Short-Term Electric Load Forecasting Using Recurrent Neural Network (Study Case of Load Forecasting in Central Java and Special Region of Yogyakarta)

Abstract

Short-term load forecasting (STLF) is a very important factor in the planning and operation of power systems. The purpose of load forecasting is to balance the demand and electricity supply. The electrical load is dynamic, changing over the time. The provision of electrical energy is also dynamic following the pattern of load changes. Load forecasting is required to ensure an accurate decision on power plant scheduling, unit commitment, and power delivery. This paper presents a recurrent neural network (RNN) model with Levenberg-Marquardt and Bayesian regularization training algorithms used for short-term electrical load forecasting. The accuracy criterion used is Mean Absolute Percentage of Error (MAPE). The results show that the RNN model can make good predictions. RNN model with the Bayesian regularization training algorithm has better accuracy. Its average MAPE in one week is 1,4792%. It implies that the RNN model is great tool for STLF.