiency in building design. Machine learning models are promising tools for predicting. The main objective of this study is to use machine learning models to effectively predict the Heating and Cooling Load of buildings. The novelty of this study lies in using a broad dataset with diverse building structures and strong influence factors along with a novel prediction method. This includes the use of both stochastic-based (Stochastic Forest and Stochastic Gradient Boosting) and non-probabilistic-based models (Random Forest and Extreme Gradient Boosting), as well as the application of advanced optimization algorithms and ensemble prediction techniques. In addition, the study develops interpretable machine learning models using SHAP and FAST analyses to enable greater interpretability. From the obtained results, the SGEH, which combines prediction results by two Stochastic Gradient Boosting-based hybrid models, achieved an excellent performance in the prediction of Heating and Cooling Load with R2 values greater than 98% and 99%, respectively. The current book contributes to current research by utilising new and explainable prediction techniques and more precise predictions for energy consumption and, thus, filling the gap emerging in the energy-efficient building management.