Surrogate modeling of deep excavation impacts on existing tunnels in soft clay using finite element simulations and machine learning

Impact of deep excavation on existing tunnels presents a common challenge in rapidly urbanizing cities with dense metro networks. As underground space development increases, especially in areas with soft soils, excavation-induced deformation of nearby tunnels becomes a critical concern for ensuring structural safety and preventing potential damage. Consequently, this study employed a calibrated PLAXIS 2D finite element model, considering factors such as excavation depth, diaphragm wall thickness, soil properties, and the relative position between the tunnel and the excavation site, to investigate the deformation responses of tunnels induced by adjacent excavation under representative soft clay conditions. Using these simulation results, machine learning models were developed to predict tunnel deformations, incorporating algorithms like XGBoost, ANN, and MARS. While XGBoost and ANN showed higher accuracy in predictions, MARS provided useful explicit expressions for practical engineering applications. Key influencing factors such as horizontal distance, vertical distance, and soil elastic modulus were identified. Based on the relationship between horizontal distance, vertical distance and excavation depth, tunnel deformation zones were defined to guide excavation planning and control. It was observed that increasing the soil elastic modulus can effectively reduce tunnel deformation, offering a practical solution for managing excavation impacts in soft soil environments. The proposed framework provides an efficient surrogate prediction approach for tunnel deformation assessment under representative soft clay excavation conditions and the adopted modeling assumptions.