RecovAir: Model-driven Airline Scheduling Tool for Disruption Recovery

MAX Z LI^1*Michael Peng²

• ¹University of Michigan, Ann Arbor, United States
• ²University of California, San Diego, La Jolla, California, United States

The Southwest Airlines scheduling crisis of December 2022 and its consequences have highlighted the importance of robust airline disruption management and recovery. A wide variety of approaches have been applied to airline schedule recovery and robustness, but they are often evaluated with respect to static snapshots of disruption scenarios, which lend little consideration toward how recovery decisions interact with emerging disruptions over time. To help future research estimate and improve the utility of airline recovery strategies, we present RecovAir, a high-performance agent-based model that simulates the flow of aircraft, crew, and passengers in an airline's flight network under disruptive departure and arrival rate limits and repeated applications of ad-hoc recovery strategies. By measuring Key Performance Indicators like On-Time Performance, cancellation count, and total delay, RecovAir supports comparisons and controlled experiments with recovery parameters. We demonstrate RecovAir's utility by synthesizing plausible scenarios for both the first day of the 2022 scheduling crisis and a day with zero cancellations in 2024 for Southwest Airlines. We simulate these scenarios while varying recovery strategies and prioritization between delays and cancellations. Our results show that a simple greedy algorithm can perform nearly as well as Southwest Airlines' actions on the first day of the scheduling crisis without initiating any ferry flights (i.e., non-revenue flights to reposition airline crew)-critically, we do not use any proprietary crew schedules. We then test a range of values for the maximum delay before cancellation parameter and discover an inversely proportional relationship between total delay and number of cancellations beyond a constant baseline. We envision RecovAir as a novel, lightweight simulation platform where airline stakeholders and researchers can rapidly evaluate schedule recovery algorithms without the burden of large-scale data collection efforts.