The adoption of ELM to the prediction of soil liquefaction based on CPT

Establishing a soil liquefaction prediction model with high accuracy is a critical way to evaluate the quality of in situ and prevent the loss caused by seismic. In this paper, considering the advantage of cone penetration test (CPT) over standard penetration test (SPT) and the suitability for dealing with the nonlinear problems of the extreme learning machine (ELM), the ELM was tried to train the prediction model. Firstly, seven prediction parameters were analyzed and determined; then 226 CPT samples were divided into the training set and test set; then the parameter of ELM model was assured by comparing the training accuracy and speed of model when setting the number of the neuron of the hidden layer from 5 to 16 and the activation function as sig, sin, hardlim. Finally, the performance of the established ELM model was tested through the test set. The results showed the accuracy of using function sin was 81.43% and 87.50% for the training set and test set, respectively; at the same time, the operation was 1.5055 s which was not much different from other two functions. The prediction model based on CPT perform better than that of SPT and can obtain a highly accurate prediction of 100% for the liquefied case and overall accuracy of 87.5%. ELM was proved to be feasible to be used and developed into the in situ evaluation.

Automated summary

This study established a soil liquefaction prediction model using extreme learning machine (ELM) trained with cone penetration test (CPT) data, which improved the prediction accuracy. Experimental results showed that using the sin function achieved higher prediction accuracy, demonstrating the applicability and feasibility of the ELM model.