| Soil is the core object of geotechnical engineering research,and the selection of its parameters is the basis for engineering design,construction,and theoretical calculation.The standard values of geotechnical parameters are the basic representative values of geotechnical engineering design,representing the reliability of geotechnical parameters.Therefore,selecting reliable parameter standard values has a crucial impact on geotechnical engineering design and construction.In geotechnical engineering,the shear strength of soil is an important engineering parameter and an important indicator for strength analysis and stability evaluation.In practical engineering,the standard value of shear strength parameters is generally estimated through classical statistical methods based on the test results of multiple samples,which will generate many errors and affect the accuracy of the determined standard values.Based on a large amount of survey data from Ningbo Rail Transit,this article statistically analyzes the parameters of soft soil in Ningbo and proposes a method for determining sample shear strength parameters based on machine learning theory,which can optimize the determination of shear strength parameters for individual samples.In addition,Bayesian theory is used to optimize and correct the method for determining the standard value of shear strength parameters.Furthermore,numerical simulation analysis of excavation processes is conducted by inputting the corrected standard value of geotechnical shear strength parameters and data provided in survey reports.By comparing the simulation results with actual monitoring data,the reliability of the selected standard values and feasibility of the proposed method are verified.Finally,a machine learning model is established using a series of dynamic triaxial test data to predict the dynamic damping ratio of Ningbo soft soil,and the model is applied for predicting dynamic parameters.The following achievements have been made:(1)By analyzing and processing survey data of Ningbo rail transit,the probability density function of typical soft soil in Ningbo was obtained,and the correlation between shear strength parameters and various physical parameters was analyzed.It was found that there is a significant correlation between shear strength parameters and natural water content,density,porosity ratio,plastic limit,and liquid limit.(2)Traditional experimental methods for determining shear strength parameters of soft soils require a lot of time and cost,and there is significant disturbance and uncertainty.Therefore,this thesis proposes a machine learning-based method to establish a prediction model and determine the sample’s shear strength parameters.Physical parameters such as natural water content,porosity,gravity density,plastic limit,and liquid limit are used as independent variables,and geotechnical shear strength parameters are used as dependent variables to establish machine learning prediction models.The article establishes K-nearest neighbor algorithm(KNN),classification and regression tree(CART),support vector regression(SVR)prediction models.Furthermore,the particle swarm optimization(PSO)is used to optimize the parameters of the optimal model,which is the SVR model.PSO-SVR achieves an RR~2 of 0.664 in predicting cohesive strength and an RR~2 of 0.818 in predicting internal friction angle.(3)Based on the PSO-SVR prediction model for determining sample shear strength parameters,this thesis introduces Bayesian theory to optimize and correct the method of determining standard values for geotechnical parameters to reduce the variability of calculated standard values for shear strength parameters from sample data.To verify the reliability of the determined standard values for geotechnical shear strength parameters combined with machine learning theory and Bayesian theory,this thesis uses the excavation process of a working well in the Ningbo subway as a case study.Combining actual monitoring data,the FLAC3D finite difference software is used to simulate the excavation process of the foundation pit,incorporating the corrected geotechnical shear strength parameter standard values and data provided in the survey report.The simulated results are then compared with actual monitoring data.The research results show that the determined standard values for geotechnical shear strength parameters are reliable and preliminarily verify the feasibility of this value determination method.(4)A series of dynamic triaxial tests were conducted to explore the application of PSO-SVR in predicting the dynamic parameters of soft soil in Ningbo.The experimental results showed that confining pressure,dynamic load frequency,and consolidation ratio have an impact on the skeleton curve and dynamic parameters of soft soil in Ningbo.A PSO-SVR model was established based on the experimental data to predict the dynamic damping ratio,and the predicted RR~2 reached 0.863,indicating that the PSO-SVR model also has good performance in predicting the dynamic parameters of soft soil in Ningbo.This thesis has 51 maps,31 tables and 80 references. |
