Deep Reinforcement Learning for Solar-Powered Electric Vehicle Charging Stations

The rapid growth of electric vehicle (EV) adoption has significantly increased electricity demand, creating operational challenges for power distribution networks and accelerating the need for sustainable charging infrastructure. Solar photovoltaic (PV)-powered EV charging stations provide a promising solution by reducing grid dependence, lowering carbon emissions, and increasing renewable energy utilization. However, the intermittent nature of solar generation and the stochastic behavior of EV charging demand require intelligent energy management to ensure efficient and reliable operation. This paper proposes a Deep Reinforcement Learning (DRL)-based energy management framework for optimizing grid-connected solar PV EV charging stations integrated with the IEEE 33-bus distribution system. The proposed DRL agent continuously determines optimal power dispatch among the solar PV array, battery energy storage system, utility grid, and EV charging units based on real-time operating conditions. Simulation results demonstrate that, compared with the base case, the proposed framework reduces the daily operating cost from USD 1254/day to USD 914/day (27.1%) and network power losses from 182.5 kW to 123.4 kW (32.4%). The minimum bus voltage improves from 0.941 p.u. to 0.978 p.u., renewable-energy utilization reaches 92.7%, peak demand is reduced by 27.5%, and daily carbon emissions decrease from 684 to 406 kg CO2/day (40.6%) compared with conventional charging strategies. These findings demonstrate that the proposed DRL-based optimization framework significantly enhances energy efficiency, operational reliability, renewable energy integration, and grid support, providing a scalable solution for future intelligent and sustainable EV charging infrastructure.