BISC Seminar, 3 October 1996, 4-5:00pm, 310 Soda Hall

Michael Lee (
Wed, 2 Oct 1996 10:58:42 +0100

Fieldwise Reservoir Management:
Integration of Reservoir Simulator, Neural Network, and Fuzzy Logic

BISC Seminar

Dr. Masoud Nikravesh
BISC Postdoctoral Researcher
Earth Sciences Division of Lawrence Berkeley National Laboratory and

Department of Materials Science and Mineral Engineering,
University of California, Berkeley, CA 94720

3 October 1996
310 Soda Hall


Fluids, such as water, carbon dioxide or steam, are injected into
reservoirs to maintain pressure and displace oil. The process of fluid
injection into petroleum reservoirs is known to exhibit an inherently
complex, nonlinear, time varying and nonstationary behavior.

Optimally, fluid injection policy should be managed to produce oil as
quickly and efficiently as possible, without damaging the reservoir
excessively. However, translating this fieldwise goal into a practical
operating procedure for each well is a very difficult and tedious task and
state-of-the-art understanding of fluid movement in low permeability
rock systems is insufficient for the design and operation of large
waterfloods and steamdrives. In this study, we focus on water and
steam injection into low-permeability diatomaceous reservoirs because a
significant volume of crude oil remains in them. Nevertheless, this
methodology is applicable directly to fluid injection into other tight
fractured reservoirs such as the Austin Chalk and the West Texas

Here we present the second generation of "intelligent" oil field
surveillance and prediction software based on Reservoir Simulator,
neural networks, and fuzzy logic. Despite our incomplete knowledge,
neural network models can match and then predict complicated reservoir
behavior when historical databases are available. They are capable of
making accurate predictions even when all mechanisms affecting
injection or production behavior are not known. Further, neural
networks and fuzzy logic provide a way to incorporate disparate
information because a structural relationship between input and output
data is not required.

In this study, we examine and compare specifically two waterflood
projects. First, we examine a waterflood project (Lost Hills I) in
Section 2 of the Lost Hills Diatomite (Kern County, California),
operated by Mobil E&P U.S. The Lost Hills I waterflood has 123
producers and 48 injectors. The data set includes 5 years of historical
injection and production rate data collected at 1 to 10-day intervals.
Second, we examine a waterflood project on the Dow-Chanslor lease in
the Middle Belridge Diatomite (Kern County, California), operated by
Crutcher-Tufts. The Dow-Chanslor waterflood has more than 125
producers and 36 injectors. The data set includes 10 years of historical
injection and production rate data collected at 30 day intervals. In both
projects, neural networks and fuzzy logic are used to divide the oil field
into regions with similar characteristic behavior. The model helps to
improve waterflood management, avoid reservoir damage, and increase
oil recovery per unit volume of injected water. Finally, the model
visualizes the global trajectory of an entire project and allows engineers
to recognize patterns of incipient reservoir damage and poor
performance. The diatomaceous fields of California, which hold an
estimated 10 billion bbl of oil-in-place, have been chosen to
demonstrate the power of neural networks and fuzzy logic to optimize
fluid injection policy.

Michael A. Lee
Post Doctoral Researcher
Berkeley Initiative in Soft Computing
387 Soda Hall                                      Tel: +1-510-642-9827
Computer Science Division                          Fax: +1-510-642-5775
University of California                    Email:
Berkeley, CA 94720-1776 USA       WWW: