| China is a mountainous country and the topography is complicated, two-thirds of which is mountains and hills. Along with flourish of Chinese highway construction project, the construction scale of tunnels is larger and larger, and stability problems of wall rock also become more prominent in tunnel construction process.Tunnels are constructed in rock mass which has some certain stress history, then the nature of tunnels will be impacted by the nature of wall rock of tunnels. Wall rocks of tunnels are tunnel geological body, which have experienced many geological processes and destruction, formed certain rock composition and structure and existed in certain geological environment. Wall rocks is called rock mass as mechanical objects, while referred to a part of rock mass which is impacted by tunnel excavation. In highway tunnel construction process, due to the equilibrium of natural stress state of rock mass is destructed, unloading rebound and stress re-distribution of tunnel surrounding rock take place. When this kind of rebound stress and re-distribution stress surpass wall rock intensity, the engineering rock mass will be unstable and damage, which brings harms to the tunnel construction and management.In view of the fact that the stability of surrounding rock is much more important in highway tunnel design and construction. In this paper,combining with the geological prospecting material, rock mechanics tests and the field monitoring gauging experiment. Then the paper comprehensive evaluated the stability of wall rock of tunnels from different aspects, obtains some valuable highway-tunnel stability appraisal theory to tunnel's design, construction and operation management. The theory has great practical and theoretical significance to successful construction and safe operation of tunnel.In the paper, taking Wuzhuling mountain range tunnel in Zhejiang Province as an example, based on detailed investigation to geological conditions of the tunnel. Carry out systematic research to tunnel's wall-rock stability using nonlinear theory as the basic idea when consider aspects of wall-rock stress, surface subsidence, rock classification and landslides. In the paper, three methods are proposed for the first time, the 1st is non-linear wall-rock stress, surface subsidence prediction method, which combined wavelet noise reduction with time series phase space reconstruction and GA-BP neural network; and the 2nd is nonlinear discriminating method, which combined GA-BP neural network with MIBARK arithmetic; the 3th is tunnel collapse forecast method combining"the uncertainty model"with "the deterministic model"which based on the unify of three-dimensional fracture network and three- dimensional discrete element numerical simulation.Paper used the date of the rock-mass pressure and ground surface settlement, based on the wavelet de-noising theory to remove the noise monitoring data information, to exclude information on the time series of noise prediction model of interference, improve accuracy of model prediction. And by the time series phase space reconstruction. Reconstruction of time series techniques, excavation of the important system information, and to set up GA-BP neural network sample set, the training of genetic algorithm to optimize neural network surrounding rock stress, surface subsidence prediction model, and prediction model was applied to Wuzhuling tunnel surrounding rock stress, monitoring of surface subsidence prediction, prediction with previous methods, the training iteration number less accurate prediction results. Paper in order to rock pressure, surface subsidence monitoring data based on wavelet de-noising theory to remove the noise monitoring data information, to exclude information on the time series of noise prediction model of interference, improve accuracy of model prediction. And by the time series phase space reconstruction reconstruction of time series techniques, excavation of the important system information, and to set up GA-BP neural network sample set, the training of genetic algorithm to optimize neural network surrounding rock stress, surface subsidence prediction model, and prediction model was applied to Wuzhuling tunnel surrounding rock stress, monitoring of surface subsidence prediction, prediction with previous methods, the training iteration number less accurate prediction results.Through a comparative study show that the rock classification are many factors, is a highly non-linear and non-deterministic problem. Such questions must be non-linear theory to study ways to get the most objective judge the results. Paper put forward the theory of nonlinear surrounding rock classification research new ideas. Since non-linear theory has a strong data mining and knowledge discovery capabilities, be able to set up a good mapping ability of the recognition model, be able to study a high degree of approximation of the problem. Through the Tunnel classification factors research, mutual information-based knowledge of the relative reduction algorithm on the factors affecting rock classification attribute reduction, the impact of rock classification drawn the smallest relative reduction to property after Reduction Set as the GA-BP model of rock classification and identification of input, the trained model was applied to WuZhuling tunnel rock classification, training iterations number small, the high rate of recognition accuracy.Tunnel as a jointed rock mass, the deformation and damage process is a nonlinear process, the study of tunnel landslide discrete element numerical simulation techniques are very appropriate. Based on joints investigate of Wuzhuling tunnel, using three-dimensional fracture network technology to get advantage jointed group, joint width and the average distance between joints, with three-dimensional discrete element numerical simulation technology to build the tunnel cave-prediction of numerical simulation