A Hybrid SSA-CNN-SVM Model for Seismic-Induced Sand Liquefaction Discrimination

Abstract Seismic-induced sand liquefaction‌ represents ‌a high-impact geohazard‌, ‌rendering‌ the discrimination‌ and ‌prediction‌ of ‌sand liquefaction states‌ ‌essential‌ for ‌geohazard mitigation research‌. For the rational discrimination of sand liquefaction states‌, this study proposes an SSA-CNN-SVM model that integrates Sparrow Search Algorithm (SSA)-optimized Convolutional Neural Networks (CNN) with Support Vector Machines (SVM) for liquefaction discrimination. This model initiates from raw sand liquefaction data, accomplishes layer-by-layer learning to extract liquefaction features and discriminate the states of liquefaction, and employs SVM in lieu of Softmax functions for liquefaction state classification. Taking the sand liquefaction case from the Tangshan earthquake as the comprehensive dataset, the evaluation metrics - standard penetration test (SPT) blow count, mean particle size, coefficient of uniformity, groundwater table depth, effective overburden pressure, seismic intensity, and cyclic shear stress ratio - are input into the SSA-CNN-SVM model for prediction. The results are compared with those from SSA-SVM, SVM, CNN, and Backpropagation Neural Network (BPNN) models, validated against actual sand liquefaction data. The results indicate that the SSA-CNN-SVM model demonstrates superior performance in sand liquefaction discrimination, achieving an accuracy of 83.33%, precision of 83.33%, recall of 83.33%, and F1-Score of 83.33% – all exceeding corresponding metrics of other comparative models. This validates the high precision of the proposed liquefaction discrimination model and provides a novel approach for practical applications.