Hierarchical Reinforcement Learning for Volt/Var and Wireless Communication Co-Scheduling in Active Distribution Network

Hong Cheng, Huan Luo, Zhi Liu, Celimuge Wu, Wei Sun, Weitao Li, Qiyue Li

Abstract: In active distribution networks (ADNs), the rapid changes in photovoltaic (PV) generation can easily lead to short-term voltage stability issues. However, achieving real-time voltage control under limited communication resources is a major challenge. This article addresses this issue by introducing a novel co-scheduling scheme for volt/var control and wireless resources allocation. We model the nonlinear dynamics between PV generation and communication delay into a co-scheduling optimization problem, targeting the minimization of system voltage deviations. To efficiently solve this problem, we propose a multiagent reinforcement learning (MARL) algorithm, termed meta-learning equivalent (ME) model-based hierarchical reinforcement learning (MEHRL). This algorithm employs a hierarchical reinforcement learning (HRL) framework to segment the complex action space and incorporates an ME model to enhance adaptability during distributed training and decentralized execution (DTDE). Simulation results validate the efficacy of the proposed co-scheduling scheme in ADNs and underscore the advanced capabilities of the MEHRL algorithm in addressing the optimization challenge.

External IDs:dblp:journals/iotj/ChengLLWSLL25