Time series forecasting and anomaly detection of bridge deformation based on PS-InSAR technique and seasonal-trend decomposition

Abstract

To achieve cost-effective long-term monitoring of bridge displacement, the permanent scatterers interferometric synthetic aperture radar (PSI) technique was applied to a large-span high-speed railway steel truss arch bridge. Between June 7, 2012, and September 7, 2019, a total of 107 descending-orbit images from the Italian COSMO-SkyMed satellite were processed using the PSI technique to derive displacement time series for 19,467 coherent targets on the bridge. A novel hybrid model, integrating improved seasonal-trend decomposition based on locally weighted regression with long short-term memory networks, was developed for multistep prediction of PSI-derived displacements. Based on a statistical analysis of the prediction errors, upper and lower thresholds were defined for early warning. The relative prediction errors for the first six prediction steps remained below 10%, and were compared against the predefined thresholds to detect abnormal displacement events. The results demonstrate that the spatiotemporal locations of anomalies can be effectively identified when prediction errors exceed the established thresholds.