Minute Ahead Wind Speed Forecasting Using a Gaussian Process and Fuzzy Assimilation

Abstract

This paper presents an intelligent data driven method for forecasting minute ahead wind speed, which is essential in predicting the power output coming from wind generators. The proposed methodology, is based on the principle that “the most recent past should be used to predict the near future”, and implements a two-stage forecasting method. In the first stage a Gaussian Process Regression model is trained multiple times on different length time window, and forecasts a set of next minute wind speed values. In the second stage, a fuzzy inference system collects the forecasts, rejects some of them and then provides a mean and a variance of a single forecast value. The proposed method is applied to a dataset of real-world data, and benchmarked against the autoregression (AR) model. Results exhibit the superiority of the proposed method over AR as well as over GPR which uses a single train set.