Soil parameter inversion modeling using deep learning algorithms and its application to settlement prediction: a comparative study

doi:10.1007/s11440-023-01935-z

Soil parameter inversion modeling using deep learning algorithms and its application to settlement prediction: a comparative study

2023-06-12

7

PRE

AI

DOI：10.1007/s11440-023-01935-z

原文链接

原文求助

摘要

En 中文

This study shows the application of the deep learning (DL) algorithm in the inversion of the crucial constitutive model parameters directly referring to the surface settlement monitoring data. Three parameter inversion models (BiLSTM, CNN1D, and ConvLSTM) based on DL algorithm are used for a comparative study. The datasets for training and testing are synthetic data that we calculated by importing the independently designed Python script into the finite element (FE) software ABAQUS. The Mohr-Coulomb (MC) model and the Modified Cam-Clay (MCC) model are used in the numerical model. The generalization abilities of the inversion models are tested using the monitoring settlement of the new airport in Xiamen, China. The main results show that: (i) the three-parameter inversion models based on DL can accurately identify the soil parameters that have a significant influence on the settlement calculation, and (ii) in terms of training time and prediction accuracy, the inversion model based on CNN1D has a better comprehensive performance than ConvLSTM and BiLSTM. The good consistency between the settlement calculated by the parameters retrieved from the DL model and the monitoring values shows that it has great potential to use DL as a meta-model for the calculation of the foundation settlement and the prediction of future development.

Keyword:

ABAQUS

Deep learning algorithms

Parameters inversion

Python

Secondary development

Settlement prediction

Soil parameter inversion modeling using deep learning algorithms and its application to settlement prediction: a comparative study

摘要

期刊

机构

引用论文