Hybrid machine learning model for building heating demand forecasting using autoencoder and deep extreme learning

Space heating load is the main energy consumption driver for residential buildings, and its estimation is critical for the efficient and economic operation of energy systems. In this paper, nine different Artificial Intelligence (AI)-based Machine Learning (ML) models — Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LGBM), Elastic Net (ENet), Extreme Learning Machine (ELM), Convolutional Neural Networks (CNN), Decision Tree (DT), Histogram-based Gradient Boosting (HGB), Long Short-Term Memory (LSTM) and a hybrid Autoencoder and Deep Extreme Learning Machine (AE + DELM) — are compared to estimate space heating load using seven independent input variables. The comparison is mainly carried out using Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), coefficient of determination (R 2 ), and Mean Absolute Percentage Error (MAPE). Additionally, Residual Prediction Deviation (RPD) and uncertainty analysis using the prediction interval coverage percentage (p-factor) and the normalized prediction interval width (d-factor) are employed to further confirm the findings. The hybrid AE + DELM method demonstrates superior performance among all modelling approaches and achieves the lowest MAPE value of 0.04. The excellent performance of the hybrid AE + DELM model is further demonstrated by its high RPD, while all models show strong correlations with R 2 scores above 0.99. Although CNN and XGBoost also exhibit strong performance, LSTM and HGB show comparatively higher uncertainty.