We describe our experiences in building CAPTAINS (Crew Availability Planning and Training System) for the Royal Dutch airline company KLM. CAPTAINS is a complex application that enables a planner to maintain strategic, tactic and operational models of pilot populations. A major problem' in building the short-term planning algorithms for CAPTAINS was prediction of pilot bid behavior. Twice a year pilots may express their preference for new seats (functions) and KLM is obliged to give a new seat to the most senior officer who is qualified. However in some cases the bids are not yet known for the season for which one is planning. If a pilot changes his bid, this can influence the planning substantially, so that a correct prediction of pilot bid behavior has vital importance for a good plan. Using genetic algorithms, we were able to produce rules that predict pilot bid behavior with an acceptable level of accuracy.
展开▼
