.png){kind=icon}
.png){kind=icon}
Return
Research on rolling schedule coordination optimization method based on distributed decision-making framework
.png){kind=icon}
PRE
AI
DOI：10.1016/j.jmapro.2025.08.052.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
To address the challenges of multi-objective, strongly coupled production processes in hot-rolled strip steel manufacturing, where traditional methods struggle to balance global and local objectives, this paper proposes a distributed multi-level, multi-stage coordination optimization framework encompassing three critical subsystems: rough rolling, finish rolling, and coiling. This framework integrates evolutionary algorithms, position-enhanced attention mechanisms, and the iSOMA-SACPEAF reinforcement learning (RL) based on the Pareto front to achieve dynamic control within each subsystem. Combined with a distributed coordinated optimization controller, it enables efficient coordination of global objectives, thereby improving product quality, process stability, and energy utilization. The AGA-BO-ATTENTION-DNN (AGA-BO-ADNN) model is proposed to achieve high-precision prediction and correction of critical process parameters (rolling force) within the subsystems. Low-error rolling force calculations ensure the accuracy of key variables such as rolling power, providing more precise objective function values for the optimization process. The results of real production data and actual application tests show that the prediction framework controls rolling force error within 0.55 %; the distributed coordination optimization framework outperforms existing solutions in improving strip quality, coil quality, energy consumption optimization, and process stability. In the optimized rolling scheme, the average crown of the strip at the exit of the finishing and roughing rolls is close to 10 μm, the thickness deviation is stabilized within 0.4 %, and the rolling tension torque fluctuation is controlled within 0.1953 %. Additionally, the overall energy consumption across the three stages is reduced by 2.13 %, providing an efficient and innovative solution for multi-objective collaborative optimization in complex industrial processes.
Keywords:
multi-objective optimization
distributed coordination
hot-rolled strip steel
evolutionary algorithms
reinforcement learning
Journal
IF:
6.8
Papers:
7.6K
Citations:
3.5W