Electric Vehicle Charging Profile Forecasting Using Hybrid Models

TL;DR

This paper introduces a hybrid, lightweight forecasting method for individual EV charging profiles, enhancing the accuracy of charging station management and departure time estimation by leveraging variable information levels.

Contribution

The paper presents a novel hybrid approach for EV charging profile forecasting that accounts for varying information availability, improving granularity and accuracy.

Findings

The proposed method performs well on multiple EV datasets.

Variable information levels impact forecasting accuracy.

Hybrid model improves granular scheduling capabilities.

Abstract

Electric Vehicle (EV) fast charging stations require forecasting techniques both at the single charger level and aggregated level. While for the latter several models exist, forecasting individual EV charging profiles is still underexplored in literature. However, such methods may be potentially used by battery-aware scheduling, leading to a more granular update of the charging station aggregated forecast and provide a more accurate estimation of EVs departure times. Nonetheless, the variable extent of available information in time and in different settings could jeopardize these benefits. For this reason, we propose a hybrid and lightweight method to estimate the EV charging profile before and during the charging process. Besides evaluating this method on multiple EVs from a public dataset, we also assess the impact of different level of information in the time transposition of the charging profile.