Inverse Analysis Method For Deep Excavation Considering Expirical Model Errors

In braced excavation,model parameters(e.g.soil parameters)are often represented by random variables because of the uncertainty of stratigraphy,measurement and so on.But the data about the model parameters is very limited and thus the distributions of model parameters are difficult to determine directly.Bayesian inverse analysis method provides an effective way to solve the problem.It can make use of the field observation data,which is easy to get in the process of excavation,to update the model parameters.There are two commonly used Bayesian methods for the inverse analysis of model parameters.The first is to use the multiple observation data in a sequential manner,namely sequential updating,but this method needs to make assumptions on the type of the posterior distribution,such as normal distribution.The second uses the multiple observation data in the Bayesian framework as a whole,namely one-step updating.The influence of the assumption in the first method needs to be examined.In the process of inverse analysis,empirical or semi-empirical model is often used to calculate the response because of the high efficiency.However,as any model is only an approximation or simplification of the real world,model uncertainty always exits.There exists the model bias factor between the observation and the response predicted by the model.In fact,the model bias factor is correlated with the model parameters(e.g.soil parameters),and the updating of model parameters cannot be independent of the model bias factor.It is necessary to update the model parameters and model bias factor at the same time.The problem how to update the model parameters and model bias factor simutaneously needs to be solved immediately.Considering the above problems,this thesis firstly concentrates on the comparative study of the two methods(i.e.sequential updating and one-step updating).And then the Bayesian inverse analysis method is proposed,which can update the model parameters and model bias factor at the same time using the observed maximum ground settlements.The implementation details are summarized as follows:(1)The existing empirical or semi-empirical models to calculate the response in braced excavation are compared.The advantage of KJHH is pointed out.(2)The two methods(i.e.sequential updating and one-step updating)are both applied to TNEC.By contrast,the influence of the assumption in the former method is revealed.(3)The Bayesian inverse analysis method is proposed,which can update the model parameters and model bias factor at the same time using the observed maximum ground settlements.And the practicability and applicability of the proposed method is illustrated through engineering examples.This thesis first briefly presents the background and purpose of this study.And then the basic theory of Bayesian inverse analysis method and calculation methods are introduced,and comparasion study of the the two methods(i.e.sequential updating and one-step updating)is carried out.Finally,the Bayesian inverse analysis method,which can update the model parameters and model bias factor simutaneously,is proposed.And the practicability and applicability of the proposed method is illustrated through engineering examples.