Short-term Load Forecasting Based on CEEMDAN-PE-GWO-LSTM

In order to obtain more accurate power load forecasting data, a combined short-term load forecasting method based on CEEMDAN-PE-GWO-LSTM is proposed in this paper. Specially, CEEMDAN and PE are first used to decompose the load sequence and expose the historical features of the data sequence at various frequencies. Then, GWO algorithm is applied to optimize the hyper parameters of LSTM, so as to improve the accuracy and robustness of the forecasting model. Consequently, the decomposed sequence is used to train the GWO-LSTM and gain the prediction result. In addition, experimental results reveal that the proposed CEEMDAN-PE-GWO-LSTM model achieves superior forecasting performance than the single LSTM model and other commonly used forecasting models.