Research On The Neural Network Prediction Method Of Large Ground Displacement On Liquefaction Site Based On Piezocone Penetration Testing

The existing earthquake disasters data indicate that the main factor of engineering structural damage in strong earthquake areas is the large ground displacement of the foundation caused by saturated sand liquefaction,including lateral deformation and settlement.The occurrence and development of large ground displacement have a very complicated process.At present,the research history of large ground displacement is short,and the mainstream prediction formulas are mostly empirical formulas based on the regression of field measured data,which cannot be very good reflects of the nonlinear relationship between large ground displacement and site parameters.In addition,most site data are obtained through the Standard Penetration Test(SPT).However,the test parameters are single,discontinuous,and the sampling point dispersion is high,which is not conducive to accurate prediction of large ground displacement of the liquefaction site.As a new in-situ test method,Piezocone Penetration Testing(CPTU)technology has the characteristics of continuous test,high precision,fast and effective test,and has obvious advantages in liquefaction site data evaluation.Artificial neural network is a multivariate nonlinear system,which has the following advantages when dealing with nonlinear relationships: good adaptability,self-organization,strong self-learning ability,association,fault tolerance and anti-interference ability.Therefore,the neural network can flexibly model the multi-parameter complex terminal system to realize the large-scale deformation prediction with various parameters.This paper mainly does the following work:(1)Studying domestic and foreign various methods of large ground displacement calculations in the past,including conventional prediction methods,large ground displacement prediction methods based on CPTU,and large ground displacement prediction methods based on artificial neural networks and studying the effects of neural networks on prediction values,highlighting the typical advantages of neural network in the process of fitting nonlinear relationships.(2)Based on the measured parameters of the liquefiable site of the Suxin Expressway and the CPTU test data,the finite element analysis model of the site ABAQUS is established to predict the large ground displacement of the surface liquefaction after the earthquake in the site,and the calculation results are compared with the GMDH(Group Method of Data Handling)neural network prediction method proposed in this paper.And the results of typical empirical formulas are compared to verify the reliability of the neural network model.(3)Using BP neural network,GA-BP neural network,RBF neural network and GMDH neural network to train the collected database of domestic and foreign sites,analyze the nonlinear relationship between site parameters and large ground displacement of the site to obtain the large ground displacement prediction model of liquefaction site.Through the comparison of conventional prediction methods and neural network prediction results,the feasibility of neural network prediction in large ground displacement of liquefaction sites is evaluated,and a GMDH neural network method for predicting large ground displacement of liquefaction sites is proposed based on the training results.Moreover,using the trained neural network model,the relationship between the soil parameters of the site and the large ground displacement value of the liquefaction site is quantitatively analyzed,and the relationship between each factor and the large ground displacement of the liquefaction site is obtained,and the results are analyzed.Compared with the influence law of conventional large ground displacement prediction method,the influence law of various factors on the large ground displacement value of liquefaction site is obtained,which further verifies the feasibility of the neural network model.(4)The method of using CPTU data to improve the model parameter acquisition is proposed and used in the liquefaction site of Suqian Expressway in Jiangsu Province.Based on the Suxin Expressway,togerher with the site's CPTU data and drilling data,the current typical empirical method and the GMDH(Group Method of Data Handling)neural network prediction method proposed in this paper are used to analyse the liquefaction possibility and the large ground displacement of the site site.