Dewpoint Pressure Estimation of Gas Condensate Reservoirs, Using Artificial Neural Network (ANN)

Abstract

Abstract Dew point pressure is a crucial parameter for evaluation of gas condensate reservoirs. Hence, the availability and accuracy of this property has been the main bottleneck for the reservoir simulation and development. Although the dew point pressure can be determined experimentally from collected laboratory samples frequently these measurements are not available. In these cases, fluid reservoir properties are determined with the use of empirical correlations (explicit methods) or they can be determined iteratively using an equation of state (EOS). Theoretical results and practical experience indicate that feed forward network can approximate a wide class of function relationships very well. This property is exploited in modeling petroleum engineering process. In this work, a set of conventional feed forward multilayer neural network have been proposed to predict dew point pressure of gas condensate reservoirs. The accuracy of the method is evaluated by its application for dew point pressure estimation of various reservoir fluids not used in the development of the model. Furthermore, the performance of the model is compared against the performance of other alternatives correlations reported as the most accurate and general correlations for dew point pressure prediction. Results of this comparison show that the proposed method out performs the other alternatives, both in accuracy and generality. The network was developed using experimentally Constant Volume Depletion (CVD) measured condensate sample of south pars reservoir and collected data from literature of 111 gas condensate samples covering a wide rang of gas properties and reservoir temperature. The network has an average absolute error of 2.573%, 3.832% and 2.612% for training, validation and test processes, respectively.