Deep Reinforcement Learning for Alleviation of Unbalanced Active Powers Using Distributed Batteries in LV Residential Distribution System

The high penetration and uneven distribution of single-phase rooftop PVs and load demands in power systems may lead to unbalanced active powers, adversely impacting power quality and system reliability. This paper introduces a strategy based on multi-agent deep reinforcement learning to address these unbalanced active powers. The approach involves deploying single-phase battery systems throughout the LV residential distribution system, subsidized by the utility. Initially, the unbalanced active powers are framed as a Markov game, which is then addressed using a multi-agent deep deterministic policy gradient algorithm. The strategy relies on local measurements, with agents' experiences centrally shared during training for cooperative tasks. Notably, information about the phase connections of the battery systems becomes unnecessary. The strategy learns from historical data, gradually mastering the process. Real data from rooftop PVs and demands in a four-wire LV residential distribution system validate the effectiveness of the proposed approach. Acting as adaptive agents, the battery systems collaboratively operate by adjusting active powers to minimize neutral current at the point of common connection.