基于特征提取和机器学习的异常数据识别算法

结构健康监测系统的大力发展每天都在产生大量的监测数据。对于结构健康监测系统来说，判断这些产生的监测数据是否正常是对结构健康状态进行分析的第一步，也是关键的一步。同时，监测数据的异常与否也是判断传感器、采集设备、传输设备等是否正常工作的关键性依据。对于一段数据进行识别，判断数据是属于什么样的异常，是一个多分类的问题。采用基于特征提取和机器学习相结合的算法，对时序数据进行分类，能够快速地判断数据是否异常和异常的类型。

Identification Algorithm of Anomaly Data Based on Feature Extraction and Machine Learning

The rapid development of structural health monitoring system will generate a mass of monitoring data every day. For the structural health monitoring system, judging whether these generated monitoring data is normal is the first and crucial step in analyzing the structural health status. At the same time, the abnormality of monitoring data is also a key basis for judging whether the sensors, acquisition equipment and transmission equipment are working normally. It is a multi-classification problem to identify whether a piece of data is normal and to judge what kind of anomaly does the data belongs to. Based on the algorithm combining the feature extraction and machine learning, the time series data are classified, which can quickly judge whether the data is abnormal and the type of abnormality.