TimesNet-Gen: Deep Learning-based Site Specific Strong Motion Generation
Baris Yilmaz, Bevan Deniz Cilgin, Erdem Akagündüz, Salih Tileylioglu
2025-12-08
Summary
This research focuses on creating more realistic earthquake simulations for specific locations, aiming to improve earthquake risk assessment and preparedness.
What's the problem?
Currently, predicting how strongly the ground will shake at a particular location during an earthquake is difficult. Standard earthquake models don't always account for the unique characteristics of the soil and geology at a specific site, which significantly influence how earthquake waves behave. This means current risk assessments might not be accurate enough for local areas.
What's the solution?
The researchers developed a new computer model called TimesNet-Gen. This model learns from actual earthquake recordings at different locations and then generates synthetic earthquake recordings that mimic the specific ground motion patterns of each site. It focuses on the 'signature' of each location, using a technique that captures the essential characteristics of the site's response to shaking. They tested it by comparing features of the generated earthquakes, like how strongly different frequencies of shaking are amplified, to those of real earthquakes at the same locations.
Why it matters?
More accurate earthquake simulations are crucial for designing safer buildings and infrastructure. By creating site-specific models, engineers can better understand the potential shaking intensity at a given location and build structures that can withstand those forces, ultimately reducing the risk of damage and casualties during an earthquake.
Abstract
Effective earthquake risk reduction relies on accurate site-specific evaluations. This requires models that can represent the influence of local site conditions on ground motion characteristics. In this context, data driven approaches that learn site controlled signatures from recorded ground motions offer a promising direction. We address strong ground motion generation from time-domain accelerometer records and introduce the TimesNet-Gen, a time-domain conditional generator. The approach uses a station specific latent bottleneck. We evaluate generation by comparing HVSR curves and fundamental site-frequency f_0 distributions between real and generated records per station, and summarize station specificity with a score based on the f_0 distribution confusion matrices. TimesNet-Gen achieves strong station-wise alignment and compares favorably with a spectrogram-based conditional VAE baseline for site-specific strong motion synthesis. Our codes are available via https://github.com/brsylmz23/TimesNet-Gen.