A Multi-Agent Reinforcement Learning Approach to the Dynamic Job Shop Scheduling Problem

In a production environment, scheduling decides job and machine allocations the operation sequence. shop system, wide variety of jobs, complex routes, real-life events becomes challenging for activities. New, unexpected disrupt schedule require dynamic updates to on an event-based basis. To solve problem, we propose multi-agent system with reinforcement learning aimed at minimization tardiness flow time improve techniques. The performance proposed is compared first-in–first-out, shortest processing time, earliest due date dispatching rules in terms tardy mean tardiness, maximum earliness, work process, makespan. Five scenarios are generated different arrival intervals jobs system. results experiments, performed 3 × 3, 5 5, 10 problem sizes, show that our overperforms as workload increases. Under heavy workload, gives best five criteria, which proportion time.