热轧C-Mn钢工业大数据预处理对模型的改进作用
|
摘要
在应用C-Mn钢工业大数据进行神经网络建模时,如果将大量原始数据不加处理或者经过简单的剔除异常值处理后进行建模,很容易建立满足一定精度要求的模型。但是,如果进一步研究模型的规律性,却常常有违背客观规律的情况。这是由于原始数据中大量的数据相互干扰和生产数据的离散分布造成的。因此在建模过程中,需要将冗余和误差较大的数据剔除,保证训练数据和预测数据的均匀分布,这样能够在减小建模的计算量的同时保证数据具有显著的规律性,从而建立出合理的模型。文章利用Bayes神经网络建立了多种牌号C-Mn钢力学性能预测模型,并对影响屈服强度的工艺参数进行了分析。经统计,屈服强度和抗拉强度的预测数据中分别有96.64%和99.16%的数据预测值和实测值绝对误差在±30 MPa之内,伸长率的预测数据中有85.71%的数据预测值和实测值绝对误差在±4%内。
It is easy to construct a model that meets a certain requirement of precision through neural network base on big data of C-Mn steels. In this case,the original industry data are usually without preprocessed or preprocessed by remove the abnormal value simply. However,there will comes a situation that is contrary to the objective laws if the regularity of the model is further studied. This is due to a large amount of data in the original data to interfere with each other and the discrete distribution of the industry data. Therefore,in order to construct a reasonable model,redundant and large error data must be removed,while the distribution of train data and prediction data must be uniform. In this way,the amount of calculation of the model is reduced while a significant regularity of data is excavated. For the sake of verify the hypothesis of ways to use big data,Bayes regularization neural network was selected to construct a model for mechanical properties of multi- steel number. At the same time,the process parameters which influence on yield strength were analyzed. By statistics,the prediction accuracies of yield strength and tensile strength data are 96.64% and99.16%,respectively,of which the absolute error between the predicted value and the measured value lies in the ±30MPa. Among the predicted data of the elongation rate,85.71% of the data absolute error between predicted value and measured value is within ±4%.
It is easy to construct a model that meets a certain requirement of precision through neural network base on big data of C-Mn steels. In this case,the original industry data are usually without preprocessed or preprocessed by remove the abnormal value simply. However,there will comes a situation that is contrary to the objective laws if the regularity of the model is further studied. This is due to a large amount of data in the original data to interfere with each other and the discrete distribution of the industry data. Therefore,in order to construct a reasonable model,redundant and large error data must be removed,while the distribution of train data and prediction data must be uniform. In this way,the amount of calculation of the model is reduced while a significant regularity of data is excavated. For the sake of verify the hypothesis of ways to use big data,Bayes regularization neural network was selected to construct a model for mechanical properties of multi- steel number. At the same time,the process parameters which influence on yield strength were analyzed. By statistics,the prediction accuracies of yield strength and tensile strength data are 96.64% and99.16%,respectively,of which the absolute error between the predicted value and the measured value lies in the ±30MPa. Among the predicted data of the elongation rate,85.71% of the data absolute error between predicted value and measured value is within ±4%.
引文
[1]贾涛,胡恒法,曹光明,等.基于组织-性能预测的集装箱热轧板工艺优化[J].钢铁,2008,43(11):54.(JIA Tao,HU Heng-fa,CAO Guang-ming,et al.Optimization of process parameters of hot rolling SPA-H steel based on prediction of microstructure-properties[J].Iron and Steel,2008,43(11):54.)
[2]庄志平.集装箱用耐候钢热轧板的性能预测系统设计与应用研究[D].镇江:江苏大学,2014.(ZHUANG Zhi-ping.The design and application of the system performance of hot rolled plate weathering steel container[D].Zhenjiang:Jiangsu University,2014.)
[3]ZHAO Yong-hong,WENG Yang,PENG Ning-qi,et al.Prediction of mechanical properties of hot rolled strip by using semi-parametric single-index model[J].Journal of Iron and Steel Research,International,2013,20(7):9.
[4]刘学伟,胡恒法.基于BP神经网络的BNS440热轧板力学性能预测[J].梅山科技,2010(4):25.(LIU Xue-wei,HU Hengfa.Prediction of mechanical properties of BNS440 hot rolled plate based on BP neural network[J].Baosteel Meishan,2010(4):25.)
[5]王丹民,李华德,周建龙,等.热轧带钢力学性能预测模型及其应用[J].北京科技大学学报,2006,28(7):687.(WANG Dan-min,LI Hua-de,ZHOU Jian-long,et al.Quality prediction model of the mechanical properties of hot-rolled steel strips and its application[J].Journal of University of Science and Technology Beijing,2006,28(7):687.)
[6]邹波,魏元,关菊.BP神经网络在热轧带钢力学性能预报中的应用[C]//2007中国钢铁年会.成都:中国金属学会,2007.(ZOU Bo,WEI Yuan,GUAN Ju.Application of BP neural network in prediction of mechanical properties of hot rolled strip[C]//Annual Steel Conference 2007 in China.Chengdu:The Chinese Society for Metals,2007.)
