Static And Dynamic Bayesian Network Prediction Models For Liquefaction-Induced Uplift Of Underground Structures

China is one of the earthquake prone countries.Underground structural engineering has been widely used in urban construction,but liquefaction uplift damage can occur when subjected to strong earthquakes.Therefore,it is important to construct a model that can accurately predict the liquefaction uplift displacement of underground structures,which can provide a basis for disaster prevention and mitigation in underground engineering.Based on this,this paper investigates the important factors of the liquefaction uplift response of subsurface structures using both numerical simulation and machine learning methods,and then constructs an uplift prediction model based on static and dynamic Bayesian networks.The main studies in this paper are as follows:(1)A numerical method was used to simulate the seismic liquefaction uplift response of subway stations under 74 sets of working conditions,and then used sensitivity analysis methods to quantify the degree of influence of each factor.In addition,the most suitable ground vibration intensity index for evaluating the uplift displacement was selected from 32 candidate indexes using correlation,feasibility,and validity analysis methods.The key factors obtained for the prediction of seismic liquefaction uplift of underground structures were relative compactness of sandy soil,Arias strength,burial depth of the structure,thickness of liquefiable soil layer,self-weight of the structure and groundwater level.(2)The 6 key factors screened for seismic liquefaction uplift of subsurface structures are cross-transformed and combined to obtain 465 sets of working conditions for numerical simulation to obtain a static database of seismic liquefaction uplift of subsurface structures.The static model for the prediction of seismic liquefaction uplift of subsurface structures is further obtained by using the conditional linear Gaussian Bayesian network method.The comparison analysis of the obtained prediction model with the traditional discrete Bayesian network model,artificial neural network model,and support vector machine model shows that the static continuous Bayesian network has better reliability and accuracy in the problem about seismic liquefaction uplift prediction of subsurface structures.(3)Based on the static Bayesian network structure constructed by conditional linear Gaussian,combined with Markov chain method and 385 sets of dynamic uplift data as samples,a dynamic discrete Bayesian network prediction model of seismic liquefaction uplift response of subsurface structures is constructed.And the influence of time slice on the dynamic discrete Bayesian network model is studied,and it is found that the dynamic discrete Bayesian network with time slice of 0.5 s has the best prediction performance.Then the static and dynamic Bayesian network prediction models are compared,and the Bayesian network structure with the addition of time information can further improve the model performance in the case of the same network structure.In addition,the validity of the model was experimentally verified using centrifuge shaker data.