Learning Memory-Enhanced Improvement Heuristics for Flexible Job Shop Scheduling

Researchers propose MIStar, a memory-enhanced improvement search framework using heterogeneous graph neural networks for flexible job-shop scheduling problems in smart manufacturing. The approach significantly outperforms traditional heuristics and state-of-the-art deep reinforcement learning methods in optimizing production schedules.

• →MIStar introduces a novel heterogeneous disjunctive graph representation to accurately model operation sequences on machines for scheduling solutions.
• →The framework employs memory-enhanced heterogeneous graph neural networks that leverage historical trajectories to improve decision-making.
• →A parallel greedy search strategy enables superior scheduling solutions with fewer iterations compared to existing methods.
• →The approach addresses key challenges in flexible job-shop scheduling including state representation, policy learning, and search strategies.
• →Extensive experiments demonstrate significant performance improvements over both traditional handcrafted heuristics and modern DRL-based constructive methods.