Short-term heat load prediction of heat exchange stations based on CEEMDAN-SE-LSTM

Abstract

To address the issue of low short-term heat load prediction accuracy due to factors such as the centralized heating system's lagging nature. A prediction approach CEEMDAN-SE-LSTM is proposed that improves the integration of fully integrated empirical modal decomposition with adaptive noise and long and short-term memory neural networks. CEEMDAN decomposes the preprocessed heat load data into smooth modal components, SE reorganizes the modalities with similar entropy values, LSTM predicts each reorganized modality, and the predicted values of all modalities are superimposed as the final prediction output. Finally, the model's accuracy in predicting short-term heat loads is demonstrated using data from a university heat exchange station in Dalian as an example.