高频监测数据的异常数据特征分析与检测

数学理论与应用 ›› 2016, Vol. 36 ›› Issue (3): 42-53.

高频监测数据的异常数据特征分析与检测

李婷，孔文佳，周娟，郑洲顺

中南大学数学与统计学院,长沙,410083

出版日期:2016-09-30 发布日期:2020-09-28
基金资助:
国家自然科学基金项目(51174236)

Characteristic Analysis and Detection of Abnormal Data in High Frequency Monitoring Data

Li Ting， Kong Wenjia， Zhou Juan， Zheng Zhoushun

School of Mathematics and Statistics,Central South University,Changsha 410083,China

Online:2016-09-30 Published:2020-09-28

摘要/Abstract

摘要： 对某火电厂10个设备的60000条高频监测数据进行基本的统计分析,得出高频数据的异常数据具有异常点和异常段两种特征,提出了一种基于频数分布和一阶向前差分的检测高频数据中的异常点和异常段的方法.根据数据的一阶向前差分绝对值的频率分布以及风险系数来确定异常数据的阈值,根据设备本身的性能和采样频率确定了异常段所包含异常点的最大个数,根据阈值和最大异常点个数给出异常点和异常段的判断规则.用该方法诊断火电厂前置泵电机的6000条数据的异常数据,结果与实际异常数据相符.

关键词: 火电厂, 异常点, 去噪, 频数分布, 向前差分

Abstract:

In this paper,by analyzing 60000high-frequency monitoring data of 10equipments in a thermal power plants,we find that abnormal data has two characteristics:outlier or abnormal segment.A denoising method based on the first-order forward difference and the frequency distribution to detect the outlier and abnormal data in the high-frequency is proposed.In this method the threshold of abnormal data is determined by the risk probability and the frequency distribution of absolute value of the first-order forward difference, and the maximum number of the outliers included in the abnormal segment is obtained according to the performance of the device and the sampling frequency. The judgment rule for detecting abnormal data is then given with the threshold and the maximum number of the outliers.The method is applied to 6000data of a thermal power plant pre-pump motor for showing its effectiveness.

Key words: Thermal power plant, Abnormal data, Denoising method, Frequency distribution, Forward difference

李婷, 孔文佳, 周娟, 郑洲顺. 高频监测数据的异常数据特征分析与检测[J]. 数学理论与应用, 2016, 36(3): 42-53.

Li Ting, Kong Wenjia, Zhou Juan, Zheng Zhoushun. Characteristic Analysis and Detection of Abnormal Data in High Frequency Monitoring Data[J]. Mathematical Theory and Applications, 2016, 36(3): 42-53.