Using an Expert System To Identify a Well-Test-Interpretation Model

Abstract

Summary We present a new approach that uses artificial intelligence to identify thewell-test-interpretation model that best describes the behavior of a reservoir. This knowledgebased (expert) system is composed of carefully extracted rulesand facts for buildup and drawdown test analysis. The rules simulate thereasoning process used by a human expert to identify the appropriateinterpretation model for a well test. The elaborate control needed to simulatea human's reasoning process was simplified with a blackboard architecture. Toexplain the application of the expert system, we use it to analyze a fieldexample. Introduction The accuracy of reservoir properties estimated from well tests depends onprior identification of a well-test-interpretation model that accuratelydescribes the underground system. Identifying the appropriatewell-test-interpretation model is the most fundamental and difficult problemthat an engineer faces when analyzing a pressure-transient test. Recent use ofinvolved mathematical models and techniques has not completely solved thisproblem because choosing an applicable model is not a completely quantitativeprocedure. The choice is based on experience and on a complex reasoning processthat is often deeply buried in the process that is often deeply buried in thehuman expert's mind. If successful, the process of interpreting a well test should provide areservoir model that predicts the behavior of the reservoir being tested. Thisprocess, known as the inverse problem, is characterized by identifying anunknown system (model) by studying the input (rate) and output (pressure)signals. Because a number of models can produce the same output responses froma given input signal, there is no single solution to this problem, suggestingthat a successful interpretation depends on the skills and experience of theinterpreters and their conception of the true model. We developed a knowledge-based system that seeks the appropriateinterpretation model by imitating the reasoning process of human well-testingexperts. The knowledge base uses interactive well-test-analysis and automatichistory-matching modules, and should minimize-and sometimes eliminate-thelack-of-uniqueness problem. Literature Review Although much literature exists on well-test interpretation, work related tothe identification of well-test-interpretation models is very limited. Gringarten discussed the difficulty and complexity of identifying thewell-test model and-validating the analysis. He concluded that, because theprocess involved subjective reasoning, this was an artificial-intelligenceproblem that required an expert system. Using a logical and systematic approach, McVay et al. developed interactivesoftware for well-test analysis. In their work, they indicated that the humanexpert's knowledge needed to be incorporated into an expert system to preventthe user from imposing preconceived modeling ideas on the data. In a recent study, Watson et al. discussed the importance of identifying theinterpretation model as a precondition to successful analysis. They used astatistical approach to choose the applicable model from a group of candidatemodels. Automatic history-matching has been used extensively in recent years toenhance the efficiency and speed of well-test analysis. The interpretationmodel in this case is preselected and a match is sought by changing the modelparameters. This solution, however, is apt to result in misleadinginterpretations because, as indicated earlier, different models can respondsimilarly to a given set of input signals. Erdle et al. described a general interactive software package for well-testinterpretation and introduced expert advice into the well-test-interpretationprocess. In this paper, we introduce a new approach that simplifies the processof determining the applicable well-test-interpretation model. Our approach isbased on artificial-intelligence technology. Problem Solving Model Problem Solving Model Blackboard Architecture. Theknowledge-based architecture consists of three principal components: aknowledge base (facts and rules), a data base (supplied data and solutionstate), and a control (problem-solving) strategy. Although knowledge is the most important part, the control strategy is acritical design part, the control strategy is a critical design component forknowledge-based systems because simulating the human expert's reasoning processrequires that the reasoning steps process requires that the reasoning steps anddomain knowledge be organized such that they can be used efficiently to solvethe problem. A number of problem-solving models have been described in theartificialintelligence literature. JPT P. 654