Ground Delay Program Planning Using Markov Decision Processes

This paper compares three approaches for selecting planned airport acceptance rates in the single-airport ground-holding problem: the Ball et al. model, the Richetta-Odoni dynamic model, and an approach based on approximate dynamic programming. Selecting planned airport acceptance rates is motivated by current practice of ground delay program planning under collaborative decision making. The approaches were evaluated using real flight schedules and landing capacity data from Newark Liberty International and San Francisco International Airports. It is shown that planned airport acceptance rates can be determined from the decision variables of the Richetta-Odoni dynamic model. The approximate dynamic programming solution, introduced by the authors, is found by posing a model that evaluates planned airport acceptance as a Markov decision process. The dynamic Richetta-Odoni and approximate dynamic programming approaches were found to produce similar solutions, and both dominated the Ball et al. model. The Richetta-Odoni dynamic model is computationally more efficient, finding solutions 10 times faster than approximate dynamic programming, although the approximate dynamic programming approach can more easily incorporate complex objectives. Surprisingly, the performance of all three approaches did not change significantly when evaluated using different models for collaborative decision-making procedures. This observation suggests that modeling collaborative decision making may not be important for selecting near-optimal planned airport acceptance rates.