军事知识服务平台
Mechanical property prediction of medium-heavy plate rolling process based on RF-BO-XGBoost
摘要 Accurate prediction of mechanical properties during medium-heavy plate rolling is of great significance for improving rolling qualification rates and optimizing alloy composition.Based on the actual production data of steel grades A and B,a hybrid approach integrating metallurgical mechanisms,data analytics,and expert knowledge was employed for feature selection.A random forest surrogate-based Bayesian optimization of extreme gradient boosting(RF-BO-XGBoost)model was developed to predict yield strength,tensile strength,and elongation.In the Bayesian optimization(BO)procedure,the conventional Gaussian process(GP)surrogate was replaced with a random forest(RF)model.This modification improves compatibility with the high-dimensional and nonlinear characteristics of rolling data and accelerates hyperparameter-search convergence.Comparisons with extreme gradient boosting(XGBoost),light gradient boosting machine(LightGBM),Bayesian optimization-LightGBM(BO-LightGBM),and Gaussian process-based Bayesian optimization of XGBoost(GP-BO-XGBoost)showed that RF-BO-XGBoost achieved the best overall performance.Relative to GP-BO-XGBoost,the proposed model reduced the iteration count and convergence time by 14.6%and 32.4%,respectively.SHapley Additive exPlanations analysis identified final rolling thickness and composition-related variables,including Ceq,Ti,Nb,and V,as important factors affecting mechanical properties,supporting the physical interpretability of the model.After implementation in the plant control system,the model increased the qualification rate of steel grade A by 2.2 percentage points,reduced its Nb content by 1.4%,and increased the elongation of steel grade B by 6.5%.
年,卷(期): 2026, 33(9)
机标分类号: S TP18 F713.36
在线出版日期: 2026年10月12日