Altitude aware trajectory prediction methods for non towered terminal airspace

Area of Science:

• Aviation Safety
• Artificial Intelligence
• Air Traffic Management

Background:

• Accurate trajectory prediction is crucial for aviation safety, especially in non-towered terminal airspace where real-time air traffic control is absent.
• Frequent altitude changes in this airspace pose significant challenges for existing trajectory prediction models.

Purpose of the Study:

• To develop an altitude-aware trajectory prediction model, ACTrajNet, designed to enhance aviation safety in non-towered terminal airspace.
• To improve the accuracy of flight path prediction by effectively incorporating vertical flight dynamics.

Main Methods:

• ACTrajNet utilizes temporal convolutional networks (TCN) for independent extraction of altitude features.
• A channel attention fusion mechanism dynamically integrates altitude features into the trajectory representation.
• The model is evaluated on the TrajAir dataset to assess its performance against state-of-the-art methods.

Main Results:

• ACTrajNet demonstrated superior performance compared to baseline models on the TrajAir dataset.
• The model achieved up to 2.47% improvement in Average Displacement Error (ADE).
• A 2.42% improvement in Final Displacement Error (FDE) was observed, highlighting enhanced prediction accuracy.

Conclusions:

• ACTrajNet effectively captures complex vertical flight patterns, outperforming existing trajectory prediction models.
• The proposed altitude-aware approach significantly enhances aviation safety in non-towered terminal airspace.
• This model shows strong potential for real-world application in air traffic management systems.