Abstract
The ability to discover and maintain multiple solutions within a single run is a desirable property for evolutionary algorithms. Researchers have previously turned to many biologically-inspired methods for inspiration to produce niching evolutionary algorithms. This paper extends previous work on the Gradient-Based Spatially-Structured Evolutionary Algorithm, which attempts to embody the concept of parapatric speciation within an evolutionary algorithm. Through an comparison of the evolved population with that of an idealised, perfectly proportioned population, we show that the distribution of population members among the niches of a given problem’s fitness landscape does not rely on the global properties of the landscape. Rather, the allocation of individuals to peaks relies on the relative values of neighbouring peaks with regard to their spatial relationship in the fitness landscape.