Seasonally Adaptive VMD-SSA-LSTM: A Hybrid Deep Learning Framework for High-Accuracy District Heating Load Forecasting

To improve the accuracy of heating load forecasting and effectively address the energy waste caused by supply–demand imbalances and uneven thermal distribution, this study innovatively proposes a hybrid prediction model incorporating seasonal adjustment strategies. The model establishes a dynamically adaptive forecasting framework through synergistic integration of the Sparrow Search Algorithm (SSA), Variational Mode Decomposition (VMD), and Long Short-Term Memory (LSTM) network. Specifically, VMD is first employed to decompose the historical heating load data from Arizona State University’s Tempe campus into multiple stationary modal components, aiming to reduce data complexity and suppress noise interference. Subsequently, the SSA is utilized to optimize the hyperparameters of the LSTM network, with targeted adjustments made according to the seasonal characteristics of the heating load, enabling the identification of optimal configurations for each season. Comprehensive experimental evaluations demonstrate that the proposed model achieves the lowest values across three key performance metrics—Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE)—under various seasonal conditions. Notably, the MAPE values are reduced to 1.3824%, 0.9549%, 6.4018%, and 1.3272%, with average error reductions of 9.4873%, 3.8451%, 6.6545%, and 6.5712% compared to alternative models. These results strongly confirm the superior predictive accuracy and fitting capability of the proposed model, highlighting its potential to support energy allocation optimization in district heating systems.