An Optimization-Based Sample Day Selection Algorithm for Future Schedule Generation

Future flight Schedules are generated using air traffic growth forecasts along with a set of baseline schedules. The baseline schedules are usually selected by sampling historical operational data for a fiscal year and choosing representative days that best reflect seasonality in terms of a given set of performance metrics. Larger sample sizes capture more accurate trends in the NAS at the expense of the computer processing time of other important elements of the model process. This trade-off is evaluated each year based on known information on computer run-time and other performance requirements of the modeling community. To maximize accuracy with a minimal sample, we propose an optimization based method for solving the sample day selection problem, which is formulated as a Mixed Integer Program (MIP). The objective of the MIP is to minimize the weighted difference between the true population and the sample to be selected in terms of the defined metrics subject to a set of constraints including the sample size limit, coverage requirements and other desired properties. This paper presents two solution algorithms which have been implemented using the CPLEX MIP solver. A standard MIP formulation is first presented followed by a decomposition formulation which partitions the problem into smaller parts in order to reduce the computation time required for larger day selection exercises.