Back Analysis Of M-value Of Metro Foundation Pit In Water-rich Sandy Gravel Stratum

With the rapid development of urban modernization,buildings are piled up one after another,and more and more underground engineering applications are being applied,especially the deep foundation pit.Undoubtedly,for geotechnical engineering,the design of excavation and support of ultra-deep foundation pit is very important,especially the selection of parameters in the design of foundation pit support,which is not only related to the rationality of design,but also affects the overall safety of a project.The m-method of elastic foundation beam theory,which is recommended by most codes,is the most commonly used method in the design of foundation pit support in China.In the application of m-method,the selection of calculation parameters is a very difficult problem for the geotechnical field.Because m-value can not be obtained in the actual survey,the corresponding codes only give a relatively broad value limit,and because different regions are specific.It is difficult to determine the optimum m value for different geological conditions.Therefore,through the corresponding research methods to determine an accurate and reasonable m-value,it can not only truly reflect the actual working status of foundation pit excavation and support,but also enrich the value of m value in special strata,which has certain engineering significance.Firstly,the research status and development of the anti-analytical method in the geotechnical field are briefly introduced.Then the theoretical knowledge of back analysis and artificial neural network is introduced,and the basic principles,structures and algorithms of BP neural network used in this paper are briefly summarized.Then the basic overview of the foundation pit project of the Beijing Metro Station was subsequently introduced,and the monitoring data of the underwater excavation project in the rich sand and gravel stratum was analyzed.The model was established by using the finite element simulation software Midas/GTS,and the reliability of the model was verified by simulation comparison.In the inversion study of this paper,the first step is to determine the learning training samples to be used later.Therefore,under the orthogonal experimental design principle,the orthogonality of each soil layer in the foundation pit excavation and support stage is established by the above-mentioned Midas/GTS model.The next part is to establish the corresponding BP neural network inversion analysis model,which is mainly based on the principle of artificial neural network and coded by matlab program.In order to optimize the inversion effect,the number of training functions and hidden layer neurons is calculated,and the data obtained by orthogonal experiment design is used to train the established neural network model,and the verification sample is used to test the reliability of the neural network.After the inversion model is established,the monitoring data obtained from the field measurement points are selected and the inversion model is used to obtain the m-value of the soil layer in the regional foundation for inversion optimization,and then the m-value is more suitable for the region under various working conditions.The value is finally compared with the norm m-value.In addition,some useful conclusions have been drawn,and suggestions for design optimization are available for future design reference.