CHARACTERIZATION OF TIGHT GAS RESERVOIRS WITH STIMULATED RESERVOIR VOLUME: AN ARTIFICIAL INTELLIGENCE APPLICATION
Open Access
- Author:
- Zhang, Yiming
- Graduate Program:
- Energy and Mineral Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- August 01, 2016
- Committee Members:
- Turgay Ertekin, Thesis Advisor/Co-Advisor
Sanjay Srinivasan, Program Head/Chair
Zuleima T Karpyn, Committee Member
Hamid Emami-Meybodi, Committee Member - Keywords:
- SRV
ANN
Characterization
Tight gas reservoir
Stimulated reservoir volume
artificial intelligence network - Abstract:
- Stimulated reservoir volume (SRV) characterization has been a practical concept developed by researchers in the petroleum industry and utilized recently for the purposes of quickly evaluating the effective well drainage area and analyzing the performance of multistage stimulated horizontal wells in tight formations. In this study, SRV modeling of tight gas reservoirs is focused for characterization purposes. Characterization of reservoir properties is an exhaustive and time-consuming process. For a dual porosity type of reservoir, the characterization of the reservoir is often much more descriptive than a typical conventional reservoir, thus making the history matching process much more complex and difficult to solve by hand. The recent applications of artificial neural network (ANN) technology in petroleum engineering applications has attracted increasingly more attention. In this study, two ANN based tool are designed and developed. The first ANN model is the forward-looking ANN model, which predicts the production profiles based on the reservoir characteristic parameters from horizontal wells in tight gas sands. The forward-looking model predicts coefficients of the hyperbolic decline curve with cumulative production error of less than 3%. The second ANN model is the inverse ANN model, which focuses on characterization of tight gas reservoir with known reservoir parameters, design parameters, and production profile. The inverse model is capable of predicting characteristic parameters with cumulative production error of less than 8% by the numerical model itself, or cumulative production error of less than 13% in conjunction with the forward-looking model.