基于遗传算法改进BP算法的精轧首轧钢卷出口厚度偏差预测  

Deviation prediction of exit thickness of finished first rolled steel coil based on improved BP algorithm of genetic algorithm

作　　者：许欣恺 陈雪姣 宋向荣[1,2] 齐正 路广洲 孙晨熙 XU Xinkai;CHEN Xuejiao;SONG Xiangrong;QI Zheng;LU Guangzhou;SUN Chenxi(Automation Research and Design Institute of Metallurgical Industry Co.,Ltd.,Beijing 100071,China;Steel Industry Green and Intelligent Manufacturing Technology Center,China Iron and Steel Research Institute Group Co.,Ltd.,Beijing 100081,China)

机构地区：[1]冶金自动化研究设计院有限公司,北京100071 [2]中国钢研科技集团有限公司钢铁绿色化智能化技术中心,北京100081

出　　处：《冶金自动化》2024年第4期46-52,共7页Metallurgical Industry Automation

基　　金：国家重点研发计划项目(2022YFB3304801)。

摘　　要：为解决目前广泛存在的热连轧精轧换辊后首卷带钢头部厚度偏差过大的问题,提出了一种基于方差选择、互信息、L1/2正则化结合专家经验的混合式特征选择方法,对国内某1 700热连轧厂换辊首卷带钢历史生产数据进行特征选择,并将特征选择结果作为基于遗传算法改进BP(genetic algorithm-back propagation, GA-BP)神经网络的精轧换辊后首卷带钢头部厚度偏差预测模型的训练集。对模型进行一系列实验,以平均绝对百分比误差(mean absolute percentage error, MAPE)、均方误差(mean square error, MSE)、决定系数(R2)等指标作为模型评价标准。结果表明,本文提出的混合特征选择方法相较于传统数学特征选择方法对于模型训练后的预测精度有显著提高。通过在某钢厂主要产品的不同钢种和不同厚度区间的数据样本上测试,验证了模型具有较高的预测精度,具有一定的泛化性。该方法在生产实践中具有很好的应用前景。In order to solve the widely existing problem of excessive head thickness deviation of the first strip after finishing roll change in hot strip mill,a hybrid feature selection method based on variance selection,mutual information,L1/2 regularization combined with experts'experiences was proposed.The method is used to select features from historical production data of the first strip after roll change in a 1700 hot strip mill in China,and the feature selection results are used as the training set of GABP neural networkbased head thickness deviation prediction model for the first strip coil after finishing roll change.A series of experiments are carried out on the model,and mean absolute percentage error(MAPE),mean square error(MSE),coefficient of determination(R2)and other indicators are used as model evaluation criteria.The results show that the hybrid feature selection method proposed in this paper has significantly improved the prediction accuracy of the model after training compared with the traditional mathematical feature selection method.Through testing on data samples of different steel grades and different thickness intervals of the main products of a steel mill,it is verified that the model has a high prediction accuracy with a certain degree of generalizability.The method has good application prospects in production practice.

关 键 词：精轧 遗传算法 BP神经网络 厚度预测 特征工程 

分 类 号：TG335[金属学及工艺—金属压力加工] TP18[自动化与计算机技术—控制理论与控制工程]