Studies On Nonlinear Prediction And Risk Control Of Ground Surface Settlement In Subway Shield Construction

In the shield construction,the factors affecting the surface settlement deformation are often more,and too many parameters will increase the difficulty of the traditional mechanical prediction model.In addition,the current risk assessment methods for surface settlement of shield construction are based on static data analysis,and rarely combine their risk assessment with construction dynamics.Therefore,this paper selects the least squares support vector regression machine(LS-VSM)which is robust to environmental changes to improve the surface settlement prediction model.For the different risk source status of shield construction,this paper uses Bayesian network to establish The dynamic evaluation model of surface settlement risk of shield construction is carried out.The construction of the tunnel section of Yuanhua exit section of Chengdu Metro Line 8 is the background of the project,and the risk of surface settlement caused by subway shield construction is studied.The research content and results of this paper mainly include:(1)Based on the analysis of the factors affecting the deformation of the ground surface settlement and the large number of data samples based on the monitoring and construction statistics of the shield construction site in the north line of the Yuanhua exit section of Chengdu Metro Line 8.In order to measure the influence of different kernel function types on model prediction accuracy and generalization ability,the Gaussian kernel function,linear kernel function and polynomial kernel function in LS-VSM are modeled and analyzed.From the perspective of kernel function and parameter,this paper establishes a nonlinear prediction model based on the least squares support vector machine(LS-VSM)regression multi-dimensional variable input and the surface center settlement variable output.The results show that the Gaussian kernel function has strong statistical learning ability and higher generalization ability,and it is suitable for nonlinear prediction regression of maximum settlement in the surface center;The prediction accuracy of the ground settlement prediction model is the highest when the tunnel geometric parameters,the structural parameters of the formation and the technical parameters of the tunneling are comprehensively considered.(2)In order to compare the accuracy of least squares support vector machine(LSVSM)regression and BP neural network in dealing with small sample and multi-factor complex nonlinear prediction problems,a BP neural network is established to establish the surface subsidence prediction model.The results show that the LS-VSM regression model has a better ability to estimate the settlement of the surface center of the shield construction.This also verifies that the LS-VSM method is more accurate than the BP neural network in dealing with small sample and multi-factor complex nonlinear prediction problems.(3)This paper conducted incomplete statistics on the shield construction accidents that occurred in 2004-2018.The key risk factors of subway shield construction settlement are identified and extracted.this paper proposes a Bayesian network-based dynamic risk rating method for surface settlement of shield construction,which is analyzed by using GeNIe software for posterior probability and sensitivity.The model is applied to the project of Yuanhua's entry and exit section of the Chengdu Metro Line 8 project.The model is used to track the surface settlement risk during the tunneling process in real time,which greatly reduces the collapse accident caused by excessive surface subsidence during the tunneling process.