A hybrid artificial-based model for accurate short term electric load prediction

This paper aims to propose a novel hybrid intelligent linear-nonlinear load forecasting model which takes into account both linearity and nonlinearity of load time series as a requirement of precise forecasting. The linear part of the time series is forecasted by the Auto Regressive Integrated Moving Average (ARIMA). The nonlinear ARIMA residuals are then modeled by the Support Vector Regression (SVR) forecaster. Since the ARIMA residuals tend to be nonlinear thus the proposed methodology tries to subdue these nonlinearities by utilizing the discrete wavelet transform in which the ARIMA residuals are decomposed into their high and low frequency components. In order to optimize the value of SVR parameters a new Modified Honey Bee Mating Optimization (MHBMO) algorithm is proposed as well. The proposed MHBMO algorithm prevents the optimization process from trapping in local optimums through a new modification phase. The veracity of the proposed methodology is corroborated by applying it to the empirical load data of Fars Electric Power Company, Iran.