| dc.description.abstract | This thesis investigates the application of Bayesian Optimization (BO) in the optimization of the Expected Net Present Value (ENPV) for oil field development. The objective is to maximize the ENPV while reducing the computation time required for the optimization process. The challenges associated with this optimization problem, such as the absence of an analytical form of the objective function, the presence of multiple local optima, and the computational demands of evaluating the ENPV, are addressed using BO.
I begin by providing a comprehensive overview of BO and its suitability for addressing these challenges. I demonstrate its effectiveness through a 1D toy problem, where the objective function is treated as a black box and iteratively approximated using a Gaussian Process. This serves as a foundation for applying BO to a realistic 3D reservoir simulation model, where I optimize water injection rates to maximize the ENPV.
To further improve the optimization process, I investigate the impact of initial value selection methods on computation time and ENPV convergence. I compare two approaches: random selection using a uniform distribution and Latin Hypercube Sampling (LHS) to ensure scattered initial injection rates for wells. The results highlight the significance of initial value selection, with LHS demonstrating reduced computation time while maintaining comparable ENPV outcomes.
Moreover, I propose a modified method to decrease the optimization time by employing an eliminating criterion. I evaluate the feasibility of using the first realization NPV as a decision point to determine whether further NPV calculations are required for other realizations. I demonstrate the effectiveness of this approach in reducing computation time, particularly in cases with low uncertainty.
Overall, this thesis contributes to the understanding of how initial value selection and modified BO methods can improve the efficiency and effectiveness of ENPV optimization in oil field development. The results provide insights for decision-making processes and can guide practitioners in maximizing the value of oil field investments while reducing computation time. | |
