Airspace Collision Risk Hot-Spot Identification using Clustering Models

Abstract

A key safety indicator for airspace is its Collision Risk estimate, which is compared against a Target Level of Safety (TLS) to provide a quantitative basis for judging the safety of operations in airspace. However, this quantitative basis fails to provide any insight regarding the magnitude, location and timing of the risk of collision, distributed within a given airspace. In this paper, we propose a methodology for identification of Collision Risk Hot-Spots in a given airspace. The proposed methodology consists of processing air traffic data and developing traffic routes based on entry and exit points within the airspace. These routes and other flight information are then used to project air-traffic crossings and cluster potential collisions. The proposed method then estimates the Collision Risk for each identified cluster, culminating in Risk Assessment for the entire airspace. The model extends and adopts state-of-the art clustering models, systemically identifies airspace collision risk hot spots and further analyses hot spots by analysing cluster features (number of points and contribution to overall risk) with flight levels and time of day. Experiments were conducted using one-month traffic data (25,440 flights) from Bahrain en-route airspace. By visualizing crossing points and clustering them in a two dimensional GIS (Geographic information system) model we are able to identify Collision Risk Hot-Spots which contribute significantly to overall collision risk.

SAPIENT: Enabling Real-Time Monitoring and Control in the Future Communication Infrastructure of Air Traffic Management

“…Such information can then be used to efficiently predict the position of ACs during the flight. In [6], instead, authors propose a method that uses air-traffic data to compute traffic routes and identify the risk of collisions in certain hot spots. All the methods falling in this category are complementary to the SAPIENT system, as information coming from the latter can be used together with knowledge of the trajectory to select the collision-free route having the best communication performance.…”

LSTM-based Flight Trajectory Prediction

“…Turning points may provide rich information, which were clustered and then used to build a stochastic model [27]. Clustering models were used to identify the hot-spot [28] where airspace collision may occur. Historic flight data was trained on collision risk.…”

A Deep Unsupervised Learning Approach for Airspace Complexity Evaluation

“…Therefore, evaluation of airspace complexity is a non-trivial problem, attracting many investigations from scholars and field practitioners [6]- [16], [22]- [40]. Currently, the state-of-the-art methods for airspace complexity evaluation fall into two main categories: (1) Airspace complexity is described through a single indicator [6]- [14]; (2) Airspace complexity is evaluated based on a multi-indicator system [15], [16], [22]- [31].…”