Numerous complex systems exist in our world and the airline schedule system is of interest to us. Since airline schedules are pre-planned and fixed, the challenges modern airlines face are to run schedules with minimum delays. Unfortunately, airline operations are exposed to many delay sources such as weather, airport capacity constraints, air traffic control delays and passenger delays. Some are unpredicted and can be only treated as operational risks, e.g. weather effects, while some are due to airline operations, e.g. passenger processing delays, baggage loading delays … etc. When we model the airline ground operations (i.e. aircraft turnaround operations) as a complex stochastic system, we will be able to calibrate the model by using historical punctuality data. Then we are able to ‘simulate’ how schedule punctuality performance will look like after operations.
Given a fixed schedule, few airlines aim at running a schedule with zero delays. Instead, airlines build in buffer times (or called slack times) somewhere in a schedule so to expect the buffer will absorb delays to an extent. Scenario A shown in the figure shows the ‘expected departure delays’ of a network system comprising a fleet of 17 B737. The expected delay levels are calculated according to schedules and historical punctuality data of the airline. We can see some delays are well controlled by either operations or built-in buffer times. Scenario B shows the real punctuality results for the same period of time (being one year). We can see clearly the gap between two lines exists. Wider gaps indicate that real operations are far from expected, and vice versa. This model allows airlines to ‘virtually run’ draft schedules and observe the expected results in simulation mode.
If the model is calibrated by changing input parameters, the simulation model can be used to model current operations. In other words, one can use the model to ‘test’ schedule changes and see how the overall network effects look like. Please be noted that given the complexity of airline schedules and operations, changing one ‘design factor’ may cause chain reactions on the whole network. Without this simulation model, it would be impossible to estimate the consequences of schedule changes or even ‘visualise’ the results. This model also gives us a chace to optimise airline schedules and operations to a higher level.
(Note: this project is carried out by Dr. Richard Cheng-Lung Wu. Results have been accepted by Journal of Air Transport Management and will appear soon in mid 2005.)
More articles containing information about research in the Department of Aviation can be found on the homepage articles web-page.