A hibernating multi-swarm optimization algorithm for dynamic environments

Abstract

Abstract-many problems in the real world are dynamic in which the environment changes. However, the nature itself provides solutions for adaptation to these changes in order to gain the maximum benefit, i.e. finding the global optimum, at any moment. One of these solutions is hibernation of animals when food is scarce and an animal may use more energy in searching for food than it would receive from consuming the food. In this paper, we applied the idea of hibernation in a multi-swarm optimization algorithm, in which a parent swarm explores the search space and child swarms exploit promising areas found by the parent swarm. In the proposed model, whenever the search efforts of a child swarm for exploiting an area becomes unproductive, the child swarm hibernates. Similar to the nature, which the change of the season awakens hibernating animals, in the proposed model hibernating swarms are awakened upon the detection of a change in the environment. Experimental results on various dynamic environments modeled by the moving peaks benchmark show that the proposed algorithm outperforms other PSO algorithms, including similar particle swarm algorithms for dynamic environments like mQSO, adaptive mQSO, and FMSO.

A survey of swarm intelligence for dynamic optimization: Algorithms and applications

“…Whenever the best particle gets better, a child swarm is generated with the best particle and particles within a given distance from the best one. Similar ideas were adopted in a multi-swarm PSO algorithm (mPSO) [158], and algorithms proposed in [152,151]. 3.…”

Multi-population methods in unconstrained continuous dynamic environments: The challenges

“…-Splitting-off approaches: Populations are created by splitting off from a main population from generation to generation [9,61,34,30].…”

“…-Different populations may have different roles: A part of populations are responsible for exploring new peaks and the others are for exploiting peaks that have been explored [9,61,34,30,43,44,21,73,74].…”

“…This guarantees that no more than one child swarm covers a single peak. Another similar idea 105 is the hibernation multi-swarm optimization (HmSO) algorithm, introduced in [30], where a child swarm hibernates if it is not productive anymore and wakes up when a change is detected. Recently, a similar hibernation scheme was employed in [70,69], in which competitive results were reported in comparison with a set of algorithms.…”

“…-Mixed relationship: Populations that play different roles cooperate with each other, but populations that play the same role compete with each other [9,61,34,30].…”