Power load forecasting by wavelet least squares support vector machine with improved fruit fly optimization algorithm

Abstract

As is affected by many factors, mid-long term power load forecasting has become the nonlinear and multi-dimension complex problem, and its accuracy affects the decision and layout of power generation sector. In order to improve the accuracy and convergence ability of the single least square support vector machine (LSSVM), this paper proposes the improved fruit fly optimization algorithm applied to wavelet least square support vector machine (IFOA-w-LSSVM). Firstly, the Gaussian kernel function of LSSVM is replaced by the wavelet kernel function and wavelet least square support vector machine (w-LSSVM) is built. Secondly, the ordinary fruit fly optimization algorithm (FOA) is improved from three aspects: (1) dividing fruit fly group into two parts: (2) improving the taste detection function; (3) using Cauchy mutation process to make fruit fly individuals variant. Finally, w-LSSVM is optimized by IFOA for seeking the optimal parameters and achieving the forecasting accuracy. Additionally, the example verification results show that the proposed model outperforms other alternative methods and has a strong effectiveness and feasibility in mid-long term power load forecasting.