关键词: 软土, 修正剑桥模型, 室内试验, PSO-BP神经网络, 参数反演

Abstract: The reasonable value of soil constitutive parameters is an important prerequisite for numerical simulation. In order to accurately obtain the parameters of the modified Cambridge model for Huzhou soft soil, a parameter inversion method for the modified Cambridge model based on laboratory experiments and neural networks was developed for two typical soft soils in the region. Firstly, laboratory triaxial consolidation undrained tests and standard consolidation rebound tests were conducted, and based on the test results, the parameter inversion interval of the modified Cambridge model for typical soft soil in Huzhou region was determined. Secondly, based on the principle of orthogonal experimental design, numerical calculations were conducted on the lateral displacement of the retaining structure at different parameter levels during the excavation process of the foundation pit. Based on the numerical calculation results, 64 sets of PSO-BP neural network training samples were constructed. Finally, the constructed training set was used to invert the parameters of the modified Cambridge model for soft soils in Huzhou area. The critical state effective stress ratio (M₁, M₂), compression parameter (λ₁、λ₂), rebound parameter (κ₁、κ₂), and void ratio (e₁、e₂) of the two typical soft soil modified Cambridge model parameters obtained through inversion were M₁=1.076、λ₁ =0.050、κ₁ =0.021、e1=1.712，M₂=1.123、λ₂=0.038、κ2=0.012，e₂=0.967. The predicted deformation values of the retaining structure calculated through inversion parameters were in good agreement with the measured values, with a relative error of no more than 5%. Based on the inversion parameters, finite element numerical calculation was used to predict the deformation of the foundation pit, and the prediction results verified the accuracy of the inversion method. The influence of the number of neural network training samples and the number of input layer nodes on the inversion results of the Cambridge model parameters for soft soil correction was analyzed. The research results can provide parameter support and technical guidance for similar foundation pit projects in Huzhou area.

Key words: machine learning, artificial intelligence, geotechnical engineering, forecast, algorithm