Open Access
Bansal, Yogesh
Graduate Program:
Petroleum and Natural Gas Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
April 14, 2009
Committee Members:
  • Turgay Ertekin, Thesis Advisor
  • Majid Al Wadhahi, Thesis Advisor
  • In-situ combustion experiment
  • heavy oil recovery
  • ANN
Artificial neural networks (ANNs), also known as expert systems, have become an increasingly important part of the petroleum industry for performance analysis of reservoirs. ANNs work similar to the biological brain which can make predictions based upon past experiences. ANNs used as a screening tool offer a low cost alternative to commercially available simulators that may require extensive data collection in order to build an effective model. The ANN serves to pare down the analysis so that a more focused data set and corresponding model can be developed. This screening tool consists of data sets usually referred as knowledge base and set of algorithms capable of predicting fact based results for an unexposed part of the input data set. Artificial neural networks are widely used in the optimization of the reservoir parameters. These include operating conditions and reservoir performance, instantaneous and cumulative production, cyclic injection of fluids into the reservoir in order to maintain reservoir pressure for better recovery, field development and well stimulation, among others. The expert system developed during this research models in-situ combustion which is a technique utilized in heavy oil recovery. In in-situ combustion process oil is ignited in the porous matrix of the reservoir to improve the mobility of viscous oil using the heat generated. It is a complex process as the reaction parameters are unknown at the reservoir conditions, but the ANN is able to predict reliable results without a formal analysis of the mechanisms at work. The ANN developed in this study is able to predict the cumulative production profile of oil, water and gas in a laboratory scale experiment utilizing the technique of in-situ combustion under simulated reservoir conditions. Peak temperatures of the combustion zone, their positions and the velocity of the combustion front in the tube at 25%, 50%, 75% & 100% of the production time are predicted. One dimensional flow was analyzed and expert systems were prepared for both dry and wet combustion. Understanding the ability of the network to predict the output parameters is a crucial task in this type of development. The complexity of the network is bound to increase with the number of the inputs and outputs. This problem becomes cumbersome because of the complexity of combustion reactions occurring in the porous media. The mechanisms of reactions have not been fully developed for different crudes with varying asphaltenes and maltenes compositions. By developing a data set for Athabasca crude, which already has a developed reaction mechanism in the literature, an expert system has successfully been developed each for dry and wet combustion processes in the combustion tube experiment.