A study of search neighbourhood in the bees algorithm

The Bees Algorithm, a heuristic optimisation procedure that mimics bees foraging behaviour, is becoming more popular among swarm intelligence researchers. The algorithm involves neighbourhood and global search and is able to find promising solutions to complex multimodal optimisation problems. The purpose of neighbourhood search is to intensify the search effort around promising solutions, while global search is to enable avoidance of local optima. Despite numerous studies aimed at enhancing the Bees Algorithm, there have not been many attempts at studying neighbourhood search. This research investigated different kinds of neighbourhoods and their effects on neighbourhood search. First, the adaptive enlargement of the search neighbourhood was proposed. This idea was implemented in the Bees Algorithm and tested on a set of mathematical benchmarks. The modified algorithm was also tested on single objective engineering design problems. The experimental results obtained confirmed that the adaptive enlargement of the search neighbourhood improved the performance of the proposed algorithm. Normally, a symmetrical search neighbourhood is employed in the Bees Algorithm. As opposed to this practice, an asymmetrical search neighbourhood was tried in this work to determine the significance of neighbourhood symmetry. In addition to the mathematical benchmarks, the algorithm with an asymmetrical search neighbourhood was also tested on an engineering design problem. The analysis verified that under certain measurements of asymmetry, the proposed ii algorithm produced a similar performance as that of the Bees Algorithm. For this reason, it was concluded that users were free to employ either a symmetrical or an asymmetrical search neighbourhood in the Bees Algorithm. Finally, the combination of adaptive enlargement and reduction of the search neighbourhood was presented. In addition to the above mathematical benchmarks and engineering design problems, a multi-objective design optimisation exercise with constraints was selected to demonstrate the performance of the modified algorithm. The experimental results obtained showed that this combination was beneficial to the proposed algorithm