Skip to content

Towards Situation-aware State Modeling for Air Traffic Flow Prediction

Home / Papers / Towards Situation-aware State Modeling for Air Traffic Flow Prediction

Towards Situation-aware State Modeling for Air Traffic Flow Prediction

Authors: Anqi Liu, Bin Wang, Jigang Zhao, Dechuan Ma, Guiyuan Jiang, Feng Hong, Yanwei Yu, Tianrui Li Date: 2026-04-13 Paper ID: openalex:2604.11198

Summary

The paper proposes AeroSense, a novel framework for air traffic flow prediction in terminal airspace that moves away from traditional aggregated time series forecasting. By modeling the airspace as a dynamic set of microscopic aircraft states, the approach effectively captures real-time kinematics and inter-aircraft interactions. Using a masked self-attention architecture with decoupled prediction heads, the model achieves state-of-the-art predictive performance on a large-scale real-world airport dataset. The results demonstrate superior robustness and interpretability compared to standard time-series paradigms.

Key Contributions

  • Introduces AeroSense, a direct state-to-flow modeling framework that processes microscopic aircraft states as a dynamic set instead of aggregate time series.
  • Develops a situation-aware state representation to capture real-time terminal airspace dynamics.
  • Outperforms traditional time series-based baselines on a large-scale real-world airport dataset, particularly during peak traffic periods.

Open Questions & Future Work

Key Concepts

  • state-to-flow-modeling: A modeling paradigm that represents complex systems as a dynamic set of microscopic states and maps them directly to macroscopic flow outputs, bypassing traditional aggregate time-series sequences.

Archivist Review

I approved ‘State-to-Flow Modeling’ as a core paradigm shift in time-series forecasting, preferring it over the specific ‘AeroSense’ implementation. The open question regarding scalability was refined to focus on the technical challenge of set-based representations in high-density, heterogeneous environments, ensuring it addresses a substantial bottleneck in the field.

Approved Concepts

  • State-to-Flow Modeling: This represents a significant methodological shift from aggregate time-series forecasting to set-based dynamical modeling in cyber-physical systems, which is highly reusable.

Approved Open Questions

  • Scalability of Set-based Air Traffic Forecasting: As a paradigm shift, validating the robust scaling of these set-based architectures is essential to determine if they can replace legacy time-series methods in operational settings.

Rejected Candidates

  • [concept] AeroSense (aerosense) - subcomponent_of_broader_mechanism: AeroSense is the paper-specific implementation of the broader ‘State-to-Flow Modeling’ paradigm; the latter is a more appropriate and reusable entry for the vault.
  • [open_question] Scalability of Set-based Air Traffic Modeling (scalability-of-set-based-air-traffic-modeling) - duplicate_existing: The proposed question was rewritten for clarity and to better align with standard naming conventions while preserving the core technical research problem.

Metadata & Links

url
https://arxiv.org/abs/2604.11198
paper_id
2604.11198
paper_source
openalex
domain
time-series
tags
architectures
datasets
concept_slugs
state-to-flow-modeling
dataset_slugs
skill
TimeSeriesSkill
created_at
2026-04-16T05:46:46Z