| Resumo : |
Passenger boarding is on the critical path for airplane turnaround time and a contributor to delays. In the pre-pandemic period, flight delays cost U.S. passengers and airlines nearly $30 billion, according to the FAA (Federal Aviation Administration), highlighting the importance of minimizing all delay sources, including boarding inefficiencies. The literature discusses strategies that minimize boarding time, focusing on pre-defined ones that usually do not consider the context of passengers' profiles and behavior. Based on that, this study investigates how passenger profile influences boarding efficiency and the perceived Level of Service (LoS), aiming to support airlines in selecting boarding strategies for a given context. For that, it combines simulation and machine learning techniques, allowing airlines to evaluate different boarding scenarios in a replicable way and in a data-driven manner. A simulation environment was developed and verified using an Airbus A320 layout (174 seats), the most common airplane in commercial fleets. Three of the most used strategies in the literature and on real-world operations (Back-to-front, Random, and Outside-in) were tested with four load factors (70%, 85%, 90%, 100%) and two scenarios that were modeled in order to allow analyzing the effect of adding the passenger profile as a decision variable when setting boarding strategies: a baseline with standard passenger behavior, and a passenger profile scenario where passengers were grouped by their respective agility based on real data and followed probabilities distributions for their behavior. Each condition was simulated 10,000 times, resulting in 240,000 flight samples. These were used to develop a model that can be helpful to predict boarding time and perceived LoS for a given context of passenger profile. From eleven machine learning techniques, the XGBoost model achieved the highest R² of 0.96 for predicting boarding time and an accuracy of 0.95 for interferences. Results show that incorporating passenger profiles increased total boarding time and interferences, except for seat interferences, which decreased because of faster groups. Between the strategies, Random was efficient (811s to 1332s) but resulted in higher interferences. Outside-in provided a more balanced trade-off solution between efficiency (877s to 1749s) and perceived LoS. Back-to-front performed the worst due to the highest aisle congestion (which is the main factor influencing boarding time for all strategies, primarily from luggage handling), with boarding times ranging from 1433s to 2551s, and should be avoided. |