Advancing Collision Risk Models in Terminal Airspace using BlueSky Simulator and Gaussian Process Regression

Abstract

In recent years, with the rapid growth in air traffic and the increasing complexity of terminal airspace operations, the need for reliable collision risk models has become paramount. This research sought to address the field of collision risk models, specifically through the BlueSky simulator and Gaussian Process Regression (GPR). Existing models require high computational power and large trajectory datasets. One of the primary struggles faced was the lack of easily accessible and relevant data. Many aviation datasets are either proprietary and could not be used to validate and refine our models. Furthermore, gaining a comprehensive understanding of collision risk models was difficult. The majority of time was spent conducting literature review to contribute to a report. This research used BlueSky, a versatile, open-source tool designed for conducting simulations of air traffic. The BlueSky simulator allows for a more controlled environment for experimenting with GPR than aviation datasets available online. BlueSky was utilized to create synthetic flight trajectories and collect corresponding data on latitude, longitude, and altitude. JupyterLab and Scikit-Learn were used to experiment with GPR kernels and their hyperparameters. While valuable insights were gained in creating synthetic trajectories and selecting functions to describe aircraft data, further investigation is needed to enhance the accuracy and applicability of these models in real-world scenarios.