| The phenomena of adaptive optimization in social creatures inspire humanity constantly. Collective behaviors of social creatures have made optimization problems with high degree of complexity to scientists perfectly solved. Research on swarm intelligence emerged out of observing and mimicing collective behaviors of social creatures. A general-purpose metaheuristic named Ant Colony Optimization algorithm, which take inspiration from real ant's behavior in finding shortest paths using as information only the trail of a chemical substance (called pheromone) deposited by other ants, boasts a number of attractive features, including adaptation, robustness and distributed, decentralized nature, and have recently been successfully applied to several discrete optimization problems. It has the latent application prospect. The artificial neural networks (ANN) is a new information processing method which developed rapidly in recent years, and is a intelligent system which simulates the human brain organizational structure and intelligent behavior based on the simulation of human brain operational mechanism.As a result of equipment and service unceasing increasement, modern data network, including wired and wireless network, is assuming diverse and isomerism characteristic gradually. The demand of unmistakably interacting between innumerable combination network component, has formed the huge challenge. The network management questions, including optimization, network traffic analysis and forecast, is a kind of special combination optimization question, and is a NP-hard problem. To find, research, application of intelligent heuristic approach is particularly important.The goal of this dissertation is to explore and further swarm intelligence models, which makes it easy to solve large-scale complicated problems. On the other hand, this thesis extends the application domains of swarm intelligence and to cope with more practical engineering problems. On the other hand, this thesis On the other hand, this thesis applies swarm intelligence and evolutionary computation methods to the research domain of distributed dynamic network routing, load balancing and network traffic forecast, so that the intelligent computation method can solve the more practical engineering problems.The contributions of this dissertation are as follows:(1) A multiple ant colony optimization algorithm is proposed. The artificial ants are partitioned into several groups. Each group of ant colony releases different types of pheromones. Attract factor and exclusion factor are introduced, and a new transition probability with multiple ant colony was given, so as to strengthen the global search capability. By tackling symmetric travelling salesman problems, this paper compares the improved algorithms implementation with the ACS and AS algorithms. The experimental results indicate that the improved algorithm is superior to the ACO and AS algorithms. The improved algorithm has excellent global optimization properties and faster the convergence speed, and it can avoid premature convergence of ACO.(2) A hybrid algorithms combining ACO algorithm with genetic Algorithm is present. The new algorithm is for solving the problem of the stagnation and slow convergence in ACO. The path genetic operators are proposed, and a new pheromone update rule is achieved. Each chromosome is encoded as a series of nodes that in the path ants have found, and is evaluated with a fitness function. Path crossover and path mutation are performed on the path chromosomes. The experimental results indicate that the improved algorithm is superior to the ACS algorithms. The improved algorithm has excellent global optimization properties and faster the convergence speed.(3) A new approach to parallel ant colony optimization (ACO) algorithms by changing the behavior of ACO is present. The principal idea is to partition each ant colony into several sub-colonies, and to propose a new transition probability, so as to strengthen the global search capability. The parallelization strategies for multiple Ant Colony Optimization algorithms are discussed. The performance of the proposed parallel algorithm, applied to the Traveling Salesman Problem, is investigated and evaluated with respect to solution quality and computational effort. The experimental studies demonstrate that the proposed algorithm outperforms the sequential Ant Colony System as well as the existing parallel ACO algorithms. The studies also indicate that the new explore scheme based on multiple ant colony improves performance, particularly in exchange strategies that exchange best solution information among all colonies. The new algorithm has a good extendibility and suit to solve large-scale problem.(4) A dynamic network simulation system has been developed using event driven method. The methods of how to simulate the network topology, traffic model, and intelligent network routing protocol have been given. The experimental result indicates that the network simulation model can simulate dynamic network and non-precise condition, and also can be applied to study intelligent routing algorithm over it.(5) Two distributed swarm intelligence routing algorithm are present. One is Adaptive distributed routing algorithm based on ant-algorithm. The network nodes are partitioned using k-means clustering, and different search strategies are implemented by different types of ants. A new routing table formation scheme is proposed. The proposed algorithm is scalable, robust and suitable to handle large amounts of network traffic, with minimizing delay and packet loss. The other method is based on the hybrid algorithms combining ACO algorithm with genetic Algorithm in (2). The AntNet algorithm is improved through the introduction of genetic algorithm strategy. A new pheromone update rule is achieved via path genetic operators. The simulation results show that the improved algorithm has faster speed of the convergence, also the network throughput is effectively improved, and the average time delay in network is reduced.(6) An effective load-balance routing strategy is present. Appling the multiple ant colony optimization algorithms in the distribution network load-balance routing. The proposed algorithm has the essentially extendibility, can effectively address the resources assignment problem in large-scale network. A new dynamic transition and search strategy based on multiple ant colony are used for load-balance routing. Three mechanisms based on ACO for load-balancing and routing are proposed. The performance of the proposed three mechanisms, applied to the simulation network SDH, is investigated and evaluated with respect to solution quality and computational effort. The experimental results indicate that, the algorithm can use the multiple feasible routing path while network in heavy load status. The algorithm enhances the network load balance, robustness as well as the Qos of network.(7) A new network traffic prediction model which combines the grey theory and neural network was presented. The recursion neural network has very strong ability of the self organization, adaptive, self learning and extraction information. The combined model is to forecast the short-term network traffic with high precision. The simulation results on real network traffic show the new model has better predictive precision.Finally, the work of this dissertation is summarized and the prospective of future research is discussed.
|
PDF Full Text Request