Research on Multi-Agent Distributed Optimization Control under the Operating Mode of the New Power System

Multi-agent systems provide distributed decision-making and self-organizing coordination for dynamic control of the new power system. This study develops a multi-agent collaborative regulation framework for this application. First, game theory, consensus protocols, and crowd-sensing mechanisms are reviewed as complementary coordination tools. Second, a fully distributed multi-agent reinforcement learning architecture based on Q-learning is proposed. A reward recorder is introduced by redesigning the incentive function, and distributed linear matrix feedback (DLMF) constrains the state and action spaces during training. The method is evaluated using a generalized constrained distributed optimization model and case simulations. The results show that the proposed framework improves economic performance, operational safety, and disturbance resilience, providing a technical route toward a safe, stable, green, and low-carbon power system.