A markov decision process model for airline meal provisioning

• Main Author: Goto, Jason Hidekazu
• Language: English
• Published: 2009
• Online Access: http://hdl.handle.net/2429/9385

An airline caterer seeks to provide a meal quantity for each flight that closely matches final on-board passenger load. Faced with preparation lead-time, the caterer must estimate required meal quantities well in advance of departure. Passenger load may vary considerably during this lead-time, thus, adjustments are often required as more information becomes available. In this thesis, we model the meal ordering processes at Canadian Airlines as a finite-horizon Markov decision process. The model generates policies that show the caterer how to adjust meal quantities at each decision point to minimize ordering costs. We evaluate the performance achieved with the optimal policies by applying them to a holdout dataset, and compare the results to those observed in actual practice. Next, we use the model to observe the multi-objective problem of excess meal provisioning and short meal provisioning, and estimate the cost of achieving high service levels. This thesis finds that current practice in meal ordering at Canadian Airlines often achieves performance close to that achieved with the minimum cost policies. Application of the model to a group of 40 flights yielded estimated savings of $4,500 per month, while reducing the number of short catered flights by 42%.