Electrical Load Prediction With A Hybrid Deep Learning Method Based on Quadratic Decomposition Aggregation Under High Fluctuation Situation

This paper has proposed a hybrid prediction model based on deep learning technologies. First, complementary ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is used to decompose the original load sequence into several intrinsic mode components. Secondly, variational mode decomposition (VMD) is used to re-decompose the strong non-stationary sequence in the sample entropy in the sequences, so as to further reduce the non-stationary of the original load data based on the concept of a CNN-BiLSTM, prediction model is constructed to reduce the influence of input on the prediction accuracy of output. Finally, by using the actual measured data from a southern China region, comparative tests have been execute. The results show that the prediction method has higher prediction accuracy by comparing other fellow prediction methods.