基于马氏深度的变点识别方法研究

 在统计过程控制的第I阶段，准确识别运行状态发生漂移的时间点是决定控制效果的关键。本文以多维空间的数据离心程度作为判定变点规则的标准，通过概率密度轮廓将单一观测值序列转化为多维空间中的数据点，运用数据深度技术构造特征变量，并建立变点定位规则。仿真性能分析的结果表明新方法能够在不需要假设过程服从正态分布的前提下对变点位置进行精确定位。在比较研究中也表现出良好的综合性能。

Change Point Identifying Based on Mahalanobis Depth

In the stage of Phase I,it is very important to identify change point accurately.We propose an idea of change point identification based on the proximity of the data center.The proposed method transfers individual observations sequences into points in multi-dimensional space,and the variable is constructed based on data depth,then a change point location rule is developed.The simulation performance analysis result shows that new method could be used without normal distribution assumption.The comprehensive performance is better than the comparative methods.