Machine Learning Model of Electric Vehicle Charging for National-Level Solar Photovoltaic Planning

Abstract

Solar photovoltaic (PV) systems have the potential to manage peak demands linked to electric vehicle (EV) charging. Given the rise in EV deployment, sizing models are needed to predict EV charging demands; yet none use national-level datasets for straightforward predictions. This study utilizes the supervised learning technique of linear regression for predicting daily charger energy, charging time, and the relationship between charger loss and energy for range of commercial EV models at a national level. Year-round behavior of target variables in Canada had a mean absolute percentage error below 4.2%. Additionally, charging requirements of the three most common EVs were modeled individually. The developed model is used to assess the total PV needed to charge EV fleet in each of the Canadian provinces and territories. The results indicate that nationally 4.9 GW of PV is required by 2030 and 27 GW by 2050 to provide for the predicted EV needs.

Keywords: machine learning, solar photovoltaic, electric vehicle, energy planning, electrification