Domain Knowledge and Representation in Genetic Algorithms for Real World Scheduling Problems

This paper discusses the issues that arise in the design and implementation of an industrialstrength evolutionary-based system for the optimization of the monthly work schedules for the pilots of Delta Air Lines. We detail the system's multiple and often conflicting goals and rules, providing the background for understanding the problem. Then, we describe the algorithm that we use to solve it. One important difference of our approach from other commonly used GA implementations is our use of the GA as a feasibility procedure: the first phase of our approach is responsible for building a very high quality partial solution based on the domain knowledge of the problem. The GA is responsible for completing this solution, finding a feasible solution to the remaining problem. We illustrate the impact that the representation can have on overall performance by comparing two implementations of the same algorithm based on two "orthogonal" encodings of the problem at hand. The comp...

• Type: Conference
• Year: 2000
• Where: GECCO
• Authors: Ioannis T. Christou, Armand Zakarian
• Keywords: Delta Air Lines, GA Implementations, GECCO 2000, Monthly Work Schedules, Optimization