model, successfully predict landslide of Wuzhuling tunnel. Systematic study of stability of tunnel surrounding rock was more objectivity, more suitable for rock materials using such nonlinear stability analysis.By the systematic non-linear analysis of tunnel rock-mass stability, some conclusions can be obtained:1. By monitoring the time-varying of the rock-mass pressure and ground surface settlement and analyzing the characteristic and principle of changing of the two parameters with time, the best time to support can be determined and the stability of the rock-mass can be judged. But, generally, there may be some"noisy"in the monitoring data of rock-mass pressure and the time series of ground surface settlement. And to delete the noise, the noise reducing of the monitored data based on wavelet technology is effective, after which more accurate character information can be provide to monitored data analysis and prediction model.2. To evaluate the effect of noise reducing on signal field, the indexes of Signal-Noise Ratio (SNR) and Signal-Noise Ratio Gain are usually used in the area of signature. The both indexes require knowing the pure signal no-polluted by noise, which can't be determined in engineering monitor. From the phase of smoothness and similarity, using the smoothness and root-mean-square error to evaluate the effect of noise reducing of the monitoring data, the optimal result of noise reducing can be found effectively. Thus the process of evaluation is more scientific and objective.3. It is confined to understand the intrinsic mechanism of temporal series for its complexity as well as some essential uncertainties existed possibly. Taking the data mining of characteristic and principle of time series as the aim, phase-space reconstruction techniques will be needed. In this paper, the phase-space reconstruction of the monitored data that is noise reduced is adopted applying the methods of mutual information and Cao which is used to determine the time delay and embedding dimension in Phase-space reconstruction techniques. The data set reconstructed act as the sample set of the GA-BP predicting model, which is used to train the network. The predicting model gained through this way above is more capable to include the characteristic of time-varying of monitored data and has a higher precision.4. BP neural network can approximate any continuous function at any precision. While some problems such as local minimum and slow rate of convergence still exist. Genetic algorithm takes the encoding of decision variable as operand, objective function value as the searching information directly; uses searching information from many search points simultaneously; thus, can approximate the optimal solution at global scope. Using the program of neural network optimized by genetic algorithm by Matlab, the weight and threshold of the BP neural network are optimized. In regard to the network training, the problem of trapping in local minimum easily is avoided, and the rate of convergence is faster. Meanwhile, the purpose of optimizing the network is reached, and the ability of solving the predicted problems is increased.5. This paper presents a method of constructing neural network sample set using wavelet de-noising and phase-space reconstruction technology for the first time, optimize the weight values and the threshold values of genetic algorithm, construct the surrounding rock stress and the nonlinear model of ground settlement, the results showed that this method can reject the noise information such as artificial factors of monitoring data, reduce the effect of artificial division time series and artificial adjusting parameters workload, improve learning speed of neural networks. The prediction accuracy is also considerable.6. This paper based on the analysis of influencing factors of surrounding rock classification, using MIBARK algorithm to reduce the influence factors of surrounding rock classification, the results showed that after the reductions,minimum knowledge reduction can fully reflect the main influencing factors of the WuZhuling tunnel rock classification, provide a most simple and optimum evaluation index set for the subsequent nonlinear identification.7. MIBARK algorithm is developed at the basis of rough Set Theory. Its reduction result is simplest and unique. The simplest knowledge reduction results from the MIBARK algorithm contains mutual information between the condition attribute and decision attribute and has the same decision-making ability with the original condition attribute. Construct sample set based on the simplest evaluation index using the MIBARK algorithm, establish rock classification model using genetic algorithm optimized neural network model. This model has low inputting sample set dimension, low training process iterative times, high recognition accuracy. 8. This paper first proposed using MIBARK algorithm-GA-BP neural network rock nonlinear identify model, the attribute reduction result does not need artificial filtrate, BP network weights and network threshold evolutionary generated, explored a new means to surrounding rockmass classification by nonlinear method .9. The paper analyzed jointed field survey data of Wuzhuling tunnel ,get width and distance information of dominant joint set group, constructed "uncertainty model" of fracture network, and combining"certainty model"of joints information near the working face, using three-dimensional discrete element numerical simulation technique, set up the tunnel cave-prediction numerical simulation model. Accurately determined the stability of the not opened segment wall rock at the direction of the working face tunneling, determined the location and scale of the block destruction happened, This method made up the shortage of the traditional finite element analysis can't analysis block control damage. Provide a study idea for tunnel landslide.
|
PDF Full Text Request