AN ARTIFICIAL NEURAL NETWORK BASED TOOL-BOX FOR SCREENING AND DESIGNING IMPROVED OIL RECOVERY METHODS

Open Access
Author:
Parada Minakowski, Claudia Helena
Graduate Program:
Energy and Geo-Environmental Engineering
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
January 18, 2007
Committee Members:
  • Turgay Ertekin, Committee Chair
  • Robert W Watson, Committee Member
  • Luis F Ayala H, Committee Member
  • Mirna Urquidi Macdonald, Committee Member
Keywords:
  • IOR
  • reservoir simulation
  • ARTIFICIAL NEURAL NETWORK
  • IMPROVED OIL RECOVERY
  • petroleum engineering
Abstract:
Typically, improved oil recovery (IOR) methods are applied to oil reservoirs that have been depleted by natural drive mechanism. Descriptive screening criteria for IOR methods are used to select the appropriate recovery technique according to the fluid and rock properties. The existing screening guidelines neither provide information about the expected reservoir performance nor suggest a set of project design parameters that can be used towards the optimization of the process. In this study, artificial neural networks are used to build two neuro-simulation tools for screening and designing miscible injection, waterflooding and steam injection processes. The tools are intended to narrow the ranges of possible scenarios to be modeled using conventional simulation, reducing the potentially extensive time and energy spent in modeling studies and analysis. A commercial reservoir simulator is used to generate the data supplied to train and validate the artificial neural networks. The proxy models are built considering four different well patterns with different well operating conditions as the design parameters. Different expert systems are developed for each well pattern. The screening networks, or forward application, predict oil production rate and cumulative oil production profiles for a given set of rock and fluid properties, and design parameters. The inverse application provides the necessary design parameters for a given set of reservoir characteristics and for the specified (desired) process performance indicators. The results of this study show that the networks are able to recognize the strong correlation between the displacement mechanism and the reservoir characteristics as they effectively forecast hydrocarbon performance for different reservoir types undergoing diverse recovery processes. The inverse proxy models are able to predict the operation conditions at the same time that accurately provide the complete oil production profiles. Both neuro-simulation applications are built within a graphical user interface to facilitate the display of the results. The project design tool-box helps in the quantitative project assessment if proper combinations of expected project abandonment time and total oil recovery are provided for the same reservoir. Its use, when combined with the screening network application, becomes a powerful tool that facilitates the evaluation and validation of the proposed production scenarios. The tools proposed in this study have the potential of providing a new means to design a variety of efficient and feasible IOR processes by using artificial intelligence. Appropriate guidelines are provided to the reservoir engineer, which decrease the number of possible scenarios to be studied and reduce the time spent with conventional reservoir simulation methodology.