Answers are grounded in this paper and cite their sources.
Hot rolling scheduling in steel plants is critical for resource utilisation, equipment efficiency, and cost control, yet conventional human- or rule-based approaches struggle with dynamic demand, strict process constraints, and multi-objective trade-offs. This study proposes a hierarchical reinforcement learning (HRL) framework that exploits historical production data to classify slab orders into main product groups and transition orders, enabling a two-level decision hierarchy: a high-level manager selects main product groups and organises rolling courses, while a low-level worker inserts transition orders. Using industrial rolling course data partitioned into training, homogeneous, and heterogeneous test sets, we evaluate performance via scheduling completion rate. On the homogeneous test set, the HRL scheduler achieves a median completion rate of about 97%, compared with roughly 65% for a genetic algorithm baseline; on the heterogeneous test set, HRL maintains over 86% versus about 69% for the baseline. Experiments with 10 random seeds confirm stable convergence, with average completion rates of approximately 96% and 88% on the homogeneous and heterogeneous test sets.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: