Ground deformation prediction based on SBAS-InSAR and RBF neural network: a case study of Zhengzhou Metro Line 10
摘要
Most existing studies on ground deformation from metro construction and operation rely on sparse leveling data, limiting spatial coverage and early warning capabilities. Using Zhengzhou Metro Line 10 as a case study, this study processed 50 Sentinel-1A satellite images (from December 2018 to June 2023) through the SBAS-InSAR technique to derive time-series ground deformation data along the metro alignment. An RBF neural network model for ground deformation prediction was developed using soil cohesion, internal friction angle, compression coefficient, and building height as input variables. SBAS-InSAR results revealed a maximum settlement of 12.1 mm and a maximum annual settlement rate of 3.63 mm a−1 at stations within the completed metro sections. The mean absolute error between SBAS-InSAR and leveling data was 0.46 mm, and the root mean square error (RMSE) was 0.61 mm, confirming the technique’s reliability in loess areas. The RBF prediction model achieved training and testing RMSEs of 0.31 mm and 0.48 mm, respectively, with a 95% confidence interval radius of 0.94 mm, and was validated to accurately predict ground deformation trends in unbuilt metro sections. The integrated application of SBAS-InSAR and RBF neural networks offers an effective approach for monitoring and predicting ground deformation induced by metro construction. Future research could improve prediction accuracy by quantifying additional influencing factors through multi-source data fusion and optimizing model algorithms. This method establishes a scientific basis for early risk warning and route optimization in metro engineering, supporting the safety and intelligent development of urban underground spaces.