Short-term wind speed forecasting based on an EEMD-CAPSO-RVM model

Abstract

Accurate wind speed forecasting can reduce the adverse effects of wind power on the power grid, ensure the safe and stable operation of the power system, and enhance the competitiveness of wind power market. A wind speed forecasting model is proposed on the basis of Ensemble Empirical Mode Decomposition (EEMD) theory and Relevance Vector Machine (RVM) optimized by Cloud Adaptive Particle Swarm Optimization (CAPSO) algorithm. First, the pretreated wind speed time series is adaptively decomposed into a series of relatively stable component through adopting EEMD technology. Then, the RVM models are developed according to the characteristics of each component, and the combination of each component forecasting results produce the final forecasting values. The kernel function has a great influence on the RVM model forecasting performance. In order to ensure the optimality of the RVM model, a typical Gaussian kernel function is selected and optimized the Gaussian kernel width by the CAPSO algorithm due to its stability and fast convergence. Study results show that EEMD resolves the mode mixing problem and the RVM model has superior forecasting results, further reducing the wind speed prediction error.