On the Performance of Explorative Artificial Bee Colony Algorithm for Numeric Function Optimization

Abstract

The Explorative Artificial Bee Colony (EABC) algorithm is a recently introduced swarm intelligence based algorithm that has been successfully tested to optimize only a limited number of multimodal functions. This paper evaluates EABC on a larger number of benchmark functions, including both unimodal and multimodal functions. EABC is an improved variant of the Artificial Bee Colony (ABC) algorithm. A major problem with the basic ABC algorithm is that it is more aligned towards exploitations, rather than explorations, which often leads to premature convergence and fitness stagnation. The improved variant -EABC tries to increase the degree of explorations of ABC by introducing more randomness during its perturbation operations. Besides, EABC customizes the degree of exploitations and explorations at the individual solution level, separately for each candidate solution of the bee population. EABC also introduces a crossover operation that assists the explorative perturbation operation of EABC. This paper extends the experimental studies on EABC by evaluating it on as many as 13 complex, high dimensional benchmark functions, including both unimodal and multimodal, separable and non-separable functions. The results are compared with the basic ABC algorithm. The comparison demonstrates that EABC often performs better optimization than the original ABC algorithm, which indicates the effectiveness of its more explorative operations.