Research On Mechanism And Comprehensive Prediction Of Rockburst Regarding Deep Buried Tunnel In Tectonic Active Area

As an important part of China’s “country with strong transportation network” strategic planning,the CZ Railway project has great value in consolidating border security,promoting regional coordinative development,and breaking through world-class engineering construction technical problems.Due to the characteristics of large burial depth,high ground stress,and hard surrounding rock,many ultra-long tunnels are extremely vulnerable to rockburst disasters.Accurately predicting the location of rockbursts and estimating the level of rockbursts are of great significance for alleviating the adverse effects of rock bursts and ensuring the safe construction of tunnels.This paper takes a deep buried tunnel in the Gangdese orogenic belt as the engineering case,and conducts systematic research on rockburst prediction in ultra-long tunnels with large burial depth in tectonic active area from some aspects such as geostress,evolution characteristics of rockburst and deformation simulation,rockburst evaluation indicators,comprehensive prediction models,etc.The main research contents and results are as follows:(1)Based on the measured stress data,tectonic history and features of stratum,the stress characteristics of the tunnel site were preliminarily analyzed.Then a three-dimensional numerical model that can reflect the real terrain features was established.Three stress inversion models were constructed by combining numerical simulation with boundary condition regression,which are boundary load + least squares regression,uniform boundary load + support vector machine(SVM)regression,variable boundary load + SVM regression respectively.After model mechanism analysis and comparison with measured stress data,it was found that the inversion results of variable boundary load + SVM model were more consistent with reality.In this model,trapezoidal stress boundary as variable load is more in line with the distribution law of measured stress,while SVM regression ensures the nonlinear mapping ability.(2)Through monotonic loading,increasing amplitude cycle loading and unloading tests and acoustic emission monitoring,the physical characteristics and crack characteristics of rockburst evolution were identified,and the rockburst mechanism was revealed.In addition,the deformation characteristics of rock under uniaxial cyclic loading and unloading(Massing effect,Ratchet effect,Felicity effect and nonlinear characteristics)were studied.Then a constitutive model composed of elastic element,plastic element and friction element was proposed to describe the stress-strain relationship regarding cycle behavior.Finally,the neural network was used to calibrate model parameters based on the experimental data.(3)The stress-strain curve and acoustic emission characteristics of rocks obtained from tunnel site under cyclic loading and unloading were analyzed to reveal energy distribution and evolution during deformation process.Then an explicit expression of elastic energy evolution was established.Based on this expression,a rockburst hazard evaluation index that can simultaneously reflect hazard degree and occurrence possibility was proposed.Moreover,a minimum loss function method was used to determine threshold values for different rockburst grades,this criterion was applied to rockburst estimation of multiple deep-buried tunnels in Gangdise orogenic belt.(4)A comprehensive model for predicting rockburst grade considering multiple factors was established by using machine learning method.In pre-processing stage,a combination method incorporating feature selection(FS),t-distribution stochastic neighbor embedding(tSNE)and Gaussian mixture model clustering(GMM)was proposed to relabel data set,its effect was compared with that of common statistical methods according to operation principles.In addition,9 machine learning models for rockburst prediction were trained and their prediction performance was evaluated and compared.The tree-structure Parzen estimator(TPE)was applied to search for optimal hyperparameters of model and compared with 3 common optimization algorithms.The optimal combination model was determined through performance evaluation and applied to actual rockburst prediction.Finally,the importance of various feature factors was analyzed based on marginal contribution rate.