基于神经网络的火电机组实时标煤耗建模

《基于神经网络的火电机组实时标煤耗建模》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、摘要封于笈魄受照寒德+爱怒援毫克南场熬丈方巍，莲本盎故适应魄力翥场球境熬变兜，重要瓣是耍准确掌握发电成本，实{于优化运行，节能降耗。本论文以神经网络理论为基础．主嚣致力予火电站夫溪餐热撬缀稼潆怒率建攥。妻簧进行了强下尼方瑟懿错究：l+首先简要介绍前向神经网络的特点和BP算法及其激进措施；然岳则出了多屡前向蹰络的逼避耱娄鞠泛化裁力驹一蓬垂胥窕簸祭，穆为驭磊建模豹瑾论裱攒；最蓐掇密了～静霹捷速避近鹩亭串经褥络穗造算法。2，嘏撼袋集数据嚎声的特点，我们给出来异常数据剔除法和数攒滤波方法，分别烈异常跳变点鞠麓祝礤声进行蹙理，铸囊蛙累表明，爨绘方法畿较好遗滤豫嗓声，探持疆好魏平

2、滔霞果。3．定性分析了影响荦元机缀标煤糍率的各个因素．然后鼹出了基于棒经阏络的贡献分析法+扶众多澎演闲豢中选取了囊献最大醵6个困豢俸必棒经网终模爨夔翰入变霪；髂冀绪聚表明，聪选敦变鬟蹙商效翡。4．提出了一种燕于增量学习的媳型样本选取方法。并与聚粪方法及仝样举BP算法进行了比较。待粪炼豢袭鞠，班聚餐鞠增量攀笼方法选数麓典鏊撵拳硼蠛弼络均鸯铡予嬉鬣镩经羁路静淄练辩阈，蠢基}遂蘩学嚣匏典整群奉选壤瑟寄燕好静泛纯缝力，觚薅鲶塞了～静嚣l予避鼗遥近穆经鼹终舞鬃揖率选取的有效方法。媛磊络瞧了标漾耗鬻颈溅缀鬻豹慧蘸设诗方寰鞠系统糕蔼，瓣器巾襄链模块侮了麓单寅缀。关键酒：狰鲢弼珞，供

3、热壤缀，撂煤麓拳，戆灭变囊选择，魏型撵零逡馥。j塑墩L———————————————一AbstractInordertograspingthedevelopmentdirectionofelectricitymarketandadaptingthechangeofpowermarket，powerplantshouldintegrateallofitsinformationresource．Itisimportanttomasterthegenerationeostexactlyandimplementdynamiccostanalysis．Areal—timesta

4、ndardcoalconsumptionratemodelofthermalgenerationisresearchedinthisthesis．ThemaincontentsandachievementsofthethesisareasfoIlows：First．wepresentabriefintroductionofmuttiplayerfeedforwardneuralnetworkandBack。PropagationalgorithmInordertoacceleratetrainingspeedofBPalgorithm，someimprovedmeth

5、odsarediscussedinthischapter．Thenwepresentasystematicintroductionofapproximationandgeneralizationresearchoffeedforwardneuralnetworkandthenpresentafastapproximationneuralnetworkconstructionalgorithm．SecondlvIbasedonthecharaderisticofnoisyrealwtimedata,wepresentamoth喇妇氇erejectionofoutlier

6、dataandanalgorithmofdigitalfilter,becausetheintogrityandaccuracyofthereal-timedataacquiredbyP1databasearethefoundationofmodeling．Andthengivethecorrespondingexperimentalresult．Thirdly,weanalyzethefactorsthataffectthestandardcoalconsumptionrate，mainlyfromboilerandturbine．Togeneralizewell，

7、itisnecessarytodeleteunimportantdatacomponentsinthetrainingsets．Weproposeavariableselectionapproachbasedonneuralnetworkarialysis，tosearchthemostimportantfactorsbycalculatingtheircontributionratios．Asaresult，weselectsixfactorsfromvariablesetsasneuralnetworkinputs．Theefficiencyof