基于SVR模型的带钢精轧宽展预测  

Prediction of lateral spread for hot strip finishing mill based on SVR model

作　　者：张峥 仲兆准 李阳 章顺虎 ZHANG Zheng;ZHONG Zhaozhun;LI Yang;ZHANG Shunhu(School of Iron and Steel,Soochow University,Suzhou 215100,Jiangsu,China)

机构地区：[1]苏州大学沙钢钢铁学院,江苏苏州215100

出　　处：《钢铁研究学报》2023年第8期1017-1024,共8页Journal of Iron and Steel Research

基　　金：国家自然科学基金资助项目(52074187)。

摘　　要：为提高带钢精轧宽展预测精度,结合实际生产数据,建立了基于支持向量机的回归预测模型。首先采用数据清洗、独热编码和主成分分析等方法对样本数据进行预处理,通过交叉验证法对惩罚参数和径向基函数核参数进行寻优,由此确立最佳模型结构。针对测试数据,分别采用回归预测模型、BP神经网络模型和经验公式进行预测,并从平均误差、最大偏差和误差分布等多个角度进行对比。结果表明,回归模型预测下的各项误差指标优势明显,绝对误差在2 mm内的分布占比达到94.52%。In order to enhance the prediction accuracy of hot strip finishing mill,combined with the actual production data of strip finish rolling process,a regression prediction model based on support vector machine was established.First,the sample data were preprocessed by data cleaning,One-Hot encoding and principal component analysis,and the penalty parameters and radial basis function kernel parameters were optimized by cross validation to establish the optimal model structure.Depending on the test data,regression prediction model,BP neural network model and the empirical formula were respectively used for prediction.Comparisons were made from the average error,maximum deviation and error distribution.The results show that the advantages of each error indicators under the prediction of the regression model are obvious,and the absolute error distribution within 2mm accounts for 94.52%.

关 键 词：支持向量机 回归 带钢精轧 宽展 预测模型 

分 类 号：TG335.5[金属学及工艺—金属压力加工]