Ultra-short-term wind power prediction model based on long and short term memory network

To improve the accuracy of ultra-short-term wind power prediction model, a deep learning-based multivariable long and short-term memory neural network (MLSTM) algorithm considering wind power data and wind speed data is proposed. First, the practical data is reconstructed based on the chaotic analysis results. In addition, the model inputs are selected based on the feature importance computed by a classification and regression tree. Next, a wind power prediction model is constructed usiug a long and short-term memory neural netuork(LSTM). Finally, an error correction strategy is proposed to improve the model accuracy. The wind power prediction other experiments on different time scales are carried out using the actual operating data of wind turbines. The results show that compared with models such as back-propagation neural networks, multilayer perceptrons, and least-squares support vector machines, the proposed model performs well at different scales. Higher prediction accuracy and more stable performnace in different data sets are achieved. © 2021, Solar Energy Periodical Office Co., Ltd. All right reserved.