An adaptive neighboring search using crossover-like mutation for multi modal function optimization

We propose a new population-based evolutionary algorithm which uses a real-coded representation and normal-distribution crossover-like mutation for generating the next searching points. This Gaussian distribution is formed based on the positional relationships between an individual and its neighbors, and is not carried with the self-adapting parameters as an inheritable trait. This algorithm causes the emergence of clusters of individuals within the population, as a result of the evolution of each individual, which does not have any actual intent to cluster. By searching independently, the emergent clusters introduce various solutions that include optima at the same time, even if the problem has strong local minima. The proposed method robustly solves a highly multi-modal 30-dimensional Fletcher-Powell function with a small population size.