Short-Term Heat Load Forecasting Based on CEEMD and a Hybrid IDBO-TCN-BiGRU Network

Given the complex nature of central heating systems, which exhibit nonlinearity, significant time lags, and strong coupling, this study proposes an enhanced short-term heat load prediction model to improve accuracy. The model integrates a Temporal Convolutional Network (TCN) with a Bidirectional Gated Recurrent Unit (BiGRU), optimized via an Improved Dung Beetle Optimization (IDBO) algorithm. Initially, the unsteady heat load sequence is decomposed into stable modal components using Complementary Ensemble Empirical Mode Decomposition (CEEMD), with relevant features selected as inputs. Subsequently, two improvement strategies are incorporated into the dung beetle optimization algorithm through a “dynamic and mutation” approach. Finally, the optimal features are used in conjunction with the IDBO-TCN-BiGRU model for prediction. The performance of the proposed model is compared with various single and combined models. Experimental results show that the CEEMD-IDBO-TCN-BiGRU method achieves superior prediction accuracy, with Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R-squared (R²) values of 0.07292, 0.19048, and 99.1%, respectively, outperforming alternative models. These findings validate the model’s effectiveness and offer valuable insights for optimizing and regulating centralized heating systems.