--> Abstract: Quantifying Uncertainty of 2-D Fluid Flow Modeling Results by Using Methods of Experimental Design, by J. Wendebourg; #90937 (1998).

Datapages, Inc.Print this page

Abstract: Quantifying Uncertainty of 2-D Fluid Flow Modeling Results by Using Methods of Experimental Design

WENDEBOURG, JOHANNES, Institut Francais du Petrole, Rueil- Malmaison, France

Multidimensional fluid flow modeling for exploration purposes usually has a well defined objective such as “is charge into prospect location X possible” or “what is the sealing potential at prospect location Y”. Calibration data are oftentimes available from wells such as maturity data, temperatures, fluid pressures, GOR etc. The aim of fluid flow modeling is then to give a predictive answer conditioned to the available data, and ideally to give a range of possible answers, or in other words, the uncertainty of the prediction.

Generally, in a 2-D fluid flow modeling study, a cross section is calibrated fitting all available data by adjusting input parameters to the model. Sometimes a sensitivity analysis is also performed on parameters that are deemed to be important to the objective of the study. However, such a sensitivity analysis is usually not done systematically because individual 2-D models may take hours to run. This paper proposes to use methods of experimental design to do a systematic sensitivity analysis with a minimum of runs because it can determine the most important model parameters as well as the uncertainty of the modeling results.

Methods of experimental design are used in the industrial world for designing, developing, and optimizing processes. They are well adapted for situations where many input parameters potentially influence the performance or the quality of the outcome. Input to the design functions of fluid flow modeling are independent parameters such as permeability, capillary pressure, viscosity etc. but they also may include boundary conditions or even alternative model formulations. The design algorithm will find those parameter combinations that cover the parameter space with as few model runs as possible. Sensitive parameters are determined and non-sensitive parameters may be discarded.

Model results are represented in a simplified form as a linear or quadratic response surface that can be used as a proxy for the complex actual fluid flow model. Optimization problems such as “what are the parameter ranges that maximize saturation in prospect location X” and uncertainty problems such as “what is the uncertainty distribution of the predicted pressure in location Y” can be solved based on the simplified proxy without having to invoke additional fluid flow model runs.

It is important to note that this method is problem dependent, not model dependent and can therefore be applied to any fluid flow problem using any flow simulator. In the poster, synthetic and real-world examples will be shown using IFP's basin modeling software to demonstrate the use of this method.

AAPG Search and Discovery Article #90937©1998 AAPG Annual Convention and Exhibition, Salt Lake City, Utah