Intelligent Prediction Method for Settlement Curves of Shield Tunnel Based on Machine Learning Algorithms

This paper combines machine learning models and heuristic intelligent optimization algorithms to propose an intelligent prediction method for surface settlement curves during shield tunnel excavation. First, a numerical analysis model for shield tunnel excavation was established, taking the effects of equivalent layer elastic modulus, soil elastic modulus, tunnel radius, soil friction angle, and cohesion into account. A database of 1 680 surface settlement curves under different working conditions was constructed. Then, the influence of geological and lining mechanical parameters, as well as tunnel geometric parameters, on the surface settlement curves was analyzed. The Peck function was used to fit the obtained surface settlement trough curves, and the maximum surface settlements and settlement trough width coefficients under corresponding working conditions were obtained. Finally, the particle swarm optimization algorithm (PSO) was used to optimize the hyperparameters or random numbers of four machine learning methods, namely multi-layer perceptron (MLP), extreme learning machine (ELM), random forest (RF), and support vector regression (SVR). Four finite element simulation proxy models for shield tunnel excavation were established, and the surface settlement curves of the shield tunnel were predicted. The prediction results, prediction errors, and evaluation indexes of the models were compared and analyzed. Results indicate that the PSO-SVR model performs best during training and testing. The established intelligent prediction method for surface settlement of shield tunnels has high computational accuracy and efficiency, and can reasonably and efficiently predict the distribution pattern of surface settlement curves.