基于局部K-距离的靶场异常数据检测算法

A new agorithm for outlier detection in rocketdrome based on local K-distance

航天靶场观测数据是鉴定运载火箭性能的重要依据,数据中的异常值严重影响数据处理的质量.传统的靶场异常数据处理方法不能适应日益提高的精度要求.为了解决这一问题,文章分析了测量数据中粗大误差的特点,提出了一种适合靶场观测数据的基于局部K-距离的异常数据检测算法LKD(Local K-Distance).该算法通过计算对象与最近k个最近邻中的最大距离来分析数据对象的稀疏程度,从而检测异常值.实验结果证明,该方法简单快速,对粗大误差的检测有效率可达90%以上.

The missile-drone experiment data is important measure to evaluation the performance of the carrier rockets. The abnormal data decreases the quality of data processing. Traditional methods can not meet the in- creasing demand of precision. To solve the problem, this study analyzed the characteristics of the thick errors, and proposed a new algorithm named LKD（LOCal K-Distance） to fit the requirement of rocketdrome. It can detect the outliers based on local K-Distance. It computes results the max distance between specified object and its nearest k-neighbors and analyses the density of the objects to determine the outliers. Experimentals show that the algorithm is efficient , and the hit rate of thick errors is greater than 90 %.