Bayesian Spatiotemporal Graph Transformer for Aircraft Trajectory Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current air traffic management systems face challenges in predicting multi-aircraft trajectories accurately, especially in congested near-terminal airspace, due to uncertainties from environmental and human factors, and lack the capability to quantify these uncertainties effectively, which is crucial for safety and efficiency.
Innovation Solution
A Bayesian Spatiotemporal Graph Transformer network (B-STAR) is developed to predict multi-aircraft trajectories by incorporating Bayesian deep learning for uncertainty quantification, using a graph-based model that encodes aviation regulations and leverages spatiotemporal correlations to infer uncertainties directly from flight track data, rather than relying on pre-defined parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional trajectory prediction methods are used, then the system is simpler to implement, but the prediction accuracy and uncertainty quantification capability deteriorate
Solution Approach 1:
The patent replaces traditional mechanical/mathematical prediction models with a Bayesian neural network system that uses deep learning to predict aircraft trajectories. The system substitutes deterministic algorithms with probabilistic deep learning models that can automatically learn complex patterns from historical flight data, achieving superior prediction accuracy while managing complexity through modular architecture.
Solution Approach 2:
The patent transforms the prediction approach by changing from fixed deterministic parameters to probabilistic distributions. The Bayesian neural network outputs not just single trajectory predictions but entire probability distributions over possible trajectories, enabling uncertainty quantification. This parameter transformation allows the system to capture the inherent variability and unpredictability in aircraft behavior.
2Reliability
If deterministic prediction models are used, then the computational process is simpler, but the ability to quantify uncertainty deteriorates
Solution Approach 1:
The patent replaces deterministic prediction models with Bayesian neural networks that inherently provide uncertainty quantification. Instead of producing single-point predictions, the Bayesian model outputs probability distributions that naturally encode uncertainty. This substitution transforms the modeling approach from certainty-based to probability-based, enabling reliable uncertainty estimation for safety-critical air traffic management.
Solution Approach 2:
The Bayesian neural network incorporates feedback loops during training and inference that continuously refine uncertainty estimates. The model learns from historical data how prediction uncertainties correlate with various flight conditions, traffic patterns, and environmental factors. This feedback mechanism enables the system to adaptively adjust uncertainty quantification based on observed patterns, improving reliability without requiring manual calibration.
3Reliability
If complex Bayesian deep learning models are used, then uncertainty quantification improves, but the training and inference time increases
Solution Approach 1:
The patent performs preliminary training of the Bayesian neural network offline using extensive historical flight data. During this pre-training phase, the model learns to quantify uncertainties for various flight scenarios. Once trained, the model can rapidly infer uncertainties for new trajectories without requiring extensive computation. This preliminary action separates the computationally intensive learning phase from the time-sensitive inference phase, reducing real-time delays.
Solution Approach 2:
The patent implements dynamic computation strategies where the level of uncertainty quantification detail adapts based on operational needs. For high-risk scenarios or edge cases, the model performs more comprehensive uncertainty analysis, while for routine predictable flights, it provides faster, coarser estimates. This dynamic approach optimizes the trade-off between accuracy and speed, allocating computational resources efficiently across different flight situations.
Data Source
AI summary
A system includes a Bayesian Spatiotemporal Graph Transformer (B-STAR) architecture that models spatial and temporal relationship of multiple agents under uncertainties. The system enables Multi-Agent Trajectory Prediction for safety-critical engineering applications, (e.g., autonomous driving and flight systems) and considers the impact of various sources, such as environmental conditions, pilot/controller behaviors, and potential conflicts with nearby aircraft. It is shown that B-STAR achieves state-of-the-art performance on the ETH/UCY pedestrian dataset with UQ competence.