EV charging station deployment on coupled transportation and power distribution networks via reinforcement learning. (15th March 2023)

• Record Type: Journal Article
• Authors: Zhao, Zhonghao; Lee, Carman K.M.; Huo, Jiage
• Is Part Of: Energy. Volume 267(2023)
• Publication Date: 2023-03-15
• Subjects: Electric vehicle -- Charging station deployment -- Coupled network -- Reinforcement learning; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2022.126555
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: This study addresses the optimal electric vehicle (EV) charging station deployment problem (CSDP) on coupled transportation and power distribution networks, which is one of the critical issues with the mass adoption of EVs in the recent years. In contrast to existing works that mainly employ heuristics and exact algorithms, we propose a finite-discrete Markov decision process (MDP) formulation defined in a reinforcement learning (RL) framework to mitigate the curse of dimensionality problem. The RL-based approach aims to determine the location of a set of EV charging stations with limited capacity by minimizing the total investment cost while satisfying the coupled network constraints. Specifically, a long short-term memory (LSTM)-based recurrent neural network (RNN) with an attention mechanism is used to train the model based on an offline strategy. The model parameters are learned by the policy gradient algorithm with a learned baseline function. Numerical experiments on multiple problem sizes are conducted to assess the efficiency and feasibility of the proposed solution method. We experimentally show that our approach is efficient to solve the CSDP and outperforms other baseline approaches in solution quality with competitive computational time. Highlights: We propose an RL framework to address the EV charging station deployment problem. An RNN-based model is developed to find the optimal policy. A policy gradient method is used to train the agent based on an offline strategy. Experiments are conducted to test the performance of the proposed method.