基于两步子空间划分的化工过程监测方法

为了解决现代化工过程采集的数据维度高、分布复杂的问题,提出一种基于两步子空间(two step subspace division, TSSD)划分的化工过程监测方法。为了降低过程分析复杂度,将具有相似特性的变量划分为同一空间。考虑数据的复杂分布问题,将第一步得到的每个子空间划分为高斯空间与非高斯空间。利用主元分析(principal component analysis, PCA)和独立元分析(independent component analysis, ICA)方法建立检测模型并构造统计量。整合每个子空间的统计量并基于局部离群因子(local outlier factor, LOF)方法构建综合统计量。结果表明:TSSD方法对于16个故障均能取得最优的漏报率,尤其是故障10和故障16,漏报率分别为15.375%和6.75%,有效验证所提出的基于两步子空间划分的过程监测方法的优越性。

Chemical process monitoring based on two step subspace division

In order to solve the problem of high dimension and complex distribution of data collected from modern chemical processes, a method for monitoring chemical process was presented based on two step subspace division(TSSD). In order to reduce the complexity of process analysis, variables with similar characteristic were divided into the same space. Considering the complex distribution of data, the subspace obtained from the first step was divided into Gaussian subspace and non-Gaussian subspace. Principal component analysis(PCA)and independent component analysis(ICA)were used to establish the detection models and construct the statistics. All statistics of subspaces were integrated and used to construct the final statistics based on local outlier factor(LOF). The process results showed that the optimal missed detection rates of TSSD can be obtained for 16 faults, especially 15.375% for fault 10 and 6.75% for fault 16. The superiority monitoring performance of the proposed two steps subspace division method was proved.