Hybrid ROP modeling: Combining analytical and data-driven approaches for drilling

Abstract

Rate of penetration (ROP) modeling has been widely employed to improve drilling efficiency, aiming to reduce both operational costs and risks. This study presents a hybrid model for predicting ROP by combining analytical and data-driven approaches, aimed at enhancing drilling efficiency in challenging formations. The model integrates operational parameters, rock strength properties, and bit design factors, using machine learning (ML) to estimate real-time rock strength at the bit for each depth interval by predicting compressional wave velocity and lithology. These predictions facilitate the calculation of uniaxial compressive strength (UCS) and confined compressive strength (CCS), inputs for ROP estimation. The study included datasets from five wells in the Norwegian Continental Shelf, with four wells used for model training and one for model testing/validation (blind test). The Random Forest regression model achieved an R2 value of 93% for compressional wave velocity predictions, while the Random Forest classification model attained 96% accuracy in lithology prediction during blind testing. Model validation showed a strong correlation between calculated and measured ROP values, underscoring its accuracy. Sensitivity analysis was performed to evaluate the influence of various parameters, such as weight on bit (WOB), revolutions per minute (RPM), drilling fluid density, flow rate, bit hydraulics, and CCS, highlighting their interdependent effects on ROP. The sensitivity analysis indicated that CCS, WOB, RPM and bit diameter have the greatest impact on ROP. Existing ROP models lack real-time integration of lithology and compressional wave velocity at the bit, limiting their ability to estimate rock strength. This study addresses these gaps by incorporating ML to predict these parameters at each depth interval, enhancing unconfined and confined compressive strength calculations used in the ROP model. The results highlight the importance of optimizing drilling parameters to maximize ROP and operational efficiency, indicating the hybrid model's potential for real-time applications in complex drilling environments.