CAJ | 학술논문

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.
토론료기우체추부분최소이승(RPLS)산법적연측량건모방법급기응용.침대과정적시변특성,채용이동창RPLS산법,재선수정양본적균치화방차,실시경신모형삼수,겸고료건모양본적포화성、양본신식적완정성.연후장연측량모형응용우공업이구화장치,재선고계대이갑분(PX)적함량.침대대량공업수거,진행방진계산,득도모형적최대상대오차、상대균방오차화근종성능지표분별위2.68%、0.17%화0.9569,설명해연측량모형구유량호적예측능력화근종성능.접착토론료건모양본장도대모형성능적영향,지출기최가적양본장도위20～50.
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.