Abstract: For the optimal scheduling problem of the aggregation in large-scale battery charging and swapping stations, a real-time optimal scheduling strategy is proposed for battery charging and swapping loads based on soft actor-critic（SAC） deep reinforcement learning. The strategy fully considers the user, system and market factors in the load control process, and can realize the friendly interaction between large-scale electric vehicles and various types of power system subjects. Firstly, considering the development scale and scheduling performance of battery charging and swapping stations, a joint operation framework is established. Secondly,an adjustable identification model considering multi-user characteristics is proposed to judge the actual adjustable performance of electric vehicles. Furthermore, considering the multiple spatio-temporal characteristics of the optimal scheduling of battery charging and swapping stations, the optimal scheduling models of adjustable battery charging and swapping loads in different scenarios are constructed. Then, the real-time scheduling scheme of battery charging and swapping loads is solved based on SAC algorithm. Finally, the case of virtual power plant load for electric vehicle load aggregation optimization verifies the economy and high efficiency of the SAC algorithm applied in real-time optimal scheduling of electric vehicle loads for large-scale battery charging and swapping.