A tree-structured random walking swarm optimizer for multimodal optimization

This paper develops a novel tree structured random walking swarm optimizer for seeking multiple optima in multimodal landscapes. First, we show that the artificial bee colony algorithm has some distinct advantages over the other swarm intelligence algorithms for accomplishing the multimodal optimization task, from analytical and experimental perspectives. Then, a tree-structured niching strategy is developed to assist the algorithm in exploring multiple optima simultaneously. The strategy constructs a weighted complete graph based on the positions of the food sources (candidate solutions). A minimum spanning tree that encodes the distribution of the food sources is built upon the complete graph to guide the search of the bee swarm. Each artificial bee sets out from a food source and flies along the edges of the tree to gather information about the search space. The dance trajectories of bees are simulated by a random walk model considering both distance and fitness information. Then, mutant vectors are selected from the trajectories to update the food source. This graph-based search method is introduced to simultaneously promote the progress of exploitation and exploration in multimodal environments. Extensive experiments indicate that our proposed algorithm outperforms several state-of-the-art algorithms.