Hybrid AI-assisted multi-criteria optimization of horizontal well trajectories under representative Caspian offshore conditions
Y.Y. Shmoncheva, S.D. Abbaszade
https://doi.org/10.65294/gpogc.2026.29
A hybrid artificial-intelligence workflow is proposed for multi-criteria optimization of a horizontal well trajectory and its drilling set-points under representative Caspian offshore conditions. The method combines a physics-based trajectory generator, engineering indicators for dogleg severity, mechanical specific energy, drilling time, effective target contact and screening-level operational risk, an extremely randomized trees surrogate, and a weighted composite-objective search. A transparent synthetic case with a target true vertical depth of 3200 m, a landing departure of 2200 m, a 20 m productive interval and a 3° reservoir dip was evaluated. The surrogate was trained on 10,000 Latin-hypercube designs and used to screen 180,000 additional combinations within predefined engineering bounds. For the independent test set, the coefficient of determination for the penalized composite objective was R² = 0.996 and the mean absolute error was 0.104 objective units. Relative to the conventional reference design, the verified optimized solution reduced measured depth by 1.1%, maximum dogleg severity by 32.0%, mechanical specific energy by 20.9%, estimated drilling time by 26.0% and the integrated screening risk index by 16.6%. Predicted rate of penetration increased by 33.7%, while effective in-zone contact increased by 57.5%. The results are illustrative outputs of a reproducible synthetic model rather than disclosed field-performance data. The workflow is intended as an engineering decision-support layer that screens feasible alternatives and updates recommendations from MWD/LWD information while retaining expert control.