충전소에서의 대기시간을 고려한 전기차량의 경로 최적화 문제에서 강화학습 기법의 적용 방안 연구

There are several technical obstacles to Electric Vehicles (EVs) adoption such as limited battery capacity, long charging time and low accessibility to EV charging stations. We extend the conventional EVRP (Electric Vehicle Routing Problem) by including charging station visits in the routing decisions. Unlike the literature, the proposed model particularly decides which charging station to visit based on the information of charging station complexity, which is specified by arrival and service rates at each charging stations. We formulated the routing optimization problem as an MDP (Markov Decision Problem) model and used a deep reinforcement learning approach (i.e., DQN; Deep Q-Network) to optimize the route including charging station visits. Numerical analysis shows that the proposed model outperforms two other benchmarks such as nearest station choice and random choice in terms of the total travel time. Moreover, we show that the charging station complexity and the resulting waiting time have significant impact on the performance.