最大熵原理的供热负荷预报研究
2008年2月第35卷 第1期
西安电子科技大学学报(自然科学版)犑犗犝犚犖犃犔 犗犉 犡犐犇犐犃犖 犝犖犐犞犈犚犛犐犜犢
Feb.2008
o.1Vol.35 N
最大熵原理的供热负荷预报研究
陈 烈,齐维贵,邓盛川
(哈尔滨工业大学电气工程及自动化学院,黑龙江哈尔滨 1)50001
摘要:根据建筑供热的特点和供热节能控制的需要,提出应用最大熵法进行负荷预报,介绍了最大熵谱对从热力站采集的历史随机负荷序列进行预处理,将其中的确定性部分和随机部法原理及Burg算法,分进行分离;并对负荷样本序列,分别用相关法和最大熵谱法进行负荷预报,对两种结果进行了分析比较,采用最大熵谱法进行负荷预报,其预报精度、自适应性和算法的实时性均能较好地满足建筑分户计量节能供热的要求.
关键词:最大熵;随机序列;负荷预报;计量供热
中图分类号:()TM921.2 文献标识码:A 文章编号:10012400200801018306
犛狋狌犱犳犺犲犪狋犾狅犪犱犳狅狉犲犮犪狊狋犻狀犪狊犲犱狅狀狋犺犲狔狅犵犫
犿犪狓犻犿狌犿犲狀狋狉狅狉犻狀犮犻犾犲狆狔狆狆
犆犎犈犖犔犻犲,犙犐犠犲犻狌犻,犇犈犖犌犛犺犲狀犮犺狌犪狀犵犵
(,SchoolofElectricalEnineerinndAutomationgga,H)HarbinInstituteofTechnoloarbin50001,China 1gy
:犃犫狊狋狉犪犮狋ccordinothecharacteristicsofarchitectureheatsulndthedemandsforenersavin Agtppyagyg
,loadforecastinasedonthemaximumentroethodisproosed.Bsinheleastsuarecontrolgbpympyugtq
,fittinndthepretreatmentaroachthehistoricalrandomloadseriescollectedfromtheheatsulgappppystationaresearatedintothecertainpartandtherandompart.Thentheloadseriesisdealtwithbhepytautocorrelationmethodandthemaximumentroethodresectivel.Comarinheresultsofthesepympypgttwomethodsshowsthattheloadforecastinasedonthemaximumentrotheorcanmeetthedemandsgbpyy
,forheatinnersavinontrolintermsoftheforecastinccuracautoadativeandrealtimeabilitgegygcgaypybetter.
:;;;犓犲狅狉犱狊aximumentrorandomseriesloadforecastinheatmeterin mpygg狔犠
Duetothelareconsumtionofenerintheheatinstem,effectiveenersavinontrolsstemsgpgygsygygcyaredesirabletodecreasethetotalcost.Accurateforecastofheatloadinheatinstemisakeoheatgsyytsulontrolsstemdesin.MattiasB.O.Ohlsson1describedamethodusinrtificialneuralnetworkppycygga
,toforecasttheheatloadofalarebuildinnAmerica.TheinutsoftheANNmodelweretimeggip,,temeraturesunlihtandwindseedwhiletheoututwastheheatload.ZhouEnze2aliedthetimepgpppp
,seriestoaccomlishshorttermforecastinfheatloadandutilizedcorrelationmethodformetahasepgopforecastin.Theresultswereusedasaninstructionoftheeneranaementofheatinetwork.ggymggnOutdooreuivalenteuationwhoseparameterscouldbeestimatedwasaliedtotheheatloadqqpp
34
CaoYuiantal.resentedamethodtoadusttheaveraetemeratureofsulaterforecastinpjgpppywqgeg.
