Research On Several Problems On Basis Of Ant Colony Optimization

Ant Colony Optimization (ACO) is a novel evolutionary algorithm, derived from the foraging behavior of real ants of nature, which can find the shortest path between a food source and their nest. Its main characteristics are positive feedback, distributing computing and the use of a constructive heuristic. And as a new branch of computational intelligence and swarm intelligence, it has attracted more and more attention.In this dissertation, the model, algorithm and performance of Ant Colony Optimization are introduced. And then some improved Ant Colony Optimizations are advanced and simulated. Finally different applications based on Ant Colony Optimization, such as navigating complex labyrinths, feature selection and HP protein folding, are discussed.The major contributions of this dissertation are presented as follows:1. According to the crucial role of pheromone intensity in Ant Colony Optimization, methods are proposed by strengthening the pheromone intensity of the ants with the best solutions in the current or global ant colonies. Experiments show that the ameliorative Ant Colony Optimizations are more efficient than the standard ACO and other evolutionary algorithms.2. On Ant Colony Optimization for the Chinese Traveling Salesman Problem, two improved versions are presented. One is Ant-F, which can expand the system's search field and avoid premature; the other is ACS+, which can speed up the system's convergence at the last stage of evolution, through magnifying the contrast between the pheromone intensity of every edge.3. As a new approach, ACO is applied to the navigation of complex labyrinths for finding the shortest paths in the traffic networks. And a proposal based on ant intelligence for the dynamic routing in the intelligent traffic system is outlined.4. How to find out the best optimal features to efficiently represent human faces is a combinatorial optimization problem in face recognition. A new hybrid method, employing Ant Colony Optimization for feature selection and Support Vector Machine for classification, is detailed, and experiments are carried out with high efficient results.5. Given a HP amino-acid sequence, finding an energy-minimizing conformation is NP-hard. A dynamically progenitive Ant Colony Optimization is brought forward, and many HP benchmarks are successfully solved.