Automatic Parameter Estimation From Well Test Data Using Artificial Neural Network

Abstract

Abstract We propose a robust way of achieving a well test interpretation by combining the sequential predictive probability method with an artificial neural network approach. The sequential predictive probability method considers all possible reservoir models and determines which candidate model or models best predict the well response. This method is dependent on obtaining good initial estimates for the parameters governing the candidate reservoir models, which is achieved by applying the artificial neural network approach. We use the neural network to identify the characteristic components of the pressure derivative curve corresponding to the flow regimes known to be in each candidate model. Reservoir parameters are then computed using the data in the identified range of the corresponding behavior. As a final step, the candidate models and their initial estimates are evaluated using the sequential probability method. The method discriminates between the candidate models and simultaneously performs nonlinear regression to compute the best estimates of reservoir parameters. The trained neural network was able to identify the characteristic components of the derivative curve in most cases. The algorithm written to interpret the neural network signals into flow regimes required special procedures to take care of the misclassification from the neural network. The initial estimates of reservoir parameters from the neural network were found to be reasonably close to the eventual estimates from the sequential predictive probability method. Introduction Traditional methods of well test interpretation are usually based on a combination of manual and automated techniques, although both techniques are usually computer based. Manual interpretation uses the pressure derivative plot introduced by Bourdet et al. The characteristics of different reservoir flow regimes can be observed from the plot. Hence, we are able to analyze the type of the associated reservoir and determine their parameters from the appropriate flow regimes. Automated interpretation by nonlinear regression is then used to determine the best estimates of reservoir parameters, and confidence intervals are used to authenticate the selected reservoir model. With new developments in pressure measurement, including permanently installed gauges, we may have an enormous amount of pressure data coming in each day. This study looked at a procedure to mechanize the interpretation of such well test data. There are three key steps in the procedure. First, all the characteristic components of the derivative plot have to be recognized. This task is accomplished by a specially trained neural network. Second, the signals from the neural network are translated into reservoir flow regimes so that initial estimates of reservoir parameters can be evaluated. Third, the sequential predictive probability procedure discriminates between candidate reservoir models, simultaneously performing nonlinear regression based on the initial estimates provided by the neural network. Previous Work Allain and Horne used syntactic pattern recognition and a rule-based system to identify the reservoir model and to estimate its parameters. The pressure derivative data were first preprocessed in order to distinguish the true response from the noise. P. 249