Self adaptation of operator rates for multimodal optimization

This work presents a niching technique for an evolutionary algorithm that adjusts the genetic operators probabilities at the same time evolves a solution for the optimization problem. Such niching technique is based on the deterministic crowding technique and a variation of the dynamic inbreeding mating restriction. Since each individual encodes its own operator rates and uses a randomized version of a learning rule mechanism for updating them according to the performance reached by the offspring (relative to its parent performance), it is possible to apply mating restriction schemes for selecting the additional parent in the crossover. Moreover, individuals are replaced according to a variation of the deterministic crowding replacement policy. The behavior of the niching technique is studied using different genetic operators for both real and binary encoding schemes on some benchmark functions.