چكيده به لاتين
Abstract:
Understanding the behavior and properties of reservoir rock and fluids is an important and integral part of managing oil and gas reservoirs. Due to the static and dynamic complexity and uncertainty of reservoirs, this understanding is not simply achieved and requires co-operation and of a group of expertise along with the petroleum engineers. Evaluation of the petrophysical parameters plays an important role for calculation of reservoir parameters such as relative permeability. In this thesis, the laboratory data of the Iranian carbonate reservoirs has been used to predict the relative permeability using smart models. Five methods including the experimental regression model, multilayer perceptron neural network, radial basis neural network, fuzzy Sugeno system and support vector regression were used to predict the relative permeability of oil and water. Moreover, in order to increase the accuracy of the models and due to the obtainability of available data from carbonate reservoirs, data classification has been considered based on the flow units as an appropriate solution to overcome the heterogeneity of reservoirs. In this thesis, 165 curves (1984 data) is used to estimate the oil-water relative permeability. Finally, after classifying the available cores based on the reservoir quality index, the collected data was divided into 3 groups. Comparison of the results showed that the classification of data into flow units improves the performance of these methods drastically. The results shows that the support vector regression is the best in prediction of the relative permeability of oil and water among the others.
Keywords: Relative Permeability, Carbonate Reservoirs, Neural Networks, Fuzzy Sugeno Inference Systems, Support Vector Regression.
