基于鲁棒规范变量分析的故障诊断方法  

Fault diagnosis method based on robust canonical variate analysis

作　　者：邓晓刚[1] 田学民[1] 

机构地区：[1]中国石油大学信息与控制工程学院,山东东营257061

出　　处：《控制与决策》2008年第4期415-419,共5页Control and Decision

基　　金：国家863计划项目(2004AA412050)

摘　　要：针对工业过程的建模数据中含有离群点的情况,提出一种基于鲁棒规范变量分析(CVA)的故障诊断方法.该方法使用相关系数的鲁棒估计代替传统的相关系数,通过基于粒子群算法的投影寻踪技术计算最大化鲁棒相关系数的规范变量,从而建立统计模型并监控统计量检测过程的变化.连续搅拌反应器(CSTR)系统的仿真结果说明,鲁棒规范变量分析方法能在含离群点数据的基础上建立准确的统计模型,比规范变量分析更有效地监控过程变化.A fault diagnosis method based on robust canonical variate analysis is presented to analyze model data with outliers in industrial processes. This method replaces the traditional correlation coefficient with its robust estimator. and applies projection pursuit technique based on particle swarm optimization algorithm to calculate the canonical variate that maximizes the robust correlation coefficient. Then statistical model is built and monitoring statistics are constructed to detect process faults. The simulation results on a continuous stirred tank reactor（CSTR） system show that robust canonical variate analysis can built accurate statistical model from process data with outliers and monitor process changes more effectively than canonical variate analysis.

关 键 词：故障诊断 离群点 鲁棒规范变量分析 粒子群优化算法 投影寻踪技术 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]