Accelerating Artificial Bee Colony Algorithm with Elite Neighborhood Learning

Abstract

Artificial bee colony (ABC) algorithm has been shown good performance over many optimization problems. For some complex optimization problems, however, ABC often suffers from a slow convergence speed, because it is good at exploration but poor at exploitation. To achieve a better tradeoff between the exploration and exploitation capabilities, we introduce the breadth-first search (BFS) framework and depth-first search (DFS) framework into different phases of ABC respectively. The BFS framework is combined with the employed bee phase to focus on the exploration, while the DFS framework is integrated into the onlooker bee phase to concentrate on exploitation. After that, an elite neighborhood learning (ENL) strategy is proposed to enhance the information exchange between the employed bee phase and the onlooker bee phase, because in ABC the employed bees cannot well communicate with the onlooker bees which may also cause slow convergence speed. Extensive experiments are conducted on 22 well-known test functions, and six well-established ABC variants are included in the comparison. The results showed that our approach can effectively accelerate the convergence speed and significantly perform better on the majority of test functions.