Thermal Load Forecasting of an Ultra-short-term Integrated Energy System Based on VMD-CNN-LSTM

In order to better excavate the effective information contained in the massive data of ultra-short-term comprehensive energy systems and find out the correlation between these information, so as to improve the accuracy of the thermal load prediction of ultra-short-term comprehensive energy systems. On characteristics of thermal load confusion, instability, noncontinuity, a VMD-CNN-LSTM prediction model based on the combination of VMD, CNN, LSTM. The model first uses VMD to decompose and denoise the input data to improve the data continuity and stability, then uses CNN for feature fusion extraction, and finally takes the extracted eigenvector as the input of LSTM. To verify the model's effectiveness, the University of Arizona (Tucson) AZMET database is used to learn and predict, and compare the model, CNN and CNN-LSTM, experimentally proving that the VMD-CNN-LSTM combined neural network has a better predictive effect.