A Multi-Agent Decision-Making Approach for Automated Inter-Airline Slot Swapping in Ground Delay Program with Airline Privacy Preservation

The inter-airline slot-swapping phase in the Ground Delay Program (GDP) allows airlines to collaboratively revise schedules and manage delayed flights. However, due to the airlines' reluctance to disclose sensitive information, most approaches to identify optimal slot swaps between airlines face practical challenges. In addition, current practices for slotswapping are largely manual, requiring submitted offers from airlines to operate. To achieve privacy-preserving and automated properties, we propose a multi-agent inter-airline slot swapping framework, where each airline has an intelligent agent assisting it to do the swaps. Two types of agents are implemented: the heuristic agent and the reinforcement learning agent. The framework operates in a multi-round setting, allowing an agent to gradually disclose information. In each round, agents submit lists of offers to the pool, and the coordinator subsequently matches the offers. An agent's objective is to minimize the airline's disruption costs: delay costs and cancellation costs. The results show that the total cost of the system is reduced by 9.4 percent after using the framework, and the reinforcement learning agent achieves a 12.8 percent cost reduction on average for an individual airline.