监护信息系统中异常值的识别和处理方法  

Identification and Processing Methods for Outliers in Information Monitoring Systems

作　　者：邹焱飚[1] 林兆花[1] 谢存禧[1] 

机构地区：[1]华南理工大学机械工程学院,广东广州510640

出　　处：《测试技术学报》2007年第6期531-535,共5页Journal of Test and Measurement Technology

基　　金：粤港关键领域重点突破项目(20054982304);广东省科技攻关项目(2004B10201010);华南理工大学自然科学基金资助项目(B01-E5050810)

摘　　要：提出了一种监护信息系统中异常值分析处理方法,以提高模型辨识的准确性.应用Hampel滤波器算法,识别监护数据中异常值出现的位置;并采用最小二乘支持向量机回归模型,基于递推预报的方法,对于出现异常值的数据点进行数据重构,实现监护信息系统中出现异常值分析处理.通过应用PhysioNet生物医学信号研究资源中的三组数据集:心率、中心静脉压以及血氧饱和度,进行实验研究,结果表明此方法在对监护数据异常值进行分析和处理中取得了很好的效果.Identification and processing methods for outliers in information monitoring systems were proposed to improve the identification accuracy of empirical models. First, outliers are located based on Hampel filter. Second, reconstruction and processing for the data with outliers are implemented based on least squares support vector machine reunification model and recursive prediction. Some real data sets including heart rate, central venous pressure and oxygen saturation of blood from PhysioNet were used for experiments. The results show the effectiveness of the methods for the outliers analysis and processing in information monitoring systems.

关 键 词：监护信息系统 异常值 Hampel滤波器 最小二乘支持向量机 预报 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]