Niching Techniques: a Study on the Cluster Geometry Optimization Problem



Niching techniques are commonly used in Evolutionary Computation to maintain population diversity and help to explore more efficiently the solutions space. In complex optimization problems, the definition of a distance measure between individuals is a serious difficulty for this type of tech¬niques. The cluster geometry optimization problem has singularities that make the use of a reliable distance function computationally prohibitive. We analyze the behavior of an alternative niching technique, the Spatially-Dispersed Genetic Algorithm, and its application on the cluster geometry optimization. Results show that the algorithm is able do efficiently discovers several niches when searching for good solutions. We also present a detailed parametric analysis that helps to understand the role played by the different parameters of the method.


Evolutionary Optimization

TechReport Number

TR 2007/002

PDF File

Cited by

No citations found