Forecasting Thermal Energy Demand: A Multimodal Approach Using Random Forest and SARIMA Models

Correct prediction of heating load is critical for maximizing energy consumption and enhancing thermal efficiency in domestic buildings. In this research, the performance of Random Forest and Linear Regression models in predicting heating energy consumption from different building features, such as wall material, number of floors, type of heating system, and external temperature, is assessed. The data set contains daily readings from residential buildings' heating meters, recording important parameters like input/output water mass, temperature differences, and energy utilized. The outcome shows that Random Forest performs dramatically better than Linear Regression, recording an R² value of 0.9878 against 0.8471 for Linear Regression. Furthermore, Random Forest has reduced error rates (MAE = 0.1491, RMSE = 0.2808) compared to Linear Regression (MAE = 0.6116, RMSE = 0.9943), reflective of its increased capability to account for intricate, non-linear relationships in the consumption of heating energy. The results affirm the effect tiveness of using machine learning models for enhancing the energy efficiency and thermal comfort control in housing.