Applicationand Research On Artificial Bee Colony Algorithm

Artificial bee colony algorithm is a novel meta-heuristic search algorithm whichsimulates the intelligent foraging behavior of honeybee swarm to solve the practicalproblems. Because the artificial bee colony algorithm has the characteristic ofsimpleness, flexibility and robustness, it has increased tremendously in many researchtopics including numerical function optimization, integer programming, thecombinatorial and multi-objective optimization problems, neural network training,image processing, etc and got good results. However, as a novel algorithm, its modelis not mature, and it is still in the initial stage in the applications of the NP-harddiscrete domain optimization problem. Hence, it has important research value andpractical significance to improve the theory research and explore the applications ofthe NP-hard discrete domain optimization problem.By learning some other heuristic algorithms and the research results of biologists,some improved approaches are proposed according to shortcomings of the artificialbee colony algorithm. In addition, this dissertation explores the applications ofNP-hard discrete domain optimization problem such as0-1multidimensionalknapsack problem and the structure learning of Bayesian network. The main workincludes three parts:1) To solve the insufficient cooperation problem in the typical artificial bee colonyalgorithm owing to single communication way, we introduced the chemicalcommunication mechanism on inductive pheromone and proposed an artificial beecolony algorithm based on inductive pheromone updating and diffusion. First, the newalgorithm introduces the mechanism of inductive pheromone and its updating; then,the inductive pheromone diffusion model is proposed which is based on associationdistances; finally, we combine the mechanism of inductive pheromone updating anddiffusion with the typical artificial bee colony algorithm. Compared with the typicalartificial bee colony algorithm for the0-1multidimensional knapsack problem, ournew algorithm can obtain global optimization solutions easier and have higherconvergence speed; compared with the other stochastic optimization approaches, ournew algorithm is superior in the solution quality.2) To overcome the shortcomings of large iteration number and blindness ofsearch in the artificial bee colony algorithm, we proposed an artificial bee colony algorithm based on elite mechanism for0-1multidimensional knapsack problem. Onone hand, the new algorithm introduces the elite mechanism in the process ofconstructing new solutions, which prevents the scouts from searching unnecessarily;on the other hand, by modifying the repair operator with the transition probability inthe determination of the neighborhood of a food source, the employed bees andonlookers can perform the neighborhood searching more effectively. Theexperimental results show that the new algorithm is greatly enhanced in the solutionquality and convergence speed compared with other approaches.3) To extend the application of the artificial colony algorithm, we proposed anartificial bee colony algorithm for learning Bayesian network. First, in light of thecharacteristics of Bayesian network, we define the solution representation andconstruction, inductive pheromone and updating rules, the probability of choosingfood sources and the determination of the neighborhood of a food source. Then, thenew algorithm is described in detail based on the K2score search framework. Finally,the experimental results based on benchmark datasets demonstrate that the newalgorithm performs better in the solution quality and computation time than the otheralgorithms. This research provides a new thought of learning Bayesian networkstructure.