Probabilistic Back Analysis Of Geotechnical Parameters For Braced Excavation Based On Multi-source Observations

With the rapid development of urban rail transit and underground space,the scale and complexity of foundation pit engineering are increasing.It is very important to improve predicted reliability of deformation response of foundation pit excavation and ensure the safety of surrounding infrastructure construction and operation.Due to various uncertain factors in geotechnical environment and limited investigation data,the deformation prediction of foundation pit and the actual response have deviation,and the predicted value is usually inconsistent with the observed value.With the development of monitoring technology,the multi-field characteristic information of foundation pit can be sensed and monitored by different types of sensors,and the multi-source monitoring data can be obtained to depict the deformation response of foundation pit.The probabilistic back analysis method based on Bayes theory can reasonably quantify all kinds of geotechnical related uncertainties,integrate multi-source monitoring data and prior information,update geotechnical parameters in real time and reduce their uncertainties,and improve the reliability of foundation pit deformation response prediction.However,update strategies composed of different data sources have a great impact on the results of probabilistic back analysis,and the mechanism of single-source and multi-source monitoring data on probabilistic back analysis is still unclear.In the current research,only one or two types of monitoring data are usually used in probabilistic back analysis,and the key information of Bayesian fusion multi-source monitoring data is not fully mined and extracted,which leads to difficulty in selecting update strategy,inadequate calibration of geotechnical parameters and unreliable prediction of foundation pit deformation response.In view of the above research deficiencies,this thesis firstly proposes a probabilistic back analysis method based on single source monitoring data,which can effectively consider the uncertainty of parameter,model bias and observation error,and make full use of the monitoring information contained in multi-stage observation.Then,a probabilistic back analysis method based on multi-source monitoring data is proposed,and the influence and mechanism of updating strategies composed of different data types on the behavior response prediction of geotechnical are explored.Finally,aiming at the difficult problem of updating strategy selection in probabilistic back analysis,a probabilistic back analysis method based on multi-source monitoring data selecting updating strategy is proposed,and multi-objective optimization method is used to determine the optimal updating strategy,which effectively improves the reliability of geotechnical behavior response prediction.The main research contents and achievements of this thesis are summarized as follows:(1)A probabilistic back analysis method which can consider various uncertainties is constructedBased on Bayesian theory,a probabilistic back analysis method is proposed which can consider the uncertainty of geotechnical parameters,model bias and observation error at the same time,and the monitoring information contained in the multi-stage observation is integrated into the likelihood function.With the increase of observed data during construction or operation,the statistical characteristics of uncertain variables(key geotechnical parameters and model deviations)are gradually updated,so that the performance prediction of geotechnical system can be timely adjusted during construction or operation,and the reliability of prediction can be improved.The three significant features of this method include: 1)multiple observations are incorporated into the Bayesian updating,2)the statistical information of the uncertain variables is updated in a stage-by-stage manner,and 3)the posterior distributions of uncertain variables are derived with Markov Chain Monte Carlo(MCMC)simulation that is based on the Hamiltonian Monte Carlo(HMC)algorithm.A case study of Formosa excavation in Taiwan Province was carried out to verify the effectiveness of the proposed approach and compare the advantages and disadvantages of the proposed method with the existing probabilistic back analysis method.(2)A probability back analysis method integrating multi-source monitoring data is constructedBased on the probabilistic back analysis method using single source monitoring data,a probabilistic back analysis framework based on Bayesian theory is proposed to integrate multi-source monitoring data,and a likelihood function which can effectively accommodate multi-source monitoring data is constructed.Taking the foundation pit project of Taiwan Enterprise Center(TNEC)as an example,three different updating strategies are constructed based on the multi-source observation information.They are respectively single-source updating strategy using only maximum ground settlement data(S1),single-source updating strategy using only maximum retaining wall deflection data(S2),and multi-source updating strategy using both maximum ground settlement and maximum wall deflection data(S3).We explore the influence and mechanism of updating strategies composed of different data sources on the behavior response of foundation pit structures,and compare the influence of single source monitoring data and multi-source monitoring data on the prediction results of foundation pit behavior response in Bayesian updating.(3)A probabilistic back analysis method based on the screening updating strategy of multi-source monitoring data is constructedAiming at the difficulty of selecting updating strategy in probabilistic back analysis,a probabilistic back analysis method based on the screening update strategy of multisource monitoring data is proposed.By maximizing prediction accuracy and minimizing prediction variability,the optimal updating strategy is identified from a set of candidate update strategies composed of multi-source monitoring data based on multi-objective optimization method.The proposed method can fully mine and extract the key information of Bayesian fusion multi-source monitoring data and improve the reliability of subsequent geotechnical behavior response prediction.This method has the following three characteristics: 1)more than two types of observations are incorporated into the Bayesian updating,2)prediction fidelity and prediction variability are both considered for determining the optimal strategy,and 3)multi-objective optimization results in a set of non-dominated optimal strategies in the pool of candidate updating strategies.Based on a numerical foundation pit case,the effectiveness of the proposed method is verified,and the influence of the optimal updating strategy and the traditional updating strategy on the probabilistic back analysis results is compared.