Artificial intelligence approach to the identification of the well test interpretation model

The accuracy of reservoir properties estimated from well tests depends on prior identification of a model which describes the underground system accurately. This model is known as the "well test interpretation model." The objective of this study is to present a new approach to solve the problem of identifying this model. Our approach is based on integrating artificial intelligence, pattern recognition, and history matching techniques.;Using artificial intelligence techniques, we developed a knowledge-based system of carefully extracted rules and facts that determines the interpretation model in a manner parallel to techniques human experts use. The knowledge in the system imitates that of an expert in well testing. The knowledge base employs Blackboard Architecture as a problem-solving model and as the control mechanism.;Pattern recognition techniques were used to identify a preliminary well test interpretation model automatically from the derivative plot. Two pattern recognition techniques were implemented: syntactic and adaptive pattern recognition techniques. Syntactic pattern recognition uses symbolic and rule-based approaches to simulate the human vision task. Adaptive pattern recognition uses a distributed approach in the form of an artificial neural network. Both techniques were applied successfully to identify the well test interpretation model. However, we found the adaptive pattern recognition approach more tolerant to noise in field data than syntactic pattern recognition.;To simulate and verify the identified well test interpretation model, we built a library of well test analytical models. This library is linked to a parameter estimation program which applies the Levenberg-Marquardt method.;We found the approach presented in this dissertation useful in assisting the engineer to identify the well test interpretation model. We successfully applied our approach to identify the well test interpretation model from field test data.