移动窗递推PLS软测量建模及其工业应用  

Recursive PLS Soft-Sensor with Moving Window of Fixed Length and its Commercial Application

作　　者：徐欧官[1] 陈祥华[1] 

机构地区：[1]浙江工业大学之江学院,浙江杭州310024

出　　处：《高校化学工程学报》2009年第6期1044-1050,共7页Journal of Chemical Engineering of Chinese Universities

基　　金：浙江省重大科技专项(2008C11130)

摘　　要：讨论了基于递推部分最小二乘(RPLS)算法的软测量建模方法及其应用。针对过程的时变特性,采用移动窗RPLS算法,在线修正样本的均值和方差,实时更新模型参数,兼顾了建模样本的饱和性、样本信息的完整性。然后将软测量模型应用于工业异构化装置,在线估计对二甲苯(PX)的含量。针对大量工业数据,进行仿真计算,得到模型的最大相对误差、相对均方误差和跟踪性能指标分别为2.68%、0.17%和0.9569,说明该软测量模型具有良好的预测能力和跟踪性能。接着讨论了建模样本长度对模型性能的影响,指出其最佳的样本长度为20～50。A soft-sensor modeling method based on the recursive partial least squares （RPLS） regression and its application to an industrial unit were discussed in this paper. A simplified RPLS method with moving window of fixed length was proposed to re-estimate the parameters of the model by updating the mean value and variance of the modeling samples with the consideration of saturation and information integrality of the samples. The model was then applied to an industrial isomerization unit for on-line estimation of the para-xylene （PX） concentration. A large number of commercial data were simulated, and the high performance of the soft-sensor is demonstrated by the fact that the maximum relative error, relative root mean square error and tracking precision are 2.68%, 0.17% and 0.9569, respectively. At the same time, the discussion about the effect of the modeling sample length on the performance of the model was performed and the optimized sample length was recommended to be 20-50.

关 键 词：软测量 递推部分最小二乘(RPLS)回归 工业异构化装置 在线估计 建模样本长度 

分 类 号：TQ018[化学工程] TP273[自动化与计算机技术—检测技术与自动化装置]