Comparison of Markov Chain Monte Carlo and Sequential Monte Carlo inference techniques for calibrating soil parameters in a braced excavation

1. Home
2. Publications
3. Online Library
4. Comparison of Markov Chain Monte Carlo and Sequential Monte Carlo inference techniques for calibrating soil parameters in a braced excavation

Publication Title
Comparison of Markov Chain Monte Carlo and Sequential Monte Carlo inference techniques for calibrating soil parameters in a braced excavation

Abstract
Data assimilation has gained significant attention for the sequential updating of soil parameters, particularly in the context of automatic back-analysis. Compared to traditional manual calibration methods, continuous Bayesian calibration provides a formal approach to merge measured data with model predictions. The performance of two prevalent inference algorithms, namely the Markov chain Monte Carlo (MCMC) and Sequential Monte Carlo (SMC), are compared in back-analyzing the parameters of a synthetic braced excavation problem. Model parameters are sequentially inferred using inclinometer data calculated during a typical excavation project based in London. The accuracy of the MCMC and SMC samplers in inferring the synthetic ground truth is compared systematically. Results reveal that back-analysis using MCMC is better for problems which involve high-dimensionality in the output space, while SMC yields faster inference results, especially in cases which involve a lower dimensional output space. This study demonstrates the effectiveness of the two methods in sequential Bayesian updating and outlines their benefits and limitations in identifying soil parameters for a synthetic problem.
Authors
N. Yang; Truong Le; L. T. Nguyen

Published In

18th European Conference on Soil Mechanics and Geotechnical Engineering (ECSMGE2024)

Theme

A - New developments on structural design

Download PDF