Abstract: There are a lot of factors that affect sand liquefaction, and then it is necessary to establish a multi-index liquefaction prediction model. At present, all of the multi-index sand liquefaction prediction models based on a hypothesis—the selected discriminant factors are independent and there is no correlation between each other, which may lead to information superimpose between the different factors and mis-discrimination. In this paper, based on 25 cases of sand liquefaction in Tangshan earthquake, eight factors influencing sand liquefaction are selected as the initial discriminant indexes. The principal component analysis(PCA) is introduced in the correlation analysis of initial discriminant indexes and dimension-reduction processing is conducted to some indexes with high correlation. The new sample data are obtained based on the 4 principal component conversions. A predictive model for predicting sand liquefaction is established under the combination of PCA and distance dscriminant analysis(DDA). The forecasting results of 18 training samples are all correct by using the established model. The liquefaction of the other 7 cases is also predicted; and the results are compared with those of the standard, Seed method, BP method and DDA method. The results show that the prediction accuracy of the forecasting model is 100%. The model is applied to a practical engineering example, and the results are consistent with the actual situation, so as to show that the model has good prediction function and can be used in practical engineering.

Key words: sandy soil liquefaction, multiple index correlation, predictive model, principal component analysis, distance discriminant analysis