ADS-B-Based Wavelet Time-Frequency Learning Model for Flight Trajectory Prediction

Flight Trajectory Prediction (FTP) is a critical task in Air Traffic Control (ATC), especially during maneuvering flights. Modern data-driven approaches often have room for improvement in terms of rapid response to control and addressing the vulnerability of prediction results. The Automatic Dependent Surveillance-Broadcast system (ADS-B) has some distinct advantages in addressing the vulnerabilities and cumulative error issues in prediction systems, particularly in terms of real-time capability, high accuracy, and directness. To address these issues, a novel time-frequency learning model is proposed—an ADS-B-based Wavelet Time-Frequency Learning Model (WTTLM). This model supports multi-step predictions in a non-autoregressive manner, utilizing wavelet transforms to capture both the global flight trends and local motion details of aircraft. It effectively addresses the lag in time-frequency modeling for rapidly changing signals and mitigates the accumulation of errors due to the vulnerability of prediction results. Experiments on real datasets validate the effectiveness of this research, showing that the WTTLM outperforms other models in terms of flight trajectory prediction performance and computational efficiency.