Evaluation Of Earthquake Liquefaction-induced Lateral Spread Displacement Utilizing SGO-RBF Neural Network

Sand liquefaction is one of major questions for study in geotechnical engineering and earthquake engineering at present.The investigation of earthquake damage shows that liquefaction,as a form of earthquake disaster,often causes the loss of foundation bearing capacity,uneven foundation settlement,slope sliding,lateral displacement and structural damage,resulting in serious disasters and casualties.Among them,the lateral displacement which is more serious than any other type of damage caused by liquefaction makes buildings（structures）,roads,bridges,pipelines and other infrastructure in the liquefaction area seriously damaged,and often causes secondary disasters such as fire.Therefore,it is very necessary to study the problem of earthquake liquefaction lateral displacement,establish a reasonable and accurate prediction model of liquefaction lateral displacement,and effectively evaluate the degree of site liquefaction lateral displacement.In this case,this paper carries out the following research:（1）On the basis of the lateral displacement database of earthquake liquefaction compiled by Professor Youd in 2002,the data of the Kocaeli earthquake in Turkey in 1999 and the Chichi earthquake in Taiwan in 1999 are added,and the distribution and correlation analysis of the parameters in the database are carried out.（2）The liquefaction lateral displacement data samples in the above database were randomly divided into two groups to train and verify the radial basis function neural network（RBF）model respectively.In order to improve the accuracy of prediction,social group optimization（SGO）algorithm is used to optimize the parameters of neural network.The prediction results of SGO-RBF neural network are compared with the prediction results of empirical formula proposed by Youd et al.In 2002 and Javadi et al.In 2006 by three prediction effect evaluation indexes of resolvable coefficient,root mean square error and average absolute error,so as to verify the learning and prediction ability of SGO-RBF neural network for seismic sand liquefaction lateral displacement.（3）This paper summarizes the related parameters of earthquake disaster prediction proposed by predecessors,and selects peak ground acceleration（PGA）,Arias Intensity（I_a）and Cumulative Absolute Velocity（CAV₅）as the representative ground motion parameters for analysis.The appropriate attenuation formula is selected to fit the parameters I_a and CAV₅,and new parameters are added to the existing database.PGA,I_a and CAV₅ are used to predict the lateral displacement of sand liquefaction instead of magnitude（M）and epicentral distance（R）to verify the effectiveness of PGA,I_a and CAV₅.The results show that CAV₅ can effectively reduce the prediction error of seismic liquefaction lateral displacement,and PGA and I_a can replace M and R well.（4）According to the degree of lateral displacement,the liquefaction lateral displacement data in the database are classified by D_h.For the two types,using SGO-RBF neural network to train and predict the lateral displacement,and comparing the prediction results with unclassified prediction results,which proves that the classification prediction can improve the accuracy of prediction;Support Vector Machine（SVM）is used to classify the data,and the classification effect is evaluated by precision and recall;the test group data classified by SVM is predicted by the SGO-RBF neural network model trained by classification,which proves that the classification prediction can improve the accuracy of prediction The feasibility of using SVM classification and SGO-RBF neural network to predict the data to be measured.