摘要
良好的铁水质量是铸铁性能可靠性和稳定性的保证,而铁水中硫(S)含量和硅(Si)含量是衡量铁水质量的主要指标,因此在出铁前精准获取铁水S含量和Si含量具有非常重要的意义。实验提出一种结合主成分分析(PCA)和最小二乘支持向量机(LS-SVM)模型的铁水S含量和Si含量的预测方法。将某钢厂大型高炉的在线采集数据作为研究对象,首先对影响铁水中S含量和Si含量变化因素的数据做主成分分析,求取主成分作为模型的输入变量,其次建立最小二乘支持向量机预测模型对铁水S含量和Si含量进行预测。在S含量预测过程中,正则化参数gam和核函数参数sig分别取20、700时,预测误差最小,其均方根误差为0.0012,仿真时间为0.423105 s;Si含量预测过程中正则化参数gam和核函数参数sig分别取40、500时预测误差最小,均方根误差为0.0238,仿真时间为0.079522 s。最后将实验结果与传统最小二乘支持向量机(LS-SVM)和结合PCA的BP神经网络预测模型(PCA+BP神经网络)的结果对比,后两组对比实验关于S含量预测的均方根误差分别为0.0015和0.0014,仿真时间分别为1.320842 s和2.245967 s;后两种对比实验关于Si含量预测的均方根误差分别为0.0316和0.0325,仿真时间分别为0.459671 s和2.061576 s。实验结果表明,实验方法更加全面地考虑了所有因素对铁水中S含量和Si含量变化的影响,具有训练时间短、预测精度高等优点。
Good quality of molten iron is the guarantee for the reliability and stability of cast iron performance.The contents of sulfur(S)and silicon(Si)in molten iron are main indicators to measure the quality of molten iron.Therefore,the accurate determination of S and Si contents in molten iron before tapping is of great significance.A method for predicting the S and Si contents in molten iron by combining principal component analysis(PCA)with least squares support vector machine(LS-SVM)model was proposed.The online data of large blast furnace collected in a steel plant was used as the research object.Firstly,PCA was performed on the data that influencing the changes of S content and Si content in molten iron.The principal component was obtained for the input variable of model.Secondly,the LS-SVM model was established to predict the S content and Si content in molten iron.During the prediction process of S content,the prediction error was minimal when the regularization parameter(gam)and kernel function parameter(sig)was 20 and 700,respectively.Meanwhile,the root mean square error was 0.0012,and the simulation time was 0.423105 s.During the prediction process of Si content,the prediction error was minimal when gam and sig was 40 and 500,respectively.Meanwhile,the root mean square error was 0.0238,and the simulation time was 0.079522 s.Finally,the experimental results were compared with the traditional LS-SVM model and PCA+BP model(PCA combined with back-propagation neural network).For the prediction of S content in traditional LS-SVM model and PCA+BP model,the root mean square error was 0.0015 and 0.0014,and the simulation time was 1.320842 s and 2.245967 s,respectively.For the prediction of Si content in traditional LS-SVM model and PCA+BP model,the root mean square error was 0.0316 and 0.0325,and the simulation time was 0.459671 s and 2.061576 s,respectively.The experimental results showed that the proposed method could more fully consider the influence of all factors on the change of S content and Si content in molten iron.This method exhibited the advantages such as short training time and high prediction accuracy.
作者
赵宁
王玉英
杨凡
杨卫轩
ZHAO Ning;WANG Yu-ying;YANG Fan;YANG Wei-xuan(College of Science,Xi′an University of Architectural and Technology,Xi′an 710000,China)
出处
《冶金分析》
CAS
北大核心
2020年第2期1-6,共6页
Metallurgical Analysis