[7]吕游.基于过程数据的建模方法研究及应用[D].北京:华北电力大学,2014.(LU?You.The Research and Application of Data-Based Process Modeling Methods[D].Beijing:North China Electric Tower University,2014.)
[8]Mehmed Kantardzic.Data Mining:Concepts,Models,Methods,and Algorithms[M].Second Edition.Hoboken:John Wiley and Sons Inc,2011.
[9]郭朝晖,张群亮,苏异才,等.关于热轧带钢力学性能预报技术的思考[J].冶金自动化,2009,33(2):1.(GUO Zhao-hui,ZHANG Qun-liang,SU Yi-cai,et al.Thoughts on mechanical property prediction of hot rolled strip[J].Metallurgical Industry Automation,2009,33(2):1.)
[10]毛新平,林振源.薄板坯连铸连轧工艺Ti微合金化高强钢屈服强度影响因素回归分析[C]//2006年薄板坯连铸连轧国际研讨会.广州:中国金属学会.2006:4.(MAO Xin-ping,LIN Zhen-yuan.Regression analysis of thin slab continuous casting and rolling process of Ti micro alloyed high strength steel yield strength influence factors[C]//2006 International Symposium on the Thin Slab Casting and Rolling.Guangzhou:The Chinese Society for Metals,2006:4.)
[11]关建东,康永林.卷取温度、冷轧压下量对SPHD钢力学性能的影响[J].材料开发与应用,2009,24(1):39.(GUAN Jiandong,KANG Yong-lin.Influence of coiling temperature and total cold rolling reduction rate on mechanical properties of SPHD steel[J].Development and Application of Materials,2009,24(1):39.)
[2]庄志平.集装箱用耐候钢热轧板的性能预测系统设计与应用研究[D].镇江:江苏大学,2014.(ZHUANG Zhi-ping.The design and application of the system performance of hot rolled plate weathering steel container[D].Zhenjiang:Jiangsu University,2014.)
[3]ZHAO Yong-hong,WENG Yang,PENG Ning-qi,et al.Prediction of mechanical properties of hot rolled strip by using semi-parametric single-index model[J].Journal of Iron and Steel Research,International,2013,20(7):9.
[4]刘学伟,胡恒法.基于BP神经网络的BNS440热轧板力学性能预测[J].梅山科技,2010(4):25.(LIU Xue-wei,HU Hengfa.Prediction of mechanical properties of BNS440 hot rolled plate based on BP neural network[J].Baosteel Meishan,2010(4):25.)
[5]王丹民,李华德,周建龙,等.热轧带钢力学性能预测模型及其应用[J].北京科技大学学报,2006,28(7):687.(WANG Dan-min,LI Hua-de,ZHOU Jian-long,et al.Quality prediction model of the mechanical properties of hot-rolled steel strips and its application[J].Journal of University of Science and Technology Beijing,2006,28(7):687.)
[6]邹波,魏元,关菊.BP神经网络在热轧带钢力学性能预报中的应用[C]//2007中国钢铁年会.成都:中国金属学会,2007.(ZOU Bo,WEI Yuan,GUAN Ju.Application of BP neural network in prediction of mechanical properties of hot rolled strip[C]//Annual Steel Conference 2007 in China.Chengdu:The Chinese Society for Metals,2007.)
[7]吕游.基于过程数据的建模方法研究及应用[D].北京:华北电力大学,2014.(LU?You.The Research and Application of Data-Based Process Modeling Methods[D].Beijing:North China Electric Tower University,2014.)
[8]Mehmed Kantardzic.Data Mining:Concepts,Models,Methods,and Algorithms[M].Second Edition.Hoboken:John Wiley and Sons Inc,2011.
[9]郭朝晖,张群亮,苏异才,等.关于热轧带钢力学性能预报技术的思考[J].冶金自动化,2009,33(2):1.(GUO Zhao-hui,ZHANG Qun-liang,SU Yi-cai,et al.Thoughts on mechanical property prediction of hot rolled strip[J].Metallurgical Industry Automation,2009,33(2):1.)
[10]毛新平,林振源.薄板坯连铸连轧工艺Ti微合金化高强钢屈服强度影响因素回归分析[C]//2006年薄板坯连铸连轧国际研讨会.广州:中国金属学会.2006:4.(MAO Xin-ping,LIN Zhen-yuan.Regression analysis of thin slab continuous casting and rolling process of Ti micro alloyed high strength steel yield strength influence factors[C]//2006 International Symposium on the Thin Slab Casting and Rolling.Guangzhou:The Chinese Society for Metals,2006:4.)
[11]关建东,康永林.卷取温度、冷轧压下量对SPHD钢力学性能的影响[J].材料开发与应用,2009,24(1):39.(GUAN Jiandong,KANG Yong-lin.Influence of coiling temperature and total cold rolling reduction rate on mechanical properties of SPHD steel[J].Development and Application of Materials,2009,24(1):39.)