Short-term wind power prediction based on EMD-LSTM

Abstract

Accurate wind power prediction plays a significant role in power system operation, safety analysis and consumption reduction. To make the most of the valid Consultation in the data and further improve the prediction accuracy of wind power, forecasting model based on empirical mode decomposition (EMD) and long short-term memory (LSTM) neural networks was submitted. Firstly, EMD was used to decompose the wind power data series into multiple sub-series to reduce the influences among diverse trend information. Then, LSTM was used to learn the historical time series of each sequence and complete the prediction. Finally, all prediction results are overlapping and reconstructed to acquire the final prediction of wind power. The proposed method was applied in the wind power prediction of the farm in the United States. The findings indicate that the proposed combined prediction method improves its prediction accuracy by 38.20% and smoothness by 39.19% compared with the current mainstream algorithm prediction, which has higher prediction accuracy and performance smoothness and reduces the interaction between information of different time scales.