An explainable ensemble machine learning model to elucidate the influential drilling parameters based on rate of penetration prediction

Abstract

Many researchers have examined the benefits of machine learning (ML) algorithms in geothermal drilling, especially for predicting the rate of penetration (ROP) of drilling. However, little attention has been given to enhancing the explainability of ML algorithms. Motivated by this, this study examines the feasibility of an explainable ML approach for elucidating the behavior of the drilling parameters to generate explainable insights regarding the surface-controllable drilling parameters that could maximize the ROP for drilling effectiveness (DE). First, this study presents a performance analysis using hyper-parameter technique of the ensemble supervised machine learning (ESML) for ROP predictions. Secondly, this study uses an explainable ML method to explain the drilling parameters' behavior based on the ESML prediction. The results show that the ensemble CatBoost model shows the best performance for ROP prediction. Then, the explainable ML successfully revealed several influential drilling parameters that could maximize the ROP for DE. In earlier drilling (0–4,000 m in depth), temperature and pressure contribute to a lower ROP, and the drilling fluid flow rate and rotary speed contribute to a higher ROP. On the other hand, as the drills go deeper into the earth (4,000–7,536 m in depth), maximum ROP can be obtained by maximizing the drilling pressure, weight on bit (WOB), surface torque, and rotary speed (RPM) of the drill bit, while drilling fluid flow rate does not significantly contribute to ROP. The explainable ML also highlighted the drilling parameters' inflection points when a switch from low to high rates of ROP occurs. Overall, the findings show the proposed model's strength in generating explainable insights that can improve the selection of appropriate ROP settings and support future decision-making processes in geothermal drilling.