Enhancing Artificial Bee Colony Algorithm with Superior Information Learning

Abstract

Artificial bee colony (ABC) algorithm has attracted much attention for its good performance and simple structure in recent years. However, its solution search equation does well in exploration but badly in exploitation, which may result in slow convergence rate for complex optimization problems. To address this defect, many ABC variants have been developed based on utilizing the global best individual or the group of elite individuals. Although the utilization of such good individuals is indeed beneficial to enhance the exploitation, it may also run the risk of causing the algorithm too greedy. In this paper, we proposed a modified ABC algorithm with superior information learning (SIL) strategy for achieving a better balance between the exploration and exploitation abilities. In the proposed SIL strategy, the individuals are expected to learn superior information from an exemplar which has better fitness value than the individuals themselves. The exemplar is no longer acted by the global best individual or the elite group. The proposed SIL strategy is designed to utilize the valuable information of good individuals while without losing the diversity. In the experiments, 22 well-known test functions and six state-of-the-art ABC variants are used. The comparison results showed that our approach significantly accelerate the convergence rate, and has better or at least comparable performance than the competitors on most of the test functions.