A Stacked Generalization Ensemble Method for Rate of Penetration Prediction

Resumo

Efforts to reduce drilling costs and duration have made accurate predictive models for rate of penetration (ROP) essential in the drilling industry. These models assist decision-making concerning parameters that affect drill efficiency. Utilizing advanced machine learning algorithms, such as ensemble methods and artificial neural networks, has become a clear trend aimed at enhancing predictive precision. In this study, a stacked generalization ensemble model is introduced to improve ROP prediction performance. The adopted approach combines four base learners, namely random forest, gradient boosting, linear regression, and artificial neural networks, into a meta-model architecture. The resulting meta-data from these models are used to make the final ROP prediction using a linear regression algorithm. Drilling data from two wells in the Volve oil field are used for training, including various operational and formation-related parameters, such as average surface torque, weight on bit, average rotary speed, mud flow rate, and delta-T compressional. The performance of the model is evaluated on an unseen well, using error metrics such as mean absolute error (MAE) and mean absolute percentage error (MAPE). The proposed approach has demonstrated superior performance compared to the base learners, as indicated by the comparative analysis. This suggests its potential to enable more accurate predictions, consequently improving the efficiency of the drilling process.