Real-Time Inversion of Formation Drillability and Concurrent Speedup Strategies for Microdrilling Time Optimization

Summary As the complexity of oil drilling engineering grows, real-time optimization of drilling parameters to improve efficiency and lower costs becomes an important task. In this research, we propose a novel combination of categorical boosting (CatBoost) and genetic algorithm (GA) for real-time optimization of synchronous drilling parameters with intelligent inversion of formation drillability. The intricate causal relationship between drilling parameters and drilling time is made clear by introducing the Peter-Clark (PC) causal discovery algorithm. A drilling time prediction model is then built using this information, comparing the performance of five supervised learning models across five performance metrics. Subsequently, an intelligent inversion model was designed utilizing GA to accurately anticipate formation drillability and dynamically alter drilling parameters. In field experiments on two wells, the approach greatly increased drilling efficiency. The CatBoost model performed well through 10-fold cross-validation, with determination coefficients (R²) of 0.986 and 0.990, and effectively inverted the formation drillability that cannot be directly obtained in real time (usually calculated from logging data after drilling a well) and reduced the drilling time by about 5% and 8%, respectively, after optimization. Furthermore, the Shapley additive explanation (SHAP) methodology fully quantified the impact of each parameter and enhanced the interpretability of the model. This method breaks through the traditional limitation of relying on engineers’ experience, realizes real-time drilling parameter optimization during the drilling process, and provides scientific decision support for improving drilling efficiency.