,alonithfluctuationofthetemeratureoutdoor.Howeverthemethodinref.[1]ismorefeasibleingwp
[]
[]
[]
[]
]macroscoicadministrationthaninrealtimecontrol.Reference[2hasthedisadvantaeoflowforecastinpgg
收稿日期:20070610
基金项目:哈尔滨市科技创新人才研究专项资金资助()RC2006XK007001
作者简介:陈 烈(),男,哈尔滨工业大学博士研究生,:1982Emailchenliehitit.edu.cn.@h
自然科学版)5卷 西安电子科技大学学报( 第3184
]]recise.Theforecastininreference[3and[4whichmerelastherelationshiithweatherisnotfitpgyhpw
,aloadforecastinethodusinforthemeasuredheatinstem.Basedontheabovebackroundgmggsyg,:’whichconsiderstwomainrandomfactorsweatherandconsumersmaximumentroheorisproosedpytypneeds.Thesimulationresultsshowthatthismethodismoreaccuratethanthemethodwhichusing
,itismoreefficientandcanmeetthedemandofrealtimeoftheheattraditionaltimeseries.Inadditionsulstem,eseciallinthecasetheheatloadisuncertain.ppysypy
1犪狓犻犿狌犿犲狀狋狉狅狉犻狀犮犻犾犲 犕狆狔狆狆
1.1 Princilesofentroppy
:,Informationprincileisdescribediftheeventishaensureltheentroilleualtozero.The pppypywqvalueofentroillincreasealoniththeuncertainofevent.pywgw
,Suosematrix犡isadiscreterandomvariableand狆stheprobabilithiletherandomvariableisppyw犻i…狓犻=1,2,3,狀.犻,
},犻=1,…,犘{犡=狓2,狀 ,犻=狆犻
:Theentroisdefinedasfollowspy
狀
()1
,2,…,)=-犎(犡)=犎(n犘狆狆犻 .1狆狀∑犘犻l
犻=1
()2
,ftherandomvariableiscontinuousdistributedanditsprobabilitensitunctionis狆(狓)the Iydyf
:entroisdefinedpy
犎(犡)=-
1.2 Maximumentroulepyr
狓)ln狆(狓)d狓 .
∫狆(
-∞
+∞
()3
T
,2,…,,Ifthesamleisatimeseries狓=(狓狓犖)theentrofthisrandomprocesscanbewrittenppyo1狓
:as
犎狓=-
,2,…,where狓)isjointprobabilitensitunction狆(狓狓犖).SuosethetimeseriesisGaussianydyfpp狆(1狓
,rocesswhoseaveraeiszerothuspg
/T-1)/2-犖12-
,2,…,)()犖狓 ,狓狓犖)=(2det犚狓(犖)exπp-狓犚狓(狆(1狓2
wherethe犚狓(犖)isautocorrelationmatrix.
)):Euation(4canalsobesubstitutedb5qy(
狓)ln狆(狓)d狓 ,
∫狆(
-∞
+∞
()4
()
()5
[/12
]()()et犚狓(犖)n2犲6lndπ .+犖l
2
(),狓(),…,(),Ifnonstationarimeserieshasbeendescribedas狉1狉thecoursewhichextraolatesytp狓0狉狓犖
()b,thenextautocorrelationvalue狉ovethepremise犎狓ismaximizedisnamedmaximumentropy狓犖+1a
犎狓=
rule.
()a狉nbeobtainedbolvinhefollowinuation:Thusysgtgeq狓犖+1c
()]/(())=0 ,det犚狓[犖+1狉狓犖+1
()狉狓1
,where
)=犚狓(犖+1
()狉狓2
()…狉(狉狓0狓1-犖()…狉(狉狓1狓2-犖
()狉狓1
=0 .
()8()7
()狉()…狓犖+1狓犖
1.3 Maximumentroectrumestimatepysp
[]
Maximumentroectrumestimate5isamodernsectrumestimatebasedonmaximumentropyspppy
第1期 陈 烈等:最大熵原理的供热负荷预报研究
185
theor.Thedataoutsidetheobservationareacanbeforecastedbxtraolatinheautocorrelationyyepgtfunctionofprocessinfinitel.Themethodcansolvethelowresolutionproblemcausedbddinindowyyagwwithclassicalsectrumestimate.p
,accordinoFeerRiestheorem,maximumentroectrumestimateiseuivalenttoARInfactgtjpyspq,andsinalanalsisandlinearforecastinrrorfiltergyge
犖
犛犿)=犘犖狓(
2(犳
犖
(1-∑犪x犽ω-jp犽e
犽=1
2
) ,
()9
where犪isalsotheARmodelarameterscanbeacuiredbolvinuleWalkereuationwhichhas犖+1pqysgYq犽autocorrelationfunctionvalues.犘犖istheoututpowerof犖ranksforecastinrrorfilter.pge
,犪,…,Thecoefficientofforecastinrrorfilter1,anbeobtainedbollowinatrix,-犪-犪yfgmge1-2犖c
()狓0狉
()狉狓1
()狉狓-1()狉狓0
…狉(1狓-犖+1犘犖…狉(-犪10狓-犖+2
= .
()狉狓0
0-犪犖犚犖)=狓(()10
()狉()…狓犖-1狓犖-2
2犻狊狅狊犪犾狅犳犾狅犪犱狊犪犿犾犲 犇狆狆
2.1 Choiceofexerimentsamlepp
atacollectedfromacertainheatsultationofHarbinaretreatedasoriinaldataforload Dppysg
forecastin.Thesamleincludes70dasdatabetweenDecember1th,2005andFebruarth,2006.Thegpyy8dataareshowninTable1.
