Mutual information based anomaly detection of monitoring data with attention mechanism and residual learning

Due to the damage of sensors or transmission equipment, abnormal monitoring data inevitably exists in the measured raw data, and it significantly impacts the condition assessment of measured structures. Detecting abnormal monitoring data is generally difficult and poses serious challenges to achieve high accuracy. Considering these critical challenges, this study proposes a residual attention network (RAN) that employs the attention mechanism and residual learning to improve classification efficiency and accuracy. With the help of mutual information correlation analysis, the hourly segmented measured data is transformed into the matrix form as the input of the deep learning model for training. The capability and effectiveness of the proposed method are vali-dated with two datasets of acceleration data, one of which is from an arch bridge, and the other is from a cable-stayed bridge. The proposed RAN model is trained with the dataset of the arch bridge, and it identifies most anomaly data types in the test dataset with excellent classification performance. The good generalization performance of the proposed model is verified by another cable-stayed bridge dataset. Compared with the existing preprocessing and deep learning models, the proposed RAN also performs better in multi-classification and classification accuracy.