基于粒子群优化算法的窗口参数确定方法分析-analysis of window parameters determination method based on particle swarm optimization algorithm.docxVIP

摘要摘要本文在概率密度函数估计的框架下对5种粒子群优化算法的性能进行了验证。这5 种粒子群优化分别是标准粒子群优化、带约束因子的粒子群优化、高斯粒子群优化、带高斯跳跃的高斯粒子群优化、带柯西跳跃的高斯粒子群优化。基于3 种不同的窗口参数表达式确定方法，即Bootstrap方法、Least-squarescross-validation方法、以及biasedcross-validation方法。本文分别使用这5种粒子群优化算法来寻找最优的窗口参数，并通过4 种常用的概率分布（均匀分布、正态分布、指数分布、瑞利分布）对这5 种粒子群优化算法的优化性能进行比较，结果表明，带有跳跃的高斯粒子群优化，即带高斯跳跃的高斯粒子群优化和带柯西跳跃的高斯粒子群优化获得最佳的求解效果。关键词概率密度函数估计窗口参数粒子群优化IAbstractAbstractThedeterminationofbandwidthparameterisacriticalfactorfortheperformanceof probabilitydensityestimationmethod.Theadvancedparameterselectionmethods,suchasthebootstrapmethod,theleast-squarescross-validation（LSCV）methodandthebiasedcross-validation（BCV）method,alwaysneedthehelpofthebrute-forcesearchor exhaustivesearchtofindtheoptimalbandwidthparameters.Inthisdissertation,weapplyfiveparticleswarmoptimization（PSO）algorithms，i.e.，standardPSO（SPSO），PSOwithaconstrictionfactor（PSOCF），GaussianPSO（GPSO），GaussianPSOwithGaussianjump（GPSOGJ）andGaussianPSOwithCauchyjump（GPSOCJ）todeterminetheoptimalbandwidths.Inordertoexperimentallyvalidatethefeasibilityandeffectiveness ofselectingtheoptimalparametersbyusingPSOalgorithms,wecarryoutsomenumericalsimulationsonfourunivariateartificialdatasets:Uniformdataset,Normaldataset,ExponentialdatasetandRayleighdataset.Thefinallycomparativeresultsshowthatour strategiesarewell-performedandGaussianPSOwithjumpmethodscanobtainthebetterestimationsthanotherPSOalgorithms.KeywordsProbabilitydensityfunctionestimationBandwidthparameterParticleswarmoptimization目录目录第1 章绪论........................................................................................................................11.1研究工作的目的与意义...........................................................................................11.2本课题的国内外发展现状.......................................................................................21.3本课题研究的主要内容...........................................................................................33.5本章小结...............................................................................................