The Study Of Artificial Bee Colony Algorithm For High Dimensional Optimization Based On Mutual Learning

Inspired by the behavior of social insects, several algorithms based on swarm intelligence have been developed for solving combinatorial and numerical optimization problems. The recently invented artificial bee colony (ABC) algorithm is just an optimization algorithm based on swarm intelligence that has been applied to solve many kinds of numerical functional and non-numerical optimization problems, and it performs well in most cases. Since ABC is a new proposed optimization algorithm, it is necessary and significative to analyze its optimization performance.Most of the current versions of ABC algorithms are based on low dimensional optimizing problems. When dealing with large scale problems, the performance of ABC algorithm declines rapidly. Hence it is particularly important to enable ABC algorithm with a proper high dimensional mechanism, so that it can optimize large scale problem efficiently.This paper presents a comprehensive introduction of basic ABC algorithm including detailed backgrounds and principles of ABC algorithm, typical improved versions of ABC algorithm, the related studies and the application fields of ABC algorithm. In order to achieve under what conditions, ABC algorithm can obtain the best optimization performance, this paper conducts a comprehensive test and study of ABC algorithm’s performance. A series of experiments including effect of dimension and colony size, effect of scout bees and effect of initial region scaling are taken and analyzed. Meanwhile, inspired by genetic algorithm, ABC algorithm is applied with typical selection mechanisms, and the performance with that of different selection mechanisms is compared. Besides, this paper applies mutual learning mechanism to improve ABC algorithm, analyze and compare its performance.For large scale optimization problem, this paper details the background and principles of high dimensional mechanism, including some traditional and typical improved methods for large scale optimization. This paper applies dynamic group mechanism, deterministic sampling selection method and mutual learning to cooperatively coevolving ABC algorithm, which make it possible for ABC algorithm to tackle high dimensional problem. The performance of the high-dimensional ABC algorithm was tested and analyzed.The experimental results show that, for most of the optimization problems, ABC algorithm can obtain the global best optimum result in conditions of setting colony size be40, dimension be10or30, initial region scaling be symmetric distributed, and the selection mechanism be deterministic sampling or remainder stochastic sampling with replacement instead of roulette wheel selection used in ABC algorithm. Meanwhile, the performance of cooperative coevolving ABC algorithm using mutual learning and dynamic group method is superior to other typical high-dimensional algorithms and basic ABC algorithm, which shows that this improved ABC algorithm can be used to tackle large scale problem efficiently.