Reinforcement Learning for Vehicle-to-Grid Voltage Regulation: Single-Hub to Multi-Hub Coordination with Battery-Aware Constraints

A research team has published a new paper detailing a reinforcement learning (RL) framework designed to turn electric vehicle (EV) fleets into a dynamic tool for stabilizing power grids. The system, developed by Jingbo Wang, Roshni Anna Jacob, Harshal D. Kaushik, and Jie Zhang, uses an intelligent control strategy based on the soft actor-critic algorithm. It coordinates power flow from EVs plugged into charging hubs—both single and multiple hubs—to regulate voltage on the grid. Crucially, it does this while respecting realistic constraints like fleet availability and battery state-of-charge, ensuring the EVs remain usable for their primary purpose.

To validate their approach, the team conducted simulation studies on the standard IEEE 34-bus test system, pitting their RL agent against a conventional Volt-Var/Volt-Watt droop controller. In nominal conditions, the AI-driven system achieved performance comparable to this established baseline. However, its true value emerged under stress: during aggressive overloading scenarios, the RL framework provided robust voltage recovery, staying within 10% of the baseline controller's performance. A key innovation is its two-phase training approach, which first focuses on learning stable control policies and then integrates battery-aware deployment. This ensures the system prioritizes the long-term health and availability of the EV fleet, making the concept of using cars for critical grid services more practical and viable.