Guided Bayesian optimal experimental design

Khodja, Mohamed R.; Prange, Michael; Djikpesse, Hugues

doi:10.1088/0266-5611/26/5/055008

Cited by 32 publications

(26 citation statements)

References 29 publications

(47 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The utility of the D N criterion lies in the fact that efficient sequential design algorithms from linearized SED (e.g., Curtis et al, 2004;Stummer et al, 2004;Coles and Morgan, 2009;Coles and Curtis, 2011;Khodja et al, 2010) exist for D optimization (the A N , E N , and T N criteria remarked upon above could also be maximized/minimized using these algorithms). When combined with a linearized sequential design algorithm (LSDA), the D N criterion renders nonlinear Bayesian design computationally feasible for large-scale industrial applications, a feature shared by no other geoscientific nonlinear design technique without recourse to cluster computing or reparameterization of the design space (e.g., Ajo-Franklin, 2009;Guest andCurtis, 2009, 2010).…”

Section: Linearized Sequential Designmentioning

confidence: 99%

See 1 more Smart Citation

Efficient nonlinear Bayesian survey design usingD_Noptimization

Coles

Curtis

2011

GEOPHYSICS

View full text Add to dashboard Cite

A new method for fully nonlinear, Bayesian survey design renders the optimization of industrial-scale geoscientific surveys as a practical possibility. The method, D N optimization, designs surveys to maximally discriminate between different possible models. It is based on a generalization to nonlinear design problems of the D criterion (which is for linearized design problems). The main practical advantage of D N optimization is that it uses efficient algorithms developed originally for linearized design theory, resulting in lower computing and storage costs than for other nonlinear Bayesian design techniques. In a real example in which we optimized a seafloor microseismic sensor network to monitor a fractured petroleum reservoir, we compared D N optimization with two other networks: one proposed by an industrial contractor and one optimized using a linearized Bayesian design method. Our technique yielded a network with superior expected data quality in terms of reduced uncertainties on hypocenter locations.

show abstract

Section: Linearized Sequential Designmentioning

confidence: 99%

“…The covariance matrices are with respect to the set of all candidate observation points N. This is by convention because very efficient LSDAs exist (Coles and Morgan, 2009;Khodja et al, 2010;Coles and Curtis, 2011) that require RðNÞ as an input (they actually require ½RðNÞ 1=2 ).…”

Section: Design Workflowmentioning

confidence: 99%

Efficient nonlinear Bayesian survey design usingD_Noptimization

Coles

Curtis

2011

GEOPHYSICS

View full text Add to dashboard Cite

show abstract

“…To complement the extensive work done in this area, we have proposed an efficient Bayesian greedy OED algorithm (Khodja et al, 2010) that aims to construct an experiment by sequentially minimizing the determinant of the forecast posterior model covariance matrix, i.e., by maximizing the posterior Shannon information on the model. This design methodology not only takes advantage of the available prior information both on the model and on the data, but it also allows the treatment of large OED problems.…”

Section: Introductionmentioning

confidence: 99%

“…Here we briefly review the Bayesian design methodology presented in Khodja et al (2010), and Djikpesse et al (2012a). For more details the reader is referred to these two references.…”

Section: Introductionmentioning

confidence: 99%

Accounting for seismic radiation anisotropy in Bayesian survey designs

2016

View full text Add to dashboard Cite

The seismic radiation patterns associated with probing the earth's subsurface are essentially anisotropic due to its ubiquitous stratified structure. This anisotropy seriously complicates formation imaging and data acquisition. This is most salient for deep-water subsalt reservoirs. Traditionally, point scatterers with isotropic radiation patterns are used in migration imaging, but in the survey design problem, these might lead to design errors caused by receivers being placed in poor locations with respect to the radiation pattern of the scattering structure. Here, we extend a framework which accounts for anisotropy in the scattered radiation for optimal geophysical survey design purposes. The propagation medium is assumed to be attenuative. The locally dipping interfaces are modeled as a discrete set of finite-size planar scattering elements. The general elastodynamic expressions for the sensitivity kernels, i.e., the vectors which mathematically represent the candidate observations, in the presence of the scattering elements are provided. The size of each element controls the width of its radiation pattern, which may in turn be used to characterize the uncertainty on the dip angle, thus complementing the information provided by the model-parameter uncertainties and ultimately leading to better geophysical survey designs.

show abstract

“…Thus, to obtain a time-lapse image of a reservoir without collecting or imaging a large data set, one needs to know what data collected on the surface (or in wells) contribute most to the reconstructed image of the reservoir region. Typically, an optimal survey design together with an illumination analysis (e.g., Curtis, 1999;van den Berg and Curtis, 2003;Khodja et al, 2010) is performed to optimize seismic acquisition before the actual data collection. Ray-based methods are conventionally used for illumination analysis (Bear et al, 2000).…”

Section: Introductionmentioning

confidence: 99%

Data-driven estimation of the sensitivity of target-oriented time-lapse seismic imaging to source geometry

2013

View full text Add to dashboard Cite

The goal of time-lapse imaging is to identify and characterize regions in which the earth's material properties have changed between surveys. This requires an effective deployment of sources and receivers to monitor the region where changes are anticipated. Because each source adds to the acquisition cost, we should ensure that only those sources that best image the target are collected and used to form an image of the target region. This study presents a datadriven approach that estimates the sensitivity of target-oriented imaging to source geometry. The approach is based on the propagation of the recorded baseline seismic data backward in time through the entire medium and coupling it with the estimated perturbation in the subsurface. We test this approach using synthetic surface seismic and time-lapse VSP field-data from the SACROC field. These tests show that the use of the baseline seismic data enhances the robustness of the sensitivity estimate to errors, and can be used to select data that best image a target zone, thus increasing the signal-tonoise ratio of the image of the target region and reducing the cost of time-lapse acquisition, processing, and imaging.

show abstract

Guided Bayesian optimal experimental design

Cited by 32 publications

References 29 publications

Efficient nonlinear Bayesian survey design usingD_Noptimization

Efficient nonlinear Bayesian survey design usingD_Noptimization

Accounting for seismic radiation anisotropy in Bayesian survey designs

Data-driven estimation of the sensitivity of target-oriented time-lapse seismic imaging to source geometry

Contact Info

Product

Resources

About

Guided Bayesian optimal experimental design

Cited by 32 publications

References 29 publications

Efficient nonlinear Bayesian survey design usingDNoptimization

Efficient nonlinear Bayesian survey design usingDNoptimization

Accounting for seismic radiation anisotropy in Bayesian survey designs

Data-driven estimation of the sensitivity of target-oriented time-lapse seismic imaging to source geometry

Contact Info

Product

Resources

About

Efficient nonlinear Bayesian survey design usingD_Noptimization

Efficient nonlinear Bayesian survey design usingD_Noptimization