| Resumo : |
In recent years, there has been a global effort to modernize the Air Traffic Management (ATM) system in order to enable the continued growth of air traffic and meet increasingly stringent targets for efficiency, operational safety and environmental impact. In addition to operational and technological improvements, the efficient use of operational data is also critical to improving ATM and bringing efficiency gains to air traffic operations. This thesis explores the increasing availability and accessibility of flight path data, generated by new surveillance technologies, to develop a data analysis framework for characterization and prediction of air traffic performance with a view to decision support in the ATM. Machine learning methods are used to identify air traffic patterns and characterize their spatiotemporal performance at multiple scales. From this knowledge, data-based models are developed for flight path prediction. Three case studies are used to demonstrate the application of the framework and discuss its impacts. The first application allowed detailed comparisons of air traffic flow efficiency in multiple flight phases for the 20 main origin-destination pairs in the domestic market, revealing the most and least efficient routes, and allowing the identification of potential causes of inefficiencies in the airspace. Brazilian. The second case study focused on a post-implementation analysis of the recent change in the airspace structure of the São Paulo terminal area. It was observed that the implementation of the Point Merge concept for arrival operations at São Paulo/Guarulhos International Airport provided a small gain in horizontal efficiency, but had a negative impact on temporal efficiency. The results emphasize the importance of implementing decision support tools for tactical sequencing in order to achieve the full benefits of innovative airspace concepts such as Point Merge. Finally, in the last application of the framework, regression and classification models are developed to predict routes and flight times for short and long-haul domestic flights to São Paulo airports. It was found that models based on machine learning had a better predictive performance than a model based on current practice, for all phases of flight. The results demonstrate the potential of the proposed trajectory data analysis framework to support airspace design and performance monitoring processes and to serve as a basis for the development of predictive capabilities in support of traffic flow management. |