Hierarchical Source-Network-Storage Coordinated Scheduling for Low-Carbon and Economic Operation of Active Distribution Networks via Carbon Flow and Reinforcement Learning

To address the coupling and timescale differences between network reconfiguration and Source-storage dispatch in active distribution networks, this paper proposes a hierarchical source-network-storage coordinated scheduling method for low-carbon and economic operation. A three-layer framework is developed to coordinate hourly network reconfiguration with 15-min source-storage dispatch, where carbon flow theory is incorporated to formulate a comprehensive low-carbon economic objective. A Trajectory-Centered Advantage-Driven Proximal Policy Optimization algorithm is proposed for high-dimensional topology decision-making, while a carbon-flow-driven three-path differential reward mechanism and curriculum learning are integrated into Multi-Agent Proximal Policy Optimization for source-storage dispatch. The coordination layer further adjusts network topology according to source-storage operating states. Simulations on a modified IEEE 33-bus system demonstrate that, across renewable penetration levels of 20–100%, the proposed method reduces daily carbon emissions by 934.64–1091.66 kg and operating costs by 212.90–246.26 CNY, while reducing active power losses and improving voltage performance. The method also demonstrates good generalization capability under unseen renewable penetration levels and achieves scheduling performance comparable to mathematical optimization, while reducing online computation time by 98.14%.