A short-term wind power prediction method based on deep learning and multistage ensemble algorithm

Abstract

Wind power prediction (WPP) is extremely important in promoting the power grid's consumption of wind power. To improve the accuracy of WPP, a three-stage multiensemble short-term WPP method based on ensemble learning and deep learning is proposed in this paper. In the first stage, variational mode decomposition and wavelet transform were applied to decompose the original data sequence into different frequency bands. In the second stage, based on the decomposition sequences, the stacked denoising autoencoder (SDAE), long short-term memory (LSTM), and bidirectional long short-term memory (BLSTM) were used to predict the sequences; and 42 submodels were obtained. In the third stage, a support vector machine (SVM) was applied to give weight to each submodel to obtain the final ensemble prediction results. Based on three-stage integration, a new multi-integration model is proposed that repeats the third-stage integration operation. A case study is presented to verify the effectiveness and superiority of the proposed three-stage multi-integration WPP method. The normalized root mean square error (NRMSE) decreased by 0.0343 compared with LSTM, decreased by 0.0336 compared with BLSTM, and decreased by 0.0323 compared with SDAE, which demonstrated the effectiveness of the proposed new multistage ensemble and deep learning WPP method.