Ant Colony Algorithm And Its Application In Clustering

Ant colony algorithm which is inspired by the behavior of ant colony is parallelism, distributed. According to the behavior characteristics of ant colony in different aspects, ant algorithm can be classified as model inspired by ant looking for food behavior and model inspired by brood sorting, model inspired by division of laborj, model inspired by cooperative transport. This dissertation focuses on the two models mentioned above.Model inspired by ant looking for food behavior is also called ants colony optimization (ACO), which is a new kind of evolution computation algorithms based on bionics, and is a new heuristic intelligent optimization algorithm after simulated annealing, genetic algorithm and tabu search. After the algorithm being proposed, ACO has applied into TSP, quadratic assignment, graph coloring, vehicle route successfully. Aiming at solving disadvantages of ACO, this dissertation proposed an ant colony algorithm with dynamic update pheromone. Improved algorithm can shorten searching time, optimize solution finally and avoid the premature fall into local optimum.Clustering is an important technique in data mining, it is a rule in accordance with the data objects into multiple categories or clusters to make the same kind of data objects have a higher degree of similarity, but not quite different kind of data object.Clustering algorithm inspired by ants looking for food behavior is also known as clustering algorithm based on the theory of ants looking for food behavior. In the abstract of the ants in nature looking for food behavior, the behavior of looking for food is divided into two areas of searching food and transporting food, while the data object as ants, the cluster center as a "food source". So the clustering process can be described as ant foraging process. The clustering algorithm does not distinguish between the different attributes of the importance of data objects, this dissertation uses the maximum deviation algorithm, for each attribute according to its importance as it gives a weight. These make similar objects fast clustering, and avoid a lot of useless computation, which improves the efficiency of algorithm. Model inspired by brood sorting can also be called ant clustering model. Many kinds of ants can closely arrange eggs and larvae in a sheaf around the middle of nest area, and the biggest larvae located at the edge of area. Deneubourg and his colleagues were the first to propose the basic model (BM) to simulate this phenomenon. LF algorithm is a successfully improved BM. Fuzzy clustering algorithm is inspired by the theory of Fuzzy partition which is proposed by Ruspini in 1969. The FCM algorithm which is used in the dissertation is one of Fuzzy clustering algorithm. The dissertation analyzes the advantages and disadvantages of the LF algorithm and FCM algorithm, and their complementary analysis, LF algorithm is proposed based on improved FCM algorithm, the forthcoming integration of these two algorithms, but also used to maximize the weighted deviation algorithm, the algorithm in the distance calculation be improved to further enhance the performance of the algorithm.