The Theoretical Research And Application Of Coupling Of Support Vector Machine And Spectral Representation Method

With the acceleration of urbanization and motorization,the problem of traffic congestion has become a problem in the development of major cities in China.An increasing number of tunnels have already been completed and are still under construction because urban space is become more unavailable and more costly.Stability analysis is crucial in tunnel engineering because uncertain soil parameters play an important role in engineering,such that their effects must be correctly assessed to meet the performance,robustness and safety requirements.Spatial variability is an essential characteristic of soil properties,particularly in performing reliability analysis of tunnels,as these properties exhibit marked differences depending on the soil surrounding.However,constraints brought about by the intrinsic property of soil and the engineering conditions hinder the accurate assessment of the uncertain soil parameters for small samples.In this thesis,a new approach that combines spectral representation method(SRM)and support vector machine(SVM)is proposed to accurately describe uncertain soil parameters.The proposed method can facilitate the accurate simulation of auto-correlation and cross-correlation for a small sample size than the conventional SRM.The proposed method was used to analyze the reliability of Wuhan Metro Line 7.The contents and conclusions of the thesis are including:1.The SVM reliability analysis method is proposed on fast Markov chain simulation,so that training samples can adapt to simulation the limits of the samples of the importance region.Simulation of the Failure Probability of Implicit Limit State Function based on alternative samples of the Markov chain consists of training samples and construct an explicit expression of the original limit state function by using SVM.Numerical and engineering examples illustrate that the proposed method owns good performance in calculating efficiency and precision.2.The covariance matrix for variables associated with different algorithms can differ considerably.Example analysis indicated that the most fitting algorithm among three algorithms(particle swarm optimization,grid search algorithm,and genetic algorithm)for parameter optimization is the genetic algorithm.3.This thesis proposes a new approach that combines the advantages of SRM and SVM for random field simulation.The proposed method can facilitate the accurate simulation of auto-correlation and cross-correlation for a small sample size.Numerical cases demonstrate that the proposed approach improves the efficiency and precision of random field discretization and reliability analysis,thereby making it feasible to be applied to highly complex engineering problems.4.Taking Wuhan Metro Line 7 as an example and modeled by ANSYS,this thesis calculates the failure probability of tunnel lining by the proposed method and the traditional methods respectively and compares the results.The results indicating that spatial variability considerably influences the stability of tunnel.