We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
A hybrid numerical–probabilistic approach for machine learning-based prediction of liquefaction-induced settlement using CPT data.
- Authors
Gupta, Tanmay; Ramana, G V; Elgamal, Ahmed
- Abstract
Traditional deterministic approaches for estimating liquefaction-induced settlement (LIS) do not account for uncertainties associated with earthquake loading and soil spatial variability. A numerical–probabilistic method for evaluating LIS that accounts for these uncertainties is developed. An advanced non-linear soil constitutive model is employed in numerical simulations and verified against VELACS experiment. Thereafter, CPT data from an alluvial plain in India together with scaled accelerograms are used to generate a big dataset. This dataset is implemented in regression machine learning algorithms for prediction of LIS. Of the adopted algorithms, it is observed that the XGBoost algorithm performs best with mean r2 of 0.980, mean MSE of 0.032, and mean EV of 0.980. The developed methodology is validated by comparing the predicted LIS with data from 21 case histories and the predicted magnitude are in reasonable agreement with reported values. A ready to use open-source program is developed that enables the practitioners to quantify the LIS.
- Subjects
INDIA; MACHINE learning; ALLUVIAL plains; EARTHQUAKES; ACCELEROGRAMS; FORECASTING
- Publication
Arabian Journal of Geosciences, 2023, Vol 16, Issue 6, p1
- ISSN
1866-7511
- Publication type
Article
- DOI
10.1007/s12517-023-11500-3