查看文章

Throughout the life of a field, the only way to make contact with the reservoir is through the wellbore by making indirect data measurements. These data acquired during the measurement period is analyzed and interpreted in other to have a better understanding of the reservoir characteristics. There are several methods of estimating reservoir properties such as core analysis and well logging which are only able to obtain properties at a local point. But, well testing present an average property of the whole reservoir as the reservoir responds to the perturbation caused during the test period. This study developed an artificial neural network algorithm that was trained to automatically identify four different reservoir models (the homogenous infinite acting radial flow with wellbore storage reservoir, homogenous reservoir with a finite conductivity fracture, homogenous reservoir with an infinite conductivity fracture and a double porosity reservoir with no wellbore storage effect) and estimate the model parameters (permeability, skin factor, reservoir radius, flowing wellbore pressure and/or the length of the conductivity fracture). The algorithm was constrained to four models because of the unavailability of well test data that represent the other plethora of reservoir models. The accuracy of the algorithm on reservoir model recognition was 99.5% and also, the reservoir model parameter estimation had an index of fitness of 1 for the four reservoirs and the mean squared error of 2.62653 e-10, 3.29122 e-8, 1.05805 e-5 and 2.19763 e-6 respectively. It is then concluded that an artificial neural network is a good tool for well test analysis and interpretation. And that it …