Model improvement and scheduling optimization for multi-vehicle charging planning in IoV

• Published in: Physica A 2023-07, Vol.621, p.128826, Article 128826
• Main Authors: Qian, Jun-Hao, Zhao, Yi-Xin, Huang, Wei
• Format: Article
• Language: English
• Subjects: Charging planning; Deep reinforcement learning; Internet of vehicles; Multi-vehicle
• ISSN: 0378-4371; 1873-2119
• DOI: 10.1016/j.physa.2023.128826

Intelligent electric vehicles (EV) can transmit their location, driving status and other information to Intelligent Transportation Systems (ITS) through the Internet of vehicles (IoV) communication. Among them, the optimization of EV charging planning has great significance in finding a suitable charging station (CS) for users. However, the constraints of State-of-Charge (SOC) and driving direction are incomplete in current planning models. Meanwhile, the existing scheduling policy based on Deep Reinforcement Learning (DRL) suffers from slow convergence due to the fixed average change rate of the reward function. This paper establishes a comprehensive EV charging planning model (CCPM) and presents an efficient multi-vehicle scheduling algorithm (EMVSA). Firstly, CCPM calculates the travel time under the SOC constraints to ensure that CS is reachable and takes into account the direction constraint by minimizing the distance of the selected CS to the user’s destination. Secondly, a novel reward shaping method, which gradually increases the average change rate of the reward function, is presented and proved theoretically to accelerate the convergence of EMVSA. On the real city road network data, experimental results show that CCPM can guarantee the reasonability of CS selection and direction, and that the convergence speed of EMVSA is significantly increased to get the optimal scheduling result. •An EV charging planning model considering state of charge and driving direction constraints is introduced.•A novel reward shaping method is presented to accelerate the convergence of the scheduling algorithm.•On the real city road network data, the model and the algorithm are verified by experiments.