电站锅炉监测数据的异常值检测算法研究  

On outliers detecting algorithm of power station boiler monitoring data

作　　者：黄景涛[1] 任志伟[1] 罗威[1] 

机构地区：[1]河南科技大学电气工程学院,河南洛阳471023

出　　处：《计算机与应用化学》2013年第10期1153-1156,共4页Computers and Applied Chemistry

基　　金：国家自然科学基金资助项目(31171685);河南省重点科技攻关项目(132102210087);河南省教育厅自然科学研究项目(2011A510018)

摘　　要：经典的基于距离的异常值检测算法对参数设置比较敏感,且当数据分布不均时检测准确性会受到较大影响。针对这些问题,本文采用一个基于全局距离和的异常值判定新规则,在此基础上通过对基于距离的经典检测算法中常用的欧氏距离进行改进,给出一种面向数据分布不均的距离度量函数,基于这两点,提出了一种基于全局距离和的异常值检测算法。在电站锅炉监测数据上进行了仿真分析,结果表明该算法具有较高的查全率和较低的误报率,而且简化了对参数设置的要求,有效降低了数据分布不均的影响,同时量化了异常值的异常程度。The classic distance-based algorithm for detecting outliers is sensitive to the parameters, and the accuracy depends on the data distribution. To solve the problem, a new rule based on the sum of global distance was designed to detect outliers, and a distance metric function for uneven data is used by improving the Euclidean distance of the classical distance-based detection algorithm. On the basis, a new algorithm based on the sum of global distance is put forwarded for detecting outhers. The simulation on a power station boiler monitoring data shows that the new detection algorithm has higher recognition rate and lower false alarm rate. The requirements of parameter setting in classical methods are simplified, and the influence of data distribution on the detection results is reduced. At the same time, the deviating degree of the outliers is quantified.

关 键 词：电站锅炉 数据预处理 异常值 检测算法 

分 类 号：TP274.2[自动化与计算机技术—检测技术与自动化装置] TK224[自动化与计算机技术—控制科学与工程]