Print Email Facebook Twitter Non-Parametric Bayesian Networks (NPBNs) versus Ensemble Kalman Filter (EnKF) in Reservoir Simulation with non-Gaussian Measurement Noise Title Non-Parametric Bayesian Networks (NPBNs) versus Ensemble Kalman Filter (EnKF) in Reservoir Simulation with non-Gaussian Measurement Noise Author Zilko, A.A. Contributor Hanea, A. (mentor) Faculty Electrical Engineering, Mathematics and Computer Science Department Applied mathematics Date 2012-08-15 Abstract Lately, the objective of reservoir engineering is to optimize hydrocarbon recovery from a reservoir. To achieve that goal, a good knowledge of the subsurface properties is crucial. The author is concerned with estimating one of the properties of the field: the permeability of a reservoir. To characterize the fluid flow, a two phase (oil-water) 2D model represented as a system of coupled nonlinear partial differential equations which is unsolvable analytically is used. Ensemble Kalman Filter (EnKF) is the most common tool used to deal with this situation. However, it is not the only way. Recently, a research on a more general approach based on a dynamic Bayesian network using the Non-Parametric Bayesian Networks (NPBNs) has been initiated. This research, which uses twin experiment, indicates that the NPBN approach appears to be a promising alternative to EnKF. However, a number of open questions emerge from this initial research. The first one is the normality assumption for the noise used in the measurements generation in the twin experiment. Even though Gaussian noise for measurements is sensible in the sense that the knowledge about the noise is unavailable, it does not mean that other noise from different distributions cannot be applied. The second one is the exclusion of saturation in the NPBN approach performed in the previous research. This may result in the loss of valuable information. Further, the previous research discovers that NPBN approach seems to work well in recovering only part of the reservoir. The entire permeability field may be approximated by means of interpolation between several approximated parts of the field. Hence, the third question relates to an interpolation method that may be used in recovering the permeability of the entire reservoir. This project aims to experiment on these three key points of interest. A fourth objective, however, is surfaced during the analysis, which is to use an alternative measure of performance to the well-known Root Mean Square Error (RMSE). Along the way, the performance of both EnKF and NPBN are going to be observed and compared one more time. Subject Ensemble Kalman FilterNon-Parametric Bayesian NetworksParameter EstimationReservoir EngineeringReservoir SimulationBayesian Networks To reference this document use: http://resolver.tudelft.nl/uuid:97469901-4724-4c0b-851e-0eefb100adc8 Part of collection Student theses Document type master thesis Rights (c) 2012 Zilko, A.A. Files PDF Thesis_Report_AZilko.pdf 4.62 MB Close viewer /islandora/object/uuid:97469901-4724-4c0b-851e-0eefb100adc8/datastream/OBJ/view