Study on abnormal detection of structural deformation monitoring data of single-tower cable-stayed bridge
Conventional abnormal data detection methods have some problems of false alarm and low detection accuracy. Therefore,this paper designs a new abnormal detection method for structural deformation monitoring data of single-tower cable-stayed bridge. Firstly,according to the correlation between deformation parameters of bridge structure and time,the sensor data is processed in time series. Then,the single-channel data collected by the sensor is encoded into a two-dimensional data image by Gram differential angular field transformation,and the abnormal morphological features such as outliers,drifts and so on between numerical values and time series in the image are extracted and marked. Finally,a convolutional neural network detection architecture with three layers of convolution layer and softmax classifier as the core is built,and the abnormal data is accurately detected through convolution,pooling and classification matching. In the experimental verification,it is found that compared with the conventional method,the false alarm rate of the new method is obviously reduced,and the detection effect is better.
single tower cable-stayed bridgestructural deformationmonitoring dataanomaly detectionconvolutional neural network
万方数据
维普
