Unbalanced Powers Compensation by Distributed Batteries in Microgrid: A Data-Driven Approach

Abstract

Unbalanced active powers, caused by uneven allocation and high penetration of single-phase rooftop PVs and loads in four-wire microgrid, can affect power quality and system reliability. This paper proposes a deep reinforcement learning (DRL) based strategy to compensate the unbalanced active powers at the point of common connection (PCC) by employing distributed single-phase battery systems in the microgrid and subsidized by a utility. An active power balancing framework based on Markov decision processes (MDPs) is proposed. To access the phase active powers at the PCC, a phase selection strategy based on labelled data is developed so that each local battery system can match the phase active power at the PCC with its own phase using a unidirectional communication link. Also, a state preparation strategy for obtaining an observation state used by the DRL algorithm is proposed. Then, the active power balancing framework is solved by employing a deep deterministic policy gradient (DDPG) algorithm. Real data of single-phase rooftop PVs and loads is utilized for training and testing on modified IEEE-13-bus feeder. The verification results show that the distributed battery systems can compensate the unbalanced active powers. Thus, neutral current and voltage unbalance are minimized while the batteries' active powers and state of charges are maintained within predefined limits.