Study On Back Analysis Method Based On Multi-objective Optimization And Bayesian Theory For Geotechnical Engineering

In recent years,China has carried out a large number of constructions on municipal,transportation and hydropower projects.The development of urban underground space requires deep foundation pit excavation construction.Large hydropower stations often need to excavate high and steep slopes during construction.Collapse of foundation pits,uneven settlement caused by excavation,landslides or partial slope collapses will cause serious engineering accidents.In order to prevent and reduce the occurrence of engineering disasters,it is necessary to accurately predict and evaluate the deformation caused by excavation.Geomechanical parameters of rock and soil are one of the main factors affecting the prediction and evaluation results.The parameters obtained by the measurement method are not suitable for direct application to numerical calculation due to the disadvantages such as scale effect and uncertainty.The back analysis method can establish an analytical model through the obtained monitoring data,and finally calculate the geomechanical parameters of rock and soil.This study focuses on the research and application of multi-objective optimization and Bayesian theory in geotechnical engineering back analysis.The high rock slope excavation project and large deep foundation pit project are the case background.Based on the existing research,complete and effective inversion methods for multiple complex excavation projects are proposed from the aspects of multi-objective back analysis and probabilistic back analysis.The main work and results of this study are summarized as follows:1.By combining orthogonal design,finite difference modeling,BP neural network and VEGA algorithm,a BPNN-VEGA multi-objective analysis model is proposed.Compared with the back analysis method using the weight coefficient,the proposed BPNN-VEGA program not only increases the calculation speed but also reduces the calculation error.In the example of the Shapingba excavation,the prediction result calculated by using the inversion parameters is consistent with the basic trend of the real deformation.The analysis results show that the proposed method can predict the deformation of deep foundation pit during excavation based on the monitoring data.2.Based on the above multi-objective inversion model,a multi-objective back analysis program for hydraulic high slopes based on three or more types of monitoring data is proposed by combining the advanced NSGA-II algorithm.Firstly,the effectiveness of the proposed method is verified by a simple numerical example.Then,the proposed method was applied to the case of the right bank slope excavation of the Dagangshan Hydropower Station,and the analysis results showed that the deformation at the important part of the slope can be effectively predicted.Therefore,the multi-objective back analysis program with NSGA-II algorithm can be used to identify the geomechanical parameters of complex high slope rock masses in large hydropower projects and to predict the deformation of subsequent excavation.3.For the excavation of high-slope of large hydropower stations in Southwest China,a probabilistic back analysis method is proposed to update the model parameters.By analyzing the sensitivity of the geotechnical material to the prediction model,the dimensions of the prediction model parameters can be reduced.Field external displacements and anchorage forces monitoring values were used as the basis for Bayesian updating.Finally,a multi-chain Monte Carlo Markov chain simulation algorithm is used to sample the posterior probability density function of uncertain variables.By comparing the results of probability back analysis under three updating conditions and predicting the displacements and anchorage forces of subsequent excavation steps,the rationality and practicability of the proposed Bayesian method are verified.As the obtained monitoring data continues to increase,the uncertainty of the parameters is reduced and the output accuracy of the prediction model is increased.4.Based on the probabilistic back analysis of the high rock slope excavation project,a slope safety warning model is proposed by combining the strength reduction method.The observed monitoring information is used to feedback and analyze the geotechnical parameters of the rock and soil,and then the updated parameters are used to calculate the safety factor of the slope.Through the sequential transformation of the monitoring data space,the inversion parameter space and the safety factor space,the process from obtaining monitoring data to warning the slope conditions can be quickly achieved.The safety factor is used as early warning indicator and is classified into four levels.The analysis process of the early warning framework is elaborated through the slope excavation project of the Lianghekou Hydropower Station.