Show simple item record

dc.contributor.advisorOkuno, Ryosuke, 1974-
dc.creatorHernandez Mejia, Jose Luis
dc.date.accessioned2019-11-27T17:57:47Z
dc.date.available2019-11-27T17:57:47Z
dc.date.created2019-08
dc.date.issued2019-09-17
dc.date.submittedAugust 2019
dc.identifier.urihttps://hdl.handle.net/2152/78612
dc.identifier.urihttp://dx.doi.org/10.26153/tsw/5668
dc.description.abstractCompositional reservoir simulation is widely used as an important tool for optimization of enhanced oil recovery processes. In compositional reservoir simulation, flash calculations are performed to solve for phase properties and amounts for each grid-block and each time step by use of a cubic equation of state (EOS). EOS flash calculation is one of the most time-consuming operations during compositional reservoir simulation. There has been a critical need for more efficient EOS flash for practical compositional reservoir simulation. The central idea tested in this thesis is to use artificial neural networks (ANNs) to replace the most fundamental, but time-consuming portion of EOS flash; that is, the evaluation of fugacity coefficients. ANNs are used for efficient feedforward approximation of the EOS fugacity coefficient function with a series of weights, bias, and activation functions. A set of weights and bias is found by using an algorithm that minimizes the mean squared error between the predicted and real values. This type of approximation is called supervised learning in machine learning applications. The thermodynamic model used is the Peng – Robinson equation of state with the van der Waals mixing rules and solved by the successive substitution algorithm for flash calculations. The implementation of the ANN-based fugacity coefficient function is straightforward because it only replaces the EOS-based fugacity coefficient in conventional flash calculation algorithms. Once an ANN-based fugacity coefficient function is built based on a cubic EOS, the EOS is required only when phase densities are calculated, usually at the final convergence. That is, ANN-based flash does not use an EOS during the iterative solution. We show comparisons between the conventional EOS flash calculations and the ANN flash calculations in terms of computational efficiency. Use of ANN flash can reduce on average 89.83% of the time needed by the conventional EOS flash for the cases studied in this thesis.
dc.format.mimetypeapplication/pdf
dc.language.isoen
dc.subjectFlash calculations
dc.subjectArtificial neural networks
dc.subjectMachine learning
dc.subjectEOR
dc.subjectReservoir simulation
dc.titleApplication of artificial neural networks for rapid flash calculations
dc.typeThesis
dc.date.updated2019-11-27T17:57:49Z
dc.description.departmentPetroleum and Geosystems Engineering
thesis.degree.departmentPetroleum and Geosystems Engineering
thesis.degree.disciplinePetroleum Engineering
thesis.degree.grantorThe University of Texas at Austin
thesis.degree.levelMasters
thesis.degree.nameMaster of Science in Engineering
dc.creator.orcid0000-0001-9511-9669
dc.type.materialtext


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record