A Neural Network based Deep Reinforcement Learning Controller for Voltage Regulation of Active Distribution Network

The increased penetration of Distributed Energy Resources (DERs) poses new challenges to grid operation, especially voltage regulation. The traditional methods towards voltage regulation typically involve solving the power flow problem to estimate voltage magnitudes. This limits the use of those methods to utility companies since they require access to data that are not available to third parties (e.g., network and consumption data). To enable third-party operators (aggregators, community microgrid operators, etc.) to perform voltage regulation, a model-free method is needed. We present a novel model-free data-driven Deep Deterministic Policy Gradient (DDPG) controller for voltage regulation of active distribution networks with high penetration of Smart Inverter (SI)-based DERs. We deploy a Deep Neural Network (DNN) to predict voltage magnitudes in lieu of solving the Power flow problem. The DNN feeds a Deep Reinforcement Learning (DRL) optimization agent, which learns through several interactions with the environment. Our results show that a properly trained DRL agent can successfully coordinate with SI for voltage regulation for IEEE 34 bus system.