基于随机森林特征选取和遗传算法优化的带钢厚度预测  

Strip thickness prediction based on random forest feature selection and genetic algorithm optimization

作　　者：刘宏旭 李旭[2] 丁敬国[2] 李晓华 张殿华[2] LIU Hongxu;LI Xu;DING Jingguo;LI Xiaohua;ZHANG Dianhua(Cold rolling plant of Anshan Steel Co.,Ltd.,Anshan 114000,China;State Key Laboratory of Rolling and Automation,Northeastern University,Shenyang 110819,China)

机构地区：[1]鞍钢股份有限公司冷轧厂,辽宁鞍山114000 [2]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

出　　处：《轧钢》2024年第6期76-85,共10页Steel Rolling

基　　金：国家自然科学基金项目(U20A20187);“兴辽英才计划”项目(XLYC2007087)。

摘　　要：在冷连轧生产过程中,带钢厚度精度是衡量产品质量和下游企业最为关心的重要指标之一。现有冷连轧带钢厚度精度预测方法主要关注带钢头部厚度命中率,不能反映整卷带钢的厚度精度和波动情况,为此,提出了一种基于深度学习算法的冷连轧带钢全长厚度连续预测模型,实现了从冷连轧工业生产数据到带钢全长厚度的映射。以某厂五机架冷连轧生产线的实际数据为数据集,首先使用随机森林法进行特征选取,以精简输入特征,然后采用遗传算法对DNN模型的初始权重和阈值进行优化,进一步提升模型性能。结果表明:此模型可较为准确和直观地反映整卷带钢厚度精度和波动随轧制时间变化的情况。在加减速和稳定轧制阶段,预测结果的相对误差均控制在±0.5%和±0.1%以内,精度满足实际生产要求。应用此模型,控制系统可以提升预设定精度,并在相应时刻进行预调节,从而达到提高带钢厚度精度的目的。In tandem cold rolling,strip thickness accuracy is one of the most important indicators to evaluate product quality and that downstream enterprises are concerned about.The existing methods for strip thickness prediction mainly focus on the hit rate of the strip head thickness,which cannot reflect the thickness accuracy and fluctuation of the whole coil.A continuous prediction model based on deep learning algorithm for the full length thickness of strip in tandem cold rolling is proposed,and the mapping from industrial data to strip thickness is realized.The data set is built by actual data of a five-stand tandem cold rolling production line.Random forest is used for feature selection to simplify the input features,and then genetic algorithm is used to optimize the initial weights and thresholds of DNN model for the further model performance improvement.The results show that the model can accurately and intuitively reflect the thickness accuracy and fluctuation of the whole coil with rolling time.In acceleration,deceleration and stable rolling stage,the relative errors of predicted results are controlled within±0.5%and±0.1%,which meet the actual production requirements.By using this model,the control system can improve the pre-setting accuracy and implement pre-adjustment at the corresponding time,so as to achieve the purpose of strip thickness accuracy improving.

关 键 词：冷连轧 厚度精度 带钢全长厚度 预测 随机森林 特征选取 深度神经网络 遗传算法 

分 类 号：TG3[金属学及工艺—金属压力加工]