基于决策树算法的钢板探伤预测模型优化  

Optimization of Prediction Model for Flaw Detection of Steel Plates Based on Decision Tree Algorithm

作　　者：王复越 任毅 赵坦 崔福祥[3] WANG Fuyue;REN Yi;ZHAO Tan;CUI Fuxiang(State Key Laboratory of Metal Material for Marine Equipment and Application,Anshan 114009,Liaoning,China;Ansteel Iron&Steel Research Institutes,Anshan 114009,Liaoning,China;Bayuquan Branch of Angang Steel Co.,Ltd.,Yingkou 115007,Liaoning,China)

机构地区：[1]海洋装备用金属材料及其应用国家重点实验室,辽宁鞍山114009 [2]鞍钢集团钢铁研究院,辽宁鞍山114009 [3]鞍钢股份有限公司鲅鱼圈钢铁分公司,辽宁营口115007

出　　处：《鞍钢技术》2022年第6期33-38,共6页Angang Technology

摘　　要：采用决策树分类算法建立钢板探伤预测模型,结合冶金学原理选取精炼与连铸关键工艺特征属性,以轧后钢板探伤结果为目标标签,通过调整决策树最大深度、叶子最小样本数以及判定阈值对模型调优,经测试集验证:优化后的决策树模型对连铸板坯对应轧后钢板的探伤结果预测具有较好的预测效果,AUC值为0.848,且模型泛化能力较强,训练集与测试集AUC差值低于0.04。In this study,the decision tree sorting algorithm was used for establishing the prediction model for flaw detection of steel plates.Then the model was optimized by selecting key technical characteristics and properties in terms of refining and continuous casting combining with principles of metallurgy,taking the flaw detection results of steel plates after rolled as the target labels,adjusting the maximum depth for the decision tree and minimum number of samples of tree leaves and deciding threshold values,all of which were verified by test sets of samples.Therefore the optimized model had a good prediction effect for the flaw detection results of continuous casting slabs corresponding to rolled steel plates,embodying that the AUC value was 0.848.And as the model generalization ability was stronger,the AUC difference value between the training sets and the test sets was less than 0.04.

关 键 词：数据挖掘 决策树算法 机器学习 连铸 探伤 

分 类 号：TP39[自动化与计算机技术—计算机应用技术] TG3[自动化与计算机技术—计算机科学与技术]