Abstract: The development of shared electric vehicles does not only provide users with a convenient way to travel but also provides efficient and flexible adjustment for urban power grid resources, large-scale shared electric vehicle（EV） travel and charge-discharge behavior, strong randomness, complex charging and discharging control model online rolling together to consider multiple stakeholders, large amount of calculation, and high real-time requirements. First, this study uses the aggregation modeling method based on the SOC interval to determine the energy transition state in time or space of shared EV; it also reduces the dimension of the optimal charging and discharging schedule of large-scale EVs. Thereafter, we consider the operator profit and utility of the limited rational users’ cumulative prospect as well as the power demand response on multiple subject gains. Moreover, the optimization model of charge and discharge depth is constructed based on reinforcement learning. Next, the deep Q net is applied to solve the network method, and real-time online sharing of EV charging and discharging aggregation optimization strategy in the different regions is achieved. In addition, the model effectively copes with the effects of randomness. Finally, combined with the actual operation data of 5000 shared EVs in 9 regions of a city, numerical example analyses verify that shared EVs within the city have the characteristics of time-space transfer of energy. The modeling method and solving strategy proposed in this study are effective in solving the optimization scheduling problem of large-scale shared EV charging and discharging while aiming to ensure the interests of multiple subjects.