Entropy-Driven Parameter Control for Evolutionary Algorithms

Every evolutionary algorithm needs to address two important facets: exploration and exploitation of a search space. Evolutionary search must combine exploration of the new regions of the space with exploitation of the potential solutions already identified. The necessity of balancing exploration with exploitation needs to be intelligent. This paper introduces an entropy-driven parameter control approach for exploring and exploiting evolutionary algorithms. Entropy represents the amount of disorder of the population, where an increase in entropy represents an increase in diversity. Four kinds of entropy to express diversity and to control the entropy-driven approach are discussed. The experimental results of a unimodal, a multimodal with many local minima, and a multimodal with only a few local minima functions show that the entropy-driven approach achieves good and explicit balance between exploration and exploitation.