Simultaneous Voltage Regulation and Unbalance Compensation in Distribution Systems With an Information-Driven Learning Approach

Abstract

High penetration and uneven allocation of single-phase and three-phase loads together with photovoltaic (PV) sources in low-voltage distribution systems can result in voltage unbalance and voltage regulation challenges. Previous approaches that treated the voltage unbalance and the voltage regulation as separate problems may have overlooked their interdependencies, thus potentially limiting the effectiveness of those methods. This article proposes a novel framework that can regulate the bus voltages and compensate for the voltage unbalance simultaneously, without requiring any information on phase connection of single-phase PV inverters. The proposed framework is designed based on a Markov game. The bus voltage deviation and the voltage unbalance factor (VUF) are used to define the reward function. Local observation states are obtained from smart meters, and the reactive current of the PV-based voltage source inverter (VSI) is defined as the action within the framework. The framework employs a multiagent deep deterministic policy gradient (known as multiagent deep deterministic policy gradient) algorithm that incorporates centralized training and decentralized execution. It is utilized to solve the framework. To ensure its effectiveness, the framework is trained and tested using real data acquired from smart meters. This data encompasses information on PV generation and load demand. The obtained results demonstrate that the bus voltages are regulated close to their nominal value and the VUF is maintained at less than 2% while the PV-VSIs operate within predefined limits.