Learning-Based Hierarchical Volt/Var and Demand Flexibility Control in Active Distribution Networks
Abstract
High PV penetration increases voltage variability in Active Distribution Networks~(ADNs). While inverter-based Volt/Var Control~(VVC) is the primary means of maintaining voltage limits, its effectiveness is constrained once reactive power capability is saturated. To address this, we propose a hierarchical voltage control scheme where Multi-Agent Reinforcement Learning~(MARL) coordinates inverter reactive power, and a single aggregator agent adjusts active power from fast flexible loads only when voltage violations persist and reactive headroom is insufficient. An activation gate ensures that flexibility is used sparingly. Case studies on a high-PV feeder show that VVC alone resolves moderate deviations, while conditional flexibility effectively mitigates severe over- and under-voltage with only 10–15% load adjustment.
Mohammad Hashemnezhad
PhD Candidate @ CUT
