Research On The Method Of Accelerating Optimization Inversion Of Initial Geostress Field In Tunnels And Its Application

Initial in-situ stress field is an important factor to evaluate the stability of surrounding rock and regional geological environment,which makes the study of initial in-situ stress field more and more valued.At present,in-situ stress measurement is the most reliable and direct method to obtain in-situ stress,However,the number of measured data is small and the data are relatively discrete,so the method of in-situ stress field inversion is mostly used to obtain the initial stress field.There are many existing methods of in-situ stress field,but each method has its own specific use conditions,or there are problems such as insufficient accuracy and low efficiency of inversion.Therefore,this paper takes the inversion of in-situ stress field in a tunnel site area in western Sichuan as the engineering application background,and proposes an accelerated optimization inversion method of in-situ stress field based on local proxy model combined with genetic algorithm(hereinafter referred to as LPM-GA inversion method).At the same time,the influence of few training samples(20 groups)and multiple training samples(50 groups)in this method is studied.Compared with the advanced inversion method of in-situ stress field based on RBF(Radial Basis Function)neural network(hereinafter referred to as RBF inversion method),it is concluded that the LPM-GA inversion method proposed in this paper has higher performance superiority,accuracy and time efficiency in practical engineering applications.Finally,LPM-GA inversion method is used to obtain the inversion results of in-situ stress field in the tunnel site area of relying on the project,and the development and distribution law of in-situ stress is analyzed.The main results and understandings obtained through the research are as follows:(1)Combined with the engineering geological conditions of the tunnel site area,a three-dimensional geological model of the tunnel site area is established by using CAD,Sufer,Ansys and other software,and the model is tested and debugged under the condition of only applying self-weight stress,and a more reasonable ground stress field inversion basic model is obtained.This model is used as the same standard model for subsequent ground stress field inversion calculation methods.(2)The range of tectonic stress boundary conditions is determined by Heim hypothesis and Jinnick hypothesis,and the combination of boundary conditions is used as training samples of neural network and local proxy model,and the stress boundary conditions are uniformly selected in a large enough distribution space.Avoid the blind construction of samples and have more reliability.(3)The RBF inversion method is constructed and applied to the inversion of the in-situ stress field in the tunnel site area of the project.The relative error between the inversion calculation value and the measured value is generally within 20%,the maximum relative error is 24%,and the magnitude difference is 3.54 MPa,indicating that the RBF inversion method of the in-situ stress field is reasonable and feasible.(4)A LPM-GA inversion method using RBF neural network as a local proxy model is proposed,and the boundary conditions of multiple groups of samples(50groups)and few groups of samples(20 groups)are used to invert the in-situ stress field in the tunnel site area.Calculation.The results show that the relative error between the calculated value and the measured value obtained by LPM-GA(50 groups of samples)inversion method is generally within 15%;The relative error between the calculated values and the measured values obtained by LPM-GA(20 groups of samples)inversion method is generally within 10%.This shows that LPM-GA inversion method is not only reasonable and feasible,but also has higher accuracy in this calculation.(5)Comprehensive evaluation of LPM-GA inversion method and RBF inversion method through fitting accuracy and time efficiency shows that the fitting accuracy of LPM-GA inversion method(20 groups of samples)is higher than that of LPM-GA inversion method(50 groups of samples)and RBF inversion method;In terms of time efficiency,the time consumption of LPM-GA inversion method(20 groups of samples)is 36000.897 s,which is less than that of LPM-GA inversion method(50 groups of samples)and RBF inversion method.On the whole,LPM-GA inversion method is better than RBF inversion method in accuracy and time efficiency,and taking a small number of samples also has higher fitting accuracy and time efficiency.(6)The better LPM-GA inversion method(20 groups of samples)is used to invert and calculate the in-situ stress field in the tunnel site area of the supporting project,and the development and distribution law of the in-situ stress in this area is obtained: there is local stress concentration at the bottom of the valley at the entrance and exit of the tunnel.There is a high ground stress,and the maximum principal stress values at the entrance and exit of the tunnel are 20 MPa and 22 MPa respectively;The stress near the fault zone increases locally,while the stress release in the fault zone decreases;The in-situ stress in the whole tunnel axis range is relatively high,and the greater the buried depth,the greater the principal stress value.The maximum principal stress at the maximum buried depth(1759 m)of the tunnel reaches 50 MPa.