Research And Application Of Swarm Intelligence Algorithm

With the fast development of economy,modern industry is steping forward in the direction of non-linear,large scale and integration,so the optimization algorithms with good robustness,hign efficiency and fast convergence speed is increasingly demanded.The emerging of Computational Intelligent(CI) give hope to the solution of these complicated problems.At present,CI has become the research hotspots in the international artificial intelligent fields,its research range includes Particle Swarm Optimization(PSO),artificial neural networks,simulated annealing,artificial immune algorithm,Ant Colony Optimization(ACO) algorithm, and Artificial Bee colony algorithm(ABC)et al.Nowadays these algorithms have been widely applied to image processing,data mining,intelligent control,network optimization.Despite of all these advantages mentioned above, CI algorithms have some common disadvantages,such as premature convergence,inferior search results,slow convergence speed,fast desending of population diversity,and sensitivity on parameters, et al,which place great restrictions on the further application.The ariticle analysed the disadvantages mentioned above,absorbed the research results of predecessors and proposed several improved algorithms:(1) Inertia weight is an important parameter in particle swarm optimization algorithm.In this paper ,evolution speed,aggregation degree and similarity is integrated into particle swarm optimization organically to control the inertia weight better,and enhance the escaping ability from local optimum when used on complicated problems.The modified PSO algorithm improves the abilities of seeking the global excellent result and convergence accuracy.The experiment results demonstrate that the proposed algorithm are superior to several typical particle swarm optimization algorithms based on dynamic change of inertia weights.(2) In order to overcome prematurity and low searching speed of particle swarm optimization(PSO) algorithm,a novel immune clone PSO algorithm with personalized mutations is proposed according to the immune clone selection theory.In this modified algorithm ,clone selection operator is applied to raise the diversity of PSO, vaccine heuristic mutation ,cauchy mutation and symmetrical mutation are applied to low fitness subpopulation to speed up the convergence and enhance the capabilities of escaping local optimum,normal distribution mutation and improved chaos disturbance are applied to high fitness subpopulation to improve the accuracy of algorithm. Meanwhile, The clone number of antibody, the mutation and crossover ratios can regulate automatically in the improved algorithm.The experiment results demonstrate that the proposed algorithm is superior to several typical modified PSO algorithms and immune clone algorithm.(3) Using each advantage of Polymorphic Ant Colony Algorithm(PACA) and Simulated Annealing(SA),a new hybrid algorithm was proposed.SA is applied to shorten the length of each path after every round of search,so that the increment of pheromone can better reflect the quality of a path.The idea of SA is also applied to the pheromone release mechanism to avert precocity and stagnation,thus the inferior paths can release the pheromone by the competition mechanism based on SA.The 3-opt strategy is applied to the best path of every round of search to improve the search efficiency because it plays the most important role in releasing pheromone.Meanwhile,the newly discovered better path can release more pheromone so that the ants can"remember"the new path in the follow-up iterations.Experiment results show the effectiveness of the proposed algorithm.(4) Artificial Bee Colony(ABC)algorithm is a newly introduced swarm-based algorithm and has been proven to be a better heuristic algorithm compared with other evolution algorithms including particle swarm optimization(PSO). Meanwhile,ABC possess a inferior quality when used to optimize nonseparable functions due to its single dimension updating characteristics .This article analysed ABC's disadvantage and proposed a hybrid algorithm named ABCPSO utilizing each superiority of PSO and ABC.The proposed algorithm modifies the learning strategy of basic ABC to enhance the global information exchange of different particles,splits optimization process into two stages--single dimesnsion updating (SDU)stage and whole updating(WU) stage, implementes the two stages by turns and balances the exploitation depths of the two stages by controlling the parameter'limit'dynamically in line with the success rate of SDU stage.In the WU stage a tentative PSO is adopted by which the swarm can avoid wrong direction to some extent.The performance comparison is carried out against different hign quality PSO variants and basic ABC on the set of standard benchemark functions with asymmetric initialization.Proposed algorithm is efficient in terms of convergence rate,solution accuracy,standard deviation compared with other PSO variants and ABC.Experiment results indicate that the proposed algorithm is an effective technique to optimize not only separable functions but also nonseparable functions.(5) Fuzzy c mean (FCM) clustering algorithm is improved and turned into fuzzy kernel level clustering algorithm. The main measures taken in this algorithm are: using the kernel function; the cut factor to be added; performance function to be optimized by the optimization algorithm studied above; the application of binary-tree split level clustering approach. The simulation results show that the algorithm can effectively overcome the lack of FCM algorithm.(6)The bee colony clustering algorithm is applied to the measuring of fabric color,and experiment result show the algorithm's effectiveness.