Application of Xgboost Algorithm in Rate of Penetration Prediction with Accuracy

Abstract In oil and gas exploration, rate of penetration (ROP) is one of the most important parameters that affect drilling costs and drilling performance. Therefore, prediction of rate of penetration (ROP) with accuracy has become an important direction of current drilling work. In order to solve the problem that the existing models are not accurate enough to predict the ROP, this paper proposes an ensemble learning approach called extreme gradient boosting (XGBoost) model. In this paper, in addition to the XGBoost model, different models were constructed through various approaches, such as traditional theoretical/empirical models (Hareland model and Motahhari model) and machine learning models (support vector regression (SVR), artificial neural network (ANN) and k-nearest neighbor (KNN)). Then, the accuracy of the constructed models was compared with each other. Aiming at the problem of ROP prediction, the historical data of a specific field that has been collected is mined. The input data for training these models includes mud logging parameters and geomechanical parameters. Before data is entered into the model, preprocessing is required. After normalizing the input data, ategorical variables, and use One-hot encoding method for converting categorical variables into a form which can be used in machine learning models. These data were used as input variables, and ROP is the output variable for modeling. Useing the GridSearchCV tool in scikit-learn to perform grid search and cross-validation to find the optimal parameters of the model. Comparison of accuracy and computational performance between above models for rate of penetration prediction, the results show that the XGBoost model with prediction accuracy of 85%, the accuracy of other models is 48% (Hareland model), 45% (Motahhari model),51% (KNN), 52% (ANN)and SVR (55%), respectively. It is obviously that XGBoost model has higher accuracy than traditional theoretical/empirical methods and other machine learning models, indicating that XGBoost model has stronger ability to deal with high-dimensional and non-linear problems. XGBoost model can be used to predict the ROP at the specific field reliably and accurately provided the model be developed with enough data, which can provide a scientific and reliable reference for reducing drilling costs and improving drilling performance.