Improved neural-network model predicts dewpoint pressure of retrograde gases
Academic Article
Overview
Research
Identity
Additional Document Info
Other
View All
Overview
abstract
Accurate prediction of dewpoint pressure is a critical element in reservoir-engineering calculations. The objective of this paper is to present a novel and highly accurate application of the neural-network model (NNM) to predict dewpoint pressures in retrograde gas reservoirs. We were able to demonstrate that the model described in this paper is more accurate than any presented to date. In addition, the model is simple and is able to duplicate with reasonable accuracy the temperature-dewpoint pressure behavior of constant-composition gas condensate fluids. The neural-network model was developed using a set of 802 experimental constant volume depletion (CVD) data points. To train the neural-network model, a set of 641 experimental data points of CVD for different gas condensate fluids was used. The model was tested with 161 experimental data points, not used during the training process, to prove its accuracy. The study also considered a detailed comparison between the results predicted by this more efficient neural-network model and those predicted by other correlations for estimating dewpoint pressure of retrograde gas. The performance of this improved neural-network model and available correlations was evaluated versus the Peng-Robinson Equation of State (PR-EOS) model for the same reservoir fluid composition, a gas condensate from the Cusiana Field, in Colombia. This improved neural-network model was able to predict the dewpoint pressure with an average absolute error of 8.74%, as a function of temperature, hydrocarbons and non-hydrocarbon compositions, molecular weight, and specific gravity of heptanesplus fraction. Neural-network models can save calculation time in the prediction of the dewpoint pressures with more reliability than available multiple-regression techniques. 2002 Elsevier Science B.V. All rights reserved.
