Estimating layers deliverability in multi-layered gas reservoirs using artificial intelligence

In this paper, an artificial intelligence (AI) model has been created to estimate the production rate of each layer in a multi-layered gas reservoir using static properties (such as those obtained from well logging)and dynamic properties (such as pressure). This helps in several reservoir engineering applications, such as understanding depletion in layers, or targeting specific layers for workover. It could also be used for PLT analysis where the measured PLT values are compared to the expected values and a variance analysis could be performed. Data were collected from more than 100 wells in a certain reservoir spanning over four fields. They were clustered in related input variables and fed to the AI model for learning purposes. To compare the AI methods, the data were fed to 4 different methods (MLP, RBF, SVM, and GRNN) and the results were optimized for each method. Between the tested AI methods, SVM and GRNN performed best with a low mean absolute error percentage and a very high correlation coefficient. This paper shows a high potential for AI methods in estimating production rate from each layer in a multi-layered gas reservoir.