Bayesian optimization algorithm applied to uncertainty quantification

Asaad Abdollahzadeh, Alan Reynolds, Michael Andrew Christie, David Corne, Brian Davies, Glyn Williams

Research output: Contribution to journalArticle

Abstract

Prudent decision making in subsurface assets requires reservoir uncertainty quantification. In a typical uncertainty-quantification study, reservoir models must be updated using the observed response from the reservoir by a process known as history matching. This involves solving an inverse problem, finding reservoir models that produce, under simulation, a similar response to that of the real reservoir. However, this requires multiple expensive multiphase-flow simulations. Thus, uncertainty-quantification studies employ optimization techniques to find acceptable models to be used in prediction. Different optimization algorithms and search strategies are presented in the literature, but they are generally unsatisfactory because of slow convergence to the optimal regions of the global search space, and, more importantly, failure in finding multiple acceptable reservoir models. In this context, a new approach is offered by estimation-of-distribution algorithms (EDAs). EDAs are population-based algorithms that use models to estimate the probability distribution of promising solutions and then generate new candidate solutions.
This paper explores the application of EDAs, including univariate and multivariate models. We discuss two histogram-based univariate models and one multivariate model, the Bayesian optimization algorithm (BOA), which employs Bayesian networks for modeling. By considering possible interactions between variables and exploiting explicitly stored knowledge of such interactions, EDAs can accelerate the search process while preserving search diversity. Unlike most existing approaches applied to uncertainty quantification, the Bayesian network allows the BOA to build solutions using flexible rules learned from the models obtained, rather than fixed rules, leading to better solutions and improved convergence. The BOA is naturally suited to finding good solutions in complex high-dimensional spaces, such as those typical in reservoir-uncertainty quantification.
We demonstrate the effectiveness of EDA by applying the well-known synthetic PUNQ-S3 case with multiple wells. This allows us to verify the methodology in a well-controlled case. Results show better estimation of uncertainty when compared with some other traditional population-based algorithms.
Original languageEnglish
Pages (from-to)865-873
Number of pages9
JournalSPE Journal
Volume17
Issue number3
DOIs
Publication statusPublished - Sep 2012

Fingerprint

Bayesian networks
Uncertainty
Multiphase flow
Flow simulation
Inverse problems
Probability distributions
Decision making

Cite this

Abdollahzadeh, A., Reynolds, A., Christie, M. A., Corne, D., Davies, B., & Williams, G. (2012). Bayesian optimization algorithm applied to uncertainty quantification. SPE Journal, 17(3), 865-873. https://doi.org/10.2118/143290-PA
Abdollahzadeh, Asaad ; Reynolds, Alan ; Christie, Michael Andrew ; Corne, David ; Davies, Brian ; Williams, Glyn. / Bayesian optimization algorithm applied to uncertainty quantification. In: SPE Journal. 2012 ; Vol. 17, No. 3. pp. 865-873.
@article{891dda959ba846b8b875b209b504ccad,
title = "Bayesian optimization algorithm applied to uncertainty quantification",
abstract = "Prudent decision making in subsurface assets requires reservoir uncertainty quantification. In a typical uncertainty-quantification study, reservoir models must be updated using the observed response from the reservoir by a process known as history matching. This involves solving an inverse problem, finding reservoir models that produce, under simulation, a similar response to that of the real reservoir. However, this requires multiple expensive multiphase-flow simulations. Thus, uncertainty-quantification studies employ optimization techniques to find acceptable models to be used in prediction. Different optimization algorithms and search strategies are presented in the literature, but they are generally unsatisfactory because of slow convergence to the optimal regions of the global search space, and, more importantly, failure in finding multiple acceptable reservoir models. In this context, a new approach is offered by estimation-of-distribution algorithms (EDAs). EDAs are population-based algorithms that use models to estimate the probability distribution of promising solutions and then generate new candidate solutions. This paper explores the application of EDAs, including univariate and multivariate models. We discuss two histogram-based univariate models and one multivariate model, the Bayesian optimization algorithm (BOA), which employs Bayesian networks for modeling. By considering possible interactions between variables and exploiting explicitly stored knowledge of such interactions, EDAs can accelerate the search process while preserving search diversity. Unlike most existing approaches applied to uncertainty quantification, the Bayesian network allows the BOA to build solutions using flexible rules learned from the models obtained, rather than fixed rules, leading to better solutions and improved convergence. The BOA is naturally suited to finding good solutions in complex high-dimensional spaces, such as those typical in reservoir-uncertainty quantification. We demonstrate the effectiveness of EDA by applying the well-known synthetic PUNQ-S3 case with multiple wells. This allows us to verify the methodology in a well-controlled case. Results show better estimation of uncertainty when compared with some other traditional population-based algorithms.",
author = "Asaad Abdollahzadeh and Alan Reynolds and Christie, {Michael Andrew} and David Corne and Brian Davies and Glyn Williams",
year = "2012",
month = "9",
doi = "10.2118/143290-PA",
language = "English",
volume = "17",
pages = "865--873",
journal = "SPE Journal",
issn = "1086-055X",
publisher = "Society of Petroleum Engineers",
number = "3",

}

Abdollahzadeh, A, Reynolds, A, Christie, MA, Corne, D, Davies, B & Williams, G 2012, 'Bayesian optimization algorithm applied to uncertainty quantification', SPE Journal, vol. 17, no. 3, pp. 865-873. https://doi.org/10.2118/143290-PA

Bayesian optimization algorithm applied to uncertainty quantification. / Abdollahzadeh, Asaad; Reynolds, Alan; Christie, Michael Andrew; Corne, David; Davies, Brian; Williams, Glyn.

