Evaluation Of Seismic Liquefaction Potential And Liquefaction-induced Settlement Of Free Ground Based On Continuous Bayesian Networks

Seismic liquefaction is one of the main forms of earthquake-induced damages,it can cause serious problems such as foundation bearing capacity failure.Therefore,preidentification of seismic liquefaction and prediction of liquefaction-induced settlement will provide strong support for earthquake disaster prevention and reduction.On the basis of full summarization of avaliable research results,this thesis employed continuous Bayesian networks and numerical simulation methods to study the evaluation of seismic liquefaction potential and the prediction of liquefaction-induced settlement.Firstly,Conditional Linear Gaussian Bayesian Networks and MCMC approximate inference algorithms were adopted to establish models for evaluating seismic liquefaction potential based on shear wave velocity data,in which continuous variables were implemented directly without discretization.The newly proposed models included continuous Bayesian Network,hybrid Bayesian Network and MCMC hybrid Bayesian Network models.By comparing the above models with Discrete BN,LR and RBF-ANN models,it was shown that treating liquefaction factors as continuous variables directly in Bayesian Networks could effectively reduce the loss of information and improve the classification performance.On the basis of liquefaction potential evaluation,this thesis also used LIQCA to carry out a series of numerical simulations on the free-field seismic liquefaction and liquefactioninduced ground settlement.The effects of groundwater table,depth of the liquefiable soil layer,thickness of the liquefiable soil layer,density of sand,seismic waves and peak ground accelerations on liquefaction-induced settlements were investigated.Then a large number of numerical simulations based on various input parameters were used to generate a liquefaction-induced settlement dataset.Taking advantage of the generated liquefactioninduced settlement dataset,a continuous Bayesian Network model was established to further predict liquefaction-induced settlement.By comparing with Discrete BN and RBF-ANN models,it was found that the continuous Bayesian Network model made a breakthrough in continuous regression analysis and achieved a good performance in predicting liquefactioninduced settlements.