Research on Torque Prediction Method of Drill String Based on TCN-KAN

Abstract

To accurately predict drill string torque, enhance drilling efficiency, and reduce operational costs, this paper addresses the issues of limited generalization capability and high complexity in torque prediction models during horizontal well drilling by conducting research on intelligent prediction methods based on deep learning. An intelligent torque prediction model based on TCN-KAN is proposed. Multi-source characteristic variables such as drilling depth, rotary speed, and weight on bit are selected for model training, and a TCN comparative model is established for experimental validation. The results show that the TCN-KAN model outperforms the comparative model in evaluation metrics such as root mean square error and mean absolute error, achieving a prediction accuracy of 98.58%. It effectively captures the complex nonlinear relationships between torque and drilling parameters. This model is suitable for drilling torque monitoring, intelligent feed optimization, and drilling decision support, providing reliable technical support for the automation and intelligence of drilling operations.