Abstract: Compared with the conventional earth pressure balance shield tunneling method, the surface deformation that results from air pressureassisted tunneling is more complex. To improve the adaptability of the prediction model, accurately predict the surface deformation under the condition of air pressure, and ensure successful shield tunneling, the theory behind the support vector machine (SVM) is introduced, and particle swarm optimization (PSO) is used to optimize the superparameter combination for the SVM. The input parameters of the model are also optimized at the same time. The thickness of the clay layer that overlies the tunnel and the air pressure are set as the input parameters, and the method for calculating the model parameters is improved for the overlying heterogeneous soil layer. A PSOSVM for model for air pressureassisted tunneling is thus established. To verify the accuracy and practicability of the prediction model, the model is applied to a section of the Longzhen entrance and exit line of Guangzhou metro line 18, and the data with optimized input parameters are used for prediction. On this basis, an analysis of the predictions of the surface deformation is conducted using the PSOSVM, SVM, and PSObackpropagation(BP)3 models. The results show that: (1) the PSOSVM model for optimizing the input parameters of airpressure assisted tunneling can meet the engineering requirements; and (2) the PSOSVM model is significantly more adaptable than the PSOBP and SVM models, which indicates its applicability to engineering.

Key words: shield tunnel, air pressureassisted tunneling, surface deformation prediction, support vector machine (SVM); particle swarm optimization (PSO) algorithm