基于M-SVR的热连轧板带宽度-厚度预测  

Prediction of strip width-thickness for hot tandem rolling based on M-SVR

作　　者：姬亚锋[1] 刘瑜 宋乐宝 黄志权[1] JI Ya-feng;LIU Yu;SONG Le-bao;HUANG Zhi-quan(School of Mechanical Engineering,Taiyuan University of Science and Technology,Taiyuan 030024,China)

机构地区：[1]太原科技大学机械工程学院,山西太原030024

出　　处：《塑性工程学报》2022年第4期58-64,共7页Journal of Plasticity Engineering

基　　金：国家自然科学基金资助项目(52005358);山西省自然科学基金面上资助项目(201901D111243)。

摘　　要：针对板带热连轧过程中存在多变量、非线性、强耦合和时变性的特点以及常规支持向量机仅一维输出的问题,提出了基于机器学习和优化算法的热连轧板带宽度-厚度预测模型。首先,综合考虑影响板带宽度及厚度的主要特征参数,并对现场采集的相应特征数据进行预处理,剔除异常数据,确保数据质量。其次,建立基于多输出支持向量回归(M-SVR)的热连轧板宽度-厚度预测模型;最后采用带有精英策略的非支配排序遗传算法NSGA-II对M-SVR模型的多个参数进行多目标优化。结果表明,所提出结合NSGA-II的M-SVR多目标优化的板带宽度-厚度预报模型对板带宽度及厚度的预测效果较好且具有较强的适用性。In view of the characteristics of multivariable,nonlinear,strong coupling and time-varying in hot tandem rolling process and the problem that there is only one-dimensional output of the conventional support vector machine(SVM),a strip width-thickness prediction model of hot tandem rolling based on machine learning(ML)and optimization algorithms was proposed.First of all,the main characteristic parameters that affecting width and thickness of strip were considered comprehensively,and the corresponding feature data collected on site were preprocessed to eliminate abnormal data and ensure the data quality.Then,a strip width-thickness prediction model based on multi-output support vector regression(M-SVR)for hot tandem rolling was established.Finally,a non-dominated sorting genetic algorithm with elite strategy NSGA-II was used to optimize the multiple parameters of M-SVR model.The results show that the proposed M-SVR multi-objective optimization model combined with NSGA-II has better effect on prediction of strip width and thickness and has strong applicability.

关 键 词：热连轧 宽度-厚度预测 M-SVR NSGA-II 

分 类 号：TG335[金属学及工艺—金属压力加工]