To address the issues of insufficient high-dimensional state perception and low multi-device collaboration efficiency in dynamic scheduling of intelligent production lines, this paper proposed a dynamic scheduling model that integrates proximal policy optimization, attention mechanisms, and a centralized training and distributed execution framework. The research first proposed a framework for the scheduling problem of intelligent production line workshops. Then, it used the proximal policy optimization algorithm to model this framework for reinforcement learning. Through the attention mechanism, the representation ability of the relationship between key equipment states and tasks was enhanced. On this basis, the study introduced a multi-agent paradigm based on centralized training and distributed execution to achieve autonomous decision-making and global goal coordination for each process node of the production line under local observation constraints. After the application of the research model, the manufacturing cycle of the intelligent production line was reduced by more than 23.7% compared to the other three algorithms. Moreover, under this scheduling scheme, the equipment utilization rate reached 86.5%. In dynamic event scenarios, the study completed 81 jobs and only required 11 re-planning times, which demonstrated more efficient and suitable scheduling capabilities compared to other methods. In the actual workshop scheduling scenarios, the delivery delay of the research method was reduced by 74.1% to 56.9%, the equipment utilization rate increased by 8.6% to 16.5 percentage points, the energy consumption intensity decreased by 46.9% to 32.1%, and the scheduling response speed was accelerated by 64.6% to 43.5%. From the above content, it can be concluded that the intelligent production line dynamic scheduling and collaborative control model proposed in the research significantly enhanced the robustness and adaptability of the flexible manufacturing system. In the actual scenarios, it could quickly respond to uncertain disturbances such as order changes and equipment failures. The research method can provide a feasible technical path and decision support for industrial intelligent production, further promoting the manufacturing industry to leap to the “perception - decision - execution” closed-loop intelligentization.
Li et al. (Fri,) studied this question.