Robust short-term prediction of wind power generation under uncertainty via statistical interpretation of multiple forecasting models

Abstract

The predictions of the autoregressive moving average model, the artificial neural network model, and the grey prediction model are comparatively studied for the wind power generation. Each predictive model is the most suitable for a certain variance of wind at a given period. In this study, the weighting method is proposed to systematically combine the predicted values of those three predictive models over time, based on their forecasting performance by the root mean square errors (RMSEs) between the actual values and the predicted values. The multiple forecasting models are applied to predict the wind power generation of a wind farm with 1 h, 3 h and 6 h ahead. The RMSEs of the multiple forecasting models are significantly the lowest values among those three predictive models and the benchmark of the persistence model. Also, the prediction interval around the predicted value is statistically determined to indicate the feasible range of the wind power generation with a prescribed percentage of confidence under uncertainty causing the historic prediction errors.