A comparison of accuracy and computational time for common and artificial methods in predicting minimum miscibility pressure

摘要

Minimum miscibility pressure is the least required pressure for complete mixing of gas and oil in the reservoir conditions. It is an important parameter in the processes of gas injection in a miscible manner and its precise determination is very vital in choosing the type of injecting gas and planning injecting equipment for increasing the recovery efficiency. The common method is determining MMP in slim tube or 1 -D simulation of slim tube. Usually determining the minimum miscibility pressure via slim tube apparatus is an expensive and time-consuming test and to carry it out it is necessary to have a sample of reservoir oil and suggested injecting gas. Occasionally it is possible that for some unknown reasons despite spending much time and money it won't bring up any result. As a result, for determining this parameter, finding another method which has a higher precision in addition to being swift and less expensive is very necessary. On the other hand, there are several simulation methods to determine minimum miscibility pressure. These methods are so fast rather than slim tube experiment and relatively precise. MMP can be estimated numerically using compositional simulation, method of characteristics (MOC), mixing-cell methods, intelligent methods, and empirical correlations. However, nowadays one dimensional (1-D) slim tube simulation based on compositional simulation is very common. In this paper a suggestive method is proposed for determining MMP. A mixing rule method coupled with artificial neural network model (ANN) based on a numerous experiment data. Accuracy and computational time of artificial neural network method were compared to common prior models and correlations. The results show although intelligent methods are so fast, 1-D slim tube simulation is still a proper method to determine MMP in high accuracy. Average absolute relative error for MMP value is 1.5% for 1-D slim tube simulation, while the number for ANN is 3.25%. However, ANN method is recommended for fast MMP estimation.