Scheduling Optimization of Hybrid Microgrid Generators Based on Deep Reinforcement Learning

Abstract

The high penetration of Distributed Energy Resources (DER) changes the direction of power flow, reduces fault currents, and makes the grid configuration more dynamic, making conventional static setting-based protection schemes vulnerable to miscoordination, misoperation, and zone isolation failure. This paper proposes a graph-based adaptive protection framework for smart grids that models the power system as a weighted multigraph, where nodes represent buses/transformer secondaries and edges represent lines, switches, and DER elements. The graph topology and weights are updated in near-real-time from SCADA/PMU/AMI, and then analyzed through graph metrics (e.g., cut-set, community detection, and betweenness) to: (i) identify the most stable protection zone boundaries against configuration changes, (ii) estimate the direction and “footprint” of relevant fault currents under grid-following and grid-forming conditions, and (iii) select a pre-computed set of protection equipment settings (OCR/ROC, directional, distance, DFR, adaptive recloser). The policy engine mechanism executes transitions between settings based on trigger events (topology changes, islanding, or voltage oscillations) with safety guards to prevent chattering. Scenario evaluations show that this approach reduces miscoordination events under inverter-limited fault current conditions, maintains selectivity during reverse power flow, and accelerates the recovery of healthy areas after fault isolation. These results emphasize the potential of the graph-based method as a scalable, adaptive protection foundation ready to be integrated into smart grid control centers with high DER.