Joint Optimization of Delay-Tolerant Autonomous Electric Vehicles Charge Scheduling and Station Battery Degradation

With the increasing use of electric vehicles (EVs) and the development of emerging transportation network services, autonomous EVs (AEVs) may play an important role in the future of transportation. AEVs can automatically plan their route, park in the charging station, and support the vehicle-to-grid (V2G) services. However, V2G services may influence user dissatisfaction due to the task delays. There is a tradeoff between the optimization of electricity cost and user dissatisfaction. In this article, we formulate the problem to minimize the electricity cost of AEVs and the degradation cost of the charging station batteries with the constraint of V2G services and user dissatisfaction, which is a nonconvex problem and is difficult to solve. To solve the nonconvex optimization problem, we design a suboptimal charging algorithm with some constraints (SCAC) based on the Lyapunov optimization technique to find a tradeoff between the total cost and user dissatisfaction. This algorithm cannot find the optimal solution but can give a selection criterion. Furthermore, in order to get a global charging schedule, we use the criterion from the SCAC algorithm as a priori knowledge to design the charging scheduling reinforcement-learning-based (CSRL) algorithm, which is more efficient than the reinforcement learning (RL) method without any particular criterion. We do simulations by using day-ahead price and practical profiles of AEVs to evaluate the proposed algorithms. The numerical results show that the CSRL algorithm has a better performance 5.12% than the SCAC algorithm and both algorithms are 12.66% and 17.14% better than the benchmark algorithm which is the shortest path (SP)-based algorithm. The CSRL algorithm has more efficiency \varepsilon (1-{\mathrm{ Pr}}(\Lambda (t)=0)) than the SCAC algorithm, where {\mathrm{ Pr}}(\Lambda (t)=0) is a selection criterion calculated from the SCAC algorithm.