Regional subsidence monitoring and prediction along high-speed railways based on PS-InSAR and LSTM

Abstract

Abstract The rapid construction of high-speed railways in China has forced more and more routes to pass through slow subsidence zones caused by groundwater extraction or underground mining. It is crucial to carry out deformation monitoring and prediction those areas along high-speed railways to ensure the safe operation. Twenty-nine periods of Sentinel-1A data between Taiyuan South Station and Taigu East Station were first processed using the PS-InSAR technique, and the ground subsidence sequences at ten typical points were selected. Then the Variational modal decomposition (VMD) was combined with the Adaptive Boosting Algorithm (AdaBoost), and the VMD-LSTMAda-LSTMAda model was constructed by combining the long short-term memory (LSTM) model for the prediction of regional ground subsidence along the high-speed railway. The results show that the cumulative subsidence within the 200 m buffer of the line centreline is − 37.58 − 89.19 mm, and the annual average subsidence rate is − 26.57 − 82.63 mm/y. The proposed model can reduce the complexity of the deformation sequence and combines with AdaBoost to improve the prediction accuracy of the LSTM model. It performs well in terms of RMSE, MAE, MAPE, and R 2 at the two feature points (3703: RMSE = 0.82 mm, MAE = 0.25 mm, MAPE = 6.31%, and R 2 = 0.94; 522: RMSE = 1.32 mm, MAE = 0.46 mm, MAPE = 4.92%, R 2 = 0.95). The model improves the accuracy of regional subsidence prediction along the high-speed railway.