A Comparison of Adaptive Operator Scheduling Methods on the Traveling Salesman Problem

Abstract. The implementation of an evolutionary algorithm necessarily involves the selection of an appropriate set of genetic operators. For many real-world problem domains, an increasing number of such operators is available. The usefulness of these operators varies for different problem instances and can change during the course of the evolutionary process. This motivates the use of adaptive operator scheduling (AOS) to automate the selection of efficient operators. However, little research has been done on the question of which scheduling method to use. This paper compares different operator scheduling methods on the Traveling Salesman Problem. Several new AOS techniques are introduced and comparisons are made to two non-adaptive alternatives. The results show that most of the introduced algorithms perform as well as Davis’ algorithm while being significantly less cumbersome to implement. Overall, the use of AOS is shown to give significant performance improvements – both in...

• Type: Conference
• Year: 2004
• Where: EVOW
• Authors: Wouter Boomsma
• Keywords: Artificial Intelligence, EVOW 2004, Operator, Operator Scheduling, Scheduling Methods