一种改进的CSCA+PCA化工过程异常检测技术

为了解决化工过程异常检测时因参数众多且数据庞杂而导致一些异常无法被有效检出的问题，在Brownlee的克隆选择分类算法（CSCA）基础上，通过引入主成分分析（PCA）技术，进行数据降维和数据重整，探讨了人工免疫算法在化工过程异常检测中的适用效果和技术方案，以TE过程数据作为样本进行异常检测和分类实验。结果表明，过程异常数据的规模、属性的数目对CSCA异常检测效果具有明显影响，而通过主成分分析进行数据降维之后，CSCA检测效果有所提高；进一步的数据重整之后，CSCA对过程异常分类辨识的准确率可提升到85%以上；基于CSCA+PCA的数据降维及重构之后的过程异常检测技术方案，可以获得较高的异常检测准确率，从而一定程度上为化工过程安全运行提供技术保障。

An improved CSCA+PCA anomaly detection method of chemical process

In the anomaly detection of chemical process, some anomalies are difficult to be effectively detected due to the multitudinous parameters and miscellaneous data. To solve this problem, based on the clonal selection classification algorithm (CSCA) developed by Brownlee, the data dimension reduction and data rectification were carried out by introducing into the principal component analysis (PCA) technology. The applicability and technical scheme of artificial immune algorithm in the anomaly detection of chemical process were discussed, and the anomaly detection and classification experiments were conducted by using the TE process data as sample. The results showed that the process anomaly data size and the number of attributes had obvious influence on the anomaly detection effect of CSCA, while the detection effect of CSCA was improved after the data dimension reduction through PCA. The accuracy of process anomaly classification and identification using CSCA was significantly promoted to above 85% through the further data rectification. Therefore, the relatively higher accuracy of anomaly detection can be achieved by the process anomaly detection technical scheme after the data dimension reduction and data rectification based on CSCA+PCA, which can provide some technical support for the safe operation of chemical process to a certain extent.