Data-Enhanced Multi-Objective Optimization for Hot Rolling Load Distribution

This paper addresses the multi-objective load distribution optimization problem in the rolling process, concerning three optimization targets of balancing rolling force margin, minimizing total rolling power, and ensuring desirable strip shape. In addition to exit thickness, bending force and roll shift position are also incorporated as decisive variables, both exerting significant influence on strip shape. Notably, while a data-driven model trained by neural networks predicts the final strip crown, the strip crown during rolling is determined using a simplified mathematical model. Such an optimization model considers a more comprehensive range of influences compared with previous studies, making it more realistic and accurate. Besides, the NSGA-II algorithm is employed to identify the Pareto optimal set. Through simulation using real product data from a domestic steel company, the effectiveness and practical applicability of the proposed approach are demonstrated.