توسعه روش تطبيق تاريخچه احتمالاتي براي مخازن هيدروكربوري به وسيله تركيب روش‌هاي كراس آنتروپي و شبكه عصبي مصنوعي

چكيده انگليسي :

71 Developing a Probabilistic History Matching Method for HydrocarbonReservoirs using a Combination of Cross Entropy Method and Artificial Neural Network Niloofar Nabavi n nabavi@in iut ac ir Date of Submission 2020 01 25 Department of Industrial Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language Farsi Supervisors Mehdi Bijari bijari@cc iut ac ir Alireza Khazal arkhazali@cc iut ac ir AbstractThrough the past early years reservoir simulation has evolved from a research area intoone of the most practical and powerful tools in reservoir engineering Usage of reservoirsimulation has been grown very fast because of its ability to predict the future performanceof oil and gas reservoirs over a wide range of operating conditions Reservoir simulation isnormally used as an important reservoir management tool and play a very important rolefor production optimization from oil and gas resources History matching is the mostchallenging and critical task during the reservoir simulation study in order to have areliable prediction of reservoir behavior in future As a brief explanation history matchingis an optimization problem which tries to minimize the difference between the predictedreservoir behavior of model and real data obtained during the production history such asproduction rates and pressures In this study a combination of the Cross Entropyoptimization algorithm and artificial neural network has been used as a novel method toadjust model s result with real history of an oil reservoir It worth mentioning that suchmethod has not been applied in any previous studies based on the comprehensive literaturereview and paper study of the researcher The results show that using of this optimiza tionalgorithm is an appropriate method for performing the history matching Duringdevelopment of the aforementioned algorithm for a simple synthetic case of two dimensional reservoir model in each step the amount of error was decreasing in a way thatthe mean square error in the first history matching step was more than 4000 and it becameless than 20 in the last step which a very good achievement to show the functionality ofnew developed algorithm After that in order to check the robustness of algorithm it wasused for history matching of a tree dimensional reservoir model and the amount of errorhas been also decreasing It was reduced from the mean square error of more than 13000 inthe first algorithm s loop and reached to the less than 210 in the last step The final resultwas very close to the reservoir history According to the obtained results utilization of theartificial neural network with the aid of Eclipse commercial simulator most probably yieldvery successful result and will significantly decrease the simulation run time compared tothe other history matching methods In this study for two dimensional model simulationrun time in the combined Cross Entropy algorithm with eclipse simulator was about 280seconds while it was about 52 seconds when the cross entropy was combined withartificial neural network Also for the tree dimensional model it was about 356 seconds forthe first method versus and it was about 53 seconds for the second one Keywords Reservoir simulation History matching of hydrocarbon reservoir Cross EntropyAlgorithm Artificial Neural Network

نبوي، نيلوفر

توسعه روش تطبيق تاريخچه احتمالاتي براي مخازن هيدروكربوري به وسيله تركيب روش‌هاي كراس آنتروپي و شبكه عصبي مصنوعي

تطبيق تاريخچه مخزن هيدروكربوري , الگوريتم كراس آنتروپي , شبكه عصبي مصنوعي