Integrating the fleet assignment model with uncertain demand

Abstract

One of the main challenges facing the airline industry is planning under uncertainty, especially in the context of schedule disruptions. The robust models and solution algorithms that have been proposed and developed to handle the uncertain parameters will be discussed. Fleet assignment models (FAM) are used by many airlines to assign aircraft to fights in a schedule to maximize profit. In the context of FAM, the goal of robustness is to produce solutions that perform well relative to uncertainties in demand and operation. In this thesis, we introduce new FAMs (i.e. DFAM1 and DFAM2) that tackles the common problem associated with aircraft utilization. Subsequently, stochastic programming (SP) is presented as a method of choice for the research. Through the use of a two-stage SP with recourse technique, the DFAMs are extended to SP-FAMs (SP-FAM1 and SP-FAM2). The main distinction of the SP-FAM compared with other FAMs is that, given a stochastic passenger demand, it gives a strategic fleet assignment solution that hedges against all possible tactical solutions. In addition, we have a tactical solution for every scenario. In generating the demand scenarios, we use a network-simulation model embedded with a time-series engine that gives a snapshot of one week that is representative of any other week of the scheduling season. We later outline the approach of solving the SP-FAMs where the schedule is compacted through several preprocessing steps before inputting it into SAS-AMPL converter. The SAS-AMPL converter prepares all the data into readable AMPL format. Finally, we execute the optimizer using a FortMP solver (integrated in AMPL) that invokes branch-and-bound algorithm. We give a proof of concept using real data from a Middle East airline. Our investigations establish clear benefits of the recourse FAM compared to alternative models. Finally, we propose areas of future research to improve SP-FAM robustness through solution algorithms, revenue management (RM) effects, calibration of network-simulation models and system integration.