Use of Artificial Intelligence in Well-Test Interpretation

Abstract

Summary. This paper describes techniques for the identification of well-test interpretation models from pressure-derivative data. Artificial pressure-derivative data. Artificial intelligence (AI) techniques are used to separate the derivative response from signal and differentiation noise, and a rule-based recognition system characterizes the response using a symbolic representation. Introduction The use of computers for well-test analysis has played an important role in the development and application of new interpretation techniques. With nonlinear regression, algorithms have been developed that can perform automated type-curve matching when perform automated type-curve matching when given a specific model and a first estimation of its parameters. Such algorithms not only speed the interpretation procedure but also increase the confidence in the results by allowing quantification of the quality of the match obtained. Correct application of these methods, however, depends on the choice of the interpretation model. An interpretation procedure that uses the pressure derivative is regarded as the pressure derivative is regarded as the most appropriate tool for the diagnosis of an interpretation model for well-test data. The derivative is not only more sensitive to the different flow regimes and flow-regime transitions but also is more generally applicable and therefore provides a substitute for all the specialized analyses. In traditional graphical analysis, the choice of an interpretation model is essentially a visual process. Based on a log-log plot of the pressure derivative, an expert can perform model identification by recognizing features that are specific to particular analytical models. The combination of the various models then gives an interpretation model. Although the human visual perception process is not fully understood, significant process is not fully understood, significant work in AI has been conducted to develop machine vision systems. These systems operate not on the image itself but on a symbolic representation containing the information needed for the task under consideration. This paper describes the development of techniques to automate the modelidentification step of well-test interpretation by use of AI. The computer's choice of a model is based on the pressure-derivative curve and simulates the visual diagnosis performed by a human expert. The reasoning performed by a human expert. The reasoning involved in such a diagnosis uses a symbolic representation of the derivative curve, which in the case of a human expert is built almost unconsciously. Techniques were developed to replicate this visual-perception step. A major difficulty in the analysis of real data, particularly when the pressure derivative is used, is the separation of the true reservoir response from signal or differentiation noise. Again, this is relatively simple for a human observer, but difficult to implement in a computer program. This paper describes an algorithm developed to overcome this problem. The algorithm was able to distinguish response from noise correctly, thereby allowing for correct model identification. In a manner analogous to a human expert, the technique constructs an interpretation model for the duration of the data by combining the features of the different flow periods of the response. The adequacy of a periods of the response. The adequacy of a model is determined by qualitative and quantitative information. Once a model is chosen, its parameters are estimated with a correlation or an appropriate table. The system developed in this work can perform model identification on real or perform model identification on real or hypothetical data. Because the method also provides parameter estimation, it can be provides parameter estimation, it can be used with an automated type-curve-matching analysis, allowing full automation of welltest interpretation. The task of identifying an interpretation model by use of the derivative of well-test data can be separated into three components:observation-extraction of the features present on the data derivative and present on the data derivative and representation of these features,knowledge of the models-methods for the construction and description of interpretation models, andmatching-criteria for choosing appropriate interpretation models for given data. The development of the specialized rule base and logical approach for these three components took considerable work before the algorithms were able to interpret real data reliably. The purpose of this paper is to describe this development as an aid to others who may wish to attempt a similar approach. JPT P. 342