Research and Application of GAN-GRU in Bridge Monitoring Data Cleaning

This is the authors' abstract. We don't add key points for this paper.

With the increase of sensors in the bridge health monitoring system, huge amount of time series data are continuously generated and recorded to provid strong support for bridge health monitoring. During the acquisition of time series data, abnormal data can easily be generated due to sensor failure or improper transmission and storage, which affects the performance of the monitoring system. There is a strong dependence between the before and after values of time series data. Therefore, this paper will propose a fusion model based on GAN and GRU by combining the generative capability of generative adversarial network (GAN) with the predictive capability of gated recurrent unit (GRU). The model is very suitable for gated recurrent neural network for sequence data modeling to build the internal network structure, while the discriminator can further improve the prediction accuracy of the network at the sequence level and reduce the impact of error accumulation on the prediction performance of the network. After the network training is completed, the generator can generate forecast values that match the original time series data, thus completing the replacement of abnormal data series. Through the analysis of the experimental results, it is found that the GAN-GRU model can effectively predict the deflection data of the bridge. In the abnormal ten-minute deflection data, the RMSE was 0.0778. Practice shows that the fusion of GAN-GRU model enhances the ability of bridge health monitoring system to deal with abnormal data, and effectively improves the reliability of bridge health monitoring system.