Shield Tunneling Induced Deformation Analysis Based On Dynamic Bayesian Network

Construction of shield tunnel has stepped into its peak time these years, and theoccurrences of ground surface subsidence also become more frequent, leading to greatdamages to personal and property safety. How to systematically analyze the variousfactors that impacting ground surface subsidence according to the characteristics ofconstruction site of shield tunnel, and then predict the level of ground surface subsidenceon the basis of data changes monitored on-site, is an important research topic on the earlywarning of shield tunnel construction. By introducing the Bayesian Network Theory,which has attracted many attentions in the field of risk management, this thesis hasanalyzed and identified the impact factors of ground surface subsidence in shield tunnelconstruction, then constructed a dynamic Bayesian Network model to predict anddiagnose the level of ground surface subsidence, thus it can provide guidance to thesafety management of ground surface subsidence induced by shield tunnel construction.First, we present the basic concepts of Bayesian Network Theory and then extend todynamic Bayesian Network Theory, after the introduction of reasoning and learningmethods in dynamic Bayesian network, the learning methods and software tools appliedin this paper have been summarized. Second, mutation characteristics of shield tunnelconstruction have been analyzed from the two dimensions of time and space, and theimpact factors of ground surface subsidence induced by shield tunnel construction havealso been analyzed from the following three aspects: engineering geological condition,engineering design conditions and construction control parameters, finally ten impactfactors have been identified for this research. Third, we conduct an applicability analysisaccording to the characteristics of ground surface subsidence in shield tunnelconstruction and the applications of dynamic Bayesian network, thus determine theprocesses, structure learning method and parameter learning method as well asinference method used to construct the dynamic Bayesian network. Forth, we takeriver-crossing metro shield tunnel as example, construct a dynamic Bayesian networkmodel of ground surface subsidence induced by shield tunnel construction, use theforward inferences to analyze it, the results have proved the validity of modelpredications. The most likely impact factors leading to ground surface subsidence havebeen identified by backward inferences, and the impact factors most sensitive to groundsurface subsidence have also been identified by sensitivity analysis.It will help us toestimate the level of ground surface subsidence and further define the level of accidentearly warning according to the on-site situation changes by using the dynamic Bayesiannetwork model constructed in this thesis. The model can reason out the most likelyreason for the ground surface subsidence according to the level of ground surfacesubsidence by backward inference. Therefore, this thesis has made some supplements for the research of ground surface subsidence risks in tunnel construction.