Abstract: With the continuous increase in renewable energy penetration rates, microgrid systems face severe challenges in source-load matching. Traditional "source-following-load" control modes struggle to adapt to the intermittent characteristics of distributed renewable energy sources, urgently requiring the construction of a new operational paradigm of "load-following-source." Through establishing a cyber-physical-social system collaborative framework, this study achieves deep integration of physical power systems, information communication networks, and social user behaviors, forming a multi-dimensional collaborative optimization mechanism. Based on reinforcement learning, multi-objective particle swarm optimization algorithms, and distributed consensus control strategies, a cloud-edge collaborative real-time rolling optimization mechanism is designed. By sensing changes in renewable energy output and dynamically adjusting controllable load response behaviors, real-time supply-demand balance is achieved. Simulation results demonstrate that renewable energy consumption rate is improved by 15.3% and system operating costs are reduced by 12.7%, providing important technical support for constructing new power systems.