New method prediction of rock mechanics parameters in drilling process based on machine learning

Abstract

During the drilling process, rock mechanics parameters (RMP) are an important basis for optimizing drilling fluid density, drill bit selection, and wellbore stability. However, during the actual drilling, traditional methods cannot obtain the RMP of arbitrary wellbore sections. To solve the problem of RMP, a novel data-driven method for predicting RMP is proposed. Initially, the Pearson correlation coefficient method is used to analyze the correlation between the drilling data of the actual well and the label of RMP obtained by logging. And the major influence factors (e.g., drilling pressure, drilling speed) of RMP are chosen during drilling. Subsequently, three machine learning models of RBF, BP and KNN integrating the major influence factors of RMP are established. Furthermore, based on actual drilling parameters and RMP data as labels, the model is trained and verified, and hyperparameters are optimized. The results show that the RBF neural network exhibited strong generalization ability, achieving an accuracy of 87.69%. Compared to BP and KNN, the optimal model RBF neural network prediction accuracy is 4.74% and 6.13% higher. The new method of data-driven RMP prediction is of great significance in guiding drill bit selection and borehole stability during drilling.