This paper proposes a microgrid optimization scheduling method based on deep reinforcement learning. First, a multi-building microgrid model considering source grid load storage was constructed, and then a static optimization of the multi-agent deep deterministic strategy gradient algorithm was carried out based on principles. The model and real data were imported into a multi-agent reinforcement learning framework suitable for grid-level objectives, and the optimized algorithm was attempted to regulate the voltage of the microgrid system. The results show that the algorithm used has eliminated the illegal peak voltage of the microgrid system and reduced the overall voltage deviation; The optimized multi-objective-oriented algorithm reduces the load generation power difference while maintaining voltage stability, allowing the load power loss to converge to a lower level. The proposed solution is assessed through experiments, and the numerical results confirmed the effectiveness of the proposed solution.