Analysis of the impact of epidemic situation on total electricity consumption of the United States under the prediction scenario

Abstract

The outbreak of COVID-19 epidemic has led to a decline in electricity consumption in the United States, the world’s second largest consumer of electricity. Based on the monthly data of the United States from August 2015 to July 2020, this paper uses the ARIMA model and the ARIMA-BP model to forecast the power consumption of the United States in the next 17 months. The Mean Absolute Percentage Error (MAPE) is 5.6% and 4.5% respectively. It shows that the prediction results have high reliability. The results of prediction research will provide a scientific basis for the normal adjustment of electricity supply and demand in the United States, and the method used in this study can be used as a reference for the study of electricity consumption in other countries.

Modeling and Forecasting Electricity Consumption Amid the COVID-19 Pandemic: Machine Learning vs. Nonlinear Econometric Time Series Models

“…They also show that COVID-19 adversely affects the prediction accuracy of the ARIMA model. Ma and Wang (2022) combined an ARIMA model with a backpropagation artificial neural network (BP) ML model to predict electricity consumption in the US, demonstrating that the combined model (ARIMA-BP) outperformed the ARIMA model. The results also showed that the COVID-19 pandemic decreased energy consumption.…”

“…The first class of models, the traditional econometric models has, up to recently, dominated the empirical applications with a more focus in using linear time series models such as ARMA, ARIMA, SARIMA, VAR model, etc... Atalay, Calis, Kus, and Kuru (2019), Cihan (2022), Ensafi, Amin, Zhang, and Shah (2022), Lehna, Scheller, and Herwartz (2022) and Ma and Wang (2022). The use of this class of models is mainly motivated by the simplicity and the ease of estimation of these models.…”

“…Despite the decent reliability of the ARIMA model, Cihan (2022) implied that the effect of COVID-19 lockdowns negatively affected the model's prediction accuracy, as it was higher in electricity consumption datasets where the impact of COVID-19 was ''smoothed out''. Ma and Wang (2022) combined an ARIMA model with a machine-learning model Back Propagation Artificial Neural Network (BP) to predict electricity consumption in the US. They showed that the combined model (ARIMA-BP) outperform the ARIMA model.…”