基于DGLPP-SVDD算法的化工过程故障检测  

Fault Detection for Chemical Processes Based on DGLPP-SVDD Algorithm

作　　者：徐静 王振雷[1] 王昕 XU Jing;WANG Zhen-lei;WANG Xin(Key Laboratory of Smart Manufacturing in Energy Chemical Process,Ministry of Education,East China University of Science and Technology Shanghai 200237,China;Electrical and Electronic Experimental leaching Center,School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University Shanghai 200240,China)

机构地区：[1]华东理工大学能源化工过程智能制造教育部重点实验室,上海200237 [2]上海交通大学电子信息与电气工程学院电工电子实验教学中心,上海200240

出　　处：《控制工程》2021年第12期2462-2468,共7页Control Engineering of China

基　　金：国家重点研发计划项目(2018YFB1701103)。

摘　　要：针对实际工业生产中数据间存在动态关系,以及高斯和非高斯混合分布特性,提出动态全局局部保留投影-支持向量数据描述(dynamic global-local preserving projection-support vector data description, DGLPP-SVDD)的故障检测算法。构建动态扩展矩阵代替原始数据集,提高动态过程的监测能力。支持向量数据描述(support vector data description,SVDD)算法能高效地处理具有非高斯特性的过程数据,故面对高斯和非高斯混合分布的过程时具有较优异的表现,且只需少量数据就能检测出系统处于非正常运行状态,提高了算法的故障检测率。通过田纳西-伊斯曼(Tennessee Eastman, TE)过程仿真研究对全局局部保留投影(global-local preserving projection, GLPP)算法和DGLPP-SVDD算法进行对比。结果表明,在故障误报率均控制在合理范围且相差不大的情况下,DGLPP-SVDD算法的平均故障检测率更高。Aiming at the dynamic relationship and Gaussian and non-Gaussian mixture distribution between data in actual industrial production, a fault detection algorithm of dynamic global-local preserving projectionsupport vector data description(DGLPP-SVDD) is proposed in this paper. The dynamic extended matrix is constructed to replace the original data set to improve the monitoring ability of dynamic processes. Support vector data description(SVDD) algorithm can process the process data with non-Gaussian characteristics efficiently, so it has excellent performance when dealing with the process of Gaussian and non-Gaussian mixture distribution, and only needs a small amount of data to detect the abnormal operation of the system,which improves the fault detection rate of the algorithm. Global-local preserving projection(GLPP) algorithm and DGLPP-SVDD algorithm are compared through Tennessee Eastman(TE) process simulation research. The results show that DGLPP-SVDD algorithm has higher average fault detection rate than GLPP algorithm when the false alarm rate is controlled within a reasonable range and the difference is not large.

关 键 词：动态特性 全局局部保留投影 支持向量数据描述 数据降维 故障检测 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]