DOI: 10.29132/ijpas.1771291 ISSN: 2149-0910

Predicting heating and cooling loads in residential buildings using a GA supported hybrid ensemble regression framework with high accuracy

Faruk Kürker
Show PDF Cite
Developing high-accuracy prediction models for complex, multi-variable energy systems is important for both academic research and industrial applications. In this study, a Genetic Algorithm (GA)-supported hybrid ensemble regression framework is proposed for two target variables (Y1 and Y2). The method combines Support Vector Regression, Decision Tree, Bagging-based Ensemble, Random Forest, and Ridge regression models using blending and stacking strategies. Hyperparameter selection was performed using 5-fold cross-validation, and blending weights were determined using performance-based inverse-proportional weighting. The model was evaluated using RMSE, MAE, MAPE, R², and the Pearson correlation coefficient under a 80–20 hold-out split. The results indicate that the Stacking–LASSO approach, in particular, delivers the best performance (Y1: RMSE = 0.5752, R² = 0.9970; Y2: RMSE = 1.8358, R² = 0.9673). Plot diagrams, residual distributions, Q–Q plots, and Bland–Altman plots indicate that model errors are at an acceptable level. The proposed method offers a reliable and interpretable solution for energy system modeling.

More from our Archive