基于控制图的马田系统马氏空间生成机理研究

马田系统是以马氏距离为测量尺度,通过选取正常样本构建马氏空间,对多元系统进行诊断和预测的分类技术。马氏距离对样本数据的变化非常敏感,因此用于构建马氏空间的正常样本的数据质量直接影响到分类的准确率。实际应用中正常样本的选取大多依据主观经验判断,缺乏客观规范的选择机制。本文提出基于控制图的马氏空间生成机理,先由专家选取的正常样本构建初始马氏空间,再以每个正常样品在初始马氏空间和对应的缩减马氏空间上的马氏距离增量作为新的测量尺度,以此建立单值控制图,利用控制图稳定性判定规则剔除异常数据,从而得到稳定状态的马氏空间。实验分析结果表明该方法的有效性且提高了马田系统分类的准确率。

Generation Mechanism on Mahalanobis Space of MTS Based on the Control Chart

The Mahalanobis-Taguchi system (MTS) is a classification technology for conducting diagnosis and prediction in multivariable system which employs Mahalanobis distance (MD) as a measurement scale and selects the normal sample to construct the Mahalanobis space (MS).MD is very sensitive to changes in the sample data,so the quality of normal data for constructing MS directly affects the classification accuracy.In the practical application,the selection of normal sample is based on experiential and subjective judgment,lacking of objective and standard selecting mechanism.In this paper,generation mechanism on MS is constructed by using the control chart.MDs based on the original MS selected by the specialists and MDs in the corresponding reduced MS are computed respectively.The MD increment is used as a new measurement scale to establish the individual control chart.The control chart steadystate distinguishing rules are employed to eliminate the abnormal data and result in a stable state of MS.The results of experimental analysis show that the proposed method is effective and can improve the classification accuracy of MTS.