SANGA II: a new approach to niche radius identification

The Self-Adaptive Niching Genetic Algorithm (SANGA) is an optimization algorithm for multimodal problems able to identify almost all local optima and their niches, also estimating niche radii. SANGA is particularly suitable for optimization of electromagnetic devices, in which the behaviour of the objective function is usually unknown, because it does not need a priori specification of a dissimilarity parameter and uses a relatively low number of objective function evaluations. This paper presents some modifications of the SANGA algorithm in order to improve its ability to estimate the niche-radius and its good coupling with the deterministic Pattern Search (PS) method.