Abstract: To control the surface settlement caused by shield tunneling, the tunneling parameters are optimized to guarantee safe construction. In this study, an improved model for tunneling parameter optimization based on long and shortterm memory(LSTM) neural networks and particle swarm optimization(PSO) is proposed. First, an LSTM neural network model is designed to predict surface settlement, and its results are compared with those of random forest(RF) and back propagation(BP) neural networks, demonstrating the superiority of the LSTM model. Second, the LSTM model is improved via combined weights, and the predicted surface settlements before and after improvement are compared. Finally, based on the improved LSTM model and PSO, the improved LSTMPSO model for tunneling parameter optimization is designed and applied to a metro shield structure project in Qingdao, China, to verify its reliability. The results reveal that: (1) The LSTM model is superior to the RF and BP models in terms of both accuracy and its ability for generalization. The improved LSTM model has a better performance in predicting the surface settlement then RF and BP models. (2) After optimizing the tunneling parameters by the improved LSTMPSO model, the surface settlement is monitored during tunneling and is found to be in a reasonable range, thereby indicating good feasibility and practicality of this improved LSTMPSO model for tunneling parameter optimization.

Key words: shield tunneling, surface settlement control, tunneling parameters, long and shortterm memory neural network, particle swarm optimization