Evaluating Artificial Intelligence and Statistical Methods for Electric Load Forecasting

Abstract

Electric Load Forecasting (ELF) is one of the challenges being faced by the Power System industry. With the ever-growing consumer demand, power generating companies struggle to manage and provide an uninterrupted power supply to the users. Over the past few decades, the introduction of smart grids and power deregulation has changed load forecasting dynamics. Most of the current research focuses on short-term load forecasting (STLF), involving an hour to a week’s time forecasting. Various techniques are being used for accurately predicting the electric load. However, gold standards are yet to be defined mainly because of the subject's variety, non-linearity, and un-predictive form. In this study critical review of 25 publications has been carried out to find the most efficient method for ELF. The novelty of this study is that comparative and scientific analyses are carried out to find the most proficient techniques for load forecasting. Also, various parameters are combined for comparison in this study after analyzing published reviews on the subject. Artificial Neural Networks (ANN) and Auto-Regressive Moving Average (ARMA) models outperform other methods basing upon statistical analysis, i.e., Mean Absolute Percentage Error (MAPE) and comparative acceptance, in the research community.