Applying Proxy Models for Reservoir Simulation in SAGD

Arash Mirzabozorg
Supervisors: Dr. Long Nghiem, Dr. John Chen

Proxy-modeling (also known as surrogate modeling or meta-modeling) is a computationally fast alternative to full numerical simulation in assisted history matching, production optimization and forecasting. It establishes a correlation between response and parameters. As every known, the numerical simulation of complex SAGD processes is characterized as long computational time and a very expensive process. So the proxy model is good way to be used in this process.

The optimization of the operational parameters is necessary to improve performance and to cut down on operating expenses. This presentation introduces three types of proxy-modes (polynomial regression model, Kriging model, Artificial Neural Network), which are applied in the real case study to optimization the operation parameters.

Conclusively, optimization can significantly improve NPV of the Model. Both Polynomial and Kriging proxy optimizations significantly improve NPV of the model. Kriging Model finds more optimal solutions than Polynomial model. Polynomial model is faster than Kriging model. In terms of neural network, it seems that applying feed-forward back propagation to small datasets will not give good prediction. Early stopping might not be useful because it sometimes decreases the performance of a method.