| dc.description.abstract | Closed Loop Reservoir Management (CLRM) provides an important concept to aid decision making to maximize reservoir production performance. CLRM consists of two main tasks; data assimilation, and control and optimization. The latter is of particular interest for this thesis. CLRM was proposed more than 10 years ago. However, there are still several practical challenges that may discourage industries from applying CLRM. This thesis aims to reduce the gap between the practical solutions and optimization formulations. We try to encourage industry to use systematic methods for optimization and control by including their knowledge and concerns into the problem formulation and optimization algorithms. In particular, we emphasize realistic field development constraints and computational resources in the well placement problem under uncertainty as well as economical uncertainty in the short-term production optimization problem.
In order to formulate realistic field development constraints into the well placement optimization problem, we interact with experts in industry. In fact, we propose to include human knowledge into the semi-automated optimization algorithm. The limitations are identified and translated into the problem formulation. The obtained problem formulation is a constrained optimization problem, which necessitates use of constraint handling techniques in the optimization formulation. Well-known constraint handling techniques (e.g., the penalty method) requires problem dependent parameter tuning. This fact usually makes industry skeptical to use such methods. In this thesis, a constraint handling technique that does not require parameter tuning is proposed and compared with the penalty method.
To embed uncertainty into the optimization problem, a robust optimization framework has been proposed in the literature. The robust optimization problem usually formulates the objective function such that the performance of all realization are considered, which necessities the reservoir simulation to run over all realizations. In this thesis, we propose a robust optimization procedure, which reduces computations significantly. Moreover, the performance of different optimization algorithms (e.g., local and global search algorithms) are also compared in several well placement optimization problems.
In the optimization problem an economical measure is usually maximized, which requires a value for oil price and different costs. However, it is difficult, if not impossible, to predict the oil price accurately. In this thesis, a framework is proposed to include time-varying oil price in short-term production optimization, while the long-term decision is also considered. | nb_NO |
