基于局部离群因子的数据异常识别方法及其在古建结构监测中的应用

Data anomaly identification method based on local outlier factor and application in monitoring data of heritage building structure

基于结构健康监测,深入挖掘监测数据信息是获取古建筑结构健康状态,是保证其耐久性与安全性的重要手段。为精准开展数据分析与结构安全评估工作,区分硬件异常、人为扰动或环境突变两类数据异常成因,分别定义两类砌体结构古建筑监测数据异常。结合砌体结构古建筑监测数据长期静态缓变的特性,提出基于密度的改进离群点识别算法以提高监测数据清洗的质量。通过对时间序列数据的压缩分割,以及子时间序列数据的转化,实现精准拾取数据中的局部离群点。经定性定量分析,基于局部离群密度的改进离群点识别算法对于识别砌体结构古建筑监测数据局部离群点的准确度高、效率高,能够解决现有数据离群点识别算法不适用于砌体结构古建筑监测数据的问题。

In-depth mining of monitoring data information based on structural health monitoring is an important means to obtain the health status of heritage buildings and ensure their durability and safety. In order to accurately carry out data analysis and structural safety assessment, as well as to distinguish the abnormal hardware, artificial disturbance or environmental mutation two kinds of data anomaly genesis, respectively define two types of data anomalies according to different causes, this paper defines two kinds of data anomalies of heritage building masonry structures. According to the characteristics of long-term static and slow change of monitoring data of heritage building masonry structures, an improved outlier identification algorithm based on density is proposed to improve the quality of monitoring data cleaning. Through compression and segmentation of time series data and transformation of series data, local outliers in data can be accurately picked up. Through qualitative and quantitative analysis, the improved outlier recognition algorithm based on local outlier density has high accuracy and efficiency in identifying local outliers of monitoring data of heritage building masonry structures, which can solve the problem that the existing data outlier identification algorithm is not suitable.