Short-time wind power prediction using hybrid kernel extreme learning machine

Abstract

A hybrid short-time wind power forecasting technique based on variational mode decomposition (VMD) and kernel extreme learning machine (KELM) is proposed in this paper. The non-stationary historical wind data is initially decomposed into various modes using VMD technique, which is subsequently passed through the proposed KELM (Gaussian and wavelet-based) and conventional ELM [without weight optimisation and with optimisation – chaotic firefly optimisation algorithm (CFA)], respectively, in order to predict the 30 minutes and 1 hour ahead wind power, respectively. It is observed that, rather than optimising the arbitrary input layer weights of VMD-ELM technique, the proposed Gaussian-based EMDKELM technique illustrates the most effective and accurate short-time wind power predictions for some diverse seasons. The overall results presented in the simulation (through MATLAB simulation platform) and result section are satisfactory and indicates the proposed Gaussian-based EMD-KELM technique as a highly potential prediction technique for real-time applications in power systems. The proposed model can be tested in the industry having wind generation or wind power plant where this can be applied in order to predict the future wind power to schedule the load profile in efficient manner. This can be also validated through wind test bench system.