An efficient wind speed prediction method based on a deep neural network without future information leakage

Wind speed has strongly stochastic and fluctuating characteristics that make accurate wind speed prediction challenging. Traditional hybrid prediction models use future wind speed data that are unknown during data preprocessing, resulting in future information leakage. To address this problem, this paper proposes a short-term wind speed prediction method without future information leakage to develop a practicable method for predicting effective information contained in wind speed series. The proposed method has three main components: an effective information screening module, a real-time rolling decomposition module (RTRD), and a prediction module (Bi-LSTM-Attention) combining an attention model and a bidirectional long short-term memory neural network (Bi-LSTM). The first module extracts an effective component of the original wind speed data using an effective information screening module and constructs the prediction module output. The RTRD module decomposes the original data and provides the prediction module input. The Bi-LSTM-Attention module obtains the final wind speed prediction results. The effectiveness of this method is verified through tests on measured data. The results show that the proposed method can accurately predict effective information without future information leakage. In addition, the Bi-LSTM-Attention model has greater accuracy than traditional neural network models. Moreover, the proposed model is compared with state-of-the-art decomposition algorithms, including empirical mode decomposition (EMD), improved complete ensemble EMD with adaptive noise, variational mode decomposition (VMD), and singular spectrum analysis. The results demonstrate that the VMD-RTRD model achieves comprehensive optimal prediction accuracy.