HR: 0830h
AN: S41B-09 [Abstracts]
TI: An Artificial Neural Network Method for Formation Permeability and Skin Coefficient With Wireline Formation Test Data
AU: * Gu, N
EM: guning7210@sina.com
AF: Petroleum University of China, School of Resources and Information, Beijing, Changping, 102249 China
AU: Tao, G
EM: taoguo@vip.sina.com
AF: Petroleum University of China, School of Resources and Information, Beijing, Changping, 102249 China
AU: Liu, S
EM: liushum@cnoocs.com
AF: China National Offshore Oil orporation,, Oilfield Services Limited, CNOOC,, Beijing, 101149 China
AB:
Wireline formation testing tools have undergone generations of evolutionary changes and have continued to advance in
formation evaluation techniques by mimicking a miniature well test. Compared with drill stem tester (DST), wireline formation testers have better acquisition ability, more detailed and reliable vertical pressure profiles, shorter test time, lower
cost and higher efficiency. With the development of new types of wireline formation testers, the flow rate can be controlled
and more information may be available. It is desirable to develop an efficient data interpretation model to determine
formation mobility and skin coefficients more accurately. In this paper we have developed a back propagation neural network
(BPNN) model to simulate the complex relation between pressure response of wireline formation testers and reservoir
parameters. The neural network model has been trained by a training set. Of the training average relative error in
permeability is 6.49%. The training results in dimensionless skin coefficients are also fairly accurate. These data shows
that the neural network model developed in this paper provides accurate mapping relation between pressure responses from
wireline formation testers and reservoir parameters, such as permeability and skin coefficient. To check extrapolated
ability of the neural network model, 178 sets of data are computed with random reservoir parameters. The predicting results
have demonstrated that the predicting average relative error in permeability is 7.75% and the errors in predicting results
of skin coefficients are less than 22%. The data interpretation examples demonstrate that the neural network model takes
full advantage of data in early to intermediate time to reduce test time. It can determine reservoir permeability and
dimensionless skin coefficient simultaneously without complex mathematics and get accurate results.
DE: 5100 PHYSICAL PROPERTIES OF ROCKS
DE: 5102 Acoustic properties
DE: 5144 Wave attenuation
DE: 5199 General or miscellaneous
SC: Seismology [S]
MN: 2005 Joint Assembly