Study On Sand Liquefaction Discrimination And Liquefaction Lateral Displacement Prediction Based On Support Vector Machine Model

The earthquake-induced sand liquefaction has brought great disaster to human beings.As one of the common damage types in the liquefaction process,liquefaction lateral displacement will cause catastrophic damage to building structures.Therefore,it is extremely necessary to judge sand liquefaction and predict liquefaction lateral displacement in the site,which is related to people’s lives and property safety.At present,the main methods to deal with liquefaction discrimination and liquefaction lateral displacement prediction at home and abroad are in-situ tests,empirical laws are obtained after statistical analysis based on the data of liquefaction sites,and finally predictions are made by empirical formulas.Earthquake sand liquefaction is a very complicated process,which covers many influencing factors and presents a highly nonlinear relationship.Therefore,it is increasingly unable to meet people’s demand for more efficient and accurate prediction only by empirical formulas.As one of the most classic intelligent algorithms,Support Vector Machine(SVM)is often used in data processing problems such as classification and regression.Especially,it is good at dealing with high-dimensional and nonlinear problems,and it has efficient small sample learning ability,which makes it widely used in all fields.In view of this,this paper applies it to sand liquefaction discrimination and liquefaction lateral displacement prediction.The main research contents and conclusions are as follows:(1)Firstly,many related literatures at home and abroad are consulted,and the most commonly used empirical methods of liquefaction discrimination and liquefaction lateral displacement prediction in practical projects at home and abroad are summarized and analyzed.It can be found that the SPT liquefaction discrimination method in China has more initial judgment steps than the SPT method in the United States,and can screen out actual non-liquefied sites more efficiently.However,the theoretical basis of SPT method in China is not sufficient,the standard penetration test has not been revised and the influence of seismic factors has not been fully considered,which leads to more conservative actual discrimination.With the development of technology and the perfection of theory,the empirical formula of liquefaction lateral displacement gradually considers more influencing factors,such as earthquake,topography and geotechnical factors,and its accuracy and applicability are also improved,among which the most representative method is the multiple linear regression method proposed by Professor Youd.(2)Considering twelve parameter factors,we use Hua Luogeng’s good point set to map the best point to the target solution space,and establish a support vector machine prediction model based on the sea squirt algorithm.Then,we use the variable importance scoring function of random forest to screen the top eight parameters of importance,and discriminate them again,and get a more accurate prediction result,which is compared with the traditional Seed simplification method.The research shows that the liquefaction discrimination model proposed in this paper combines the advantages of Chinese and American SPT methods,which not only covers more liquefaction influencing factors,but also can effectively screen out the actual non-liquefaction situation,and the accuracy meets the needs of practical projects.(3)Using the support vector machine model,eight factors,including terrain,earthquake and soil,are used to predict and estimate the liquefaction lateral displacement caused by earthquake,and the predicted results of this paper are compared with those of traditional linear regression and some neural network methods.The research results show that the predicted results of RF-SSA-SVR model have the smallest error and the highest fitting degree,which verifies the accuracy and reliability of this method.(4)The measured data of Niigata earthquake in 1964,Shisheng earthquake in 1968 and Miyagi earthquake in 1978 in Japan,the liquefaction data of an expressway site in Suqian,Jiangsu Province under seismic fortification intensity and the field measured data of the 2011 Darfiled earthquake in New Zealand were collected and selected as the verification data of this method to further verify its practicability.The results show that the predicted results of this method are close to the measured data,which can meet the needs of practical engineering,and also provide a new idea for further research.