基于改进FCM和协方差的数据异常识别与评定

Data Abnormal Identification and Assessment Based on Improved FCM and Covariance

目前,传感器已广泛应用于桥梁施工监控、健康监测等领域,然而其在工作过程中常受到诸如温度变化、其他电子设备干扰等复杂环境因素影响,使采集到的数据出现异常,这为桥梁状况评估带来了干扰。为对桥梁传感器数据进行异常识别与总体质量评定,提出了一种基于改进FCM(模糊C均值)和协方差矩阵的数据异常识别方法。该方法首先对同类型传感器数据进行提取和预处理,包括对缺失数据的填充、标准化和平滑化处理。然后,利用改进FCM算法对差分后的各维数据进行聚类分析,将其划分为不同簇,从而找出簇中心间差异较大的维度,并将识别维度判定为序列异常通道。通过计算协方差矩阵,评估各维数据间的正负相关性。最后,设定阈值后再对序列中的异常进行检测,并对提取的数据质量进行评估。通过分析成都市金堂县某斜拉桥半年监测数据,验证了本文算法能有效筛选出异常传感器及其异常序列和异常点,可为桥梁状态分析和决策提供可靠支撑。

At present,sensors are widely used in bridge construction monitoring,health monitoring and other fields,but they are often affected by factors such as temperature changes and other electronic equipment during the working process,which is easy to make the collected data abnormal,which brings interference to the analysis of bridge conditions.In order to identify abnormality and evaluate the overall quality of bridge sensor data,a method based on improved FCM(fuzzy C⁃means)and covariance matrix is introduced.This method first extracts and preprocesses the sensor data of the same type,including filling,normalizing and smoothing of missing data.Then,the improved FCM algorithm is used to cluster the differentiated dimensional data and divide them into different clusters,so as to find out the dimensions with large differences between cluster centers,and determine the identification dimension as a sequence abnormal channel.The covariance matrix is calculated to assess the positive and negative correlation between the data in each dimension.Finally,the threshold is set before anomalies in the sequence are detected and the quality of the extracted data is evaluated.Through the analysis of the half⁃year monitoring data of a bridge in Jintang,Chengdu,the results of the identification of the two methods verify that the proposed algorithm can accurately and quickly screen out abnormal sensors and their abnormal sequences and abnormal points,which can provide reliable support for bridge status analysis and decision⁃making.