Research On Prediction Of Seismic Liquefaction Of Sand Based On Shear Wave Velocity And Support Vector Machine

More than half of land is located in the areas of high seismic activity induced by earthquake in China and the liquefaction hazards caused by earthquake cannot be ignored.The scientific and effective identification of sand liquefaction is of great significance in earthquake disaster planning of prevention and mitigation and the people's life and property when we take the engineering survey and the construction of project.Firstly,this paper reviews the research results of sand liquefaction at home and abroad and found that there is very limited researches on sand liquefaction prediction based on shear wave velocity and support vector machine.So the in-situ test files about the shear wave velocity of the after-earthquake site has been collected and collated in recent years after studying the theory of machine learning and mastering the principle of suppor vector machine.Then the models are established to predict the liquefaction of sand by using the SVM and Python language in the machine learning method.At the meantime,in order to study the influence of principle component analysis(PCA),particle swarm optimization(PSO)algorithm and the change of training samples on the performance of SVM model,four basical models were formed: M1 based on SVM,M2 based on PSO-SVM,M3 based on PCA-SVM and M4 based on PSO-PCA-SVM.Finally,twelve models from M1 to M4 series were formed with the change of training samples and made a comparison among the NCEER method,the recommended method by the Code for Investigation of Geotechnical Engineering,the method proposed by Sun Rui and the method of this paper.The results reveal that:1.The PCA and PSO method can increase the predictive accuracy and improvethe predictive performance of the M1?M4 model.2.The M4 model based on PCA-PSO-SVM performs best in the comprehensive evaluation.3.The series models of M1 which do not use PSO are not sensitive to the changing samples when training samples increase slightly by 3%,4% and 5%,and the series models of M3 do not learn enough from the new training samples while the series models of M2 and M4 using the PSO algorithm can obtain information from new data more efficiently,and slightly improve model predictive performance.4.By contrasting the commonly used shear wave velocity discriminant method at home and abroad,the optimal model M4 based on PCA-PSO-SVM gets much higher prediction success rate than the NCEER method,the method of Chinese code and Sun Rui method.M4 model confirmed the validity and versatility of this paper method which have achieved high accuracy both at home and abroad.