Summary
Experimental design and analysis (EDA) methods can be used practically to
minimize the number of 3D geological models that must be built to capture and
assess the significant effects in multiple deterministic (or scenario)
modeling. This study investigates the feasibility of EDA methods by using three
examples. It includes discussions and guidelines on how to select efficient
design matrices by using expert knowledge (the possible effects of an
experiment) and a decision tree, and how the experimental response can be
fitted accurately with the response surface method to develop a good surrogate
equation.
Introduction
EDA methods have been shown in the literature to have significant potential
in recoverable reserves uncertainty studies. For example: screening and
sensitivity studies in recoverable reserves1–4 and in history matching5;
production forecasting and estimating ultimate recovery (UR) curves1–3,6–13;
and field development optimization.14,15 In these studies, a design matrix is
used to obtain the experimental response (i.e., UR). A surrogate equation,
which is in the form of a simple mathematical function (often with
nonlinearities), is then developed to replace the experiment (3D numerical
reservoir simulator). The challenge is to generate an accurate surrogate
equation using a design matrix with a small number of design runs. In this
study, it is found that expert knowledge can be used effectively to achieve
this objective.
This study shows how the EDA methods should be used in multiple
deterministic (or scenario) modeling16 to study the hydrocarbon in-place volume
(VHCIP) of a reservoir. This is important especially during the exploration or
early appraisal stage, where the amount of data is not sufficient for
meaningful 3D numerical reservoir simulations. Multiple deterministic modeling
is being used more frequently as higher-risk marginal fields are developed.
Theoretically, it is better than a probabilistic approach (e.g., Monte Carlo
simulation17) in the investigation of VHCIP because it is based on a geological
representation of the reservoir, which can be used for field development
planning and the like. However, it is not practical because a large number of
models must be built to generate a VHCIP distribution curve (similar to that
derived from the probabilistic approach).
© 2005. Society of Petroleum Engineers
View full textPDF
(
773 KB
)
History
- Original manuscript received:
4 June 2003
- Revised manuscript received:
5 April 2005
- Manuscript approved:
18 April 2005
- Version of record:
15 September 2005
View Cart
