Cooperative Multi-Agent Reinforcement Learning Method for Fast Voltage Regulation using Distributed Wind Turbine Generators

This paper proposes a Multi-Agent Reinforcement Learning (MARL) voltage regulation method for distribution systems with presence of Wind Turbine Generators (WTGs). The control employs reactive power injection or absorption in three WTGs to regulate feeder voltages. A weighted global voltage metric is calculated to summarize nodes' voltages according to their relevance. Three independent RL Agents are trained using policy parameter sharing. Models are trained and tested on real-time Hardware-In-the-Loop setup. The proposed control effectively regulates voltage through the system, bringing median and mean voltages closer to 1 p.u. and reducing standard deviation.