自回归模型时序识别系统的判别函数分类性能分析  

Classification Quality Analysis of Discriminators of an AR Model Based Time Series Identification

作　　者：程懋华[1] 高亹[1] 

机构地区：[1]火电机组振动国家工程研究中心,南京210096

出　　处：《振动．测试与诊断》2002年第4期277-282,共6页Journal of Vibration,Measurement & Diagnosis

基　　金：国家重点基础研究发展规划项目 (编号 :G199980 2 0 30 0 )

摘　　要：从理论上对 5种常用自回归模型时序识别系统的判别函数分类性能进行了定性分析 ,同时定量地比较了它们对试验摩擦振动状态的分类概率。指出了它们各自的特点 ,对其分类能力进行了评价。在常用的 5种判别函数中 ,Kullback- L eibler信息量适用的范围广 ,分类性能好 ,在状态信号特征信息不充分时 ,应优先选用。反之 ,应根据信号特征 ,选择相应的判别函数。The classification quality of five commonly used discriminators of an AR model based time series identification system was qualitatively analyzed in theory and evaluated by a comparison between their probabilities of classifying four different operating conditions of a rotor rub vibration test. Comparison results show that among the five discriminators the Kullback Leibler information distance has a better quality of classification and is of wide application. The essential features of each discriminator are outlined. If there is no enough information extracted from the state signals, priority may be given to the Kullback Leibler information distance, otherwise the choice of a discriminator should be based on the feature of the signal. The work in this paper may be helpful for the choice of a discriminator of an AR model based thime series identification system.

关 键 词：时间序列 自回归模型 判别函数 分类 

分 类 号：TK267[动力工程及工程热物理—动力机械及工程]