Investigation on the CNN-LSTM-MHA-based model for the heating energy consumption prediction of residential buildings considering active and passive factors

Accurate heating energy consumption prediction is crucial for demand response in heating systems, thus for building energy efficiency achievement. In this paper, a new approach aimed at accurately forecasting heating energy consumption of residential buildings was presented. The prediction model was established by the combination algorithms (CNN-LSTM-MHA) and the critical parameters of the model were identified based on the actual energy consumption of users. To incorporate the subjective adjustment intentions of individuals in heating energy consumption prediction, which is vital to improve the prediction predictability and interpretability, both active (weather conditions, etc.) and passive factors (occupancy) were considered. The obtained results suggest that the CNN-LSTM-MHA model is superior. On average, it reduced training time by 26.67 % and achieved R2 values of more than 0.9. Furthermore, a heating elasticity margin model was constructed based on occupancy characteristics, which can be used for determining users' thermal demand by finding the adjustable margin meeting users’ heating preferences for the target heating energy consumption. Based on the model, the operation status of the heating pipe network can be quickly evaluated. Combining the two models enables heating companies to formulate operational decisions in accordance with users' willingness, thereby enhancing decision-making efficiency and user satisfaction.