Loading...

An Artificial Neural Network Surrogate Model Development for GAGD Process

Rafiee, Javad | 2011

602 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 42679 (06)
  4. University: Shairf University of Technology
  5. Department: Chemical and Petroleum Engineering
  6. Advisor(s): Pishvaie, Mahmoud Reza; Jamshidi, Saeed
  7. Abstract:
  8. Gas-Assisted-Gravity-Drainage (GAGD) is a new enhanced oil recovery method which utilizes the natural tendency of gas and oil to segregate due to gravity. Higher recovery of GAGD, compared to continuous gas injection (CGI) and Water-Alternating-Gas (WAG) makes it the focus of researches in recent years. This work is trying to propose a surrogate model based on Artificial Neural Networks for GAGD process. The proposed model should be able to mimic the behavior the process in a short time in order to be used as an alternative computationally expensive simulator where a large number of simulations are needed, such as sensitivity analysis, optimization of the process, or risk analysis. A synthetic 3D and sandstone model is used to simulate the process of GAGD using IMEX – black oil module of CMG (Computer Modeling Group). The first step toward developing a surrogate model is to find the effective parameter on the process at hand and data gathering. To do that, first, model parameters were manipulated manually to check if the parameters have any effect on the performance of the process. Second, among those parameters which has an effect on the performance of the process those which are most effective are found using a sensitivity analysis based on a Box-Behnken experimental design. Finally, for the most effective parameters obtained, a combination of Box-Behnken and Face-Centered-Cube was used for experimental design to make sure that the effect of each individual parameter as well as their interaction is captured. The second step of developing a surrogate model is to select the model (Artificial Neural Network) and finding the adjustable parameters of the model (training the network). A multilayer perceptron (MLP) neural network with one or two hidden layers and two time lags was used as the surrogate model. All the networks were trained using Artificial Neural Network Toolbox of MATLAB software, and the results reveals that due to complexities and non-linearity of the process the more sophisticated two-hidden-layered network is able to predict the performance of the process more accurately. Ultimately, the developed fast surrogate model was used to perform a risk analysis regarding the effective parameters using 10000 runs
  9. Keywords:
  10. Gas-Assisted Gravity Drainage (GAGD) ; Artificial Neural Network ; Risk Analysis ; Experimental Design ; Surrogate Model

 Digital Object List

  • محتواي پايان نامه
  •   view

 Bookmark

No TOC