High and low frequency wind power prediction based on Transformer and BiGRU-Attention

Abstract

An accurate and reliable wind power prediction model has important significance for the operation of power systems and large-scale grid connection. This paper proposes a hybrid deep learning model, CEEMDAN-SE-TR-BiGRU-Attention, for high and low frequency wind power prediction by combining complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), sample entropy (SE), Transformer (TR) and bidirectional gated recurrent unit with attention mechanism (BiGRU-Attention). Firstly, the CEEMDAN decomposes the original wind power sequence into multiple sub-modes and a residual, and the sample entropy of each sub-sequence is calculated by restructuring the sequence, which can effectively alleviate the impact of the original non-stationary series on the accuracy and computational complexity. Next, the reconstructed sequences are further divided into high and low frequency sequences according to the sample entropy value of the original sequence. The Transformer and BiGRU-Attention models are respectively applied to the prediction of high frequency and low frequency sequences according to the characteristics of each sequence. Finally, the predicted values of all components are superimposed to obtain the final prediction results. Experiments are carried out on four datasets with different seasons, and different models are compared to illustrate the effectiveness and superiority of the proposed model. The experimental results show that the proposed model achieves better prediction accuracy.