Short-Term Load Forecasting for Jordan Power System Based on NARX-ELMAN Neural Network and ARMA Model

Abstract

Over the past few years, there is a vast expansion of the Jordan National Energy Sector. Hence, National Electrical Power Company (NEPCO) sheds more light on load forecasting. It tries to build a rigid bridge between the academic and industrial fields. Subsequently, this work presents a study of short-term load forecasting (STLF) for the Jordanian power system. Three techniques are used: the nonlinear autoregressive exogenous model (NARX) recurrent neural network, the Elman neural network, and the autoregressive moving average (ARMA). These proposed techniques are trained, validated, and tested using the historical record of hourly load data for the whole year 2018, which is obtained from NEPCO. Besides, these techniques show a satisfactory forecasting accuracy and improve the predicted load shape performance of a week ahead (January 1, 2019, to January 7, 2019). Error is reduced based on optimizing the number of hidden layers and the number of neurons. Moreover, the mean absolute percentage errors (MAPEs) are estimated at 5.53%, 3.42%, and 10.28% for NARX, Elman, and ARMA, respectively. Finally, this work is implemented using neural network toolbox and MATLAB code in Mathworks.