30 25 20 15 10-5 1,950 2,025 2,100 2,175 2,250 2,325 2,400 2,475 2,550 2,625 2,700 2,775 2,850 2,925 3,000 3,075 3,150 3,225 3,300 3,375 3,450 3,525 3,600 MW 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Frequency Cumultive Frequency Usg Lod Reserch Dt Model Wer Response Sturt McMenm EFG Meetg My 3-4, 2007 2005, Itron Inc. Frequency Cumultive Freq
Overview 1. Fncil ecstg models generte ecsts monthly sles. y estimted usg billg dt. Sles = F m ( HDD,CDD, EconVrs, Pr ices ) m m 2. HDD CDD vribles ten defed sgle brekpot, exmple CDD65. 3. System lod dt tells us tht reltionship between wer lod non-ler tht not ll degrees eqully powerful. 4. Lod reserch dt cn be used see th t clss level construct multi-prt DD vribles tht improve models. 5. Proly constructed models cn n be used :. Estimte models tht expl hricl sles vritions b. Clendrize: Estimte clendr month sles by clss c. Normlize: Estimte cycle clendr month sles norml wer d. Estimte unbilled sles e. Construct vrince nlys m m 2006, Itron Inc. 2
Wht Model A model model n n eqution eqution tht tht summrizes summrizes reltionships reltionships dt. dt. coefficients coefficients eqution eqution summrize summrize level level dt dt verge verge impct impct chnge chnge drivg drivg fcrs fcrs ( ( X s) X s) on on vrible vrible beg beg expled. expled. Sttticl Sttticl methods methods fd fd prmeters prmeters ( ( b) b) tht tht mke mke errors errors (e) (e) smll. smll. When When ler ler regression regression used, used, it it importnt importnt construct construct X vribles vribles so so tht tht reltionship reltionship between between Y X ler. ler. Y Y = + b X + t t e t X 2006, Itron Inc. 3
Wht We Lern from Net System Lod Gs Dy Sendout (MCF) Ech Ech Pot Pot One One Dy Dy Dt Dt from from 2000-2004 2000-2004 Weekdys Weekdys Blue Blue Sturdys Sturdys Ornge Ornge Sundys Sundys Red Red Holidys Holidys -- -- Green Green Dily Zone Output (GWh) Dily Averge Temture Averge Gs Dy Temture re re stble stble predictble predictble reltionship reltionship between between dily dily system system lods lods wer. wer. Clendr Clendr impcts impcts lso lso cler. cler. reltionship reltionship between between wer wer system system lod lod nonler. nonler. 2006, Itron Inc. 4
Focus on Electricity Dily Zone Output (GWh) reltionship reltionship between between temture temture lod lod not not sgle sgle stright stright le. le. It It cn cn be be pproximted pproximted by by series series connected connected slopes. slopes. Models Models cn cn estimte estimte reltive reltive slopes slopes ech ech segment. segment. Dily Energy 2002-2004 Dily Averge Temture 2006, Itron Inc. 5
Wht We Lern from Lod Reserch Dt Lod Lod Reserch Reserch Dt. Dt. Ech Ech pot pot one one dy. dy. Dt Dt 2003 2003 Res SCI LCI Lod Lod Reserch Reserch dt dt cn cn help help clrify clrify how how wer wer effects effects work work t t revenue revenue clss clss level. level. se se dt dt suggest suggest different different HDD HDD CDD CDD triggers triggers different different clsses. clsses. reltive reltive power power degrees degrees ech ech rnge rnge cn cn be be estimted estimted from from se se dt dt used used models models monthly monthly sles. sles. 2006, Itron Inc. 6
Lod Reserch Power Dily Dt Lod Lod reserch reserch dt dt very very powerful powerful understg understg wer wer reltionships. reltionships. re re 30 30 pots pots month, month, strong strong dvntge dvntge over over monthly monthly dt. dt. A dy dy dy dy cn cn be be mtched mtched clenly clenly wer wer tht tht dy. dy. Monthly Monthly sles sles dt dt cn cn be be mtched mtched pproximtely pproximtely dys dys wer wer bsed bsed on on cycle cycle dtes. dtes. re re more more wer wer vrition vrition t t dily dily level. level. Billg Billg month month wer wer dt dt ggregted ggregted weighted weighted over over roughly roughly 60 60 clendr clendr dys dys tht tht impct impct sles sles billg billg cycle. cycle. extremes extremes lost lost both both X Y directions. directions. 2003 2003 Lod Lod Reserch Reserch Dt. Dt. Ech Ech pot pot one one dy. dy. 2003 2003 Billg Billg Dt. Dt. Ech Ech pot pot one one month. month. 2006, Itron Inc. 7
Inmtion from Neurl Network Dily Energy (GWh) Dily Dily energy energy from from lod lod reserch reserch hourly hourly priles priles cn cn be be used used estimte estimte neurl neurl network network models. models. Dily Averge Temture Neurl Neurl network network models models flexible flexible nonler nonler models models tht tht work work well well th th type type dily dily dt. dt. denergy/dtemp (Gwh/deg) Estimted Estimted neurl neurl network network models models hve hve derivtives derivtives respect respect X vribles. vribles. A simple simple model model Averge Averge Temtures Temtures gives gives derivtives derivtives shown shown here. here. An extr CDD dds 5 GWh An extr HDD Blnce dds.7 Pot t GWh 60 Dily Averge Temture 2006, Itron Inc. 8
