Evolving an accurate model based on machine learning approach for prediction of dew-point pressure in gas condensate reservoirs

Over the years, accurate prediction of dew-point pressure of gas condensate has been a vital importance in reservoir evaluation. Although various scientists and researchers have proposed correlations for this purpose since 1942, but most of these models fail to provide the desired accuracy in prediction of dew-point pressure. Therefore, further improvement is still needed. The objective of this study is to present an improved artificial neural network (ANN) method to predict dew-point pressures in gas condensate reservoirs. The model was developed and tested using a total set of 562 experimental data point from different gas condensate fluids covering a wide range of variables. After a series of optimization processes by monitoring the networks performance, the best network structure was selected. This study also presents a detailed comparison between the results predicted by this ANN model and those of other universal empirical correlations for estimation dew-point pressure. The results showed that the developed model outperforms all the existing methods and provides predictions in acceptable agreement with experimental data. Also it is shown that the improved ANN model is capable of simulating the actual physical trend of the dew-point pressure versus temperature between the cricondenbar and cricondenterm on the phase envelope. Finally, an outlier diagnosis was performed on the whole data set to detect the erroneous measurements from experimental data.