Prediction error adjusted Gaussian Process for short-term wind power forecasting

As one of the most affordable and widely available renewable energy resources, wind power has been recognized as the main promising form of renewable energy in many countries for achieving the targets of cutting greenhouse gas emission according to the Kyoto agreement. However, due to the intermittent nature of wind power, it is imperative to forecast the wind generation hours even days ahead to enhance the flexibility of the operation and control of real-time power systems including economic load dispatch. In this paper, a variant of Gaussian Process is proposed and applied to make short-term prediction of the overall wind power production in the island of Ireland. Only measurement data of power generation is utilized. A number of modifications are made to the standard Gaussian Process during the model training and testing procedures to reduce the computational complexity of modeling and forecasting wind power generation. The prediction results are compared to those of standard Gaussian Process and persistence model to confirm the effectiveness of the proposed method. It is shown that not only the computation complexity is greatly reduced, but also poor local optima could also be avoided due to the reduction of matrix dimension.