Dynamic Niche Clustering: a fuzzy variable radius niching technique for multimodal optimisation in GAs

摘要: This paper describes the recent developments and improvements made to variable radius niching technique called Dynamic Niche Clustering (DNC). DNC is fitness sharing based that employs a separate population of overlapping fuzzy niches with independent radii which operate in decoded parameter space, are maintained alongside normal GA population. We describe speedup process can be applied initial generation greatly reduces complexity stages. A split operator also introduced designed counteract excessive growth niches, it shown this improves overall robustness technique. Finally, effect local elitism documented compared performance basic on selection 2D test functions. The concluded view future work undertaken