.png){kind=icon}
.png){kind=icon}
Return
Artificial seismic waves generation for complex matching conditions based on diffusion model
.png){kind=icon}
PRE
AI
DOI：10.1016/j.soildyn.2025.109290.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In the process of random seismic response analysis of structures, a large number of artificial seismic waves compatible with the design response spectrum are required. The use of numerical methods can accurately generate artificial seismic waves that meet the matching conditions, but numerical methods have the problem of long-time consumption. A feasible solution is to learn the patterns of the current seismic wave dataset through a generative model, then generate a large number of seismic waves similar to the original dataset through the trained generative model quickly. However, under complex matching conditions and existing small datasets, the generative model may lose effectiveness. The paper proposes a method for quickly and accurately generating artificial seismic waves under complex matching conditions, which achieves precise compatibility with matching conditions through an existing small dataset of artificial seismic waves and a constructed diffusion model. Numerical example shows that the method proposed in this paper improves computational efficiency by two orders of magnitude compared to numerical methods without sacrificing accuracy, and the performance of the model is better than that of existing generative adversarial models. The method proposed in this paper is applied to the expansion process of an artificial seismic wave dataset for a nuclear power structure, achieving accurate matching under complex matching conditions and improving the diversity of the artificial seismic wave dataset. By reducing the correlation coefficient between the curves in the training dataset or increasing the scale of the training dataset, the generation efficiency of DDPM can be improved. It is also essential to ensure sufficient training epochs and sampling steps to maintain the generation efficiency of DDPM.
Keywords:
Earthquake engineering
Artificial seismic wave
Generative model
Diffusion model
Dataset expansion
Journal
IF:
4.6
Papers:
7.6K
Citations:
2.5W