Subsidence prediction of high-fill areas based on InSAR monitoring data and the PSO-SVR model

Abstract

Based on SBAS-InSAR technology and machine learning knowledge, the monitoring and prediction of surface settlement in high-fill areas have important guiding significance for construction, maintenance, and operation of engineering projects. This study takes the Chongqing Donggang Container Terminal as the research object, and utilizes 31 scenes of Sentinel-1A data from 2018 to 2019. The surface subsidence data of the area is obtained by SBAS-InSAR technology, and the internal and external accuracy is evaluated. The topography characteristics of the prone areas of surface subsidence were analyzed through an information quantity model to select prediction points. Grey Relational Analysis （GRA） was used to calculate the grey correlation degree between dynamic influencing factors and subsidence. Principal component analysis was used to extract principal components from influencing factors, and training and testing sets were constructed. PSO-SVR prediction model was used to predict the testing set data. To verify the reliability and superiority of the model in subsidence prediction in high-fill areas, the ARIMA model was used as a comparative model, and the prediction results of the PSO-SVR model and the ARIMA model were compared with the testing set. The results show that the prediction accuracy of the PSO-SVR model is better than that of the ARIMA model, and it has better practicality in predicting surface subsidence in high-fill areas.

Keywords

• high fill area
• particle swarm optimization algorithm
• support vector machine regression
• deformation prediction