Multi-horizon flight trajectory prediction enabled by time-frequency wavelet transform.

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Data Science

Background:

• Flight trajectory prediction is crucial for air traffic control.
• Previous WTFTP framework faced error accumulation in long-horizon forecasts due to its iterative single-horizon method.

Purpose of the Study:

• To enhance multi-horizon prediction performance for flight trajectories.
• To explore the potential of time-frequency analysis in improving trajectory prediction.
• To mitigate cumulative errors in long-term flight path forecasting.

Main Methods:

• Proposed the WTFTP+ framework with an encoder-decoder neural architecture.
• Employed a direct multi-horizon prediction paradigm to avoid iterative errors.
• Introduced a time-frequency bridging mechanism to capture correlations in flight patterns.

Main Results:

• WTFTP+ maintains the high single-horizon prediction accuracy of the original WTFTP.
• WTFTP+ significantly improves multi-horizon prediction accuracy.
• Achieved over 40% reduction in mean deviation error at the 5-minute horizon compared to WTFTP.

Conclusions:

• WTFTP+ effectively addresses the limitations of previous models for long-horizon flight trajectory prediction.
• The enhanced framework offers superior accuracy and robustness in air traffic control applications.
• Time-frequency analysis, when integrated with advanced neural architectures, shows great promise for complex trajectory forecasting.