Research on Deep Interactive Collaborative Control Strategy of Source Grid Load Storage for New Power Systems

Driven by dual-carbon goals, the power system is shifting to a renewable-led, digital, and distributed architecture. Deep source-grid-load-storage coordination is key to handling renewable volatility and ensuring security. This paper proposes a multi-agent reinforcement learning-based control method and validates it via simulation. It analyzes the interaction characteristics and theoretical basis, constructs a multi-agent model, and designs a collaborative control framework using centralized training and decentralized execution (CTDE). Simulation on the IEEE 33-node system shows the proposed strategy improves renewable absorption, reduces costs, minimizes load peaks and valleys, and enhances voltage stability compared to traditional methods.