A Self-adaptive Artificial Bee Colony Algorithm Based On Large Scale Optimization Problems

Because the Evolutionary Algorithms (EAs) have few parameters and they can be easily applied to various problems, many researchers have devoted to the study of EAs over the last three decades. Compared with previous conventional means, EAs have become well-established global optimization methods, and they all have high robustness and broad applicability. Possessing the characteristics of self-organizing and self-learning, EAs can deal with those challenging optimization problems which otherwise cannot be effectively solved by traditional optimization methods. The Artificial Bee Colony (ABC) algorithm is an optimization algorithm inspired by the foraging behavior of bee swarms. Existing research has shown that the ABC algorithm is an effective and robust population-based method which can be used to settle numerous down-to-earth optimization problems. However, similar to some probabilistic algorithms,there is a main limitation in ABC, i.e., in many problems, ABC is apt at exploration but is not deft at exploitation. Thus, in order to overcome this limitation and improve the performance of ABC when dealing with various kinds of optimization problems, the primary research work of this paper is as follows:(1) First, illuminated by Differential Evolution algorithm (DE), several local search operations are embedded into the ABC algorithm. The new algorithm could discover the balance point between diversity and convergence rate through this improvement. Not only could the speed of algorithm be accelerated, but also the more precise solutions can be found.Such an improvement can be advantageous in many real-world problems. Moreover, we also focus on the performance of artificial bee colony algorithm on the large scale global optimization problems. The experiment results indicated that the performance of the improved ABC algorithm is better than that of some other classical algorithms, however, as the dimensions of the test functions increase, the performance on most of the algorithms deteriorate gradually.(2) The former researches have displayed that the ABC algorithm is a multifarious and efficient heuristic means. However, the solution search equation used in ABC is insufficient,and the strategy for generating candidate solutions results in good exploration ability but poor exploitation performance. Although some complex strategies for generating candidate solutions have recently been developed, the robustness and versatility of them are always insufficient. This is mainly because only one strategy is adopted in the modified ABC algorithm. In this paper, we propose a Self-adaptive ABC algorithm based on the Global Best(SABC-GB) for global optimization. In SABC-GB, the evolutionary strategies can not only be selected dynamically according to their search performance, but also be enhanced. We choose the 25 benchmark functions to conduct an experiment and the result set of the data show that SABC-GB is an excellent optimization algorithm. To further validate the feasibility of SABC-GB in real-world application, we demonstrate its application to a real clustering problem. The results demonstrate that SABC-GB is superior to the traditional algorithms for solving complex optimization problems. It means that it is a new technique to improve the ABC by introducing self-adaptive mechanism.(3) We keep trying and propose a Self-adaptive Artificial Bee Colony algorithm with Symmetry Initialization (SABC-SI). In our SABC-SI algorithm, a new population initialization method based on half space and symmetry is designed, and such method can increase the diversity of initial solutions. Then,the selection operator is improved by eliminating some of the poor solutions and making good use of the two best solutions in both the current and previous generations. Besides, we optimize the self-adaptive search mechanism which is employed in ABC and several new candidate solution generating ,strategies have also been developed. The no,vel SABC algorithm is tested on 25 different benchmark functions. The experimental results show that SABC-SI outperforms several state-of-the-art algorithms, which indicates that it has great potential to be applied to a wide range of optimization problems. Moreover, the SABC-SI algorithm is used for classification problem directly and we also propose a new mathematical optimization model.The results show that the test sets can be identified accurately, and the performance of classification optimization model based on SABC-SI is wonderful. In other words,SABC-SI can be used for classification feasibly and efficiently.