Characterization of Reservoir Heterogeneity Using Inverse Model Equipped with Parallel Genetic Algorithm

Abstract

This study presents the development of reservoir characterization inverse model equipped with ANN type of ISGA and MSGA parallel processing algorithms. In order to run the developed model efficiently, homogeneous PC-cluster was constructed by four connecting PCs that have the same features. The model adapted ANN to automatically determine optimum GA parameters of operator, number of individuals and the operation rate that is appropriate for the reservoir heterogeneity. By utilizing the developed model in this study, inverse calculation was conducted for the synthetic reservoir system with the aid of an ISGA-PP. As a result, it was found that convergence is stably progressed. In the result of permeability distribution, it shows that low permeable zone in the central area for the system studied appeared to be little different compared to the result obtained by Kriging method, which is used only as static data. In the matching result of pressure, maximum relative error of 1.54% was presented at OP-4, and hence, the calculated permeability distribution is thought to be quite reliable. When MSGA-PP was applied to the same reservoir system as ISGA-PP, it converged stably similar to ISGA-PP. The difference between ISGA-PP and MSGA-PP appeared only at convergence rate and the resulting permeability distribution is very similar to each other. In the evaluation of computing efficiency of ISGA-PP and MSGA-PP against GA-SP, the result shows that the efficiency of parallel processing system is greater as the number of individual increases. Also, regardless of the number of individuals, the calculating time in parallel processing system was greatly reduced by 3.6 times compared to serial processing system of GA-SP. Finally, inverse calculation was carried out with MSGA-PP-ANN. As a result, it converged much faster than MSGA-PP, which does not have an artificial neural network system. This is the reason why the superior individuals are selected by the optimum operators, which are determined by ANN in the early time of the inverse calculation.