基于深度森林的热轧带钢凸度预测模型  

Hot rolled strip crown prediction model based on deep forest

作　　者：汪龙军 丁成砚 范宇超 孙杰[1] 彭文[1] 张殿华[1] WANG Longjun;DING Chengyan;FAN Yuchao;SUN Jie;PENG Wen;ZHANG Dianhua(State Key Laboratory of Rolling and Automation,Northeastern University,Shenyang 110819,China;Jiangsu Shagang Group Co.,Ltd.,Zhangjiagang 215625,China)

机构地区：[1]东北大学轧制技术及连轧自动化国家重点实验室,辽宁沈阳110819 [2]江苏沙钢集团有限公司,江苏张家港215625

出　　处：《轧钢》2023年第1期90-96,共7页Steel Rolling

基　　金：国家自然科学基金重点项目(52074085,51634002);国家重点研发计划项目(2018YFB1308700);辽宁省兴辽英才计划项目(XLYC1907065)。

摘　　要：在热轧现场轧制规格切换或工况异常的情况下板凸度控制模型偏差较大,为了提高模型精度,提出了一种基于深度森林的热轧带钢凸度预测模型。深度森林模型融合了集成学习和深度学习的思想,采用了多粒度扫描增加数据特征多样性,采用级联森林逐层处理,使得模型具备强大数据拟合能力。将热轧数据经前期预处理导入模型,并对模型参数进行了网格搜索寻优,对比随机森林模型,深度森林模型的效果更优。基于深度森林的热轧带钢凸度预测模型得到了MSE值为6.537,MAE值为1.587,MAPE值为2.903%和R值为0.985的预测性能。In the case of changing gauges or abnormal rolling conditions, the strip crown control model has a large deviation.In order to improve the precision of the model, a prediction model of hot rolled strip crown based on deep forest was proposed. The deep forest model integrates the ideas of ensemble learning and deep learning, adopts multi-granularity scanning to increase the diversity of data features, and the cascade forest processing layer by layer makes the model have strong data fitting ability. The hot rolling data were pre-processed and imported into the model, and the model parameters were grid-searched to find the best results Compared with the random forest model, the deep forest model was more effective, and the prediction performance of the deep forest-based hot rolled strip crown prediction model was obtained with MSE value of 6.537, MAE value of 1.587, MAPE value of 2.903% and R value of 0.985.

关 键 词：热轧带钢 带钢凸度预测 深度森林 机器学习 

分 类 号：TG335.11[金属学及工艺—金属压力加工]