A Real-Time Decision-Support Tool for the Integrated Airline Recovery using a Machine Learning Approach

Airlines frequently deal with unexpected disruptions, which have to be resolved in order to resume their operations again. Decision-support tools help airlines with disruption management. However, long computational times are associated with integrated recovery models that compute globally optimal solutions, which makes these tools unfit for large airlines. Faster sequential approaches have been proposed, which recover one resource at a time, but often only provide local optima. This paper presents a real-time decision-support tool to solve the integrated airline recovery problem. The model simultaneously recovers the schedule and allocates aircraft and pilot pairs to the flights, while minimizing additional incurred costs. Passengers are implicitly recovered by considering missed connections. Recovery actions include delaying and cancelling flights, swapping aircraft, and swapping, deadheading or using reserve crew. A machine-learned ranking algorithm reduces the computational complexity of the problem by selecting a subset of the resources that are likely to be involved in the recovery plan, such that only part of the network is considered. The decision-support tool was evaluated on disruption scenarios of one of the largest airlines in the world: Delta Airlines. The results show that the machine learning selection reduces the average computational time 15- fold compared to the integrated recovery model that uses the complete network, increasing the percentage of solutions computed under two minutes from 13% to 96%. The proposed model is able to find globally optimal solutions in 58% of the cases and yields similar results in terms of delays, but more flight cancellations in comparison with the globally optimal solutions. The proposed integrated model found a feasible solution to all disruption instances, while a benchmarked sequential model returned infeasible solutions in 4.4% of the cases. Besides, the sequential model produces 68% more flight cancellations, 4
Aerospace Engineering | Air Transport and Operations