Back Analysis Of Soil Parameters Of Deep Foundation Pit In Haixiang With Soft Soil And Dynamic Prediction Of Foundation Pit Excavation

Due to the complexity,randomness,dispersion and uncertainty,the numerical simulation analysis of geotechnical parameters is often unable to accurately and accurately predict the mechanical behavior and mechanical performance of soils.Therefore,the displacement analysis of soil parameters based on intelligent algorithm is analyzed.This paper used neural network and genetic neural network to carry out soil displacement back analysis,and used the strong nonlinear mapping ability of neural network to deal with the nonlinear mapping relationship between displacement and soil parameters.The soil mechanics parameters of the inversion analysis were applied to the forward analysis in the finite element software to achieve better guidance for the smooth construction of the foundation pit.Based on the analysis of soil parameters in the Chaoyang area of Guangdong Chaoshan area,the paper analyzed the parameters of the marine body and explored the tunnels through the monitoring analysis,finite element simulation,neural network parameter inversion and finite element forward analysis.Firstly,the article introduced in detail the layout of the horizontal pits and the vertical surveying of the foundation pits,and discussed the deformation mechanism of the retaining structure.The finite element software was used to simulate and predict the behavior of foundation pit mechanics.The selection of soil constitutive relations,the selection of soil parameters and the enclosure structure were introduced.The numerical analysis results were compared with the measured data to introduce the necessity of the intelligent algorithm displacement inversion analysis.Secondly,in order to determine the parameters of the soil displacement back analysis,the paper analyzed the parameter sensitivity and introduced the orthogonal test method to reduce the number of tests in the parameter sensitivity analysis process and ensure that the selected data can represent the change of the whole test sample.The parameters in the orthogonal experiment were brought into the finite element model for analysis and the variance analysis method was used for parameter sensitivity analysis.The soil displacement back analysis was carried out by selecting the parameters that are sensitive to the deformation of the retaining structure in the soil mechanics parameters.Thirdly,the nonlinear mapping ability of neural networks is an important factor for selecting displacement inversion analysis.In order to improve the nonlinear mapping accuracy of neural networks,this paper discussed from several aspects: introduceing the method of neural network hidden layer node value,and then useing training samples and prediction samples to calculate the possible range of hidden layer nodes.Trial calculation and selection of the node parameters of the better fitting effect from the trial results;the article introduced the commonly used transfer function and its principle,and based on the research,four suitable combinations of transfer functions were selectsed for neural network simulation.From the analysis of the results of the four groups,the appropriate network topology transfer function was selected.For the selection of the number of hidden layers,the article compared and analyzed the simulation generalization ability of the single hidden layer and the double hidden layer.The article is based on the above three points without changing the network topology,and selects the most suitable parameters for displacement inversion analysis.The soil mechanics parameters of the displacement inversion were applied to the numerical simulation analysis for forward analysis,and the measured data were compared.Finally,based on the neural network’s lack of global search ability,genetic optimization algorithm was used to improve the generalization ability of neural network.The introduction of genetic algorithm optimizes the connection weight and threshold between neural network structures and improves the search ability.The feasibility and integration of neural networks were described by introducing the principles and steps of genetic algorithms,and the specific steps of genetic neural networks were detailed.The reciprocal of the error was used as the fitness value and the optimal solution in the population was found by means of selection,intersection and mutation,and the optimal solution was applied to the neural network for displacement inversion analysis.The inversion results of the genetic parameters of the soil neural network and the neural network were compared and analyzed,and the precision of the genetic neural network is relatively improved.The thought article based on dynamic change prediction used genetic neural network for isochronous analysis.The results revealed that the prediction results are better and the prediction results are more accurate.