A data-driven integrated framework for collaborative prediction and optimisation of hot-rolled steel strip shape

Optimising steel strip shape is critical for quality control in the steel industry. Traditional optimisation methods, primarily based on mechanistic models, are often constrained by equipment limitations and costs, hindering the full utilisation of real-time production data. To address those issues, a Data-driven Integrated framework for process parameter Prediction and Optimization (DIPO) is proposed in this paper, which considers both plant data and expert knowledge. Unlike traditional data-driven prediction models that rely on real-time process state variables, the DIPO framework uses operating setpoints as input, enabling direct optimisation of parameter settings and exploring a broader range of potential solutions. The framework first employs a Stacked Nonlinear Ensemble model to improve prediction accuracy for highly nonlinear Industrial Processes (SNE-IP). Additionally, a Composite Improved Differential Evolution (CIDE) algorithm is introduced, integrating several advancements to address the limitations of traditional DE and account for practical industrial constraints. These improvements enhance optimisation performance, improve computational efficiency, and provide more reliable, implementable solutions for industrial applications. The effectiveness of the framework is demonstrated in the hot rolling process, highlighting its potential for strip shape optimisation and other industrial optimisation tasks.