Abstract: One of the important tasks in the study of sand liquefaction is to assess soil liquefaction likelihood. Based on the field liquefaction investigation cases, the correlation degree between each factor and liquefaction was given by grey relational degree method, and the relationships between factors from seismic load, soil environment, soil properties and liquefaction were analyzed. The support vector machine (SVM) algorithm and Platt scaling principle were then used to construct a probabilistic model for liquefaction assessment, and comparative study was conducted with existing deterministic and probabilistic liquefaction assessment methods to analyze its effectiveness. The model was then applied to actual engineering cases and its practicality was explored finally. The results indicate that it has a significant correlation with liquefaction and the influence relationship between each factor and liquefaction is complex. The overall classification accuracy of the SVM-based soil liquefaction assessment model was 89.41% for the test set and 86.67% for the validation set, higher than that of existing liquefaction assessment methods. The depth range of applicability is extended, and the model can also provide the soil liquefaction probability. The results provide a reference for probabilistic assessment of soil seismic liquefaction.

Key words: earthquake, grey relational degree, support vector machine, liquefaction assessment