Distributed Real-Time Workflow Scheduling for Collaborative Operation of Heterogeneous Industrial Equipment

In modern industrial environments, the collaborative operation of heterogeneous industrial equipment is crucial for enhancing production efficiency and resource utilization. However, achieving efficient workflow scheduling and cooperative actions necessitates addressing challenges related to distributed environments, heterogeneity, and realtime requirements. To tackle this issue, this paper addresses the workflow scheduling problem by representing scheduling states with disjunctive graphs. We treat task selection and component allocation as a combined decision and introduce an improved PPO algorithm, termed the PPOM algorithm. This algorithm incorporates an action mask into the traditional PPO framework to indicate valid or invalid actions for each state. By using the action mask, the PPOM algorithm proves to be more stable and practical in real-world applications, enhancing its convergence speed and adaptability. Experimental results show that the proposed PPOM algorithm achieves a reasonable improvement in efficiency compared to priority dispatching rules (PDRs) and other deep reinforcement learning (DRL) methods.