Simulated site amplification models for central and eastern North America using deep learning technique
Simulated site amplification models for central and eastern North America using deep learning technique
This paper presents a suite of deep learning/Artificial Neural Network (ANN)-based response spectrum (RS) site amplification models for Central and Eastern North America trained through large-scale one-dimensional (1D) site response simulations. ANNs significantly reduce the standard deviation of the residuals of simulated amplification estimations at CENA relative to conventional functions regressed using the identical amplification database. These observations indicate the inherent limitations of traditional relationships fitting a priori functional forms to simulated data as opposed to ANNs learning the actual behaviour of the dataset. To lend credence that ANNs might be an alternative to conventional models, the ANNs performance in capturing site-specific responses is evaluated in this study. This evaluation shows that ANNs can account for the features of site-specific amplification (e.g., the amplitude and location of peak amplification) and can better reproduce the period elongation behaviour observed in nonlinear analyses as compared to their traditional counterparts.
O. Ilhan; Youssef M.A. Hashash
10th European Conference on Numerical Methods in Geotechnical Engineering (NUMGE2023)
6. Machine learning and artificial intelligence
https://doi.org/10.53243/NUMGE2023-63
Contact Information
- Secretary General - Dr A M McNamara
- Email: [email protected]
- Phone: +44 20 7040 8154
- City, University of London
- Northampton Square
- London EC1V 0HB. United Kingdom