Broadband Ground Motion Synthesis via Generative Adversarial Neural Operators
Description:
We present a data-driven framework for 3-component ground motion synthesis intended for engineering applications. Leveraging the increase of ground-motion data from seismic networks and recent advancements in Machine Learning, we train a Generative Adversarial Neural Operator (GANO) to produce realistic three-component velocity time histories conditioned on moment magnitude, rupture distance, Vs30, and tectonic environment type of earthquakes. Neural operators allow for the sampling of functions by learning push-forward operator maps in infinite-dimensional spaces, rendering our model resolution-invariant. We train the GANO on two 50K ground motion datasets harvested from the Japanese Strong Motion Network Kik-Net (M 4.5-8.0) and the Southern California Earthquake Data Center (M 4.0-7.5) correspondingly; and show that the framework can recover the imposed magnitude, distance, and Vs30 scaling of Fourier and response spectral acceleration components. We evaluated our model through residual analysis with the empirical dataset as well as by comparisons with conventional GMMs for selected ground motion scenarios; results show that our model recovers both the mean value and aleatory variability of the evaluated ground-motion parameters. In particular, quantitative measures of Fourier and response spectral amplitude residuals for the said datasets indicate that the GANO generated ground motions are unbiased in the frequency range of 0.1-20Hz for the horizontal component and 0.1-30Hz for the vertical component. Potential applications of the presented framework include: (i) design ground motions for earthquake scenarios not represented in empirical datasets, (ii) risk-targeted ground motions for site-specific and system-level engineering applications, and (iii) high-frequency components of simulated ground motions when regional-scale velocity models lack the appropriate spatial resolution in the shallow crust.
Session: High-frequency Ground Motion Measurements, Assessments and Predictions
Type: Oral
Date: 4/18/2023
Presentation Time: 08:45 AM (local time)
Presenting Author: Grigorios Lavrentiadis
Student Presenter: No
Invited Presentation:
Authors
Yaozhong Shi yshi5@caltech.edu California Institute of Technology |
Grigorios Lavrentiadis Presenting Author glavrent@caltech.edu California Institute of Technology |
Zachary Ross zross@caltech.edu California Institute of Technology |
Domniki Asimaki Corresponding Author domniki@caltech.edu California Institute of Technology |
|
|
|
|
|
Broadband Ground Motion Synthesis via Generative Adversarial Neural Operators
Category
High-frequency Ground Motion Measurements, Assessments and Predictions