Predicting Electricity Consumption in Microgrid-Based Educational Building Using Google Trends, Google Mobility, and COVID-19 Data in the Context of COVID-19 Pandemic

Abstract

Electricity demand has been disrupted in various countries since many governments imposed comprehensive social restriction policies to control the COVID-19 pandemic. Obtaining accurate electricity consumption predictions in this highly uncertain period is particularly important for building operators to improve the corresponding operational planning efficacy. Nevertheless, developing accurate electricity consumption prediction models for buildings within the COVID-19 context is a nontrivial task. Correspondingly, this research focuses on incorporating publicly available internet data (i.e., Google Trends, Google Mobility, and COVID-19 data) to develop accurate electricity consumption prediction models for microgrid-based buildings during the COVID 19 pandemic. For this purpose, we developed extreme gradient boosting (XGBoost), support vector regression (SVR), and autoregressive integrated moving average with explanatory variable (ARIMAX) models. As a case study, we analyzed a real-life electricity consumption dataset of a six-floor microgrid-designed educational building at a technological university in Bandung, West Java, Indonesia. The findings show that incorporating publicly online data positively impacts prediction accuracy. The accuracy increases, even more when we use the lagged value of the predictors. XGBoost models utilizing lagged historical values of the electricity consumption, Google Trends, and COVID-19 data of the previous days is the best performing model. However, adding more lagged predictors does not necessarily increase SVR models’ accuracy. Lastly, the ARIMAX models become the worst-performing models compared to XGBoost and SVR models.

Keywords

• Electricity consumption
• microgrid
• prediction
• machine learning
• COVID-19
• Google mobility