GCTrajectory: A CNN-Transformer Adversarial Learning Framework for Flight Trajectory Prediction

Accurate flight trajectory prediction is critical for modern air traffic management. We propose GCTrajectory, a novel adversarial learning framework that integrates CNN-Transformer architectures for enhanced spatiotemporal modeling. Our generator employs convolutional layers to extract local spatial features, positional encoding to preserve temporal order, and Transformer encoders to capture global dependencies. A bidirectional LSTM discriminator provides adversarial feedback through sequence-level authenticity evaluation. The joint loss function combines adversarial training with L2 regularization to balance prediction accuracy and physical plausibility. Experimental results demonstrate superior performance over conventional baselines, achieving favorable outcomes on our collected real-world dataset.