Toward ensure the accuracy of extraction of features from the electrical charge sequence and further improve the model prediction accuracy, A CFVMD-PE-CBAM-BiLSTM prediction method based on the combination of Chebyshev filter (CF), variational mode decomposition (VMD), permutation entropy (PE), Convolution Block attention module (CBAM) combined with bidirectional long short-term memory (BiLSTM) is proposed. Firstly, the variational mode decomposition combined with Chebyshev filter is used for reconstruction and high frequency denoising, and then the entropy of the decomposed subsequence is reconstructed by permutation entropy. Then, the decomposed and reconstructed components are used as the input of the convolution block attention module and combining bi-directional long and short-term memory networks to complete feature extraction and compression of the prediction sequence data and complete the load forecasting process. Results of the experiment indicate that the MAE index, MAPE index and RMSE index are 9.254, 0.433% and 11.726 respectively. Therefore, the accuracy of this prediction method has been significantly improved.