Abstract: difficulties in the production and logistics links of manufacturing plants，this paper proposes a multi-agent proximal policy optimization algorithm based on event-aware credit assignment for collaborative production-logistics scheduling. Under the centralized training- distributed execution paradigm，the event-aware credit allocation mechanism is designed to realize the differential response to key business events，which effectively alleviates the non-stationary problem of heterogeneous agent collaborative training. The negotiationarbitration mechanism is introduced to solve the conflict of multiple AGV paths and reduce the invalid driving of the system. An adaptive reward weight module is constructed to enhance the scene adaptability of the scheduling strategy. The simulation results show that the proposed method has strong advantages in on-time delivery of orders，comprehensive efficiency of equipment and anti-disturbance ability，and can effectively improve the overall efficiency of production-logistics collaborative scheduling.