摘要
针对丁苯橡胶聚合转化率需在线实时预测,考虑实际工况的复杂性,首先采用具有较强非线性特征提取能力的核主元分析(KPCA)对数据进行前期处理,然后将其结果作为具有学习速度快、泛化能力强的最小二乘支持向量机(LSSVM)的输入,并以交叉验证法对LSSVM参数寻优,从而获得丁苯橡胶聚合转化率软测量模型.经采用工业现场数据仿真研究,聚合转化率预测绝对误差大于1.5的比例小于样本总数的10%,说明该模型预测精度能满足生产要求.
Aimed at the necessity of real-time prediction of polymerization conversion rate of styrene-butadiene rubber and taking account of the complexity of actual working condition,the kernel principal component analysis（KPCA） method with a strong ability of nonlinear feature extraction was used to process the data firstly.And then its result was taken as input of the least squares support vector machines（LSSVM） model with the ability of fast learning and good generalization.Cross validation method was used for the parameters optimization of the LSSVM model.Finally,a prediction model was obtained for polymerization conversion rate of SBR.It was shown by the simulation result of industrial field data that the ratio of polymerization conversion amount to the overall amount of the samples was less than 10% when the absolute error of polymerization conversion prediction was greater than 1.5,showing that the model＇s prediction accuracy could satisfy the production requirements.
出处
《兰州理工大学学报》
CAS
北大核心
2012年第2期73-77,共5页
Journal of Lanzhou University of Technology
关键词
核主元分析
最小二乘支持向量机
丁苯橡胶
聚合转化率
kernel principal component analysis
least square support vector machine
styrene-butadiene rubber
polymerization conversion rate
