Machine Learning-Based Measurement and Prediction of Ground Settlement Induced by Shield Tunneling Undercrossing Existing Tunnels in Composite Strata

Abstract

To address the issue of insufficient accuracy in traditional settlement prediction methods for shield tunneling undercrossing in composite strata in Hangzhou, this paper proposes a particle swarm optimization (PSO)-based Bidirectional Long Short-Term Memory neural network (Bi-LSTM) prediction model for high-precision dynamic prediction of ground settlement under small-sample conditions. Shield tunneling is a key method for urban tunnel construction. This paper presents the measurement and prediction of ground settlement caused by shield tunneling undercrossing existing tunnels in composite strata in Hangzhou. The longitudinal ground settlement curve resulting from shield tunnel excavation was analyzed using measured data, and the measured lateral ground settlement was compared with the Peck empirical formula. Using PSO, the performance of three machine learning models in predicting the maximum ground settlement at monitoring points was compared: Long Short-Term Memory neural network (LSTM), Gated Recurrent Unit neural network (GRU), and Bi-LSTM. The linear relationships between different input parameters and between input parameters and the output parameter were analyzed using the Pearson correlation coefficient. Based on this analysis, the model was optimized, and its prediction performance before and after optimization was compared. The results show that the Bi-LSTM model optimized with the PSO algorithm demonstrates superior performance, achieving both accuracy and stability.