A two-phase genetic algorithm for large-scale bidline-generation problems at Delta Air Lines

We developed a two-phase algorithm for solving Delta Air Lines' bidline-generation problem: assigning trips (task) to montly schedules for crew members (called bidlines). The system must produce bidlines that conform to all rules, and it should maximize average total value and the quality of the bidlines as measured by their purity. The first phase of the algorithm (the purity phase) constructs as many high-quality lines as possible, and the second (the GA phase) completes the assignments by constructing high-total-value valid lines from the remaining open trips. Delta obtains significant savings in staffing by using this algorithm. The bidlines produced are of comparable quality to those built with the semiautomatic process that Delta Air Lines used before.