Answers are grounded in this paper and cite their sources.
Optimizing the attachment point, azimuth, and length of each fishbone branch independently lets the well design adapt to reservoir heterogeneity, but many candidate designs then violate geometric constraints because branches intersect or come too close. This study combines repair-based constraint handling with a two-stage workflow: a genetic algorithm (GA) selects the branch attachment points, and particle swarm optimization (PSO) then optimizes the azimuth and length of each branch. Out-of-bounds coordinates are reflected back into their ranges, and infeasible candidates are repaired by Gaussian mutation or regenerated by Latin hypercube sampling with opposition-based learning (hybrid regeneration). The operators are established; the contribution is their combination for clearance constraints among the branches of one well and its systematic evaluation. The workflow was coupled with CMG STARS and tested on eight synthetic models based on the Lloydminster heavy oil and Bakken tight oil formations. Relative to the best of twelve fixed-pattern designs, which use the full 2000 m branch-length limit, the optimized five-branch designs produced 13.2–29.4% (Lloydminster) and 5.8–20.8% (Bakken) more oil over 15 years. Across 16 combinations of reservoir model and branch number, the proposed workflow outperformed GA–PSO without regeneration in all 16 cases (by 0.5–18.8%) and PSO-only with evenly spaced attachment points in 15 of the 16 cases (by up to 18.4%). In five independent runs on each of four selected models, its mean production exceeded that of GA–PSO without regeneration by 1.4–9.6% and that of PSO-only by 5.8–16.4%. In the tested models, fewer, longer branches (five versus ten attachment points) outperformed more, shorter branches of equal total length.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: