Deep learning for nonlinear seismic responses prediction of subway station

A novel and computationally inexpensive method for predicting the nonlinear seismic response of subway stations using deep learning approaches is developed to reduce the computational cost in stochastic seismic responses analysis. The proposed method takes the deformation of the free field where the subway station is located as the input to predict seismic responses of the subway station according to the characteristic of seismic responses of underground structures. One-dimensional convolutional neural network (1D-CNN) and long-short term memory (LSTM) network are adopted for seismic responses modeling of a two-story and three-span subway station in a data-driven fashion as opposed to the computational expensive physics-based finite element model. The prediction performance and extrapolating ability of both models are evaluated and compared with a baseline multi-layer perceptron (MLP) model. With the same training samples, the 1D-CNN has better prediction performance and extrapolating ability than both LSTM and the baseline MLP model and the LSTM model has the worst performance among the three models. The good prediction performance of 1D-CNN makes it suitable to be applied in the stochastic seismic responses analysis using the probability density evolution method (PDEM) which is solved by the finite-difference method (FDM). The evolution characteristics of the probability density function of the layer drift and the distribution characteristics of the peak value of layer drift can be captured by a low computational cost with the proposed method.

• Record URL:
  https://doi.org/10.1016/j.engstruct.2021.112735
• Record URL:
  http://www.sciencedirect.com/science/article/pii/S0141029621008853
• Availability:
  • Find a library where document is available. Order URL: http://worldcat.org/issn/01410296
• Supplemental Notes:
  • © 2021 Elsevier Ltd. All rights reserved. Abstract reprinted with permission of Elsevier.
• Authors:
  • Huang, Pengfei
  • Chen, Zhiyi
• Publication Date: 2021-10-1

Language

• English

Media Info

• Media Type: Web
• Features: Figures; References; Tables;
• Pagination: 112735
• Serial:
  • Engineering Structures
  • Volume: 244
  • Issue Number: 0
  • Publisher: Elsevier
  • ISSN: 0141-0296
  • Serial URL: http://www.sciencedirect.com/science/journal/01410296

Subject/Index Terms

• TRT Terms: Machine learning; Neural networks; Predictive models; Seismicity; Subway stations
• Subject Areas: Public Transportation; Railroads; Terminals and Facilities;

Filing Info

• Accession Number: 01778294
• Record Type: Publication
• Files: TRIS
• Created Date: Jul 29 2021 12:02PM