基于多阶段多核支持向量数据描述的间歇过程监控方法  

Batch process monitoring based on multi-phase and multi-kernel support vector data description

作　　者：王晓慧[1,2] 王延江[1] 邓晓刚[1] 曹玉苹[1] 王平[1] WANG Xiaohui;WANG Yanjiang;DENG Xiaogang;CAO Yuping;WANG Ping(College of Control Science and Engineering in China University of Petroleum(East China),Qingdao 266580,China;College of Application Technology in Qingdao University,Qingdao 266061,China)

机构地区：[1]中国石油大学(华东)控制科学与工程学院,山东青岛266580 [2]青岛大学应用技术学院,山东青岛266061

出　　处：《中国石油大学学报（自然科学版）》2020年第4期182-188,共7页Journal of China University of Petroleum(Edition of Natural Science)

基　　金：国家自然科学基金项目(61403418,21606256);中央高校基本科研业务费专项(17CX02054);山东省重点研发计划项目(2018GGX101025)。

摘　　要：针对间歇过程数据的多阶段特性及复杂非线性特性,提出一种基于多阶段多核支持向量数据描述(MPMK-SVDD)的间歇过程故障检测方法。为充分挖掘间歇过程数据的多阶段信息,首先提出一种基于互信息相似矩阵的改进谱聚类方法,解决间歇过程数据集的多阶段划分问题。进一步考虑到单一核函数难以充分描述过程数据的复杂非线性问题,设计一种基于多重核函数和核参数的SVDD监控模型,并通过贝叶斯推理构造全局监测统计量,以实现过程故障的有效监控。以青霉素发酵过程为仿真研究对象,验证方法的有效性。结果表明,提出的方法比传统的SVDD方法能更有效地检测过程故障,具有更高的故障检出率。Aiming at the multi-phase characteristics and complex nonlinear characteristics of batch process data, a fault detection method using multi-phase and multi-kernel support vector data description(MPMK-SVDD) is proposed. In order to fully mine the multi-phase information of batch process data, an improved spectral clustering method based on mutual information similarity matrix is proposed to solve the problem of multi-phase partition of batch process data set. Considering the complexity and non-linearity of the process data that cannot be fully described by a single kernel function, a SVDD monitoring model based on multiple kernel functions and kernel parameters is designed, and the global monitoring statistics are constructed by Bayesian inference to detect process faults. The proposed method is verified by the simulation of penicillin fermentation process. The results show that the proposed method can detect process faults more effectively than the traditional SVDD method and has a higher fault detection rate.

关 键 词：故障检测 支持向量数据描述 贝叶斯推理 谱聚类 间歇过程 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]