Search

Tunnels buried in liquefiable deposits are vulnerable to liquefaction-induced uplift damage during earthquakes. This paper presents support vector machine (SVM) and artificial neural network (ANN) models to predict the liquefaction-induced uplift displacement of tunnels based on artificial databases generated by the finite difference method. The performance of the SVM and ANN models was assessed using statistical parameters, including the coefficient of determination R2, the mean absolute error, and the root mean squared error. Applications for the above-mentioned approaches are compared and discussed. A relative importance analysis was adopted to quantify the sensitivity of each input variable. The precision of the presented models is demonstrated using centrifuge test results from previous studies.

Underground structures located in areas with saturated soil are vulnerable to floating behaviour during earthquakes. This article aims to develop a simple approach for estimating the uplift displacement of underground structures. A large number of artificial data are generated using the finite difference method. These data, consisting of various features of structures, soils, and earthquakes, provide the basis for the development of a multivariate adaptive regression spline (MARS) model. The relative importance is used to quantitatively investigate the influence of each parameter and coupled interaction on the results. The developed method is validated through centrifuge tests. [ABSTRACT FROM AUTHOR]