A hybrid heat load forecasting model based on multistage decomposition and dynamic adaptive loss function

Accurate and stable heat load forecasting is very important for heating systems. However, due to the high nonlinearity of heat load data, dynamic characteristics and noise interference, the existing forecasting methods still face many challenges. To conclude, we propose an innovative hybrid forecasting model that integrates adaptive quadratic decomposition (ASD), deep time series modeling, and dynamic loss optimization strategies. The proposed model mainly includes four core modules: (1) ASD strategy, The Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) and the improved K-means based on sample entropy (SE) are combined the clustering method and Successive Variational Mode Decomposition (SVMD) achieve efficient noise reduction and feature extraction. (2) The rolling window mechanism dynamically adjusts the training data range to prevent data leakage and enhance the model's adaptability to time changes; (3) Based on the improved GRU combinatorial forecasting model, the exponential gating mechanism and memory hybrid strategy are introduced to improve the GRU, so as to improve the model's learning ability of long-term dependence and short-term dynamic features; (4) Dynamic adaptive loss function (DAL) can effectively suppress the influence of noise and outliers, and improve the robustness and generalization ability of the model. To verify the validity of the proposed model, the actual heat load data of three different time scales were collected for experimental evaluation. The experimental results show that the proposed model is superior to the existing methods in terms of forecasting accuracy, stability and computational efficiency. It provides a novel and effective solution for energy optimization management of an intelligent heating system.