Graph-Aware Multi-Agent Power Dispatch in V2G-Enabled Distribution Networks

Abstract

Large-scale integration of renewable energy and electric vehicles (EVs) increases voltage regulation and coordination challenges in active distribution networks (ADNs). This paper proposes a graph-aware multi-agent coordinated active and reactive power dispatch method for vehicle-to-grid (V2G) enabled ADNs. The proposed framework combines day-ahead scheduling with intraday residual control. In the day-ahead stage, active power references, reactive power references, energy trajectories, and flexible regulation boundaries are generated for EV aggregators. In the intraday stage, each EV aggregator acts as an agent and corrects the reference plan within the day-ahead feasible domain. To characterize time-varying regulation capability, a dynamic service availability model is introduced by considering charging availability, V2G availability, and discharging service risk. A graph neural network (GNN)-based multi-agent proximal policy optimization (MAPPO) method is then developed to capture network topology coupling and short-term operational dynamics. Case studies on the IEEE 33-node ADN show that the proposed method reduces voltage violation events and improves terminal-node voltage support.