Abstract:Sand liquefaction is a harmful natural disaster, and it is of great importance to evaluate and predict sand liquefaction in the field of geological disaster prevention and control. In this paper, the rough set theory (RS) was used to perform attribute reduction on six initial evaluation indices, including magnitude, depth of soil, epicentral distance, groundwater level, standard penetration test blow count, and earthquake duration, all of which affect sand liquefaction. After removing redundant or interference information, we obtained a data set based on four core predictors. The principal component analysis (PCA) method was then used to extract the principal component from the four-core evaluation indices. The support vector machine (SVM) was used to train the data set, and the genetic algorithm (GA) was used to optimize the parameters. Finally, the RS-PCA-GA-SVM prediction model for sand liquefaction was established. Combined with the actual data of sand liquefaction, the predicted result of the proposed model was compared with that of the back propagation (BP) neural network model based on the improved Levenberg-Marquardt algorithm (LM-BP). The calculated results showed that the accuracy of sand liquefaction prediction results based on a RS-PCA-GA-SVM model are much better than those of the LM-BP neural network. The discriminant results were in good agreement with the actual situation and can be applied in practical engineering.