A Surrogatebased Adaptive Sampling Approach for History Matching and Uncertainty Quantification
Abstract
A critical procedure in reservoir simulations is history matching (or data assimilation in a broader sense), which calibrates model parameters such that the simulation results are consistent with field measurements, and hence improves the credibility of the predictions given by the simulations. Often there exist nonunique combinations of parameter values that all yield the simulation results matching the measurements. For such illposed history matching problems, Bayesian theorem provides a theoretical foundation to represent different solutions and to quantify the uncertainty with the posterior PDF. Lacking an analytical solution in most situations, the posterior PDF may be characterized with a sample of realizations, each representing a possible scenario. A novel sampling algorithm is presented here for the Bayesian solutions to history matching problems. We aim to deal with two commonly encountered issues: 1) as a result of the nonlinear inputoutput relationship in a reservoir model, the posterior distribution could be in a complex form, such as multimodal, which violates the Gaussian assumption required by most of the commonly used data assimilation approaches; 2) a typical sampling method requires intensive model evaluations and hence may cause unaffordable computational cost. In the developed algorithm, we use a Gaussian mixture model as the proposalmore »
 Authors:
 Publication Date:
 Research Org.:
 Pacific Northwest National Lab. (PNNL), Richland, WA (United States)
 Sponsoring Org.:
 USDOE
 OSTI Identifier:
 1178496
 Report Number(s):
 PNNLSA106708
KJ0401000
 DOE Contract Number:
 AC0576RL01830
 Resource Type:
 Conference
 Resource Relation:
 Conference: Society of Petroleum Engineers 2015 Reservoir Simulation Symposium, February 2325, 2015, Houston, Texas, Paper No. SPE173298MS
 Country of Publication:
 United States
 Language:
 English
Citation Formats
Li, Weixuan, Zhang, Dongxiao, and Lin, Guang. A Surrogatebased Adaptive Sampling Approach for History Matching and Uncertainty Quantification. United States: N. p., 2015.
Web. doi:10.2118/173298MS.
Li, Weixuan, Zhang, Dongxiao, & Lin, Guang. A Surrogatebased Adaptive Sampling Approach for History Matching and Uncertainty Quantification. United States. doi:10.2118/173298MS.
Li, Weixuan, Zhang, Dongxiao, and Lin, Guang. 2015.
"A Surrogatebased Adaptive Sampling Approach for History Matching and Uncertainty Quantification". United States.
doi:10.2118/173298MS.
@article{osti_1178496,
title = {A Surrogatebased Adaptive Sampling Approach for History Matching and Uncertainty Quantification},
author = {Li, Weixuan and Zhang, Dongxiao and Lin, Guang},
abstractNote = {A critical procedure in reservoir simulations is history matching (or data assimilation in a broader sense), which calibrates model parameters such that the simulation results are consistent with field measurements, and hence improves the credibility of the predictions given by the simulations. Often there exist nonunique combinations of parameter values that all yield the simulation results matching the measurements. For such illposed history matching problems, Bayesian theorem provides a theoretical foundation to represent different solutions and to quantify the uncertainty with the posterior PDF. Lacking an analytical solution in most situations, the posterior PDF may be characterized with a sample of realizations, each representing a possible scenario. A novel sampling algorithm is presented here for the Bayesian solutions to history matching problems. We aim to deal with two commonly encountered issues: 1) as a result of the nonlinear inputoutput relationship in a reservoir model, the posterior distribution could be in a complex form, such as multimodal, which violates the Gaussian assumption required by most of the commonly used data assimilation approaches; 2) a typical sampling method requires intensive model evaluations and hence may cause unaffordable computational cost. In the developed algorithm, we use a Gaussian mixture model as the proposal distribution in the sampling process, which is simple but also flexible to approximate nonGaussian distributions and is particularly efficient when the posterior is multimodal. Also, a Gaussian process is utilized as a surrogate model to speed up the sampling process. Furthermore, an iterative scheme of adaptive surrogate refinement and resampling ensures sampling accuracy while keeping the computational cost at a minimum level. The developed approach is demonstrated with an illustrative example and shows its capability in handling the abovementioned issues. Multimodal posterior of the history matching problem is captured and are used to give a reliable production prediction with uncertainty quantification. The new algorithm reveals a great improvement in terms of computational efficiency comparing previously studied approaches for the sample problem.},
doi = {10.2118/173298MS},
journal = {},
number = ,
volume = ,
place = {United States},
year = 2015,
month = 2
}

A surrogatebased uncertainty quantification with quantifiable errors
Surrogate models are often employed to reduce the computational cost required to complete uncertainty quantification, where one is interested in propagating input parameters uncertainties throughout a complex engineering model to estimate responses uncertainties. An improved surrogate construction approach is introduced here which places a premium on reducing the associated computational cost. Unlike existing methods where the surrogate is constructed first, then employed to propagate uncertainties, the new approach combines both sensitivity and uncertainty information to render further reduction in the computational cost. Mathematically, the reduction is described by a range finding algorithm that identifies a subspace in the parameters space,more » 
An EPGPTbased approach for uncertainty quantification
Generalized Perturbation Theory (GPT) has been widely used by many scientific disciplines to perform sensitivity analysis and uncertainty quantification. This manuscript employs recent developments in GPT theory, collectively referred to as ExacttoPrecision Generalized Perturbation Theory (EPGPT), to enable uncertainty quantification for computationally challenging models, e.g. nonlinear models associated with many input parameters and many output responses and with general nonGaussian parameters distributions. The core difference between EPGPT and existing GPT is in the way the problem is formulated. GPT formulates an adjoint problem that is dependent on the response of interest. It tries to capture via the adjoint solution themore »