Discrimination Between Reservoir Models in Well-Test Analysis

In well-test analysis with nonlinear regression, discrimination between candidate reservoir models is a key procedure. Graphical analysis now performs this step, with the pressure derivative plot and confidence intervals used to select an adequate model. Human bias may influence the selection by graphical analysis, however, and the result may vary according to the interpreters. Confidence intervals can give a quantitative evaluation of model discrimination, but it sometimes provides inappropriate results. The authors propose a new quantitative method, the sequential predictive probability method, to discriminate between candidate reservoir models. This method is based on Bayesian inference, in which all information about the reservoir model and, subsequently, the reservoir parameters deduced from well-test data, is expressed in terms of probability and is regarded as a direct extension of the idea of confidence intervals. This method gives a unified measure of model discrimination. They demonstrate the applications of this method to synthetic and actual field well-test data.