Loading...

Improvement of Production Prediction in Reservoir Simulation Using Artificial Neural Networks

Golzari, Aliakbar | 2012

706 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 43437 (06)
  4. University: Sharif University of Technology
  5. Department: Chemical and Petroleum Engineering
  6. Advisor(s): Jamshidi, Saeid; Badakhshan, Amir
  7. Abstract:
  8. By far, the most expensive part of the production optimization process is the evaluation of the objective function because this requires computationally expensive reservoir simulations to be performed.One way to reduce this high computational cost isby using surrogates or proxies for the reservoir simulator. There are different methods for constructing a surrogate that their aims are mimicking the reservoir behavior with high accuracy and low computational cost. In reservoir engineering surrogate modeling has been used for the problem of well placement optimization,while in the context of production optimization it has not yet been investigated in the literature. Moreover, most of surrogate models constructed in the literature have used one-shot approach. In this study, the use of black box based surrogate models are investigated for a production optimization problem. The main challenge of a production optimization problem is the high number of input and output variables, making Artificial Neural Network (ANN) the best option for approximation function. For this purpose, a dynamic Neural Network is used as approximation function and trained through adaptive sampling algorithm. This network accepts the well control parameters (e.g., well pressures, injectionrates) as inputs and generates the cumulative production curves. A space filling sequential design is used for selecting initial and candidate training points. Neural Network is trained by initial design points and for increasing its accuracy an adaptive sampling algorithm adds training points in a step wise manner. This algorithm selectsnew design points from candidate points by the use of cross validation and jackknifing. Then the accuracy of surrogate model is assessed by cross validation method. The surrogate model is developed for different reservoir models and then used for production optimization. Optimization is done by Genetic Algorithm. Results obtained from surrogate based optimization had a good consistency with results obtained from reservoir model optimization.
  9. Keywords:
  10. Reservior Simulation ; Surrogate Model ; Production Optimization ; Artificial Neural Network ; Adaptive Sampling

 Digital Object List

 Bookmark

No TOC