An Adaptive Encoding Learning For Artificial Bee Colony Algorithms And Its Application

Artificial bee colony algorithm(ABC)is a swarm intelligence algorithm which is simulated the intelligent foraging behavior of honey bees in recent years.Due to its few control parameters,simple structure,and ease of implementation,ABC has attracted wide attention by domestic and foreign researchers,and improved algorithms with various characteristic have been proposed.In the process of improving and designing ABC,however,we still need to focus on the following two aspects:how to analyze the mathematical characteristics of ABC search equation;how to improve the search equation by using the characteristics of the problem.Thereby,this paper aims at developing a new class of ABCs with adaptive encoding learning,and solving numerical optimization problems with variable linkages.The main contributions can be summarized as follows.1.The mathematical characteristic of search equation of ABC is analyzed.The search equation is the core operator of ABC,which directly determines its performance.Therefore how to mine the mathematical characteristic of search equations is the basic premise of designing efficient ABC.For this reason,firstly,the search equation is described as the process of matrix transformation by the related matrix theory analysis.Secondly,the search equation is proved to be the rotationally variant operator according to the algebraic property of matrix transformation.Finally,the geometric interpretation based on(he matrix transformation search equation is given.2.An adaptive encoding learning(AEL.)framework is designed.For perform poorly in solving numerical optimization problems with variable linkages,an adaptive coding learning ABC is proposed.Firstly,the data association characteristic of population distribution is identified by covariance matrix learning,thus lose the interactions among the variables by identifying the properties of the fitness landscape,which is useful for directing the evolution of the population toward the directions of weak linkage.Secondly,an adaptive selection mechanism is designed to control the computational resource assigned to the eigen coordinate system and natural coordinate system,which balances the exploration and exploitation abilities of AEL+ABCs,thus improves the computation efficiency.3.The effectiveness of adaptive encoding learning(AEL)is studied.In order to verify the effectiveness of AEL,the proposed AEL strategy is integrated into eight existing ABCs:ABC,ABCM,ABCVSS,EABC,GABC,MABC,OPIABC and qABC.They are termed as AEL+ABCs,and tested on 30D,50D and 100D CEC2014 test platform,respectively.Besides,the proposed improved ABC is further applied to the Ford vehicle identification problem.The experimental results have shown that the classification results based on the training model of the AEL+ABC are better than those of the original ABC in accuracy,sensitivity and specificity.In conclusion,an adaptive coding learning(AEL)framework based on the mathematical characteristics of the search equation of ABC not only significantly improves the optimization performance of the existing ABC in solving problems with variable linkages,but also has a positive role in enriching the theory and method system of evolutionary computation.