Two-layer Optimization Scheduling for Off-grid Microgrids Based on Multi-agent Deep Policy Gradient

To address the voltage limit violations and bidirectional power flow problems arising from high-penetration integration of distributed renewable energy, this paper proposes a two-layer active-reactive power cooperative optimization method to achieve cooperative optimal dispatch of active and reactive power in off-grid microgrids, ensuring the secure and stable operation of the system while enhancing operational economy. The lower-level model optimizes slow-regulating discrete devices based on mixed-integer second-order cone programming, while the upper-level model optimizes fast-regulating continuous devices using a multi-agent deep policy gradient algorithm. The two-layer model coordinates both active and reactive power flows of the microgrid, enabling real-time monitoring of the microgrid's status and online decision-making for the optimization of device regulation, without reliance on precise power flow models or complex communication systems. Finally, the feasibility and effectiveness of the two-layer optimization model are validated in the improved IEEE 33-bus microgrid system.