A new probabilistic prediction approach based on local v-support vector regression

In this paper, a general prediction methodology is proposed which can provide a good service to the related investigations in probabilistic prediction. In particular, the proposed model has the ability to deal with both the deterministic prediction and probabilistic prediction of noisy time series. By means of the proposed approach, local nu-support vector regression (L-nu-SVR) model is exploited to suppress noise disturbance in deterministic prediction (points prediction), and the error intervals, which avoid the distributional assumptions of error, can be gained by using nonparametric kernel estimation (NPKE). Then forecasting confidence intervals (FCIs) are obtained by combining the deterministic prediction results and error intervals. Furthermore, joint forecasting confidence intervals (JFCIs) are proposed to improve the prediction reliability. Finally, a comparison of the proposed model and normal distribution-assumed model is performed through simulations by applying them to a real power system, and the validity and practicability of the proposed model is illustrated.