DIAGNOSIS OF RELATIVE PERMEABILITY CHARACTERISTICS FROM RATE TRANSIENT DATA: AN ARTIFICIAL INTELLIGENCE APPROACH
Open Access
- Author:
- Case, Nathan Joseph
- Area of Honors:
- Petroleum and Natural Gas Engineering
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Turgay Ertekin, Thesis Supervisor
Turgay Ertekin, Thesis Honors Advisor
Zuleima T. Karpyn, Faculty Reader - Keywords:
- Artificial Neural Networks
Relative Permeability
Reservoir Engineering
Rate Transient Data
Artificial Intelligence
Petroleum Engineering - Abstract:
- The world’s dependence on oil and natural gas is evident by the constant coverage they receive in the media regarding their pricing. These two hydrocarbons power the economies of the world despite the recent alternative energy movement. At the heart of hydrocarbon production lies the study of fluid flow. Permeability serves as a critical property to provide information on the rock’s ability to transmit fluids, but in the case of multiphase hydrocarbon reservoirs, relative permeability plays a crucial role in predicting and interpreting flow regimes. Several studies have been conducted to predict relative permeability in reservoirs, and numerous equations, models, and correlations have been developed to provide numerical solutions. One common way to obtain relative permeability data involves obtaining core samples of the reservoir and running saturation tests to provide the permeability data. Coring proves to be an expensive investment for companies, while also posing a high potential for data errors in terms of locality of the core, as core samples only provide a snapshot of the relative permeability data from one specific location in the reservoir, therefore leaving the representation of the relative permeability of the entire reservoir in question. In addition, the removal of the core from the reservoir provides data that fails to represent in situ conditions. This study seeks to develop representative permeability curves for the entire drainage area of gas-water reservoirs through the use of artificial neural networks (ANNs). The networks were developed using Corey’s parameters for the representation of relative permeability characteristics. By generating multiple datasets from reservoir simulations, the network was trained to understand the relationship between the inputs and the outputs (Corey’s Model parameters) to develop representative permeability curves based upon rate transient data. The network performed well in these predictions and the final network produced relative permeability graphs that closely match the relative permeability characteristics utilized in performing the simulation runs.