Two Supervised Machine Learning Approaches for Wind Velocity Estimation Using Multi-Rotor Copter Attitude...

Area of Science:

• Atmospheric Science
• Robotics and Control Systems

Background:

• Wind velocity estimation is crucial for various applications, including renewable energy and aviation.
• Traditional methods often rely on fixed infrastructure or complex dynamic models, limiting their applicability in diverse terrains.

Discussion:

• This research evaluates two machine learning models, K-nearest neighbor (KNN) and Long Short-Term Memory (LSTM) neural networks, for wind velocity estimation.
• The study utilizes on-board inertial drone data and wind tower measurements, focusing on the drones' stabilization response to wind fields.
• The LSTM model demonstrated superior performance, especially in variable wind conditions, achieving an average root mean square error of 0.61 m·s⁻¹.

Key Insights:

• LSTM networks effectively predict wind speeds by learning the direct relationship between drone attitude and wind fluctuations.
• The LSTM model significantly outperforms linear regression in variable wind regimes.
• This method bypasses the need for complex drone dynamic models, simplifying wind speed estimation.

Outlook:

• The developed approach shows promise for wind velocity estimation in challenging environments such as mountainous regions and offshore locations.
• Further research could explore hybrid models or advanced deep learning architectures for enhanced accuracy and robustness.
• This drone-based method offers a flexible and cost-effective solution for distributed wind monitoring.