A Step Size Preserving Directed Mutation Operator

Using a directed mutation can improve the efficiency of processing many optimization problems. The first mutation operators of this kind proposed by Hildebrand [1], however, suffer from the asymmetry parameter influencing the mutation step size. Extreme asymmetry can lead to infinite step size. The operator presented here overcomes this drawback and preserves the step size. The main idea of the directed mutation is to focus on mutating into the most beneficial direction by using a customizable asymmetrical distribution. In this way the optimization strategy can adopt the most promising mutation direction over the generations. It thus becomes nearly as flexible as with Schwefel’s correlated mutation [2] but causes only linear growth of the strategy parameters instead of quadratic growth. A normalization function is introduced to decouple asymmetry from the variance, i.e. the step size. By incorporating the normalization function the variance becomes independent of the asymmetry parame...

• Type: Conference
• Year: 2004
• Where: GECCO
• Authors: Stefan Berlik
• Keywords: Directed Mutation, GECCO 2004, Mutation, Step Size