系统的控制策略对多变量统计监控性能的影响  

Impact of control strategies on multivariate statistical monitoring performance

作　　者：叶鲁彬[1] 刘育明[1] 梁军[1] 

机构地区：[1]浙江大学工业控制技术国家重点实验室,浙江杭州310027

出　　处：《浙江大学学报（工学版）》2009年第1期47-53,134,共8页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(60574047);国家教育部博士点专项基金资助项目(20050335018);国家"863"高技术研究发展计划资助项目(2007AA04Z168)

摘　　要：为分析系统控制策略对复杂工业过程多变量统计监控的影响,以模拟实际复杂系统特征的标准问题Ten-nessee Eastman(TE)过程为测试对象,在基本控制、分散控制和厂级控制3种典型控制策略下,比较了各种控制策略的异同及其对于故障的抑制能力.采用故障检出率作为监控性能指标,详细分析了不同控制策略对基于主元分析(PCA)和基于数据相异度的两种具有代表性的多变量统计监控方法的影响.对TE过程20种故障的仿真结果表明,控制策略对监控方法性能的影响十分明显,针对控制策略的特点来选择合适的建模变量和模型参数能有效地提高故障检出率.The Tennessee Eastman （TE） process simulating the features of actual complex systems was taken as testing object to analyze the impact of system＇s control strategies when applying multivariate statistical monitoring methods in complex industrial processes. The similarities and differences between control strategies and the ability of restraining faults were compared under three control strategies including base control, decentralized control and plant-wide control. Fault detection rate was adopted as performance index for monitoring methods. Then the impact of control strategies on monitoring performance was detailedly analyzed in two typical multivariate statistical monitoring methods respectively based on principal component analysis （PCA） and dissimilarity of data. The simulations of 20 faults of TE process indicate that control strategies have significant impact on the monitoring performance, and choosing appropriate modeling variables and model parameters according to the characteristic of control strategies can effectively improve fault detection rate.

关 键 词：控制策略 Tennessee EASTMAN 主元分析 相异度 监控性能 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]