The Development of an Artificial-Neural-Network-Based Toolbox for Screening and Optimization of Enhanced Oil Recovery Projects
Open Access
- Author:
- Sun, Qian
- Graduate Program:
- Energy and Mineral Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- September 29, 2017
- Committee Members:
- Turgay Ertekin, Dissertation Advisor/Co-Advisor
Turgay Ertekin, Committee Chair/Co-Chair
Luis F. Ayala H., Committee Member
Eugene Morgan, Committee Member
C. Lee Giles, Outside Member - Keywords:
- Artificial neural network
Enhanced oil recovery
Proxy models
Particle swarm optimization
Graphical user interface
Artificial neural network
Enhanced oil recovery
Proxy models
Particle swarm optimization
Graphical user interface - Abstract:
- Enhanced oil recovery (EOR) refers to sophisticated hydrocarbon production techniques which aim at recovering residual oil after primary and secondary recovery stages. Broadly, EOR techniques are categorized into thermal recovery, chemical EOR and miscible gas injection. A successfully implemented EOR project may produce up to 30 to 60 percent of original oil in place. As the demand on oil increases, studies targeting EOR procedures become increasingly more attractive. Numerical reservoir simulations implementing high fidelity compositional and thermal recovery models play a significant role in understanding of the recovery mechanisms and project optimization of EOR processes. An artificial neural network (ANN) based tool can be considered as a powerful subsidiary tool for high-fidelity numerical simulation models to study EOR processes for its extremely fast computational speed, especially when a quick screening or large volume of simulation runs are required. The objective of the research is to develop an artificial-neural-network-based ‘toolbox’ covering extensive types of EOR projects. The toolbox is established on the basis of assembling and improving the existing ANN models, and the development of new ANN models that completes the missing tools. This work proposes generalized ANN models which are capable to be implemented in various type of reservoirs. Training of the ANN models employs synthetic production histories generated from a commercial numerical reservoir simulation package. In this work, we develop and utilize a robust ANN topology optimization workflow to proceed efficient and accurate training. A graphical user interface (GUI) is designed to provide the users with straightforward access to the ANN tools. Particle swarm optimization (PSO) is successfully coupled with the ANN models developed in this work to optimize EOR related objective functions.