基于模式识别的结构健康监测异常数据诊断

Data anomaly is inevitable in field monitoring, leading to interference and misjudgment in the structural safety assessment. To address the problems of low efficiency and low accuracy in detecting multiple data anomalies in field monitoring, this study proposed a multiple data anomalies identification method based on feature extraction and pattern recognition neural network (PRNN). A set of features were established based on the characteristics of different data anomalies, transforming the long raw data samples into short feature vector samples, leading to significantly improved efficiency of data processing and anomaly detection. Moreover, the polarized AUCs curve was introduced to accurately describe the anomaly detection performance, improving in the optimization efficiency for the feature selection and the adjustment of network parameters. A structural health monitoring system was built on the Wuhan Yangtze Shipping Center (335 m). The accuracy and efficiency of the proposed method were verified using the monitoring data of the super high-rise building. The results show that six types of data anomalies are recognized with a 99. 7% detection accuracy using the PRNN-based data anomaly detection method, and the operation time is only one-tenth of the time of deep learning methods.