Prediction of Grain-Size Distribution in Clastic Reservoirs for Gravel Pack Design
| Abraham Tativ Faga (afaga@petroleumjournals.com) |
| School of Engineering, Robert Gordon University, Aberdeen |
| April, 2001 |
| |
Dr. Faga has over 16 years experience in drilling and well completion operations. He has worked for Shell, C-FER Technologies, and Schlumberger where he held various engineering, operations, research and commercial positions around the world. He has delivered significant success in numerous onshore and offshore oil and gas field developments including in Canada, Nigeria, Algeria, UK and Russia. Dr. Faga is the Managing Editor of Petroleum Journals Online Ltd. (the publisher of the first fully refereed e journals of petroleum engineering).
He holds a PhD from Robert Gordon University, Aberdeen, and a Bachelor of Engineering degree in petroleum engineering from University of Benin. He is a chartered engineer (UK) and is registered with the Association of Professional Engineers, Geologists and Geophysicists of Alberta (APEGGA). |
| |
Abstract |
Gravel packing and screen completions, the most popular mechanical techniques of sand control, have sustained oil and gas production in wells that would otherwise have been shut in. When properly designed and installed gravel packs can prove to be both cost effective and reliable, yielding long-term performance. However, poor practices especially with respect to gravel pack design are common and have led to cases of poor well productivity and early pack failure. Selection of gravel-size is one of the key factors in gravel pack design. Proper gravel-size selection is carried out in relation to formation grain-size distribution obtained by sieving conventional cores but the data is often not available primarily due to cost. In the absence of well specific data, it has become accepted practice to use either offset data or other less representative alternative sources of data such as bailed samples and produced samples creating a potential for ineffective gravel packs. This thesis presents the results of a research aimed at developing a method for the prediction of grain-size distribution in clastic reservoirs.
The scope of this dissertation covers the identification of a prediction technique and the investigation of its suitability for grain-size prediction using well logs. It also includes a study of the feasibility of grain-size prediction in diagenetically modified formations and use of a locally trained prediction model in a different depositional environment. The additional scope was defined to assess the effects of factors affecting the natural variability between textural properties and well log responses.
A review of literature identified the relationship between the response of the gamma ray tool and grain-size and the relationship between porosity, permeability and texture as good theoretical basis for a quantitative approach to grain-size prediction using the pattern recognition capabilities of backpropagation neural networks. This theoretical basis is further strengthened by the possibility of qualitatively deriving textural attributes such as sorting, packing, grain shape and level of consolidation from gamma ray, resistivity and the porosity logs. The selection of a backpropagation neural network methodology was additionally justified on the basis of its reported superiority over conventional linear and non-linear regression techniques in dealing with complex non-linear relationships such as those existing between petrophysical properties.
Backpropagation neural networks were found to be both capable of learning the petrophysical patterns in the data presented and generalising grain-size distribution when blind-tested with new data. The networks performed well in the diagenetically modified Brent Group. The study demonstrated that a network trained using data from the fluvio-deltaic depositional environment of the Brent Group could be applied for prediction in the marine environment of the Statfjord sands which have mean grain-sizes that are higher than those of the Brent by at least 200 microns. The study also showed that gravel selected using well specific grain-size predicted across an entire completion interval could be superior to that selected using discontinuous and scanty offset data and could result in better well productivity and longevity.
Other areas of potential application of the predicted grain-size were discussed and include the mapping of its spatial variation for the study of depositional models and the spatial variation of reservoir quality. The potential benefits of these applications include the possible identification of previously undiscovered stratigraphic traps within larger fields. |
|
