Short-term wind power prediction based on DWT-PSO-LSTM

Abstract

Abstract With the “carbon peaking and carbon neutrality” target, quite a few people are focusing on the development of renewable energy, especially wind power, which is the largest source of renewable energy generation and is still expanding. In aiming to make the wind power forecasts more accurate by limiting the errors between the predicted and actual values, a model combining Particle Swarm Optimization, Discrete Wavelet Transform, and Long Short-Term Memory (DWT-PSO-LSTM) is developed to predict the wind power for short periods. First, the wind power is decomposed and normalized by DWT. Then, each decomposition result is input into LSTM prediction, and the model with the good result is obtained by finding the number of nodes and learning the efficiency of the best LSTM through PSO. Finally, the full prediction results are output and then all components are summed. Simulations using open wind power data from Yalova province, Turkey, indicate that the model advanced in this article can reduce the prediction errors better than other models.