A Hybrid Distributed AI Framework for Voltage Regulation in DER-Rich Distribution Feeders

Abstract

High penetration of inverter-based distributed energy resources (DERs) can lead to overvoltage conditions and reduced operational margins in distribution feeders. Conventional distributed Volt-VAR control provides limited coordination and may not adequately balance voltage regulation and control effort under varying operating conditions. This paper proposes a Distributed Artificial Intelligence (DAI) controller that combines sensitivity-based coordination with a bounded reinforcement learning (RL) adaptation layer. The deterministic backbone computes feasible active and reactive power adjustments, while the RL layer refines control actions to improve voltage regulation and coordination performance. The approach is evaluated on the IEEE 13-bus feeder using quasi-static time-series simulations under high DER penetration. Results show that the proposed method maintains DER-bus voltages closer to nominal values than a baseline distributed strategy, remains effective under communication disturbances, and achieves this improvement mainly through coordinated reactive-power support with only minimal active-power curtailment. The framework provides a promising distributed approach for voltage regulation, with scalability supported by its local information structure.