Summary
Using actual field cases, a neural network model was developed to identify
candidate wells and predict well performance for water shutoff treatments using
polymer gels. A feed forward back propagation algorithm was used to develop the
neural networks. The before and after treatment data for 22 wells treated with
polymer gels in the Arbuckle formation in central Kansas were used to train and
verify the neural networks.
Polymers and gels have been used extensively in field applications to
suppress excess water production and improve oil productivity. Field experience
has demonstrated that candidate-well selection is critical to the success of
gel-polymer treatments. To date, most candidate-well selections are on the
basis of anecdotal screening guidelines, which often results in inconsistent
treatment outcomes. With only pretreatment well data as input parameters, the
neural networks developed in this work can accurately predict the
post-treatment cumulative oil production of the well one month after treatment
with an average error of 16%, and the post-treatment cumulative oil production
three months after treatment with an average error of 10%. This is a dramatic
improvement over the current method of using anecdotal screening guidelines for
candidate-well selections.
This method represents a major breakthrough where the candidate selection
can now be on the basis of the accurate predictions of treatment outcomes
without having to use complicated reservoir models to simulate the well
performance after treatment.
Introduction
Excess water production is a major issue in oil field operations worldwide,
currently averaging three barrels of water for each barrel of oil produced
(Bailey et al. 2000). The situation is even worse in the US where more than
seven barrels of water are produced for each barrel of oil (EPA 1999). The
annual cost of treating and disposal of this water is estimated to be USD 40
billion (Bailey et al. 2000). Water shutoff and conformance control, therefore,
represents a significant financial and environmental challenge/incentive for
the petroleum industry. Polymers and gels have been used extensively in
field applications to suppress excess water production and improve oil
productivity (Seright et al. 2003). Field experience has demonstrated that
candidate-well selection is critical to the success of gel-polymer treatments
(Seright et al. 2003). To date, most candidate-well selections are on the basis
of anecdotal screening guidelines, which often results in inconsistent
treatment outcomes (Seright et al. 2003).
Reservoir simulation can potentially be used as a screening tool to predict
the post-treatment performance of a candidate well using the pre-treatment
historical data (Barati et al. 2006). However, this method is usually expensive
and also requires extensive knowledge of the target reservoir (including the
rock and fluid properties) the historical production data, and the geological
reservoir model. Unfortunately, they are not always available for older
reservoirs where the well records are often incomplete or lost (Barati et al.
2006).
Another method that has been investigated is to correlate the historical
pre- and post-treatment performance data of the wells treated with polymer gels
in the target reservoir. In this method, multivariate analysis is used to
correlate the post-treatment performance of the treated wells with the
pre-treatment data such as the geographical location of the wells, the depth of
the wells, and the production history of the wells, and so on. The correlation
could then be used as a predictive tool for candidate selection in the target
reservoir. The problem with this method is that the physical processes involved
in gel polymer treatment downhole are too complex to be accurately represented
by correlations generated between pre- and post-treatment data using
multivariate analysis (Alhajeri et al. 2006).
Prediction of the performance of a well after treatment, using the
pre-treatment data, is a pattern recognition problem. Neural networks have
shown great capabilities in solving pattern recognition problems (Ali 1994;
Mohaghegh 1995; Ahmed et al. 1997). The objective of this study is to develop a
methodology using neural networks to identify candidate wells on the basis of
the predicted outcomes for gel-polymer treatments. The before and after
treatment data for 22 wells treated with polymer gels in the Arbuckle formation
in central Kansas were used to develop the neural networks (Saeedi 2005).
Arbuckle Formation
The Arbuckle formation is the main oil producer in Kansas, responsible for
approximately 36% (~2.2 billion barrels) of the total produced oil in Kansas
(Franseen et al. 1999) (see Fig. 1). Arbuckle reservoirs are
fracture-controlled karstic reservoirs with porosity and permeability
influenced by basement structural patterns and subaerial exposures. The
subaerial exposure has resulted in weathering and secondary dissolution of the
upper beds of the Arbuckle. It is believed that these processes have
significantly increased porosity and permeability and created petroleum
reservoirs in these strata. (Franseen et al. 1999). Shallow-shelf dolomites
predominantly constitute the Arbuckle formation. Porosity of the Arbuckle
reservoirs is enhanced by the dolomitization process (Franseen et al. 1999).
Most of the Arbuckle’s oil and gas zones are perforated in the top 25 ft of the
Arbuckle, while some are perforated at a depth of 25 to 50 ft within the
formation (Franseen et al. 1999). High initial oil productivity, a rapid
decline in oil production rate, and the production of large amounts of water at
high water to oil ratios (WOR) are characteristics of Arbuckle wells (Willhite
and Pancake 2004). On the basis of these characteristics, Arbuckle reservoirs
have been visualized as a column of oil on top of a strong aquifer. To prevent
water coning, most of the wells in the Arbuckle were drilled relatively shallow
into the formation(less than 10 ft) and completed open hole (Franseen et al.
1999). Because of the absence of field cores and the lack of well log data for
the full productive intervals, the reservoir characteristics of the Arbuckle
formation are not well understood.
© 2007. Society of Petroleum Engineers
View full textPDF
(
1,507 KB
)
History
- Original manuscript received:
2 June 2006
- Meeting paper published:
11 September 2006
- Revised manuscript received:
27 November 2006
- Manuscript approved:
8 February 2007
- Version of record:
20 November 2007
View Cart
