.png){kind=icon}
.png){kind=icon}
Return
Gaussian Process with Bayesian Kernel Selection for Seismic Attenuation Modeling
.png){kind=icon}
PRE
AI
DOI：10.1061/AJRUA6.RUENG-1579.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Modeling seismic attenuation relationships aims to estimate expected ground motion at a given site from earthquakes, which is important for seismic hazard assessment and disaster mitigation. The complex nature of the seismological mechanisms poses challenges in deriving reliable parametric seismic attenuation models. Gaussian process (GP) regression is a Bayesian nonparametric approach inferring both the estimations of the concerned quantity and the associated uncertainty. Since it does not rely on prescribed functional equations, GP regression is flexible for seismic attenuation modeling. However, subjective designation on the GP kernel is required, and an improper kernel may distort the modeling performance. To tackle this problem, we propose a Bayesian optimal kernel selection approach. In particular, a kernel candidate pool that includes potential kernels is established. It comprises the typical kernel types with different combinations of the concerned input variables. Bayesian inference is employed to select the optimal kernel that strikes the optimal balance between fitting capacity and robustness. Based on the selected optimal kernel, GP regression can be conducted properly for estimating the concerned quantity. Moreover, the uncertainties of the parameters and all estimates can be quantified. The efficacy of the proposed approach is demonstrated using the data set of the Ms 8.0 great Wenchuan earthquake. In addition, a comparative study is conducted with a parametric modeling approach.
Keywords:
Bayesian inference
Gaussian process (GP)
Kernel selection
Nonparametric modeling
Seismic attenuation
Peak ground acceleration
Journal
A
IF:
2.7
Papers:
65
Citations:
0
View more organizations