Improving the accuracy of dynamic inclination measurement by machine learning

Area of Science:

• Petroleum Engineering
• Geophysics
• Machine Learning

Background:

• Diminishing oil and gas resources necessitate advanced drilling technologies like Rotary Steerable Systems.
• Vibrations and shocks in drilling impact Measurement While Drilling (MWD) accuracy.
• Limited research exists on precise drilling tool attitude measurement, crucial for directional drilling.

Purpose of the Study:

• To improve the accuracy of dynamic inclination measurement for drilling tools.
• To address challenges posed by vibrations and shocks in MWD.
• To enhance data support for drilling operations through precise attitude measurement.

Main Methods:

• Utilized drilling tool attitude sensor data combined with artificial neural networks.
• Employed machine learning, integrating real-time z-axis acceleration and magnetic induction signals.
• Applied a deep learning model (Long Short-Term Memory) to invert x and y-axis acceleration signals for inclination calculation.

Main Results:

• Achieved high-precision dynamic inclination angle measurements.
• Reduced dynamic inclination curve errors to 0.4°-0.7° under simulated conditions.
• Demonstrated strong adaptability to varying rotational speeds.

Conclusions:

• The proposed method significantly enhances the accuracy of dynamic inclination measurement.
• The technique offers improved data reliability for drilling operations, especially in challenging conditions.
• This advancement supports more efficient and precise directional drilling in the oil and gas industry.