Ensemble Neural Network Method for Wind Speed Forecasting

Abstract

Wind power generation has gradually developed into an important approach of energy supply. Meanwhile, due to the difficulty of electricity storage, wind power is greatly affected by the real-time wind speed in wind fields. Generally, wind speed has the characteristics of nonlinear, irregular, and non-stationary, which make accurate wind speed forecasting a difficult problem. Recent studies have shown that ensemble forecasting approaches combining different sub-models is an efficient way to solve the problem. Therefore, in this article, two single models are ensembled for wind speed forecasting. Meanwhile, four data pre-processing hybrid models are combined with the reliability weights. The proposed ensemble approaches are simulated on the real wind speed data in the Longdong area of Loess Plateau in China from 2007 to 2015, the experimental results indicate that the ensemble approaches outperform individual models and other hybrid models with different pre-processing methods.