Transmission Characteristics and Machine Learning Prediction of Lateral Vibration in Deep Vertical Drill Strings

Severe drill string vibration can accelerate tool fatigue and failure, but direct measurement near the drill bit is difficult. Using a 6376 m vertical shale-gas well in western Sichuan as a case study, this study established and validated a coupled drill string dynamic model. Lateral acceleration signals under three operating conditions were generated, filtered, and normalized. XGBoost models were then trained using vibration signals measured at shallow and intermediate locations to estimate deep-level lateral acceleration. The results showed that vibration originated near the drill bit and propagated upward with a measurable delay, while its amplitude was generally higher in the middle section of the drill string. For single-point monitoring, the coefficient of determination (R2) ranged from 0.013 at 500 m to 0.64 at 10 m from the bit. Combining multiple monitoring locations increased R2 to 0.7071. XGBoost achieved the best balance between accuracy and computational efficiency among the tested models. The 10–300 m interval from the bit is recommended for monitoring, whereas 300–500 m is considered an uncertain zone.