A hybrid intelligent-based linear-nonlinear model for accurate daily peak load forecasting

Abstract

The objective of this work is to suggest a new hybrid intelligent-based linear-nonlinear model to capture both the linearity and nonlinearity of load time series to reach more accurate forecasting results. In the proposed method, the autoregressive integrated moving average (ARIMA) model is used to forecast the linear part of the time series. After that, the ARIMA residuals as the nonlinear component are modeled by the support vector regression (SVR) forecaster. In order to reduce the nonlinearity of the residuals, the discrete wavelet transform is used to decompose the ARIMA residuals into its high and low frequency components. Moreover, a new adaptive modified optimization tool based on particle swarm optimization (PSO) algorithm is proposed to find the optimal values of the SVR parameters suitably. The proposed adaptive modified PSO algorithm makes use of an adaptive framework to update the inertia weighting factor and the acceleration coefficients during the optimization process. The accuracy of the proposed method is examined by the empirical load data of Fars Electric Power Company, Iran.