CAJ | 학술논문

核函数的全影结构投影(total kernel projection to latent structures,T-KPLS)最近在故障监控领域取得了广泛应用，其实质是对数据矩阵的协方差矩阵进行分解，没有利用数据的高阶统计量等有用信息，在进行特征提取时会造成数据有用信息的丢失，导致故障识别效果差。为了解决此问题，提出了统计量模式分析(statistics pattern analysis, SPA)与核函数的全影结构投影法(total kernel projection to latent structures, T-KPLS)相结合的多向统计量模式分析的核函数的全影结构投影法(multi-way statistics pattern analysis total kernel projection to latent structures, MSPAT-KPLS)。该方法首先构造样本的不同阶次统计量,将数据从原始的数据空间映射到统计量样本空间，然后利用核函数将统计量样本空间映射到高维核空间并在质量变量的引导下将特征空间分为过程变量与质量变量相关、过程变量与质量变量无关、过程变量与质量变量正交和残差4个子空间；最后针对与质量变量相关和残差空间建立联合监控模型，当监控到有故障发生时进行故障变量追溯。最后将该方法应用到微生物发酵过程中，并与传统方法进行比较，发现该方法具有更好的监控性能。
핵함수적전영결구투영(total kernel projection to latent structures,T-KPLS)최근재고장감공영역취득료엄범응용，기실질시대수거구진적협방차구진진행분해，몰유이용수거적고계통계량등유용신식，재진행특정제취시회조성수거유용신식적주실，도치고장식별효과차。위료해결차문제，제출료통계량모식분석(statistics pattern analysis, SPA)여핵함수적전영결구투영법(total kernel projection to latent structures, T-KPLS)상결합적다향통계량모식분석적핵함수적전영결구투영법(multi-way statistics pattern analysis total kernel projection to latent structures, MSPAT-KPLS)。해방법수선구조양본적불동계차통계량,장수거종원시적수거공간영사도통계량양본공간，연후이용핵함수장통계량양본공간영사도고유핵공간병재질량변량적인도하장특정공간분위과정변량여질량변량상관、과정변량여질량변량무관、과정변량여질량변량정교화잔차4개자공간；최후침대여질량변량상관화잔차공간건립연합감공모형，당감공도유고장발생시진행고장변량추소。최후장해방법응용도미생물발효과정중，병여전통방법진행비교，발현해방법구유경호적감공성능。
Total kernel projection to latent structures (T-KPLS) has been widely used in the fault detection control field, its core idea is to conduct the covariance matrix decomposition of the data matrix, without using the higher-order statistics and other useful information of the data, which will cause an information loss in the feature extraction process, then result in a bad fault recognition performance. Aiming to solve the problem, a statistics pattern analysis (SPA) combing with the T-KPLS based multi-way statistics pattern analysis total kernel projection to latent structures (MSPAT-KPLS) is proposed. First, different order statistics of the data samples are constructed to map the data from the original data space into the statistic sample space, then utilize kernel function to map the statistic sample space into the higher dimensional kernel space, and according to the quality variable, the feature space will be divided into 4 subspaces, namely: process variable related to quality variable space, process variable not related to quality variable space, process variable orthogonal to quality variable space and residual error space; Lastly, aiming at the process variable related to quality variable subspace and the residual error space, different detection models are constructed, which will trace the fault variables when faults are detected. In the end, apply the proposed method on the microbial fermentation process, and the comparison results with the traditional methods show that the proposed method could achieve a better detection.