Artificial Bee Colony Algorithm with UCB-Guided Multi-Strategy Coordination

Abstract

Artificial Bee Colony (ABC) algorithms can suffer from an imbalance between exploration and exploitation, particularly when the search space becomes large or complex. To alleviate these problems, this study proposes an artificial bee colony algorithm with Upper Confidence Bound (UCB)-guided multi-strategy coordination, referred to as UMABC. The proposed method integrates a ranking-based selection probability model, UCB-guided adaptive search strategy selection, and an elite-guided local search mechanism. The ranking-based model regulates selection pressure according to solution quality while retaining the participation of lower-ranked solutions. The UCB mechanism adaptively coordinates different search operators according to their observed performance, and the elite-guided search further improves local refinement around promising regions. Experimental results on 22 benchmark functions show that UMABC obtains the lowest Friedman average rank among the eight compared algorithms at D=30, 50, and 100, with average ranks of 2.98, 3.16, and 3.30, respectively. According to the predefined acceptance thresholds, UMABC reaches the target accuracy in all 30 independent runs on 18, 16, and 16 of the 22 benchmark functions at D=30, 50, and 100, respectively. On the CEC2014 test suite, UMABC obtains an average rank of 3.28 under D=30, the lowest among the compared algorithms, and an average rank of 3.52 under D=50, where it remains statistically competitive with several compared ABC variants. The pressure vessel design problem is further considered as a constrained engineering application, where UMABC achieves the lowest Mean, Std, and Worst objective values among the six compared algorithms. These results indicate favorable overall performance across standard benchmark functions, the CEC2014 test suite, and the investigated engineering design problem.