Research On Liquefaction Discrimination And Lateral Displacement Prediction Based On Cyclic Neural Network

In recent years,a large number of destructive earthquakes have been reported at home and abroad.The liquefaction of a large area of sand and the lateral displacement caused by the liquefaction during the earthquake process have caused huge property losses to local residents.At present,the discrimination of liquefaction and the prediction of lateral movement of sand at home and abroad mainly rely on SPT and CPT field investigation tests,but the obtained liquefaction discrimination method and the empirical model of lateral movement is relatively simple,so it is necessary to provide more accurate prediction model.The shear wave velocity method has the advantages of no damage,fast test speed and low cost,and can also be used for the identification of sand liquefaction.Recurring Neural Network(RNN)is a recursive Neural Network model for processing and predicting sequence data.Its advantage lies in solving the problems of identification,classification and fitting prediction,so it is applied to the discrimination of earthquake liquefaction and the lateral movement prediction after liquefaction.The main research contents of this paper are as follows:1?Refer to relevant literature at home and abroad to determine the discriminant formula of shear wave velocity liquefaction commonly used in four kinds of engineering.According to the relevant data of Bachu,Taiwan and Tangshan earthquakes in China,different methods are used to discriminate liquefaction.The results show that the overall discriminant rate of Andrus method is stable in Bachu,Taiwan and Tangshan regions.Rock gauge method and hyperbolic method have high discriminant rate and obvious regional applicability in Tangshan area.The discriminant rate of liquefaction site by hydrogeological survey method in three areas is more than 90%.2?Eight parameters were selected as liquefaction discriminant factors,and a prediction model of seismic liquefaction(RA-RNN model)based on Rectified Adam(improved adaptive algorithm)and cyclic neural network(RNN)was proposed.Four indexes were used to evaluate the model,and compared with ADAM-RNN and SVM models.The results show that when the number of hidden layer nodes of RA-RNN model is 20 and the learning rate is 0.01,the prediction effect is the best,and the prediction effect is better than that of ADAM-RNN and SVM model.In the identification process,RA-RNN model is most sensitive to the number of penetration strokes and fine grain content.3?According to the influencing factors of lateral movement proposed by Youd,the improved adaptive algorithm optimized cyclic neural network(RA-RNN)was used to evaluate the lateral movement of gentle slope liquefaction site.The performance of the RA-RNN model was verified by predicting the lateral movement of gentle slope in Kocaeli,Turkey and Chi-ji,Taiwan,and the results were compared with those predicted by multiple linear regression method.The results show that the lateral displacement predicted by the RA-RNN model is mostly between 0.7and 1.3 times the measured value,which proves the rationality and reliability of the proposed method.Compared with the traditional model,the RA-RNN model has a great improvement in the prediction of small or zero liquefaction lateral displacement,and has a better prediction effect.4?Seismic liquefaction data in Darfield,New Zealand were collected,and 6 representative boreholes were selected as validation data of the proposed method.The results show that the overall discriminant rate of Andrus method in this area is stable,and the discriminant rate of RA-RNN model is higher than that of shear wave velocity method,and the prediction results of lateral movement of the RA-RNN model are close to the measured values,and the prediction accuracy is higher than that of the traditional model.