Comparative study of reformed neural network based short‐term wind power forecasting models

Abstract

Short-term prediction of wind power plays a vital role in wind power application. In order to improve the accuracy of wind power forecasting, this paper investigates neural network combined forecasting models to forecast the wind power, the data of a real wind farm, the Pacific Wind Farm, is used. In view of the difficulty of predicting the large fluctuations of wind power, the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) algorithm is used to decompose the wind power time series which can reduce the complexity of the forecasting process, and then the intrinsic mode function (IMF) signal is predicted by the BP Neural Network, wavelet neural network (WNN) and long short-term memory (LSTM) neural network respectively, and the final result is obtained through wavelet reconstruction. By comparing with a single model, the combined prediction model has better prediction accuracy and stability, among them, the NMAE predicted by CEEMDAN-GA-BP in January was 4.167%, and the NRMSE was 6.590%. Reformed neural network based short-term wind power forecasting models proposed in here provides very useful information for operation and control of high renewable energy penetrated power systems.