Prediction of Dew Point Pressure for High-Pressure Gas Reservoirs Using Artificial Intelligence Techniques

Abstract

Abstract Dew point pressure is a curial parameter in characterizing gas reservoirs. Several methods can be used to determine the dew point pressure, including laboratory measurements and empirical models. However, laboratory determinations are expensive and time-consuming, especially for studying high-pressure tight reservoirs where more caution and procedures will be required. While empirical correlations do not accurately reflect the complexity of fluid behavior, and limited models were developed for high-pressure reservoirs. The goal of this work is to develop a reliable tool for predicting the dew point pressure for tight and high-pressure gas reservoirs. This work was carried out using five main phases; data collection, quality control, model construction, development of new correlation, and model validation. The data used in this work were obtained based on 250 laboratory measurements. All data were evaluated and the noises and outliers were removed. Different types of artificial intelligence methods were examined to come up with the best determination model. Artificial neural network (ANN) technique, support vector machine (SVM) approach, and adaptive fuzzy logic (AFL) systems were investigated. The hydrocarbon compositions and the molecular weights were used as inputs to estimate the dew point pressure. Different types of error indices were employed to measure the prediction performance of the developed equation. Average percentage error and correlation coefficient values were determined for the different models. The developed model predicts the dew point pressure with a percentage error of 4.85% and an R2-value of 0.94. The ANN model developed in this study has 4 neurons and one hidden layer. An empirical equation was proposed based on the best ANN program to provide a direct estimation of the dew point pressure. The extracted equation can provide an average error of 5.74% and an R2-value of 0.93. Overall, the proposed model can reduce the cost and time required for determining the dew point pressure and help to improve reservoir management by providing fast and reliable estimations.