,,Table1 DaAveraeHeatLoadFromDec.1th2005ToFeb.8th2006yg
/Dateday/Load
-1
·h)(MJ
,D,D,Dec.1stec.2ndec.3nd200544.250
200545.750
200545.625
…
Dec.26th,Dec.27th,200548.750
200545.250
…
Feb.7th,Feb.8th,200659.250
200659.625
……
2.2 Pretreatmentofrandomprocess
Thecertainpartcanbeestimatedbsinleastsuareandtherandompartcanbesearatedfro
mtheyugqpsamepartsimultaneousl.y
,Beforemodelintherandompartshouldbepretreated.g)(1)zeromeanvaluemethodorthetimeseries狓(犻犻= F
…,,1,2,狀)thedatashouldbedisosedberomeanvaluepyz)]≠0.methodif犈[狓(犻2)Stabilizationuntestisutilizedtorealizesteadtest. Ry’spracticabletosmallsamleobservationserieswithoutItpconsiderinhedistributionruleofthedata.gt
Thetimeseriesneedstobestabilizediftheheatloadisnonstationareries.Inthispaertrenditemsareremovedbyspyusinirstorderdifferencemethod.gf
Therandomsinalsearatedaboveisdealtbhezerogpytmeanmethodandstabilization.AndthefinalrandomsinalgsearatedisshowninFi.1.pg
Fi.1 Randomsinalafterpretreatment.gg
自然科学版)5卷 西安电子科技大学学报( 第3186
3犲犪狋犾狅犪犱犳狅狉犲犮犪狊狋犻狀犻狋犺犿犪狓犻犿狌犿犲狀狋狉狅犲狋犺狅犱 犎犵狑狆狔犿
3.1 Burlorithmgag
[]
ccordintoBurlorithm6,reflectanceiscalculatedundertherulethatboththeaveraepowerof Aggagg
theforwardandbackwardforecastinrrorsareminimized.Thusthepowersectrumofsinalcanbegepgacuiredrestrictedlevinsonrecursivealorithm.Thedetailedaroachinvolvessixstes(seeqybyLgppp:below)
):1Theinitialvalueofforecastinrrorpowercanbecalculatedbgey
犘0=
犖
犖
狀=1
∑
2
狓(狀 ,()11
andtheinitialvalueofforwardandbackwarderror
()()()犳0狀=犵0狀=狓狀 .)Let犿=1,thenthereflectanceis 2
狀2+犵犿-1(狀2] .犓犿=-∑犳犿-1(狀)狀-1犳犿-1(犵(∑2狀=犿+1狀=犿+1
):Coefficientoffiltercanbeexressedasfollows 3p
()()犿犿-1犿-1
犪犪犿-犽 ,+犓犿犽=犪犽
犿-1
()12
犖
犖
)
()13
()14
andthenletthereflectance犓犿=犪犿.
):4Forecastinrrorpowercanbecalculatedbgey
犿
犘犿=(1-犓犿
2
犘)
犿-1
.
()15()16()17
)O:ututsoffilterareobtained 5p
)狀)=犳犿-1(狀)狀-1 ,+犓犿犳犿(犵犿-1(
)狀)=犵犿-1(狀-1狀) .+犓犿犵犿(犳犿-1(
)))Inaddition,犿←犿+1,andreeatsteuntil犿=狆. 6~5pp2
3.2 Forecastinlinaximumentroethodgappygmpym
,Thealorithmbasedonmaximumentrotheorwhichisutilizedtoforecasttheloadofheatsulgpyyppysstem,solvestheinconseuenceproblemthatthevaluesoftimeseriesoutside“window”aresettozeroyqbraditionalalorithm.Thustheforecastinoneisextendedtoinfinite
.ytggz
Fi.2 Loadforecastinasedonmaximumentroethod.ggbpym
,ThetimeseriesfromDec.30th,2005toFeb.8th,2006isdealtwithBurlorithmforecastingagg
第1期 陈 烈等:最大熵原理的供热负荷预报研究
187
whilethenumberofthetimeseriesis40.ThesimulationresultsareshowninFi.2.g
,,totheheatsulstemwhichhasthecharactersofrandom,nonlinearandtimeAccordintoFi.2ppysygg,varintheaccuracfheatloadforecastinisimrovedobviouslsinaximumentrorincile.ygyogpybyugmpypp3.3 Resultscomarisonbetweenmaximumentroethodandautocorrelationmethodppym,ThrouhsimulatintheonestendtwosteeatloadforecastinithautocorrelationmethodareggpaphgwshowninFi.3.AndcomarisonofthosetwomethodsisshowninTable2
.gp
Fi.3 Loadforecastinithautocorrelationmethod.ggw
Table2 Comarisonofloadforecastinsinaximumentroethodpgugmpym
andautocorrelationmethod
