| The tunnel engineering and underground engineering would be rapid development in the twenty-first century. In recent years, countries around the world continue to build the tunnel, which has promoted the largest development of the world's tunnel industry. With the rapid development of China's national economy and mining technological advances, domestic tunnel construction has been rapid development, mainly at increasing the length and the increasing depth of the tunnel. In the tunnel engineering, landslides, karst collapse, gushing water, pouring water, cave necking, mountain deformation, cracking pit, mud debris flow and rock burst is a common problem of geological disasters. Even though the conditions of these problems happened vary, the harm caused by tunnel construction is similar.Baihe tunnel is one of the long tunnels in the Zhurong Expressway. By many times the role of structural geology, the tunnel area have a larger change in geological conditions, and during the construction process of tunnel, there were landslides, swap block, deformation, seepage and so on. For the requirement of Baihe tunnel geological prediction, thesis focused on studying stability of Baihe tunnel surrounding rock.Tunnel surrounding rock stability has its own law of development and a complex nonlinear process. The stability analysis of tunnels is a complex systems engineering problem. Because the tunnel surrounding rock has formed in different geological environments and many times crustal movement, and coupled with the stress and groundwater and other geological role of environmental factors, the tunnel surrounding rock formation and the physical and mechanical properties shown macro, micro, discontinuity and high non-linear features. It can be said that the rock is an uncertain system. Therefore, stability analysis of tunnel surrounding rock can be seen as uncertain, nonlinear dynamics and complex system. At the application, a single study ways and method are not suitable for the complex stability analysis of the tunnel, and combination and comprehensive comparison of many kinds of theories and methods will be the correct analysis and effective way to solve the problem.In the paper, nonlinear analysis methods of highway tunnel surrounding rock stability based on the wavelet theory were first proposed. Using wavelet thresholding and wavelet modulus maxima method, the tunnel surrounding rock stress measurement data is pre-treated and monitoring data in the outliers is analyzed. The wavelet denoising method is improved. The wavelet neural network analysis model for stability of surrounding rock of tunnel is proposed and surrounding rock stability is analyzed. The wavelet neural network analysis model is improved by using rough set theory. In the paper, the calculation of fractal dimension by using wavelet analysis is first proposed on the cross-section outline, and the relationship between the stability of tunnel surrounding rock and the fractal dimension of cross-section outline is analyzed.Through in-depth and systematic researching, the main results and conclusions can be obtained:(1) At the basis of a wide range of regional geological and meteorological and hydrological information, geological overview and physical geography is summed up in the tunnel area. Through different methods to investigate the detailed analysis of the tunnel engineering geology environmental conditions and a lot of reading, the status quo of tunnel surrounding rock stability studying at home and abroad has been an overall grasp.(2) By reading the literature, application of the tunnel stability research using artificial intelligence and non-linear theory and application using wavelet theory has been analysis. Thesis proposes to combine wavelet theory artificial intelligence and non-linear theory, and applied to tunnel stability analysis. The results show that combining wavelet theory artificial intelligence and non-linear theory is appropriate.(3) The tunnel monitoring data usually contain many random error, and monitoring data are often interfered by a variety of random and uncertainty factors. The tunnel monitoring data is treated by using wavelet analysis method, and through of data analysis and evaluation of the effect of de-noising method, the final selection of wavelet function is Db3 function, and the largest-scale is J=3 and threshold function is heursure function.(4) The two kind's original signal of points T100421 in the Baihe tunnel K127 +121 cross-section (observation of contact pressure and the value of stress changes) have been de-noising treated and analyzed. Through the high-frequency signal analysis of the contact pressure of tunnel surrounding rock and the construction of the tunnel, the tunnel construction and rainfall are the main reasons for emergence of noise in monitoring data. Through de-noising analysis of the changes in the value of the day stress, it is considered that the stress change value of day is swinging around in a random way, which is part of the impact of the observational error. At the same time, in fact, the tendency of stress change value has slowly decreased. Through extracting the stress of tunnel surrounding rock and stress change value, it is considered the trend is stable, and it means that the tunnel is stable after the initial support.