Reinforcement-learning-based parallel genetic algorithm for the multi-objective hot-rolling scheduling problem of wide-thick slab

Hot rolling is the core process of steel production, and its scheduling problem is the key to determining the rolling rhythm. Compared with traditional hot-rolling scheduling, hot-rolling scheduling optimisation of wide-thick slabs needs to consider the characteristics of cross rolling and the influ...

Full description

Saved in:
Bibliographic Details
Published in:Applied soft computing Vol. 179; p. 113355
Main Authors: Zhang, Zhuolun, Wang, Bailin, Yuan, Shuaipeng, Li, Yiren, Wang, Xiqing, Li, Tieke
Format: Journal Article
Language:English
Published: Elsevier B.V 01.07.2025
Subjects:
Hot rolling scheduling
Island model
Multi-objective asymmetric vehicle routing model
Parallel genetic algorithm
Q-learning
Multi-objective asymmetric vehicle routing model
Parallel genetic algorithm
Q-learning
Island model
Hot rolling scheduling
ISSN:1568-4946
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Hot rolling is the core process of steel production, and its scheduling problem is the key to determining the rolling rhythm. Compared with traditional hot-rolling scheduling, hot-rolling scheduling optimisation of wide-thick slabs needs to consider the characteristics of cross rolling and the influence of the reheating furnace, which significantly increase the difficulty of problem-solving. In this paper, the characteristics of cross rolling and the influence of the reheating furnace are analysed, and the hot-rolling scheduling problem of the wide-thick slab (WTS-HRSP) is mapped to a multi-objective asymmetric vehicle routing problem (MAVRP) model. Combined with the multi-objective characteristics of MAVRP, a reinforcement-learning-based parallel genetic algorithm (RPGA) was designed, which implements parallel computation of multiple populations based on the master-slave island model, designs three priority rules to initialise populations, and optimises the parameters of the genetic operation using Q-Learning. Based on the rolling data of the wide-thick slab hot-rolling mill of a Chinese iron and steel group, experiments were conducted to compare RPGA with five advanced multi-objective optimisation algorithms, and the results showed that RPGA had better convergence and stability and was more suitable for the WTS-HRSP. The findings can help optimise hot-rolling scheduling. •The multi-objective hot-rolling scheduling model of wide-thick slab is proposed.•Designing a reinforcement-learning-based parallel genetic algorithm.•Designing a parallel computation based on the master-slave island model.•Designing the priority rules and Q-Learning to improve the genetic algorithm.•Experiments demonstrate the proposed scheduling algorithm has the best performance.
ISSN:1568-4946
DOI:10.1016/j.asoc.2025.113355