Stochastic deep collocation method based on neural architecture search and transfer learning for heterogeneous porous media

Research output: Contribution to journal › Article › Research › peer review

Authors

  • HW Guo
  • XY Zhuang
  • Pengwan Chen
  • N Alajlan
  • T Rabczuk

Research Organisations

External Research Organisations

  • Tongji University
  • Beijing Institute of Technology
  • King Saud University
View graph of relations

Details

Original languageEnglish
Pages (from-to)5173-5198
Number of pages26
JournalEngineering with computers
Volume38
Issue number6
Early online date18 Jan 2022
Publication statusPublished - Dec 2022

Abstract

We present a stochastic deep collocation method (DCM) based on neural architecture search (NAS) and transfer learning for heterogeneous porous media. We first carry out a sensitivity analysis to determine the key hyper-parameters of the network to reduce the search space and subsequently employ hyper-parameter optimization to finally obtain the parameter values. The presented NAS based DCM also saves the weights and biases of the most favorable architectures, which is then used in the fine-tuning process. We also employ transfer learning techniques to drastically reduce the computational cost. The presented DCM is then applied to the stochastic analysis of heterogeneous porous material. Therefore, a three dimensional stochastic flow model is built providing a benchmark to the simulation of groundwater flow in highly heterogeneous aquifers. The performance of the presented NAS based DCM is verified in different dimensions using the method of manufactured solutions. We show that it significantly outperforms finite difference methods in both accuracy and computational cost.

Keywords

    Deep learning, Neural architecture search, Error estimation, Randomized spectral representation, Method of manufactured solutions, Log-normally distributed, Physics-informed, Sensitivity analysis, Hyper-parameter optimization algorithms, Transfer learning, PARTIAL-DIFFERENTIAL-EQUATIONS, GLOBAL SENSITIVITY-ANALYSIS, FLOW SIMULATION, RANDOM-FIELDS, TRANSPORT, ALGORITHM, NETWORKS, POROSITY, LAW

ASJC Scopus subject areas

Cite this

Stochastic deep collocation method based on neural architecture search and transfer learning for heterogeneous porous media. / Guo, HW; Zhuang, XY; Chen, Pengwan et al.
In: Engineering with computers, Vol. 38, No. 6, 12.2022, p. 5173-5198.

Research output: Contribution to journal › Article › Research › peer review

Guo HW, Zhuang XY, Chen P, Alajlan N, Rabczuk T. Stochastic deep collocation method based on neural architecture search and transfer learning for heterogeneous porous media. Engineering with computers. 2022 Dec;38(6):5173-5198. Epub 2022 Jan 18. doi: 10.1007/s00366-021-01586-2
Download
@article{26c441937c2e47469f5c8a3814d11317,
title = "Stochastic deep collocation method based on neural architecture search and transfer learning for heterogeneous porous media",
abstract = "We present a stochastic deep collocation method (DCM) based on neural architecture search (NAS) and transfer learning for heterogeneous porous media. We first carry out a sensitivity analysis to determine the key hyper-parameters of the network to reduce the search space and subsequently employ hyper-parameter optimization to finally obtain the parameter values. The presented NAS based DCM also saves the weights and biases of the most favorable architectures, which is then used in the fine-tuning process. We also employ transfer learning techniques to drastically reduce the computational cost. The presented DCM is then applied to the stochastic analysis of heterogeneous porous material. Therefore, a three dimensional stochastic flow model is built providing a benchmark to the simulation of groundwater flow in highly heterogeneous aquifers. The performance of the presented NAS based DCM is verified in different dimensions using the method of manufactured solutions. We show that it significantly outperforms finite difference methods in both accuracy and computational cost.",
keywords = "Deep learning, Neural architecture search, Error estimation, Randomized spectral representation, Method of manufactured solutions, Log-normally distributed, Physics-informed, Sensitivity analysis, Hyper-parameter optimization algorithms, Transfer learning, PARTIAL-DIFFERENTIAL-EQUATIONS, GLOBAL SENSITIVITY-ANALYSIS, FLOW SIMULATION, RANDOM-FIELDS, TRANSPORT, ALGORITHM, NETWORKS, POROSITY, LAW",
author = "HW Guo and XY Zhuang and Pengwan Chen and N Alajlan and T Rabczuk",
note = "Funding Information: The authors extend their appreciation to the Distinguished Scientist Fellowship Program (DSFP) at King Saud University for funding this work.",
year = "2022",
month = dec,
doi = "10.1007/s00366-021-01586-2",
language = "English",
volume = "38",
pages = "5173--5198",
journal = "Engineering with computers",
issn = "0177-0667",
publisher = "Springer London",
number = "6",

