Predictive Analysis Of Optimal Neural Network For Deep Foundation Pit Deformation

With the acceleration of urbanization,a large number of foundation pit projects are developing in a deep and dangerous direction.In reality,the excavation of foundation pits is affected by various environments.For example: natural geographical environment,social environment,human factors,etc.Furthermore,foundation pit construction has multiple characteristics such as concealment and technical complexity,so that during the entire construction process,continuous monitoring and prediction are required to avoid uncertainty risk.Therefore,in the excavation process of the foundation pit,the BP neural network is used to continuously perform dynamic prediction and analysis on it,and scientifically determine the deformation risk of the foundation pit,which has certain scientific guiding significance for the construction process of the foundation pit.This paper takes the foundation pit project of Song Culture Experience Museum in Shangheyuan of Qingming as the research object to monitor and predict the deformation law.Using the traditional BP neural network as the basic research method,two new optimization algorithms are proposed to improve the prediction accuracy of deep foundation pit in the excavation stage.The main conclusions are as follows:(1)In terms of single time influencing factors,in view of the shortcomings of traditional BP neural network,the problems such as slow convergence speed and easy to fall into local minima exist in the optimization network by adjusting weights and additional momentum terms.A five-step,one-out,multi-step rolling optimization method is proposed,that is,five input factors and one output factor are set to perform multi-step rolling training and prediction on the foundation pit sample data.It is verified that the average relative errors of the surface settlement of the two monitoring points ZQ-23 and ZQ-30 are 4.24%and 5.30%,respectively,and the relative errors of the horizontal displacement prediction are 12.82% and10.40%,respectively.The dynamic simulation of foundation pit excavation shows a high performance.It can predict the risk of foundation pit in advance and avoid foundation pit engineering accidents.(2)In terms of various influencing factors of foundation pit,a combination of genetic algorithm and Bayesian regularization is proposed to optimize BP neural network,and the prediction application in deep foundation pit is proposed.Objective environmental factors,human subjective factors,and human operation behavior factors are considered in the safety judgment of the excavation stage of the foundation pit,and the deformation of the deep foundation pit is analyzed comprehensively.The 16 main factors that affect the deformation of foundation pit support are initially selected,and the evaluation index system is established by the expert evaluation method to reduce the shortcomings of subjective judgment,and the quantitative standards are listed and assigned respectively.Finally,SPSS statistical analysis software is used to carry out multiple linear regression analysis,so as to simplify the redundant quantity of the influencing factors of the foundation pit,and further optimize the input structure of the Bayesian regularized neural network.(3)The Bayesian regularization method analyzes the variation trend of deep foundation pit monitoring values by adaptively adjusting the training parameters of the BP neural network.The training results show that the average relative errors of the surface subsidence of the two monitoring points ZQ-23 and ZQ-30 are 0.32% and 1.54%,respectively,and the average relative errors of the predicted horizontal displacement values are 0.59% and 5.15%,respectively,which improves the performance of the neural network.Generalization ability and security of monitoring data for deep foundation pit engineering.