Machine Learning Involvement in Reservoir Simulation by Optimizing Algorithms in SAGD and SA-SAGD Processes
Abstract
Machine learning, as a subset of Artificial Intelligence, has invaded many industries in recent years, thanks to the advancement of the computing power. Over the past decade, the use of machine learning, predictive analytics, and other artificial intelligence-based technologies in the oil and gas industry has grown immensely.
Global optimization techniques are useful tools for process optimization and design in various petroleum-engineering disciplines. In this research, the genetic algorithm, one of the global optimization branches, acts as the optimizer for the Steam-Assisted Gravity Drainage (SAGD) process and for the Solvent-Assisted Steam-Assisted Gravity Drainage (SA-SAGD) process.
Both binary and continuous encoding techniques that are embedded in a Computer Modeling Group (CMG) simulator STARS perform the primary role to optimize the simulation results. In the end, the comparison with a gradient-based optimization algorithm is studied. The genetic algorithm optimizer coupled with a reservoir simulator is employed to optimize the steam injection rates over the life of a steam-assisted gravity drainage process in a particular reservoir. A concise comparison between the genetic algorithm and a back propagation method discloses the advantages of the genetic algorithm.
In the end, the simulation results for various scenarios illustrate the impacts brought by the optimization tool that is coded in programming language Python. The conclusion may be effective in the specific reservoir condition only, though it indicates a worth-trying approach to optimize the SAGD and SA-SAGD operations for various reservoirs.
Description
Keywords
Engineering--Petroleum
Citation
Zhang, Y. (2017). Machine Learning Involvement in Reservoir Simulation by Optimizing Algorithms in SAGD and SA-SAGD Processes (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca. doi:10.11575/PRISM/26806