Convolutional neural network-based seismic fragility analysis of subway station structure considering spatial variation of site shear-wave velocity

Abstract

Prior research demonstrated that the uncertainty of soil parameter can significantly impact the seismic response analysis and seismic performance evaluation of underground structures. To this end, the seismic response of a three-story, three-span subway station embedded in a typical layered engineering site was investigated in this study and the uncertainty of soil parameters was explicitly considered using the shear-wave velocity interlayer correlation model. In addition, an incremental dynamic analysis considering the uncertainty of the shear-wave velocity was performed using the one-dimensional convolutional neural network (1D-CNN) model as a surrogate model for the traditional finite element method. The results indicate that the uncertainty of soil shear-wave velocity increases the average PGA values for minor, moderate, and severe failures with a probability of exceeding 50% by about 10%, and increases the logarithmic standard deviation by about 40%. When considering the uncertainty of shear-wave velocity, the discreteness of the fragility curve will increase. Moreover, the rightward shift of the fragility curve midpoint leads to a comparatively more safety seismic performance evaluation of the structure under significant seismic input.