In: SPE Journal, Vol. 17, No. 3, 09.2012, p. 865-873.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Bayesian optimization algorithm applied to uncertainty quantification

AU - Abdollahzadeh, Asaad

AU - Reynolds, Alan

AU - Christie, Michael Andrew

AU - Corne, David

AU - Davies, Brian

AU - Williams, Glyn

PY - 2012/9

Y1 - 2012/9

N2 - Prudent decision making in subsurface assets requires reservoir uncertainty quantification. In a typical uncertainty-quantification study, reservoir models must be updated using the observed response from the reservoir by a process known as history matching. This involves solving an inverse problem, finding reservoir models that produce, under simulation, a similar response to that of the real reservoir. However, this requires multiple expensive multiphase-flow simulations. Thus, uncertainty-quantification studies employ optimization techniques to find acceptable models to be used in prediction. Different optimization algorithms and search strategies are presented in the literature, but they are generally unsatisfactory because of slow convergence to the optimal regions of the global search space, and, more importantly, failure in finding multiple acceptable reservoir models. In this context, a new approach is offered by estimation-of-distribution algorithms (EDAs). EDAs are population-based algorithms that use models to estimate the probability distribution of promising solutions and then generate new candidate solutions. This paper explores the application of EDAs, including univariate and multivariate models. We discuss two histogram-based univariate models and one multivariate model, the Bayesian optimization algorithm (BOA), which employs Bayesian networks for modeling. By considering possible interactions between variables and exploiting explicitly stored knowledge of such interactions, EDAs can accelerate the search process while preserving search diversity. Unlike most existing approaches applied to uncertainty quantification, the Bayesian network allows the BOA to build solutions using flexible rules learned from the models obtained, rather than fixed rules, leading to better solutions and improved convergence. The BOA is naturally suited to finding good solutions in complex high-dimensional spaces, such as those typical in reservoir-uncertainty quantification. We demonstrate the effectiveness of EDA by applying the well-known synthetic PUNQ-S3 case with multiple wells. This allows us to verify the methodology in a well-controlled case. Results show better estimation of uncertainty when compared with some other traditional population-based algorithms.

AB - Prudent decision making in subsurface assets requires reservoir uncertainty quantification. In a typical uncertainty-quantification study, reservoir models must be updated using the observed response from the reservoir by a process known as history matching. This involves solving an inverse problem, finding reservoir models that produce, under simulation, a similar response to that of the real reservoir. However, this requires multiple expensive multiphase-flow simulations. Thus, uncertainty-quantification studies employ optimization techniques to find acceptable models to be used in prediction. Different optimization algorithms and search strategies are presented in the literature, but they are generally unsatisfactory because of slow convergence to the optimal regions of the global search space, and, more importantly, failure in finding multiple acceptable reservoir models. In this context, a new approach is offered by estimation-of-distribution algorithms (EDAs). EDAs are population-based algorithms that use models to estimate the probability distribution of promising solutions and then generate new candidate solutions. This paper explores the application of EDAs, including univariate and multivariate models. We discuss two histogram-based univariate models and one multivariate model, the Bayesian optimization algorithm (BOA), which employs Bayesian networks for modeling. By considering possible interactions between variables and exploiting explicitly stored knowledge of such interactions, EDAs can accelerate the search process while preserving search diversity. Unlike most existing approaches applied to uncertainty quantification, the Bayesian network allows the BOA to build solutions using flexible rules learned from the models obtained, rather than fixed rules, leading to better solutions and improved convergence. The BOA is naturally suited to finding good solutions in complex high-dimensional spaces, such as those typical in reservoir-uncertainty quantification. We demonstrate the effectiveness of EDA by applying the well-known synthetic PUNQ-S3 case with multiple wells. This allows us to verify the methodology in a well-controlled case. Results show better estimation of uncertainty when compared with some other traditional population-based algorithms.

U2 - 10.2118/143290-PA

DO - 10.2118/143290-PA

M3 - Article

VL - 17

SP - 865

EP - 873

JO - SPE Journal

JF - SPE Journal

SN - 1086-055X

IS - 3

ER -

Abdollahzadeh A, Reynolds A, Christie MA, Corne D, Davies B, Williams G. Bayesian optimization algorithm applied to uncertainty quantification. SPE Journal. 2012 Sep;17(3):865-873. https://doi.org/10.2118/143290-PA