}

Download

TY - JOUR

T1 - Stochastic deep collocation method based on neural architecture search and transfer learning for heterogeneous porous media

AU - Guo, HW

AU - Zhuang, XY

AU - Chen, Pengwan

AU - Alajlan, N

AU - Rabczuk, T

N1 - Funding Information: The authors extend their appreciation to the Distinguished Scientist Fellowship Program (DSFP) at King Saud University for funding this work.

PY - 2022/12

Y1 - 2022/12

N2 - We present a stochastic deep collocation method (DCM) based on neural architecture search (NAS) and transfer learning for heterogeneous porous media. We first carry out a sensitivity analysis to determine the key hyper-parameters of the network to reduce the search space and subsequently employ hyper-parameter optimization to finally obtain the parameter values. The presented NAS based DCM also saves the weights and biases of the most favorable architectures, which is then used in the fine-tuning process. We also employ transfer learning techniques to drastically reduce the computational cost. The presented DCM is then applied to the stochastic analysis of heterogeneous porous material. Therefore, a three dimensional stochastic flow model is built providing a benchmark to the simulation of groundwater flow in highly heterogeneous aquifers. The performance of the presented NAS based DCM is verified in different dimensions using the method of manufactured solutions. We show that it significantly outperforms finite difference methods in both accuracy and computational cost.

AB - We present a stochastic deep collocation method (DCM) based on neural architecture search (NAS) and transfer learning for heterogeneous porous media. We first carry out a sensitivity analysis to determine the key hyper-parameters of the network to reduce the search space and subsequently employ hyper-parameter optimization to finally obtain the parameter values. The presented NAS based DCM also saves the weights and biases of the most favorable architectures, which is then used in the fine-tuning process. We also employ transfer learning techniques to drastically reduce the computational cost. The presented DCM is then applied to the stochastic analysis of heterogeneous porous material. Therefore, a three dimensional stochastic flow model is built providing a benchmark to the simulation of groundwater flow in highly heterogeneous aquifers. The performance of the presented NAS based DCM is verified in different dimensions using the method of manufactured solutions. We show that it significantly outperforms finite difference methods in both accuracy and computational cost.

KW - Deep learning

KW - Neural architecture search

KW - Error estimation

KW - Randomized spectral representation

KW - Method of manufactured solutions

KW - Log-normally distributed

KW - Physics-informed

KW - Sensitivity analysis

KW - Hyper-parameter optimization algorithms

KW - Transfer learning

KW - PARTIAL-DIFFERENTIAL-EQUATIONS

KW - GLOBAL SENSITIVITY-ANALYSIS

KW - FLOW SIMULATION

KW - RANDOM-FIELDS

KW - TRANSPORT

KW - ALGORITHM

KW - NETWORKS

KW - POROSITY

KW - LAW

UR - http://www.scopus.com/inward/record.url?scp=85123125774&partnerID=8YFLogxK

U2 - 10.1007/s00366-021-01586-2

DO - 10.1007/s00366-021-01586-2

M3 - Article

VL - 38

SP - 5173

EP - 5198

JO - Engineering with computers

JF - Engineering with computers

SN - 0177-0667

IS - 6

ER -