Hybrid relevance vector machine model for wind power forecasting

Wind power forecasting is one of the effective ways to mitigate the challenges for power system and wind farm owners. It is significant to make reliable prediction for the safety and stability of power system, the quality of electric power, and the increase of wind farm capacity coefficient. With good prediction and generalization abilities, relevance vector machine (RVM) is able to provide accurate single point prediction as well as additional probabilistic fluctuation information. However, traditional RVM based forecasting model is implemented with one kernel function. This impairs the forecasting adaptability for variable conditions and incurs mathematical solving problems. To solve this problem, this paper presents a hybrid RVM wind power probabilistic forecasting (WPPF) model with five kernel functions, which are Gaussian kernel, Laplacian kernel, Cauchy in distance kernel, R(distance) kernel, and Tps (Thin-plate spline) kernel. Besides, support vector machine forecasting with different kernels is also proposed to investigate the single kernel performance in machine learning forecasting. To take a wind farm in Eastern China as example, the adaptability and forecasting performance of each single kernel model are demonstrated. And, the proposed hybrid RVM model is proved to be capable of expanding the range of kernel parameter and have better adaptability and accuracy than those of a single kernel model.