The Study And Application Of The Relevance Vector Machine Based Hybrid Models For Wind Speed Forecasting

With the rapid decrease in the reserves of non-renewable energy worldwide, countries and regions are actively looking for and exploit sustainable energy. Wind energy, as an eco-friendly, reproducible resource, has garnered many countries’attention. Wind speed is one of the key parameters in the prediction of wind power. Due to the complex characteristics of wind speed, the accurate prediction of wind speed helps with the reliability, stability of the power system when wind energy enters into power grid. To forecast the wind speed more effectively, this paper proposes two new hybrid models, the two proposed models are:(1)The short term wind speed forecasting model which combines wavelet decomposition and relevance vector machine. In this model, wavelet analysis is used to decompose and de-noise the original wind speed series, and the relevance vector machine is used for the final prediction of wind speed. In the use of relevance vector machine, cuckoo search algorithm is used to optimize the kernel bandwidth.(2) The wind speed forecasting model which incorporates the extended Kalman filter with the relevance vector machine. The relevance vector machine is used to establish the equation of state for extended Kalman filter and the extended Kalman filter is employed to forecast based on the established state equation of wind speed. The cuckoo algorithm searches the optimal bandwidth of kernel function of relevance vector machine.In order to verify their effectiveness, the two models are used to forecast wind speed in the wind fields of Chinese territory. The forecasting results, compared with other models, the proposed hybrid models perform better in terms of various statistical indexes.