Electricity Load Prediction Using Machine Learning

Abstract

The prediction of household or building electricity load energy consumption is a problem that is being tackled and many solutions such as statistical, machine learning and physical modes have been developed to better improve prediction accuracy. Electricity consumption predictions play an important role for both consumers (end users) and producers of electricity. The Electric utility industry uses electric load predictions to assist in electric power supply and load balancing, hence ensuring that the electricity provided reaches its customers, while meeting the standards of the quality set. For consumers, understanding and prediction of their electricity consumption offer consumers the capability to plan and manage their electricity expenses, even more so for off-grid systems where power cuts due to in availability of or insufficient energy supply lead to disruptions of planned activities and supplement with other energy sources is not possible or costly to do so. Using the dataset of an individual household electricity consumption, this article analyses the dataset and explores the application of statistical and machine learning methods to develop regression models for electric load energy consumption prediction. The study further compares the accuracy of the different regression model designs. The developed load prediction models are designed to perform electric load consumption prediction on 24h ahead. Results of the study show low prediction accuracy in all of the models. This is evident in the high MAPE scores close to 100%, which indicate that the prediction error of the model is large, furthermore, low <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^{\wedge}2$</tex> scores close to 0 or below, of which indicate that the prediction models provide little explanation of the variation in the target variable, hence resulting in low accuracy.