基于控制图的马田系统马氏空间生成机理研究

发布时间：2018-12-17 07:33

【摘要】：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。
[Abstract]:The Martian system is a classification technique which uses Markov distance as the measurement scale and selects normal samples to construct the Markov space to diagnose and predict the multivariate system. Markov distance is very sensitive to the variation of sample data, so the quality of normal samples used to construct Markov space directly affects the accuracy of classification. In practical application, the selection of normal samples is mostly based on subjective experience, and lacks objective and normative selection mechanism. In this paper, the generating mechanism of Markov space based on control chart is proposed. First, the initial Markov space is constructed from the normal samples selected by experts. Then the increment of the Markov distance of each normal sample in the initial Markov space and the corresponding reduced Markov space is taken as the new measurement scale, and the single-value control chart is established, and the abnormal data is eliminated by using the stability rule of the control chart. The stable Markov space is obtained. The experimental results show that the proposed method is effective and improves the accuracy of classification.
【作者单位】：南京理工大学经济管理学院;南京师范大学泰州学院;
【基金】：国家自然科学基金项目(71271114)
【分类号】：O213.1

【相似文献】