Optimization of drilling rate based on genetic algorithms and machine learning models

Abstract

During the oil and gas exploration phase, the drilling rate is a key indicator for assessing efficiency, and its accurate prediction is crucial for optimizing exploration and production. By constructing multiple data-driven intelligent drilling rate prediction models, including Decision Tree (DT), Random Forest (RF), Support Vector Regression (SVR), LassoCV, K-Nearest Neighbors (KNN), Multi-Layer Perceptron (MLP), Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LGBM), and combining them with Genetic Algorithm (GA) to explore the globally optimal model parameter combinations, the accuracy of drilling rate predictions is enhanced. The models are compared and analyzed based on Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R-squared (R²), with GA-LGBM identified as the optimal intelligent drilling rate prediction model. The GA-LGBM model demonstrated good generalization ability and robustness in field tests on two wells. SHapley Additive exPlanations (SHAP) plots are used to analyze the contribution and impact of parameter features on the predictions. Adjustments to positively impactful parameters are made to optimize the drilling rate. Additionally, two-dimensional contour plots illustrate the variation trends of drilling rate under different Weight on Bit (WOB) and RPM conditions, providing reliable data support and visual guidance for optimizing drilling rate. This research provides engineers with reliable data support and strategic guidance, aiding them in strategy control and optimal parameter adjustments for drilling operations under complex conditions.