Enhancing building energy efficiency: An integrated approach to predicting heating and cooling loads using machine learning and optimization algorithms

Abstract

Predicting cooling and heating loads is essential for efficient building energy management in order to maintain indoor comfort. This study employs machine learning methods containing Support Vector Regression (SVR), Extreme Gradient Boosting (XGB), and a Dempster-Shafer Theory-based ensemble model using a dataset of 768 samples. 5-fold cross-validation selects 80 % of the dataset for training, with optimization algorithms like Coot, Dingo, and Sea-Horse Optimizers fine-tuning model applicability. Evaluation metrics such as R2, RMSE, n10-index, MARE, SI, WAPE, and NSE assess model performance. XGB outperforms SVR in heating and cooling load prediction. The XGSH model, combining XGB with the Sea-Horse Optimizer, obtains the highest performance, with an RMSE of 0.9729 and R2 of 99.17 % for heating load (HL) estimation. Similarly, for cooling load (CL) estimation, the XGSH model shows the highest R2 (99.36 %) and lowest RMSE (0.7723) during training. The results suggest practical usability for energy management and indoor comfort maintenance, benefiting engineers and decision-makers alike.