| Summary: | Water saturation determination in core laboratory is known as a cost and time consuming labor. Hitherto, many scientists attempted to estimate accurately water saturation from well-logging data which has a continuous record without losing information. Therefore, various model were introduced to relate reservoir properties and water saturation. Since carbonate reservoir is very heterogeneous in shape and size of pore throat, the relation between water saturation and other carbonates reservoir properties is very complex, and causes considerable overall errors in water saturation calculation. By increasing the usage and improvement of soft computing methods in engineering problems, petroleum engineers have been attended them to measure the petrophysical properties of the reservoir.
In this study, a radial basis function neural network (RBFNN) improved by genetic algorithm has been employed to estimate formation water saturation by using conventional well-logging data. The used logging and core data have been gathered from a carbonated formation from one of oilfield located in south-west Iran, and finally their results of the proposed model were compared with the core analysis results. By checking the testing data from another well, it showed this method had a 0.027 for mean square errors and its correlation coefficient is equal to 0.870. These results implied on high accuracy of this model for oil saturation degree estimation. While the common methods like Archie, had a 0.041 mean square error and 0.720 of the correlation coefficient, which indicate a high ability of RBF model than the other usual empirical methods.
|
|---|
