一种基于DLPP的动态过程故障检测方法

A fault detection method based on DLPP for dynamic processes

针对动态过程故障检测问题,提出一种基于局部保持投影(locality preserving projections,LPP)和扩展矩阵的动态局部保持投影(dynamic LPP,DLPP)新算法.相比动态主元分析(dynamic principal component a-nalysis,DPCA)方法,该算法可以提取隐藏于过程数据中的低维流型信息,建立更精确的模型.首先选择合适的动态步数,构造扩展矩阵;然后使用LPP算法提取信息,将扩展矩阵空间划分为特征空间和残差空间;最后针对这2个空间分别构造T2和SPE统计量对工业过程进行监测.通过在田纳西-伊斯曼(Tennessee-East-man,TE)模型上的仿真研究,表明了该算法是有效的.

In order to handle the fault detection problem for dynamic processes,a novel method named as DLPP(dynamic locality preserving projections) was proposed,which was based on LPP(locality preserving projections) and extending matrix.Compared with DPCA(dynamic principal component analysis),the new method can capture the low-dimensional manifold information hidden in the process data and build more accurate model.In this study,first we chose the proper value of dynamic step for constructing extending matrix;then the LPP algorithm was used to extract information and divide extending matrix space into two parts: feature space and residual space;finally,T2 and SPE(squared prediction error) statistics corresponding to these two spaces were built to monitor the process.A case study of TE(Tennessee Eastman) process illustrated the efficiency of the proposed method.