A Metaheuristic-Trained Wavelet Neural Network for Predicting of Soil Liquefaction Based on the Standard Penetration Test Results

Abstract

Earthquake-induced soil liquefaction is one of the natural hazards that can cause irreparable damages to buildings and infrastructure in addition to threatening human lives. Therefore, accurate prediction of liquefaction has always been one of the important tasks of geotechnical engineers. In the present study, a hybrid approach is introduced to predict liquefaction using the Standard Penetration Test (SPT) results. The approach involves combining Wavelet Neural Network (WNN) and Teaching-Learning Based Optimization (TLBO) techniques to create a binary classification model. The dataset used for this study contains 288 data points, ensuring reliability. The optimal architecture for the WNN is determined as 2-8-1, indicating that it consists of two neurons in the input layer, eight neurons in the hidden layer with the Shannon wavelet function, and one neuron in the output layer with the Hardlim activation function. The developed WNN model is able to predict liquefaction with an overall accuracy of 96.52% based on two input variables, namely modified cyclic stress ratio (CSR7.5) and SPT number (N1,60). Comparing the accuracy of the WNN-TLBO model developed in this research with that of other artificial intelligence models developed in this field shows the superior accuracy of the WNN-TLBO model compared to previous models.

Keywords

• liquefaction
• wavelet neural network (wnn)
• teaching-learning based optimization (tlbo)
• standard penetration test