(5) In this paper, wavelet de-noising algorithm is improved by EMD. The curves using EMD-wavelet de-noising are smooth than only using wavelet de-nosing, and the tendency of tunnel surrounding rock is suitable of the development of reality. The de-noising effect is better than the effect of directly using Db3 function, heursure rules and 3-scale, so the favorable method is proposed for determine the stability of tunnel surrounding rock.(6) In this paper, the surrounding rock contact pressure values of T100421 point in Baihe tunnel K127 +121 cross-sections were trained and predicted by using wavelet neural network and BP neural network, and the results of two methods were compared: in training speed, the wavelet neural network convergence speed is quicker than the BP neural network convergence speed, and the wavelet neural network training steps is fewer than the BP neural network training steps; at precision , whether it is training accuracy or prediction accuracy, wavelet neural network is better than the BP neural network. Wavelet neural network training convergence speed is fast, but in the process of approximation, the error oscillation phenomenon is emerged. In general, the wavelet neural network is better than BP neural network in the convergence and generalization performance.(7) The Baihe tunnel rock mass quality evaluation of six indicators is proposed, and its attributes are reduced by using rough set theory.(8) The four types of tunnel surrounding classification neural network recognition model is built and application analysis for the rough set reduction data by using wavelet neural network and BP neural network. The training sample set of neural network is the sample set of rough set reduced, and the neural network structure is effectively simplified, and the training steps and the training time is decreased, and the network learning speed and accuracy is improved. The accuracy using rough set WNN is highest, and the recognition results are near expert discriminant system. For non-reduction WNN model, the discriminnant accuracy is worse than reduced WNN except the 1st sample, and is large difference with the expert discriminant system value. The results using BP neural network is largest difference with the expert discriminant system value. In general, the prediction accuracy using the wavelet neural networks is higher than the prediction accuracy using BP neural network, and the prediction accuracy using rough set is higher.(9)The ability of wavelet network function approximation has been strictly proved, but it encountered difficult in actual implication. Prediction decides that the effect of the prediction accuracy can not be 100% and there is no error. If there is no way at the optimal prediction instrument, only second best, wavelet network is one of sub-optimal prediction instrument.(10) The research of the rock cavern cross-section outline is major in the theoretical model, and is less in the actual site monitoring of the cross-section outline. In this paper, it is found that Baihe tunnel cross-section curves have statistical self-similarity and better fractal characteristics after reading a substantial literature and studying many Baihe tunnel sections data.(11) In this paper, a statistically self-similar fractal signal was to be calculated by wavelet analysis which is successfully applied to estimate and calculate the tunnel cross-section contour fractal dimension, and the fractal dimension of the cross-section outline was calculated. With reduction of the surrounding rock designed classification, the overall fractal dimension of cross-section curve was increased. Sidewall curve fractal dimension should be slightly larger than the overall contour curves, and it is not obvious inⅡ,Ⅲtype of rock and obvious inⅣtype rock. There is primary linear relationship between cross-section profile fractal dimension and separation angle, and the effect is more pronounced for different separation angle. There is not a linear relationship between sectional profile fractal dimension and joint dip angle. With increasing of dip angle, the fractal dimension was first increased and then decreased, and large at 30-60°. With increasing of the ultra-digging percentage, the fractal dimension of cross-section outline was increased, and there is the basic linear relationship between the fractal dimension and the ultra-digging percentage. But because of influence of the individual points by the impact of unusual geological conditions, the correlation coefficient is little. There was a linear relationship between fractal dimension an RMR and Q rock mass classification. With increasing of the RMR and Q values, the fractal dimension of cross-section outline will decrease. Multiple correlations among sectional profile fractal dimension, quality evaluation of tunnel surrounding rock and the super-dug percentage are researched.
|
PDF Full Text Request