结构健康监测系统的数据异常识别

结合机器学习方法对结构健康监测系统采集的原始数据进行初步的自动化分析，以达到降低进一步分析的计算量、提高分析子系统精度的目的.以上海中心和兰州西站监测系统为背景，利用机器学习方法研究数据异常识别问题，优化数据分析预警子系统.使用单变量特征选择提取利于识别的特征向量, 对比分析在结构健康监测中各类支持向量机（Support Vector Machine，SVM）的优劣，组合利用不同SVM的优势减少异常数据的漏报和误报.该方法已被应用于上海中心和兰州西站的结构健康监测系统中.

Abnormal data recognition in structural health monitoring system

To reduce the computation of further analysis and increase the accuracy of analysis subsystem, combining with the machine learning method, a preliminary automatic method that analyze the original data collected by a structural health monitoring system is proposed. Taking the monitoring systems of Shanghai Tower and Lanzhou West Railway Station, the abnormal data recognition is researched by the machine learning method, and the subsystem for data analysis and early warning is optimized. The single variable feature selection is used to extract the feature vector; the advantages and disadvantages of several kinds of Support Vector Machine(SVM) in structural health monitoring are compared, and they are combined to reduce false alerts and lost alerts of abnormal data. The method has been applied into the structural health monitoring systems of Shanghai Tower and Lanzhou West Railway Station.