基于CART算法的管线钢铸坯探伤结果预测与工艺诊断
本研究针对超洁净管线钢产品需求,采集某中厚板生产线管线钢连铸坯生产数据,结合冶金学原理和皮尔逊相关系数选取关键工艺特征属性,以管线钢板的探伤结果为目标标签,采用决策树算法建立管线钢连铸坯质量预测模型。经过对模型结构的调整和优化模型评价,得到了预测效果好(测试集的AUC值为 0.848)、泛化能力强(△AUC值为 0.042)的模型。本决策树预测模型提供了一种简单、高效、可解释性较强的预测方法。此外,根据重要度得分准确定位了工艺的关键点以及阈值。该方法的应用为企业工艺智能调整和产品质量智能管理提供帮助。
This study aimed at the need for ultra-clean pipeline steel products.The production data of pipeline steel billet was collected and used in an iron and steel enterprise,combined with the principle of metallurgy and Pearson correlation coefficient selection of crucial process characteristics of casting properties,for inspection results of pipeline steel plate as the label,the decision tree algorithm is adopted to establish the quality prediction model of pipeline steel continuous casting billet.After adjusting the model's structural parameters and model evaluation,the model with good prediction effect(AUC value of the test set is 0.848)and good generalization ability(△AUC value is 0.042)was obtained.The decision tree prediction model generated in this study provides a simple,efficient,and intense interpretation of the results of the prediction method.The key points and thresholds of the process were accurately located according to the importance score.It provides help for intelligent adjustment of enterprise processes and intelligent management of product quality.
王复越;任毅;田永久;崔福祥;张哲睿;付成哲;
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##鞍钢集团钢铁研究院,辽宁鞍山 114009;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##鞍钢集团钢铁研究院,辽宁鞍山 114009;鞍钢股份有限公司鲅鱼圈分公司,辽宁 营口 115007;鞍钢股份有限公司鲅鱼圈分公司,辽宁 营口 115007;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##鞍钢集团钢铁研究院,辽宁鞍山 114009;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##鞍钢集团钢铁研究院,辽宁鞍山 114009;
Decision Tree Algorithm Machine Learning Pipeline Steel Continuous Casting Prediction Model
320-325 / 6
评论