Research on Drill String Vibration Prediction Based on Bidirectional Long Short-Term Memory Network

Drill string vibration is a key factor affecting drilling efficiency, safety, and costs. Its complex nonlinear characteristics limit the accuracy of traditional prediction methods based on simplified physical models or empirical formulas. To address this, this study proposes a drill string vibration prediction method based on Bidirectional Long Short-Term Memory (Bi-LSTM) networks. By exploring the dynamic correlations in multi-source time-series data, the model effectively captures the nonlinear coupling effects of parameters such as Weight on Bit (WOB), Rotations Per Minute (RPM), and torque on vibration. Compared to models such as LSTM, XGBoost, Random Forest, and Linear Regression, Bi-LSTM shows significant advantages in performance, with an R2 value of 0.9581, Mean Squared Error (MSE) of 0.0064, Mean Absolute Error (MAE) of 0.0627, and Root Mean Squared Error (RMSE) of 0.0801, validating its superiority in complex drilling vibration prediction. Further analysis using boundary contour maps of vibration quantifies the interactive effects of WOB and RPM on vibration intensity, providing a visual basis for drilling parameter optimization. The results show that the Bi-LSTM-based prediction model can achieve high-precision real-time vibration forecasting. By dynamically adjusting drilling parameters, it can effectively reduce the risks of equipment damage, drilling failure, and wellbore instability caused by vibration, thereby improving operational safety and economic efficiency. This study provides a new method for active vibration control and decision optimization in intelligent drilling, with potential for future integration with multi-physics models and real-time data streams to expand its applicability in complex geological conditions.