資源描述:
《改進(jìn)的粒子群優(yōu)化算法及其在石油性質(zhì)預(yù)測(cè)中的應(yīng)用》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。
1、摘要在石油化工領(lǐng)域,隨著競(jìng)爭(zhēng)的加劇和企業(yè)對(duì)經(jīng)濟(jì)效益的不斷追求,如何利用已有的信息準(zhǔn)確預(yù)測(cè)石油性質(zhì),已經(jīng)成為一個(gè)值得研究的問題。人工神經(jīng)網(wǎng)絡(luò)(ANN)模仿人腦結(jié)構(gòu)及智能行為,具有并行處理、自組織、自適應(yīng)等特性,是一種強(qiáng)大的非線性建模手段,目前已在石油性質(zhì)預(yù)測(cè)領(lǐng)域得到了應(yīng)用,但是由于傳統(tǒng)人工神經(jīng)網(wǎng)絡(luò)自身存在的局限性,對(duì)石油性質(zhì)預(yù)測(cè)的精度和泛化能力有待于迸一步提高。粒子群優(yōu)化算法(PSO)源于對(duì)鳥群捕食行為的研究,是一種基于群體智能的演化計(jì)算技術(shù)。由于它具有較強(qiáng)的全局搜索能力,較少的參數(shù)設(shè)置,簡(jiǎn)單容易實(shí)現(xiàn),所以一經(jīng)提出,
2、就引起了許多學(xué)者的關(guān)注,并得到了迅速的發(fā)展,目前,該算法已成功應(yīng)用于函數(shù)優(yōu)化、神經(jīng)網(wǎng)絡(luò)訓(xùn)練、模式識(shí)別、模糊系統(tǒng)控制等諸多領(lǐng)域。針對(duì)PSO存在易陷入局部極值、進(jìn)化后期收斂速度緩慢的缺點(diǎn),提出一種基于速度夾角的粒子群協(xié)同優(yōu)化算法(V-PSCO),并且引入了一種基于柯西分布的累積分布函數(shù)的慣性權(quán)重調(diào)整策略。用兩個(gè)典型測(cè)試函數(shù)的優(yōu)化問題測(cè)試V-PSCO,實(shí)驗(yàn)結(jié)果表明,V-PSCO優(yōu)化性能明顯提高。將V.PSCO引入神經(jīng)網(wǎng)絡(luò),優(yōu)化連接權(quán)值和閾值,給出了優(yōu)化過程的原理和流程。用基于Vopsco的神經(jīng)網(wǎng)絡(luò)建立石油性質(zhì)的預(yù)測(cè)模型,
3、實(shí)驗(yàn)結(jié)果表明,與傳統(tǒng)BP算法和標(biāo)準(zhǔn)粒子群優(yōu)化算法(SPSO)相比,基于V-PSCO神經(jīng)網(wǎng)絡(luò)模型預(yù)測(cè)精度更高,泛化性能更好,為準(zhǔn)確預(yù)測(cè)石油性質(zhì)提供了一種有效的方法。關(guān)鍵詞:粒子群優(yōu)化,測(cè)試函數(shù),神經(jīng)網(wǎng)絡(luò),石油性質(zhì),預(yù)測(cè)模型ImprovedParticleSwarmOptimizationAlgorithmandItsApplicationtoPredictingPetroleumPropertiesSongYongqiang(ControlTheoryandControlEngineering)DirectedbyP
4、ro£XiaBokaiAbstractInthefieldofpetrochemicalindustry,asthemarketcompetitionbetweenenterprisesissevereandtheycontinuouslypursueeconomicefficiency,it’Sveryworthytopredictpetroleumpropertiesaccuratelybasedonnearinflateddatainproduction.Artificialneuralnetwork(ANN
5、)callsimulatethestructureofhumanbrainandintelligentbehavior,andhasthepropertyofparallelprocess,self-organization,self-adaptationandSOon.Itisallimportantmeansofnonlinearmodeling.Atpresent,theANNhasbeenapplicationinpredictionforpetroleumproperties.Butduetoitsinh
6、erentimperfections,theprecisionofpredictionandgeneralizationoftraditionalANNforpetroleumpropertiesneedtobeimproved.ParticleSwarmOptimization(PSO)algorithmisanevolutionarycomputationtechniquebasedonffWarlnintelligenceoptimizationalgorithm,whichwasinspiredbysoci
7、albehaviorofbirdflocking.Becauseofitsstrongabilityto酉obalsearch,lessparametersandsimplicity,itsmuchattentionhasgainedsinceitisproposed,andalotofPSOalgorithmsaredevelopedrapidly.UptiUnow,thesealgorithmshavebeensuccessfullyappliedinmanyareassuchasfunctionoptimiz
8、ation,neutralnetworktraining,modelidentification,fuzzysystemcontrol,etc.Aimingatthedisadvantageofthesealgorithmsthatareeasilytrappedinthelocaloptimizationandconvergencespeedisslowi