| In this dissertation, we study problems in both revenue management (RM) and irregular operations recovery. The first chapter is devoted to RM in airline passenger. We consider a problem that allocates seats to fare classes, and captures capacity nesting and customer upsell. While capacity nesting allows airlines to sell seats allocated to low-yield classes to high-yield passengers, customer upsell allows low-yield passengers to purchase seats reserved for high-yield classes. We adopt an approximate dynamic programming algorithm to iteratively approximate the complicated objective function with piecewise linear functions. We observe that the resulting allocation policy outperforms a popular bid-price policy up to 35% when demand and upsell probability are high.;The second chapter is about RM in air cargo. We study the underlying capacity allocation problem in the mid-term capacity allocation process, in which shippers bid flight capacity on multiple flights to receive discounted shipping rates and guaranteed space. The model minimizes demand covariance between the bids and future volatile free-sales demand subject to a revenue lower bound and all necessary allocation requirements. Due to its complexity, we decompose the problem by a flight partition and a set of demand clusters. We show using simulation that our partitioning algorithm is robust, and the resulting allocation increases revenue by 2% if the revenue lower bound is high and demand covariance is captured.;In the final chapter, we study a fully integrated recovery problem that recovers disrupted flight schedules by iteratively and simultaneously recovering resources (aircraft, crews, and passengers). The problem is solved by Benders decomposition, where the master problem is an extended fleet assignment problem, and the subproblems are the resource recovery problems. Several decomposition and algorithmic strategies are developed to reduce the total running time. We show that our solution can outperform a partially integrated solution used in practice by as much as 8%, which accounts for one million dollars in saving per disruption. |
