Comparative Study of Single and Multi-Objective PSO for Real-Time Drilling Optimization Using Stacked Ensembles

Abstract

This study presents a data-driven, real-time drilling optimization framework and a direct comparison between single-objective PSO and multi-objective PSO (MOPSO) for optimizing WOB, RPM, and flow rate (Q). A stacked ensemble surrogate model was trained to predict ROP and HMSE, achieving high generalization accuracy (ROP: R2_test = 0.9918; HMSE: R2_test = 0.9879). The optimizer was executed in a depth-wise manner across 3576–3866 m with an update interval of 10 m, enabling adaptive setpoint recommendations along the well section. Single-objective PSO consistently increased penetration rate but tended to generate more aggressive parameter trajectories. In contrast, MOPSO produced Pareto-optimal solutions that explicitly balance drilling speed and energy efficiency, enabling practical “knee-point” selection. Quantitatively, the baseline interval exhibits a mean ROP ≈ 23.93 m/hr, while MOPSO identifies feasible improvements up to ∼33.16 m/hr (≈+38.6% uplift) and simultaneously reaches HMSE as low as ∼8.62 × 107 J/m3, demonstrating that significant ROP gains can be achieved without excessive energy loading when trade-offs are enforced. Overall, the results indicate that MOPSO is more suitable for field implementation because it supports decision-making under realistic performance–efficiency trade-offs and yields more operationally defensible setpoints than pure ROP-maximization.