ARTIFICIAL INTELLIGENT FOR PREDICTION OF CONTINUOUS GAS LIFT PARAMETERS

9th International Multidisciplinary Scientific GeoConference - SGEM2009, www.sgem.org, SGEM2009 Conference Proceedings/ ISBN 10: 954-91818-1-2, June 14-19, 2009, Vol. 1, 571-578 pp

ABSTRACT

The purpose of this study has been assessment of capability of Artificial Neural
Networks in Gas Lift Operations optimization, which is in use for improved oil
recovery from oil wells. This in detail is in fact estimation of the two most important
parameters of the process, i.e. optimal injection depth and optimal gas injection rate. For
gas lift, gas is injected continuously or intermittently at selected location(s), resulting in
a reduction in the natural flowing gradient of the reservoir fluid, and thus reducing the
hydrostatic component of the pressure drop from the bottom to the top of the well. In
this article two Neural Network models are presented for prediction and optimization of
both the optimal injection depth and gas injection rate, using this methodology. For this
purpose four-layer neural networks have been designed and trained using real data of 36
wells. After the training step, four real data were also used for the model test step and as
a reliability check. The outputs of models for test data are compared with the wellflo
v3.6d software analysis. It has been concluded that Artificial Neural Networks approach
has an excellent competing capability for this purpose compared to the conventional
methods and can be used interchangeably. This methodology and design can
significantly help in the prompt optimal design of gas lift operations. This appears to be
the first report of applying ANNs to continuous gas lift optimization problem.

Keywords: Gas Lift, Optimization, Multi Phase Flow, Nodal Analysis, Oil, Artificial Neural Network