基坑开挖参数反分析贝叶斯方法比较研究

Comparative study of Bayesian methods for back analysis of deep excavation

根据基坑开挖过程中获取的现场监测数据可以反分析现场土层的土体参数.贝叶斯方法为充分利用现场监测数据分析土体参数提供了一条有效的途径.实现贝叶斯更新一般有两种方法:一是各开挖阶段的观测值依次对土体参数进行更新,即分步更新,更新过程中需假设后验分布类型,如正态分布;二是将多阶段的观测值作为整体对土体参数进行更新,即整体更新.为了研究贝叶斯更新方法对更新结果的影响,在先验分布为正态或均匀分布时分别采用这两种方法对台北国营企业中心开挖工程的土体参数进行更新.结果表明:当先验分布为正态分布时,假设后验分布为正态分布进行分步更新,得到的参数分布基本合理,能够准确预测开挖引起的最大沉降和失效概率;相反,当先验分布为均匀分布时,假设后验分布为正态分布进行分步更新,不能够准确预测开挖引起的最大沉降和失效概率,此时应采用整体更新方法.

In the process of excavation, the field ob used to update the soil parameters using Bayes’ th parameters using field observation data. multiple o sequential manner, or as a whole. It needs to mak servation data can be obtained constantly, which can be eorem. There are two common ways to update the soil bservation data are used in the Bayesian framework in a e assumptions on the type of posterior distribution （e. g. , normal） in the updating process when using multiple observation data in a sequential manner. In order to study the effect of the Bayesian updating methods, the two methods are, respectively, used to the Tai- pei National Enterprise Center case when the prior distribution is normal or uniform. It is found that, for normal prior distribution, the normal posterior distribution characterizes soil parameters reasonably, lead- ing to proper estimation of maximum ground settlement and failure probability. On the other hand, for u- niform prior distribution, assuming normal posterior distribution in Bayesian sequential updating may re- sult in overestimation of maximum ground settlement and failure probability. In. such a case, it is prudent to use all the observation data as a whole to update the soil parameters in one step.