A Comparative Analysis of Supervised Machine Learning Algorithms for Electricity Demand Forecasting

Abstract

Accurate electricity demand forecasting is crucial for efficient, economical and stable operation of power system grid. With the increasing integration of intermittent sources, need of reliable electricity demand forecasting has become more evident. In this paper, supervised machine learning (ML) algorithms are explored and analysed for medium-term prediction of electricity demand. The demand forecasting has been implemented for the New South Wales (NSW) area of Australian electricity market for data of one year ranging from January to December 2020. The machine learning algorithms compared in this study are - linear regression (LR), multivariate polynomial regression (MPR), support vector regression (SVR), elastic net regression (ENR), gradient boosting regression (GBR), decision tree regression (DTR), random forest regression (RFR), and K- nearest neighbour regression (KNNR). The evaluation criteria used for the performance judgement are coefficient of determination (R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> ), maximum error (ME), mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE) and median absolute error (MedAE). The machine learning algorithms are compared on the basis of evaluation criteria, and nature of prediction algorithm. The non-parametric machine learning algorithms have performed better than parametric methods.