Model Predictive Control-Guided Graph Convolutional Network for Voltage Optimization by Photovoltaic Generation and Battery Energy Storage in Distribution Networks

Abstract

: High penetration of distributed photovoltaics (PVs) introduces substantial variability in distribution networks, which may cause bidirectional voltage violations and increased network losses. To address this issue, this paper proposes an MPC-guided graph convolutional network (GCN)-based voltage optimization control strategy for active distribution networks with PVs and battery energy storage system (BESS). First, a centralized model predictive control (MPC) voltage optimization model is developed to coordinate PV reactive power and BESS active-reactive power under multiple time-series source-load operating scenarios. The MPC-derived optimal control actions are used to construct a high-quality voltage-regulation dataset. Second, a topology-aware GCN controller is trained offline to learn the mapping from network operating states, forecast information, and graph topology to coordinated control commands. During online operation, the trained GCN simultaneously generates reactive-power commands for PVs and active-reactive power commands for BESSs through a single forward propagation, avoiding repeated online optimization. PV active power is retained to maximize renewable-energy utilization, whereas BESS provides four-quadrant power support for voltage regulation. Case studies on a modified IEEE 33-bus and IEEE 141-bus distribution system show that the proposed method effectively mitigates bidirectional voltage violations and achieves control performance and network-loss levels close to those of the model-driven MPC benchmark, while substantially reducing online computation time.