Distributed multi-agent control of an EV-integrated renewable microgrid

This paper presents a distributed multi-agent control (DMAC) framework for an electric-vehicle (EV)-integrated renewable microgrid with photovoltaic generation, a wind turbine, a battery energy storage system, 100 EV agents, and a grid-forming inverter. Power-balance estimation uses a two-tier average-consensus scheme: the EV fleet first reaches consensus on its own aggregate charging demand over a 100-node communication graph, and the five physical agents then reach consensus on the system power balance over a separate 5-node graph. The battery setpoint is set by a two-agent incremental-cost dispatch between the battery and the grid (or the battery and curtailment, depending on the direction of the power imbalance). The inverter’s frequency and voltage dynamics are integrated at 1-second resolution beneath a 60-second supervisory loop, and voltage is computed over three radial feeder laterals rather than a single bus. Every disturbance event (islanding, communication outage, line outage, battery fault, cyberattack/false-data injection) is physically wired into the simulation and evaluated separately. Results are reported as the mean and 95% confidence interval over 10 independent 24-hour simulations for five cases: the full DMAC framework, a centralised first-come-first-served baseline, a centralised baseline using the same EV policy as DMAC, and two DMAC ablations. Consensus converges on 97.9% of steps; dispatch converges on 99.95% of steps; voltage standard deviation is $$0.0187\pm 0.0002$$ p.u.; and holding the EV policy fixed, distributed and centralised coordination are statistically indistinguishable, indicating that the observed performance differences between DMAC and a conventional baseline come from the EV charging policy and dispatch method rather than the coordination architecture itself.