Research On Settlement Response And Prediction Of Double Line Shield Tunneling In Xuzhou Composite Stratum Based On Machine Learning Method

The rapid development of China’s economy has driven the high-speed growth of transportation infrastructure,and the demand for subway construction in major cities has been continuously increasing.The majority of subway construction adopts the shield construction method,which has led to growing concerns about subway construction safety.Among all the safety indicators for shield construction,the most important is the ground settlement caused by shield construction.Therefore,the prediction of ground settlement during tunnel construction has significant research implications.This article takes the shield tunneling project of a section of Xuzhou Subway Line 6 as the basis,and through field visits,on-site measurements,numerical simulations,and big data modeling research methods,it analyzes the settlement deformation laws and development process of the twin tunnels caused by shield construction.Combining massive shield monitoring data for big data modeling,more accurate prediction results have been obtained,effectively solving the problem of insufficient dynamic prediction of ground settlement caused by shield construction in the past.This provides engineers with a new perspective and means for predicting ground settlement during shield construction.The specific research content and conclusions are as follows:(1)Building on previous research,this study systematically summarizes the disturbance process of shield construction on soil and the resulting time-space development process of settlement.A three-dimensional finite element model of the twin-tunnel construction is established based on the background project,which systematically demonstrates the development pattern of settlement in twin-tunnel construction.It is found that the settlement development of the twin tunnels has obvious stages:(Taking the left tunnel as the leading line for illustration)the excavation of the left tunnel causes the right tunnel to experience an advanced settlement before excavation,accounting for about 30% of the total settlement;the excavation of the right tunnel causes the stable settlement of the left tunnel to redevelop and increase by about30%.During the excavation of the twin tunnels,the maximum settlement point of the tunnel gradually shifts from the initial left tunnel axis to the middle of the twin-tunnel axis.The simulation results show a small difference compared to the actual measurements,indicating that the finite element simulation method can accurately predict ground settlement during tunnel construction.(2)A sensitivity analysis of geological parameters,tunnel depth,and some shield tunneling parameters affecting tunnel settlement is conducted using a finite element model.The study finds that when the tunnel depth is less than a specific value,an increase in tunnel depth leads to an increase in ground settlement,and the settlement trough shape transforms from a "W" shape to a "U" shape.When the tunnel depth exceeds the specific value,an increase in tunnel depth results in a slight decrease in ground settlement.For parameters such as soil elastic modulus,Poisson’s ratio,cohesion,internal friction angle,and slurry elastic modulus,an increase in these parameters leads to a reduction in ground settlement.However,the reduction magnitude gradually weakens,and the change exhibits non-linear characteristics.This indicates that tunnel settlement is the result of complex multi-factor coupling,and single factors experience a transition from "significant influencing factors" to "non-significant influencing factors".(3)To address the limitations of the finite element prediction method,a big data modeling prediction method is proposed,and the essential elements for big data modeling are extracted,introducing a complete data processing workflow.The relationships and correlations between the parameters are analyzed.The results show that the variations in shield tunneling parameters are mainly influenced by geological conditions,tunnel depth,and human operations.(4)Based on the shield tunneling parameter database,three different big data machine learning models with distinct model structures and prediction principles are employed—Neural Networks(BPNN),Support Vector Machines(SVR),and Random Forests(RF).Grid search and cross-validation methods are used to compare and optimize hyperparameters for prediction tasks.The Root Mean Square Error(RMSE)is used to determine the optimal prediction model.The study finds that,apart from SVR,the other two prediction models achieve relatively good prediction results for ground settlement.The ranking of model performance is as follows: RF > BP > SVR.Research has found that both prediction models,except for SVR,have achieved good prediction results for surface subsidence,effectively compensating for the shortcomings of previous prediction methods for continuous surface subsidence prediction.