Self-adaptation of mutation step size in Artificial Bee Colony algorithm for continuous function optimization

Abstract

This paper introduces a novel adaptation scheme of mutation step size for the Artificial Bee Colony algorithm and compares its results with a number of swarm intelligence and population based optimization algorithms on complex multimodal benchmark problems. The Artificial Bee Colony (ABC) is a swarm based optimization algorithm mimicking the intelligent food foraging behavior of honey bees. The proposed scheme dynamically adapts the mutation step size for better exploration and exploitation of the search space. Mutation with large step size is likely to produce large variations which would facilitate better exploration of the undiscovered regions of the search space while small step size usually produces small variations that are better for exploitation of the already found solutions. The appropriateness of small or large steps changes dynamically depending on the current stage and maturity of the ongoing search process as well as the properties of the search space. So, dynamic adaptation of mutation step size is a promising and interesting research direction that has not been explored so far with the ABC algorithm. This paper introduces Artificial Bee Colony with Exponentially Distributed Mutation (ABC-EDM) that incorporates exponential distributions to produce mutation steps with varying lengths and suitably adjusts the current step length. ABC-EDM is compared on a number of benchmark functions with the original ABC algorithm, Genetic Algorithm (GA), Particle Swarm Optimization (PSO) and Particle Swarm Inspired Evolutionary Algorithm (PS-EA). Results demonstrate that ABC-EDM performs better optimization with lower dimensionality, but the improvement fades away with increased number of dimensions.