Data-Driven EV Charging Load Forecasting and Smart Charging

Abstract

Electrical Vehicles (EVs) have been proposed as a solution for decarbonizing road transport. Smart charging is essential to coordinate EV energy demand with the requisite peak power supply. The performance of smart charging highly depends on understanding the EVs’ charging behaviour (charging patterns and energy demands), and an accurate forecasting of the EV energy demands are essential for designing a smart charging scheme. This paper presents findings from analysing 3 years’ data of an Oslo Vulkan parking garage pilot, one of the largest hybrid public/commercial/residential parking garages for EV charging in Norway/Europe. A long-short-term-memory (LSTM) regression network is developed to predict hourly EV charging demand with a Weighted Absolute Percentage Error of 30.5%. The analysis suggests that a smart charging strategy is needed to shave the peak demand during 19:00-21:00.