Physics-Informed CNN-BiGRU Model for Downhole Weight-on-Bit Prediction Using Surface Measurement-While-Drilling...

Area of Science:

• Petroleum Engineering
• Machine Learning
• Drilling Engineering

Background:

• Accurate downhole weight on bit (DWOB) is vital for optimizing drilling parameters and ensuring safety, especially in extended-reach horizontal wells.
• Surface weight on bit (SWOB) measurements are unreliable for DWOB estimation due to attenuation from drill-string friction, wellbore trajectory, and borehole contact.

Purpose of the Study:

• To develop a robust and accurate method for continuous DWOB estimation.
• To address the limitations of traditional SWOB measurements in complex wellbore environments.

Main Methods:

• A physics-informed convolutional neural network (CNN) combined with a bidirectional gated recurrent unit (BiGRU) prediction model was developed.
• The CNN-BiGRU model was optimized using the whale optimization algorithm (WOA) for hyperparameter tuning and physics-constrained weighting.
• A drill-string frictional attenuation relationship was integrated into the model's loss function as a physical prior.

Main Results:

• The proposed CNN-BiGRU-WOA model demonstrated high accuracy in DWOB prediction, achieving an R² of 0.9245 on the test set.
• The model yielded low error metrics: RMSE of 0.1856, MAE of 0.1135, and MRE of 0.0124.
• Validation using field data from a horizontal well confirmed the model's effectiveness.

Conclusions:

• The developed data-physics hybrid framework significantly improves the accuracy and physical consistency of DWOB predictions.
• This approach offers a reliable solution for real-time DWOB estimation in challenging drilling operations.
• The integration of physical principles enhances the generalizability and trustworthiness of machine learning models in petroleum engineering.