Low Carbon Operation Method for Electric Vehicle Charging Stations Using Improved Deep Reinforcement Learning

In response to the current problem for large scale electric vehicles with disorderly charging behavior was not fully leveraging their low-carbon benefits, the paper provide a carbon reduction strategy to electric vehicle charging stations based deep reinforcement learning (DRL). Firstly, starting from the multiple interactive entities of vehicle, station and power grid, a carbon emission reduction model for charging stations is established with the goals of minimizing carbon emissions from charging stations, minimizing variance in distribution network load, and maximizing user satisfaction. Secondly, to response to the real-time and randomness issues brought about by large-scale electric vehicles, the proposed model was transformed into a finite Markov decision process and proposed an Improved Depth Deterministic Strategy Gradient Function for the sake of training and solving. Finally, the paper designed a multi scenario simulation with distribution network system in the IEEE 33 node, and the simulation data showed that the method can guide the charging behavior of the electric vehicle group effectively, the system's carbon emissions level can be reduced, and provide a new approach for the carbon emission reduction path of urban charging stations.