Average Reward Reinforcement Learning for Optimal On-route Charging of Electric Buses

Nowadays, due to the severe environmental concerns caused by the emissions from the conventional transportations which use traditional fossil sources, people want to find an alternative method imperatively. Under this condition, electric buses attract transit service providers' attention. However, since the degradation of the battery and charging load for the local grid, the charging schedule is a critical issue to be addressed. In this paper, in order to obtain the optimal on-route charging schedule, a specific physical model and battery degradation model is built for the calculation of the energy consumption and the cost. Semi-Markov decision process (SMDP) is utilized to simulate the running process of the EBs, and the average reward reinforcement learning (ARRL) is introduced to optimize the on-route charging schedule for the EBs. The charging policy and the performance is compared with the default charging schedule according to the real EB operation data provided by the St. Albert Transit, AB, Canada.