钢材力学性能的影响因素众多且存在复杂交互作用,合理筛选性能预报模型的影响因素,将有助于提高模型精度。采用随机森林算法以及数据与机理分析相结合的力学性能建模方法,对热轧带钢力学性能预报与影响因素筛选问题进行了研究。首先,基于收集到的大量热轧生产过程实测数据,采用随机森林算法获得各影响因素的重要性排序;接着,基于各因素的重要性排序,逐个增加自变量建立一系列力学性能预测模型,并根据各模型预测误差的变化趋势,判断各因素对模型预测精度影响的大小,进一步筛选出更为重要的影响因素,最终建立以少量重要性较高的影响因素作为自变量的性能预报模型。最后,对国内某大型热连轧机组生产的热轧含铌高强钢产品进行了抗拉强度建模试验,实践表明,基于Mn、Cs、FDH、NbC、NbN、RT、Si、CT以及FET等因素建立的抗拉强度预测模型具有较高的预测精度,平均绝对百分误差为2.52%,均方根误差为21.65 MPa。
Abstract
Many factors affecting the steel mechanical property and there are complex interactions among different factors. It is helpful to improve the model accuracy by reasonably selecting the influence factors of the property prediction model. Using the random forest algorithm and mechanical property modeling method combining data and mechanism analysis,the mechanical property prediction of hot rolled strip and selection of the influence factors are studied. Firstly,based on a large amount of production data of hot rolling process,the variable importance ranking was obtained by using random forests algorithm.Based on the importance ranking of influence factors,a series of mechanical property prediction models are established by increasing the number of independent variables. After that,according to the variation trend of the prediction error of each model,the influence of various factors on the precision of the model are determined,thereby,the more important factors are selected,and the prediction model with a small number of important factors as independent variables is established in the end. Finally,the tensile strength modeling experiment of Nb high strength steel products produced by a large hot strip mill in China was carried out,practice results shows that the mechanical property prediction model based on Mn,Cs,FDH,NbC,NbN,RT,Si,CT,FET has a high predicted accuracy,mean absolute percentage error is 2.52% and root mean square error is 21.65 MPa.
关键词
热轧带钢 /
力学性能预测 /
随机森林 /
影响因素筛选
{{custom_keyword}} /
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1]王蕾, 唐荻, 宋勇. 热轧带钢组织性能预报模型及应用[J]. 钢铁, 2016, 51(11):73-78.
[2]苏理云, 邓燕, 冉雪竹,等. 低合金高强度钢力学性能与化学成分之间的统计建模与分析[J]. 重庆理工大学学报, 2009, 23(2):36-40.
[3]于子金. 钢材力学性能预测系统的研发[D]. 东北大学, 2010.
[4]吴思炜, 刘振宇, 周晓光,等. 基于大数据的力学性能预测与工艺参数筛选[J]. 钢铁研究学报, 2016, 28(12):1-4.
[5]Breiman L. Random Forests[J]. Machine Learning, 2001, 45(1):5-32.
[6]Williams J K. Using random forests to diagnose aviation turbulence[J]. Machine Learning, 2014, 95(1):51-70.
[7]Singh K, Guntuku S C, Thakur A, et al. Big Data Analytics framework for Peer-to-Peer Botnet detection using Random Forests[J]. Information Sciences, 2014, 278(19):488-497.
[8]崔东文, 金波. 基于随机森林回归算法的水生态文明综合评价[J]. 水利水电科技进展, 2014, 34(5):56-60.
[9]李维刚, 胡石雄, 刘斌,等. 热轧含Nb高强钢力学性能预报模型[J]. 冶金自动化, 2017, 41(2):15-21.
[10]Breiman L, Friedman J H, Olshen R A, et al. Classification and Regression Trees (CART)[M]. California: Wadsworth, 1984.
[11]Liaw A, Wiener M. Classification and Regression by randomForest[J]. R News, 2002, 2(3):18-22.
[12]Gr?mping U. Variable Importance Assessment in Regression: Linear Regression versus Random Forest[J]. American Statistician, 2009, 63(4):308-319.
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
基金
国家自然科学基金资助项目;国家自然科学基金资助项目;湖北省教育厅科学技术研究计划重点项目;武汉市青年科技晨光计划资助项目
{{custom_fund}}