Answers are grounded in this paper and cite their sources.
The dynamic flexible job shop scheduling problem (DFJSP) demands improved scheduling efficiency, shortened production cycles, and balanced machining status of job shop machines. To address this, a dual-agent proximal policy optimization (PPO) algorithm integrating graph neural networks (GNN) and multi-head attention mechanisms, with the objectives of minimizing the maximum makespan, reducing machine idle rate, and enhancing machine load balancing rate is proposed. The DFJSP is modeled as a Markov decision process (MDP) where the state space is constructed in the form of the heterogeneous graph. It defines the state and action spaces of the job and machine, designs a novel multi-objective reward function to minimize completion time, reduce machine idle rate, and improve machine load balancing rate, and establishes a dynamic hierarchical scheduling framework with two-stage collaborative processing of job selection and machine allocation. A novel multi-objective reward function is designed, and a dynamic hierarchical scheduling framework featuring two-stage collaborative processing—job selection and machine assignment—is established. The multi-head attention mechanism is introduced to strengthen the capability of capturing correlation features between job states and machine states, thereby improving the generalization performance of the policy and robustness against dynamic disturbances. Ablation experiments demonstrate that the proposed multi-head attention mechanism, dual-agent approach, and novel advantage estimation all improve the algorithm’s performance. Comparative experiments show that the proposed algorithm has significant advantages in terms of solution quality and generalization performance.
No takes yet. Share an insight, caveat, or question.
Wu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: