4D trajectory prediction for inbound flights.

Area of Science:

• * Geospatial analysis and artificial intelligence
• * Advanced algorithms for trajectory forecasting

Background:

• * Existing 4D trajectory prediction methods face challenges with cumulative errors, complex spatiotemporal features, and computational limitations.
• * Accurate modeling of dynamic movements is crucial for various applications.

Purpose of the Study:

• * To develop a high-precision, robust method for 4D trajectory prediction.
• * To overcome limitations in current models regarding accuracy, efficiency, and generalization.

Main Methods:

• * A hybrid model, SVMD-DBO-RCBAM, integrating Sequential Variational Modal Decomposition (SVMD), Dung Beetle Optimization (DBO), and ResNet-CBAM was developed.
• * Key innovations include frequency-domain feature decoupling, dynamic parameter optimization, and enhanced spatiotemporal feature extraction.

Main Results:

• * Achieved a low longitude Mean Absolute Error (MAE) of 0.0377 in single-step prediction, a 38.5% improvement over baseline models.
• * In multi-step prediction, the longitude R2 reached 0.9844, with a 72.9% reduction in cumulative error rate.
• * Demonstrated high accuracy and stability with prediction error Interquartile Range (IQR) less than 10% of traditional models.

Conclusions:

• * The proposed SVMD-DBO-RCBAM model offers superior performance in 4D trajectory prediction.
• * The method effectively addresses cumulative errors and enhances the modeling of complex spatiotemporal dynamics.
• * Results indicate significant improvements in prediction accuracy, stability, and generalization ability.