LoadforecastinethodgmOnesteaximumentroethodpmpymTwosteaximumentroethodpmpymOnesteutocorrelationmethodpautocorrelationmethodTwostepa
AccumulativeAbsolute
-1
/(·h)ErrorMJ
AveraeAbsoluteg
-1
/(·h)ErrorMJ
AveraeAbsoluteg/%RelativeError
7.697.808.969.13
150.89152.71175.59178.92
3.773.824.394.47
,hrouhthecomarisonoftheaccumulativeabsoluteerrortheaverae accumulativeabsolute Tgpg
,errorandtheaveraeabsoluterelativeerrorinTable2,itcanbeconcludedthattheaccuracfthegyomethodthispaerintroducedishiherthanusinutocorrelationmethod.pgga
4狅狀犮犾狌狊犻狅狀 犆
,Inthispaertheheatsulorecastinasedonmaximumentrorincileisproosed.Thepppyfgbpypppresultsshowthismethodcanbeadvantaeouslmloedinheatsulontrolsstem.Theconclusiongyepyppycy:canbeobtainedasfollows
)1Thesamledataarepretreatedtoimrovetheaccuracfforecastin.ppyog
,2)Thealorithm,whichbasedonmaximumentrotheorcanbeprocessedonline.Itcanmeetthegpyydemandsoftherealtimesstem.y
)C,homarinheresultseatloadforecastinasedonmaximumentroethodhashiher3pgtgbpymgaccurachanusinutocorrelationmethod.ytga
)4TheforecastinethodbasedonthemaximumentroasthesameeffectasARmethodingmpyh’stationarrocess.Furthermoreitsalsosuitabletoparametersestimatefornonstationarrocess.ypyp
自然科学版)5卷 西安电子科技大学学报( 第3188
:Reference
[]O1hlssonMBO.PredictinstemLoadswithArtificialNeuralNetworkmethodsFromtheGreatEnerredictorgSygyP
[],():ShootoutJ.ASHARETrans1994,100210631074.
[]Z,[]2houEnzeFaniumu.TheStudnHeatLoadForecastoftheHeatinoilerHouseJ.JournalofHarbinUniversitgXyogBy
,():2000,3335660.ofCivilEnineerinndArchitecturegga
[]J],3onssonGR,JonssonVK.EuivalentOutdoorTemeratureforDistrictHeatinstem[J.AshareTrans1995,101qpgSy
():1394399.
[]C,]4aoYuianShenJinan.AnEnersavineIntellientHeatinontrolSstem[J.IndustrialInstrumentation&qggygTypggCy
,():2002,1414446.Automation
[]B[//,5uraximumEntroectralAnalsisC]37thAnnIntExlorGeohsicsMeetin.NewYork:IEEEPressgJP.MpySpyppyg
1978:3439.
[]Z[:,6hanianda.ModernSinalProcessinM].BeiinTsinhuaPublishinoman2002.gXggjgggCpy
(编辑:齐淑娟)
(上接第124页)
[]R,A]2oerdinkJBTM,MeisterA.TheWatershedTransform:DefinitionslorithmsandParallelizationStrateies[J.jgg
,():FundamentalInformaticae2001,412187228.
[]J],():3ackwaGradientWatershedinMorholoicalScalesace[J.IEEETransonImaeProcessin1996,56913yPT.pgpgg
921.
[]G]4auchJH.ImaeSementationandAnalsisViaMultiscaleGradientWatershedHierarchies[J.IEEETransonImaeggyg
,():Processin1999,816979.g
[]王养利,吴成柯.基于V氏图的图像分割及表示[]():5J.西安电子科技大学学报,2000,273340344.
,WuC[]Wananlihenke.ImaeSementationandReresentationBasedonVoronoiDiaramsJ.JournalofXidiangYggggpg,():Unviersit2000,273340344.y
[]任获荣,王家礼,张平.形态金字塔图像分割算法[]():6J.西安电子科技大学学报,2004,312248252.
,W,[]RenHuoronanialiZhanin.ImaeSementationAlorithmBasedontheMorholoicalPramidJ.JournalgJgPggggpgyg,():ofXidianUnviersit2004,312248252.y
[]崔明,孙守迁,潘云鹤.基于改进快速分水岭变换的图像区域融合[]:7J.计算机辅助设计与图形学学报,2005,17(3)
546552.
,,CuiMinSunShouianPanYunhe.AnImaeReionMerinlorithmBasedonModifiedFastWatershedTransformgqggggAg[],():J.JournalofComuterAidedDesin&ComuterGrahics2005,173546552.pgpp
(编辑:郭 华)