Abstract: To improve the accuracy and reliability of sand liquefaction prediction, according to its characteristics,7 factors including the seismic intensity, groundwater level, standard penetration number, average particle size, non-uniform coefficient, overburden effective pressure and dynamic shear stress ratio were selected as a basis for discrimination. PSO was used to optimize the parameters of SVM, and a PSO-SVM model was built for predicting sand liquefaction. Fifty samples were chosen to train the model. The model was used to predict 14 test samples and all the samples were returned to the test. The prediction accuracy was 100%. The result shows that the PSO-SVM model is better in classification and higher in efficiency in solving the problem of sand liquefaction prediction.

Key words: earthquake, sandy soil liquefaction, data normalization, support vector machines (SVM), particle swarm optimization (PSO)