Parameter tuning of EV drivers' charging behavioural model using machine learning techniques

The charging behaviour of electric vehicle (EV) drivers significantly influ-ences planning of the future deployment of public charging stations (CSs). Thus, identifying the EV drivers' charging behaviour plays a major role in CS management and development. In this regard, some para-metric behavioural Markov models (BMMs) introduced in the literature have acceptable performance with tuned parameters. Enjoying the ben-efits of these BMMs needs accurate and feasible parameter tuning. To address this challenge, we propose a machine learning-based method to tune the parameters of such a BMM dynamically. A Deep Q-Network (DQN) algorithm is an appropriate solution in which the reward function is designed based on the statistical resemblance between the EV plug-in and charging times derived from CS simulation with their equivalents derived from the CS charging data. The evaluation results based on the real charg-ing data demonstrate the convergence of the proposed algorithm and validate the accuracy of the adapted behavioural parameters. Accurately adapting the model parameters is an essential prerequisite for designing a system that identifies the EV drivers' behaviour. This novel system helps control the CS congestion and predict the CS requirements when the EV population grows in the future.

Automated summary

This article proposes a machine learning-based method to optimize the charging behavior model of electric vehicle drivers through parameter tuning, in order to help control congestion at charging stations and predict future demand.