基于统计量模式分析的MKPLS间歇过程监控与质量预报  

Batch process monitoring and quality prediction based on statistics pattern analysis and MKPLS

作　　者：常鹏[1] 王普[1] 高学金[1] 齐咏生[2] 张亚潮 

机构地区：[1]北京工业大学电子信息与控制工程学院,北京100124 [2]内蒙古工业大学电力学院,呼和浩特010051

出　　处：《仪器仪表学报》2014年第6期1409-1416,共8页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(61174109;61364009);高等学校博士学科点专项科研基金(20101103110009)资助项目

摘　　要：传统MKPLS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(SPA)与多向核偏最小二乘(MKPLS)相结合的多向统计量模式分析的核偏最小二乘方法(MSPAKPLS)。该方法首先引入滑动窗技术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进行偏最小二乘分析,并对产品质量进行预测。最后将该方法应用到工业青霉素发酵过程中,并与传统方法进行比较,发现该方法具有更好的监控性能和预测性能。Traditional MKPLS method conducts the covariance matrix decomposition of the data matrix, and some useful high-order statistics are not used, which will cause the loss of the useful data information in the feature extraction process and lead to poor fault recognition performance. Aiming at this issue, a multi-way statistics pattern analysis kernel partial least squares method （MSPAKPLS） is proposed, which combines the statistics pattern analysis （SPA） with multi-way kernel partial least squares （MKPLS）. This method first introduces a slide window technique to construct different order statistics of the data sample; the data are mapped from the original data space into the statistic sample space, then the kernel function is used to map the data from the statistic sample space into the high dimensional kernel space, and the PLS analysis and product quality prediction are conducted. At last, this method was applied in the industrial penicillin fermentation process and compared with some conventional methods; the results show that the proposed method has better monitoring and prediction performance.

关 键 词：过程监控 多向核偏最小二乘(MKPLS) 多向统计量模式分析的核偏最小二乘(MSPAKPLS) 间歇过程 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置]