Ph.D. Dissertation (Georgia Tech)
Research Motivation
As long-term forecasts for aviation indicate significant growth, most airlines are concerned with the associated potential for expensive delays. Since the delays are primarily caused by weather, the airlines typically gather all available weather information before departure to generate routes that avoid hazardous weather while minimizing operating expenditures. However, they potentially have to perform in-flight re-planning because weather information can significantly change after departure. Given the advent of new communication and technology brings more information in the near future, these in-flight re-planning activities may have the potential to increase the flight crew workload, which may adversely impact safety. Another potential issue is that weather forecasts being used in the current in-flight re-planning system provide relatively unreliable information and are not accessible fast enough so that it challenges pilots to perform in-flight re-planning more accurately and frequently.
Time-series weather visualization on May 18, 2019 (Red polygons represent areas of convective weather where pilots should avoid)
Key Idea
This research attempts to resolve the potential issues by developing an automated framework that leverages various machine learning techniques. More specifically, this research 1) uses a supervised machine learning technique to yield a continues wind prediction model, 2) utilizes an unsupervised machine learning technique to forecast reliable and up-to-date convective activity, and 3) combines an instance-based learning technique with the A* search algorithm to optimize flight routes.
Overview of the proposed methodology
Results
The results show that 1) the simulated optimum flight route generated by the proposed methodology reduced by approximately 8 percent the duration of the historical flight cases and 2) the proposed methodology provided a better picture of the nearby convective weather activity compared to the most well-known convective weather product.
Publications
[6] ICRAT 2020
Comments