Font Size: a A A

Study On Learning Strategy Combined Particle Swarm Optimization And Its Application

Posted on:2014-01-17Degree:MasterType:Thesis
Country:ChinaCandidate:L LvFull Text:PDF
GTID:2268330401453878Subject:Electronics and Communications Engineering
Abstract/Summary:PDF Full Text Request
Inspired by nature phenomenons and Darwin evolutionary theory, genetic algorithm,ant colony algorithm, particle swarm algorithm, artificial immune system, differentialevolutionary algorithm, etc., have emerged and found their nationwide applications innumerical optimization, decision making, process controlling, circuit designing, imageprocessing, data classification, robot technique and automation industry. Particle swarmoptimization is a relatively new intelligent optimization method which imitates fishschooling and bird flocking. Concise concept, easy implementation and fastconvergence make particle swarm optimization receive tremendous amount ofapplications, however, standard particle swarm optimization is easy to get into localoptima and prematurity appears, consequently, many improved approaches have comeinto been. Due to its immaturity of theory and practice, how to better avoid thealgorithm from getting into local optima and better improve the optimization efficiencyhas long been an important research topic. This thesis has introduced and analyzed bothparticle swarm optimization algorithm and its successors, two improved particle swarmoptimization methods are newly proposed. The work of this thesis is as follows:(1) In terms of the drawbacks of the existing improved particle swarm optimizationmethod, a new algorithm based on co-chaos is proposed. This method makes use ofchaos theory, introduces a chaos search operator which is initialization insensitive, easyto jump out of local optima, fast searching, high computational efficient and canconverge to global optima. The method also adopts chaos trajectory to guide thesearching. Based on the co-evolutionary idea, by exchanging the internal informationbetween the same component of particles’, every dimension of the generated elitepopulation is optimized. To preserve diversity, the rest particles remain unchanged,since all the particles are guided by the elites to conduct evolution, thus, theperformance of the algorithm is greatly improved by applying the co-evolution tactic tothe elite population. Contrastive experiments on benchmark functions show that thismethod speeds up the converge.(2) This thesis has put forward a local learning based hybrid discrete particle swarmoptimization method for data feature selection. The particle swarm optimizationalgorithm is widely used for solving continuous optimization problems, however, inreality many problems are discrete, so designing a proper discrete particle swarm optimization techniques has aroused scholars’ attention. This thesis has proposed aneighbor learning based discrete particle swarm optimization method for featureselection, mutation operator in genetic algorithm is adopted, combined with singleelitism strategy and forward greedy local learning strategy, the proposed method hasstrong robustness. The method shows excellent performances on eight UCI data sets,contrastive experiments with three typical methods demonstrate the high efficiency ofthe proposed algorithm.(3) This thesis has given a brief introduction of community detection in complexnetworks, a particle swarm optimization based method is proposed. By introducing theneighbor dominated local search mechanism, the performance of the algorithm isgreatly improved. Experiments on both synthetic and real-world networks demonstratethe effectiveness of the proposed algorithm.
Keywords/Search Tags:particle swarm optimization, chaos theory, cooperation mechanism, feature selection, community detection, local learning
PDF Full Text Request
Related items