Slope deformation monitoring and prediction based on InSAR and deep learning model

Abstract

Slope instability hazards pose significant risks to transportation lines and infrastructure safety. Slope deformation monitoring results provide insights into hazard development. To retrospectively monitor slope deformation and predict its deformation trends, we propose a slope deformation monitoring and prediction method based on interferometric synthetic aperture radar (InSAR) and deep learning. First, InSAR is used to obtain the deformation characteristics of the target slope from January 2019 to February 2020. Next, the deformation rates in the study area and the characterization of spatio-temporal deformation on the target slope are analyzed. Then, the adaptive boosting support vector regression (AdaBoost-SVR) algorithm is used to continuously process the slope time-series deformation data and establish a data set. The whale optimization algorithm (WOA) is used to optimize the hyperparameters of four deep learning models. Subsequently, the prediction performance is assessed to determine the optimal model. Finally, the discussion verifies WOA's effectiveness and compares its performance to traditional prediction models. The results reveal an overall sliding trend in the target slope, with deformation rates predominantly between 0 and -20 mm/yr. Cumulative deformation varies spatially and temporally within the target area, exhibiting higher values at higher elevations compared to lower elevations. The fitness values of the four models rapidly decrease and then stabilize, indicating the WOA algorithm’s effectiveness in minimizing prediction errors. Based on training and test assessments, WOA-BiGRU is identified as the optimal model for slope deformation prediction, outperforming traditional models in both prediction accuracy and error. The findings could provide a reference for slope deformation prediction and hazard prevention.