Artificial bee colony algorithm based on adaptive neighborhood topologies

Abstract

During the past few years, many neighborhood-based ABC variants have been developed to utilize the valuable information of neighbors for guiding searches instead of using the best individual or elite individuals. However, neighbor selection is determined by the neighborhood topology, and different neighborhood topologies are suitable for different problems. Unfortunately, previous neighborhood-based ABC variants have often used a single type of neighborhood topology, which significantly affects algorithm performance. Hence, to take advantage of different neighborhood topologies, we propose a new neighborhood-based ABC variant using adaptive neighborhood topologies, called ABC-ANT. In ABC-ANT, to determine which type of neighborhood topology should be selected, the fitness distance correlation technique is first used to identify the feature of the fitness landscape for a given problem. Then, according to the identified feature, the most suitable neighborhood topology is adaptively selected, which is beneficial to adapt the search to the fitness landscape. Moreover, in ABC-ANT, the scout bee phase is modified by developing a dual-elite search strategy to save the search experience. Extensive experiments are conducted on two test suites, i.e., CEC2013 and CEC2017, and one real-world optimization problem, i.e., the dynamic economic dispatch problem. Seven ABC variants and six non-ABC variants are included in the performance comparison. The results verify that ABC-ANT has very competitive performance.