Using Smart Meter Data and Machine Learning to Identify Residential Light-duty Electric Vehicles

The growing adoption of electric vehicles (EVs) poses new challenges to power grids. To upgrade the grids with the increasing demand from charging EVs and from the change in customers consumption behaviors, utilities need to know where EV customers are. However, ownerships of EVs are not always known to utilities. This paper presents a methodology on how to use advanced metering infrastructure (AMI) data and apply machine learning to identify residential customers with EVs. It focuses on such aspects as how to perform sampling to reduce effects of external factors associated with other high-usage home appliances, how to create and evaluate variables for enhancing modeling, and how to apply the ensemble method to arrive at the estimation or forecasting needed for grid enhancement.

I. Introduction

The age of electric vehicles (EVs) has arrived. With the increasing need for reduction in emission for slowing down global warming, EVs have attracted more and more attention from various levels of governments across the globe as well as interest from consumers. According to California New Car Dealers Association, California battery electric vehicle (BEV) and plug-in hybrid electric vehicle (PHEV) new registrations increased from 3.6% of all vehicles to 8% in 2020. By 2030, it is estimated that EV sales share will reach 57% in California. The same trend is observed globally. The International Energy Agency (IEA) estimates that the number of EVs will reach about 130 million by 2021, a sharp increase from 5.1 million in 2018.