Research And Engineering Application Of Uncertain System In Geotechnical Engineering

As the national economy sustained and rapid development, the government continually increaseinvestment in infrastructure in China, and at the same time many problems in geotechnical projectswill also come forth. As we know, tunnels are usually built in the highly complex geologicalmaterials,which shows great randomness, vagueness and uncertainty. Due to the particularity ofnatural rock, it was cut into half continuous by various kinds of geological structure such as joints,fault, etc. Usually rock has an order from granular to continuum. It is a complicated problem whichwas coupled by multiphysics and heterogeneous, and also was influenced by the complex externalenvironment. For those reasons, it was more difficult to acquire the exact mechanics parameters andconstitutive model.Considering the complex dynamic system which was composed by the rock, soil and lining isnonlinear. It is difficult for us to find out the evolution rules by component a mathematical model ora physical model, if we are lacking in understanding about the system. But we can do it by analyzingthe monitoring time series which reflect the system's state. Data mining is an appropriate tool inanalyzing time series. By using it we can pick up the potential useful information and knowledgewhich contain characteristics of the system from lots of incomplete, noisy, random data. Theinformation and knowledge we picked up could be feedback to the fields of application to guidancethe corresponding jobs.In this thesis the experience mode decomposition technique and Hilbert-Huang transformtheory were used to analyze the monitoring data. In order to separate the time series containingfeature information, the monitoring data is noise reduced and decomposed, and also carried finiteelement analysis and regression analysis. The inherent law of the rock-mass and soil-body'sdeformation and response of construction are due to the actual situation. The needs of the projectcould be better met.In this thesis the data mining technology is introduced to the monitoring data analysis. In orderto comprehensively and systematically analyzing problems, it was necessary to take large amountsof variables. By using multivariable time series analysis, we can pick up most of information fromonly a small number, irrelevant new variables. The thesis provides a representation method of thepattern of multivariate time series based on the analysis of the principal component on the basis ofthe analysis of the features of multivariate time series and its application. We studied the data miningtasks in time series, such as clustering, similarity search, distance measures, classification, discordsdetect, etc. and we also illustrated that the pretreatment and clustering of the multivariate time series.According to the statistical and numerical analysis of lots of monitoring data, the intelligent back analysis model was built combined with dynamic displacement analysis method in finite elementtheory.In order to improve the utilization of the monitoring data, expert experience and other relatedtechnical, and make timely decisions based on the development of the surrounding rock's stress anddisplacement, we combined neural network technology with traditional expert system. By using thedistributed connecting mechanism of neural network technology in knowledge representing, weconstructed the neural network expert system. Existing data or observations are used in reasoningapplication and provided as a scientific basis for the practical analysis, forecasting anddecision-making.