Optimizing Well Trajectory Using Sequential, Hybrid Sequential, and Fully Concurrent Method Utilizing Machine Learning: A Case Study of a Tight Limestone Reservoir

Abstract

This paper describes the process and results of using machine learning to automatically determine the optimum well trajectories for a tight limestone reservoir. The study aimed to find the best trajectories to maximize gas production while following certain constraints related to safe drilling operations and surface limitations. These constraints included the available surface locations, maximum dogleg severity, maximum inclination, clearance distance to the water contact, maximum well length, and the use of hydraulic fracturing (acid fracturing).

In this study, a history-matched model was used to simulate the behavior of a reservoir with 12 components, including various gases and hydrocarbons (C1-C7+). New wells were introduced using an automatic workflow, and their trajectories were controlled by five variables: α, β, R2, R3, and clearance. The first segment of the well was a vertical section from the surface to the kick-off point, and the subsequent segments were controlled by α, β, R2, R3, and clearance. α determined the azimuthal degree of the second segment, β was the deviation of the third segment angle from the endpoint of the second segment, R2 and R3 were the surface distances of the ends of the second and third segments, respectively, and clearance specified the distance of the well to the water gas contact. The Particle Swarm Optimization (PSO) algorithm was used to optimize the well trajectories based on the objective function of cumulative gas production. Three different approaches were used for optimization: sequential, hybrid sequential, and fully concurrent. The sequential approach optimized one well at a time, the hybrid sequential approach optimized groups of wells, and the fully concurrent approach optimized all wells while including the number of wells as a variable.

The algorithm successfully found the optimum well trajectories automatically based on the objective function. As more wells were added, the recovery factor (RF) and the plateau duration increased. However, the incremental increase in RF decreased due to the reduced amount of gas that could be recovered with additional wells. At the end of the optimization process, 15 new wells were created, and the cumulative gas production reached a recovery factor of 62% and maintained a constant plateau for 165 months. A workflow for well trajectory optimization was successfully developed to find the best well trajectories that maximize gas production in this field. All new well trajectories satisfied the constraints based on the field conditions and safety limits of the drilling operations.

The well trajectory optimization results demonstrated the developed algorithm’s effectiveness in finding optimal well trajectories that maximize gas production. The successful workflow implementation and satisfaction of constraints demonstrate this approach’s feasibility and potential value for any real-world problems.