Spectrum-compatible artificial accelerograms via conditional variational autoencoder with generative adversarial networks

The need for spectrum-compatible ground motions in structural seismic design has driven the development of artificial seismic waveform generation techniques. This study proposes a Conditional Variational Autoencoder with Generative Adversarial Networks (CVAE-GAN) framework to generate artificial acc...

Full description

Saved in:
Bibliographic Details
Published in:Engineering structures Vol. 348; p. 121766
Main Authors: Hu, Xiaohu, Chen, Su, Ding, Yi, Fu, Lei, Li, Xiaojun
Format: Journal Article
Language:English
Published: Elsevier Ltd 01.02.2026
Subjects:
Acceleration response spectra
Artificial accelerograms
Generative adversarial networks
Variational autoencoder
Variational autoencoder
Artificial accelerograms
Generative adversarial networks
Acceleration response spectra
ISSN:0141-0296
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The need for spectrum-compatible ground motions in structural seismic design has driven the development of artificial seismic waveform generation techniques. This study proposes a Conditional Variational Autoencoder with Generative Adversarial Networks (CVAE-GAN) framework to generate artificial accelerograms conditioned on acceleration response spectra, addressing the demand for spectrum-compatible ground motions in earthquake engineering. Utilizing Japan’s K-NET and KiK-net seismic records, preprocessed with PhaseNet for P-wave detection, the model generates diverse accelerograms that preserve the temporal and spectral characteristics of real earthquakes. Validation on a test set (4.5 ≤ Mw ≤ 7.5) demonstrates that over 99 % of the generated spectra achieve an R² above 0.8 with an RMSE below 0.1 m/s2, confirming its high accuracy and realism. Furthermore, the model exhibits generalizability to out-of-range magnitudes (Mw < 4.5 or Mw > 7.5). •Developed CVAEGAN to generate spectrum-compatible seismic accelerograms.•Over 99 % generated spectra with R² > 0.8, RMSE < 0.1 m/s2 for Mw 4.5–7.5.•Demonstrated generalizability to out-of-range magnitudes (Mw < 4.5 or Mw > 7.5).•Captured P/S-wave times, coda attenuation in generated accelerograms.•Matched real records' fourier spectra, phase, and phase differences.
ISSN:0141-0296
DOI:10.1016/j.engstruct.2025.121766