Enhanced Black Hole Particle Swarm Optimization in Well Placement Optimization

Abstract

Engineers and geoscientists work within an asset team on defining well type, well control, well placement and facility design towards optimizing the development planning of oil and gas fields. Well placement optimization plays a critical role in field development planning, since it accounts for a major portion of the capital expenditure and significantly affects hydrocarbon recovery. This thesis presents an updated version of the black hole particle swarm optimization (BHPSO), a well placement evolutionary optimizer introduced by Harb et al. [1]. BHPSO simultaneously optimizes the well count, location, type, and trajectory of wells. The importance of the proposed algorithm is that while well placement optimization involves a large number of optimization parameters, the BHPSO drastically reduces the number of optimization variables as its computational complexity is independent of the number of optimized wells. Harb et al. [1] focused on pattern water injection. In this research project, we address a different injection scheme: peripheral water injection. Here, for each particle in a BHPSO “iteration”, the particle swarm optimization (PSO) defines the location of the first producer based on a net hydrocarbon thickness (NHCT) map, and the location of the first injector is determined based on the permeability thickness (Kh) map of the field. The number of the remaining wells is decided by the PSO and, then, using the black hole (BH) operator, producers are automatically and optimally placed using a NHCT map, followed by the placement of the injectors on a Kh map. After every well placement, the maps are updated by eliminating a black hole (a disk and/or a cylinder) around the producers and injectors, each on the relevant map respectively. The radius of the black hole is defined by the well spacing. The method was extensively tested on both the synthetic Olympus reservoir model and the PUNQ-S3 reservoir model. Additionally, a new black holing technique was introduced and tested in the process of further optimizing the BH operator results. Furthermore, the convergence criteria of PSO is addressed in this study by conducting a sensitivity analysis on its algorithmic parameters including the acceleration factors, and the swarm size. The Inertia weight, another algorithmic parameter of PSO, was changed from a Constant Inertia Weight to a Linearly Decreasing Inertia Weight (LDIW). The latter showed improved results over the former in terms of convergence as faster convergence was obtained using a LDIW.