Dynamic scheduling for hot rolling production by hierarchical reinforcement learning and heterogeneous graph neural networks

Publication Type

Journal Article

Publication Date

10-2026

Abstract

Dynamic hot rolling scheduling is a challenging problem in steel manufacturing due to the coupling between reheating and rolling, multiple conflicting objectives, and uncertainties in real-world production environments. Existing optimization methods typically struggle to deliver high solution quality within real-time computational limits and fail to handle the long-term decision dependencies in dynamic environments. To address these limitations, we propose a novel dynamic scheduling approach based on hierarchical reinforcement learning (HRL) and heterogeneous graph neural networks (HGNNs) for hot rolling production under stochastic slab arrivals. Our HRL framework features two coordinated agents: a high-level agent that learns to dynamically partition newly arrived slabs into a sequence of subproblems while capturing their long-term dependencies, and a low-level agent designed as a neural solver to solve each subproblem in real time. For efficient decision-making, the low-level agent adopts a discrete–continuous hybrid decision scheme, where the discrete decision process is structured over a heterogeneous graph representation, and an HGNN-based policy network is designed to exploit the structured relational information between slabs and furnaces to construct high-quality scheduling plans. Furthermore, we introduce a simple yet effective auxiliary loss encouraging feasibility awareness to stabilize training. Experiments based on a real-world production case demonstrate that our approach outperforms existing methods in both solution quality and computational efficiency, and remains robust across diverse instance distributions, objective configurations, and production settings.

Discipline

Artificial Intelligence and Robotics | Operations Research, Systems Engineering and Industrial Engineering

Research Areas

Intelligent Systems and Optimization

Publication

IEEE Transactions on Industrial Informatics

First Page

1

Last Page

12

ISSN

1551-3203

Identifier

10.1109/TII.2026.3738890

Publisher

Institute of Electrical and Electronics Engineers

Additional URL

https://doi.org/10.1109/TII.2026.3738890

This document is currently not available here.

Share

COinS