Ultra-short-term combined prediction approach based on kernel function switch mechanism

Abstract

A novel combined prediction approach including nonlinear data analysis, data decomposition and reconstruction, and parameter optimization is proposed to improve the accuracy of ultra-short-term wind power prediction. The distance correlation coefficient (DCC) is applied to analyze the non-linear correlation between wind power and numerical weather prediction (NWP) data. The original wind power is decomposed by the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) technique to eliminate the noise of the data, and then the decomposed wind power is reconstructed into new subsequences by weight permutation entropy (WPE) according to time series complexity. The least squares support vector machine (LSSVM) prediction model based on the kernel function switch mechanism is developed by the use of the new subsequence and key factor as input variables. Furthermore, an improved cuckoo search (ICS) strategy is established to optimize the penalty factor and kernel function parameters of the prediction model. The validation of the proposed model is tested with data from four actual wind farms at a large wind farm base in Northeast China. The results show that the proposed approach is effective with higher accuracy compared with other prediction approaches.