Niche allocation in spatially-structured evolutionary algorithms with gradients

This paper extends previous work exploring gradient-based spatially-structured evolutionary algorithms (GBSSEAs). GBSSEAs complete the parapatric speciation concept in SSEAs by introducing local fitness through the introduction of an ideal phenotype at each location in space and introducing local competition to match these phenotypes. This paper explores the theoretical niching properties of GBSSEAs, and demonstrates that their niche allocation behaviour differs from traditional niching algorithms in that allocation of individuals depends of the relative location of optima in the fitness landscape. The paper concludes with an examination of the parameter sensitivity of GBSSEAs, demonstrates the robustness of these parameters in the context of global multimodal optimisation, and provides indications for good parameter values for searching for optima of varying fitness.