Improving Metaheuristics by Natural Selection

Abstract

In this paper, the famous phrase "survival of the fittest" by Darwin is applied to modify the design of metaheuristic algorithms for improving their performance. Put simply, coined in Darwin's The Origin, the concept of 'natural selection' (NS) is about stronger species in nature will have better chances of survival and reproduction, allowing the species to carry forward their viable offspring's to future generations. Genetic algorithm is a direct implementation of this concept. However, for population-type of swarming algorithms such as particle search optimization (PSO) and wolf search algorithm (WSA), for the first time this concept is formulated as a strategy called NS strategy for controlling the lifespans of the search agents. PSO and WSA represent two typical kinds of metaheuristics, whereas a group of search agents follow some moving patterns of fully swarm with global and local velocities and semi-swarm respectively. In both kinds, guided by the NS strategy, the search agents will have a differential lifetimes depending on the fitness values that they can generate during the search. Productive agents are granted longer lives and vice-versa. Superior results are observed from benchmarking experiments for metaheuristics algorithms that are programmed with the NS strategy over their original versions.