Study on Downhole Torque Prediction Method Combining Machine Learning and Mechanism Model

Long horizontal Wells have been widely used in the development and utilization of unconventional oil and gas resources such as shale oil and gas. However, excessive torque has been a limiting factor for the length of horizontal Wells. Accurate prediction of downhole torque is a key technique to improve the rate of penetration and achieve safe drilling in the horizontal section. However, it is currently difficult to directly measure downhole torque due to the limitations of downhole measuring tools. Therefore, a new downhole torque prediction method based on soft sensing ideas combining artificial intelligence with string mechanics is proposed in this paper. Firstly, GA-BP with field-measured data is used to predict rotary torque. Then, the downhole torque is inverted by the soft rope model. Finally, the efficiency of torque transfer during drilling is evaluated. The results show that the downhole torque only accounts for 27.6% of surface torque due to the presence of drag and torque. The research results have important guiding significance for the safety control and optimization of long horizontal well drilling.