A parallel genetic algorithm with niching technique applied to a nuclear reactor core design optimization problem

Abstract Genetic algorithms (GAs) are global optimization techniques inspired by the mechanisms of natural evolution. In many complex problems, GAs have demonstrated to overcome traditional techniques. The key for such success is the efficient exploration of the search space obtained by maintaining diversity during the search process. In standard GAs, the crossover operator is the main responsible for exploration. Its function is to generate new candidate solutions (offspring) by exchanging information between existing ones (parents). In order to provide extra diversity, quite helpful when population is drifting to local optima, mutation operator is used. However, many times it is not able to avoid premature convergence to local optima (due to “genetic drift”). More efficient techniques, such as distributed approaches (island GAs) and niching techniques, have been proposed. Island Genetic Algorithms (IGAs) are able to delay efficiently (but not to avoid) the genetic drift. The advantages of this approach are the simplicity of implementation and the efficiency attained by parallel processing. Niching Genetic Algorithms (NGAs), by their turn, prevent genetic drift by the maintenance of subpopulations in the so-called “niches”. However, the computational cost is considerably increased in relation to the canonical GA. Motivated by the advantages offered by both techniques, we investigated a hybrid approach – the Niched-Island Genetic Algorithm (NIGA). The NIGA was applied to a nuclear reactor core optimization problem and compared to previously used GA-based techniques. NIGA and NGA obtained the best results. The NIGA achieves results similar to those obtained by the NGA, but in far less processing time.