Research And Application Of Multi-group Artificial Bee Colony Algorithm Based On Cuckoo Algorithm

Artificial Bee Colony algorithm(ABC)is formed by simulating the swarm-intelligence behavior embodied in bee collecting honey.As a original swarm-intelligence algorithm,ABC algorithm which has many advantages,including simple structure,few control parameters and implementation simplicity has been proved to be an effective global optimization algorithm.It has been studied by more and more scholars and is widely used in neural networks,wireless sensor networks,image processing,planning and scheduling,etc.,which has achieved good research results and application effects,However,the ABC algorithm is still in the early stage of research,and there are still some problems to solve,such as improving the performance of the algorithm on various optimization problems and extending the application range.Through the research on the internal mechanism of artificial bee colony algorithm and other scholars'improved methods,together with the advantages of other heuristic algorithms,improving the ABC algorithm from multiple angles,and exploring the application of improved algorithms in Bayesian Network Structure Learning has been done.The main work of the paper are summarized as follows:Firstly,a multi-group artificial bee colony algorithm(CMABC)based on cuckoo algorithm is proposed for the purpose of overcoming the premature convergence and the slow convergence.At first,a multi-group hybrid search method is designed.The employer is divided into three subgroups according to the fitness.Three kinds of solution search equations are made for different sub-groups,and the self-selection mechanism of the individual in the search process is adopted.Combining three different search strategies to balance the global search and local search capabilities of the algorithm improves the anti-early maturity and convergence speed of the algorithm.Then,based on the location update mechanism in the Cuckoo algorithm(CS),a new food source update strategy is created for the problem that the information of current optimal solution in the food source update mechanism is used to cause the loss of the potential better solution.The effective utilization of the local optimal solution is realized.Secondly,in order to extend the application range of ABC algorithm on problems of discrete domain,a Bayesian network structure learning method based on discrete CMABC algorithm(CMABC-BNL)is proposed.At first,the structural learning problem is abstracted into the problem of finding the optimal food source.Then,based on the CMABC algorithm.,combined with the cross and variations of the differential evolution algorithm,a discrete CMABC algorithm is designed and a labeled depth-first search strategy is proposed to correct the illegal structure.Finally,the effectiveness of CMABC algorithm on the functions of unimodal,multi-peak,translation,rotation and noise is fully proved by comparing the simulation experiments of 8 classical test functions and 14 CEC05 benchmark functions in high and low dimensions.Comparative experiments show that the CMABC algorithm has advantages in solving accuracy and convergence to some degree.Throughout the comparison experiments of two standard Bayesian network datasets,the result shows that the proposed structure learning method based on discrete CMABC is better.