Cpped vs Uncpped DD Vribles Sles slope = 10 slope = 25 slope = 15 Uncpped Exmple CDD60 = Mx(T-60,0) CDD70 = Mx(T-70, 0) Sles = + 10 CDD60 + 15 CDD70 60 70 Avg Temp Sles slope = 10 slope = 25 Cpped Exmple CDD60Cp10 = M(Mx(T-60,0),10) CDD70 = Mx(T-70, 0) Sles = + 10 CDD60Cp10 + 25 CDD70 60 70 Avg Temp 2006, Itron Inc. 9
Estimtion Multiple CDD & HDD Vribles Uncpped Dily Dily energy energy from from lod lod reserch reserch hourly hourly priles priles cn cn be be used used estimte estimte reltive reltive strength strength CDD CDD HDD HDD vribles. vribles. Models Models cpped cpped uncpped uncpped vribles vribles sme. sme. Predicted Predicted vlues vlues ll ll residul residul stttics stttics identicl. identicl. However However coefficient coefficient strd strd errors errors t-stts t-stts stronger stronger cpped. cpped. Cpped Accumulte Tke Differences 2006, Itron Inc. 10
Constructg Multi Prt (Sple) Vribles lod lod reserch reserch slope slope estimtes estimtes cn cn be be used used construct construct multi-prt multi-prt hetg hetg coolg coolg vribles. vribles. We We cll cll se se HDDSple HDDSple CDDSple. CDDSple. HDD Sple.107*Wthr.HDD60 +.718*Wthr.HDD55 +.175*Wthr.HDD50 -.213*Wthr.HDD30 CDD Sple.061*Wthr.CDD60 +.675*Wthr.CDD65 +.058*Wthr.CDD70 +.206*Wthr.CDD75 2006, Itron Inc. 11
How Regression Sees Dily Dt Residentil Residentil Lod Lod Reserch Reserch Dt. Dt. Ech Ech pot pot one one dy. dy. Dt Dt 2003 2003 Dily Energy (GWh) Dily Energy (GWh) Dily Energy (GWh) Multi Multi prt prt sples sples turn turn nonler nonler reltionship reltionship two two ler ler reltionships. reltionships. Th Th very very cler cler dily dily dt dt becuse becuse ech ech dy dy hs hs CDD CDD or or HDD HDD but but not not both. both. HDD Sple CDD Sple 2006, Itron Inc. 12
How Defe Averge Temture July 8 Avg24 = 88.0 Hi/Low = 88.5 July 10 Avg24 = 82.4 Hi/Low = 87 Averge Mx & M hs been dustry strd decdes. Averge hourly vlues will work better on dys where re sudden temture chnges, such s cused by fternoon thundersrms summer. 2006, Itron Inc. 13
Defition Averge Temture Actul Dily Energy (GWh) Act Act vs. vs. Pred Pred usg usg Avg Avg 24 24 hourly hourly vlues vlues Dily Dily energy energy from from lod lod reserch reserch hourly hourly priles priles more more strongly strongly relted relted verge verge temture temture thn thn (Mx+M)/2. (Mx+M)/2. Actul Dily Energy (GWh) Act Act vs. vs. Pred Pred usg usg (Mx+M)/2 (Mx+M)/2 Predicted Dily Energy (GWh) Neurl Neurl Network Network Actul Actul Predicted Predicted vlues vlues usg usg dily dily verge verge temture temture (verge (verge 24 24 hourly hourly vlues) vlues) MAPE MAPE = = 3.45% 3.45% Predicted Dily Energy (GWh) Neurl Neurl Network Network Actul Actul Predicted Predicted vlues vlues usg usg verge verge Mx Mx M M temtures. temtures. MAPE MAPE = = 3.87% 3.87% 2006, Itron Inc. 14
Sesonl Wer Response System Lod Dily Zone Output (GWh) Dily Zone Output (GWh) Dt Dily Energy 2002-2004 Jn July Hot wer hs bigger impcts lter summer months thn sprg months. Dily Zone Output Dily Zone Output My Th suggests models slopes tht vry by month or seson. Oct Averge Temture Averge Temture 2006, Itron Inc. 15
Sesonl Response from Lod Reserch Res Dily Energy (GWh) Res Dily Energy (GWh) July & Aug Lod Lod reserch reserch dt dt supports supports sesonl sesonl differences differences system system lod lod dt. dt. Jn & Feb Res Dily Energy (GWh) Res Dily Energy (GWh) Apr & My Oct & Nov Averge Temture Averge Temture 2006, Itron Inc. 16
Trnsition Monthly Dt 1. lessons lerned nlys from dily lod reserch dt cn be used strengn models tht use monthly dt. 2. followg models monthly billg dt. monthly dt billg cycles, so it necessry construct pproprite wer vribles. 3. Averge temture (verge hourly vlues) computed ech dy n HDD CDD vlues ccumulted cross dys ech cycle. cycle vlues n weighted cross cycles get monthly HDD CDD vlues. 4. gol estimte models tht:. Use multi-prt HDD CDD vribles (sples) constructed usg prmeters from lod reserch dt nlys. b. Allow HDD CDD slopes vry by month/seson. c. Cn be used wer normliztion, clendr month, unbilled clcs. 2006, Itron Inc. 17
Detils on Monthly Wer Vribles 1. Compute verge temture by dy 2. Compute CDD60, CDD65,... billg dys ech cycle. 3. Compute weighted verge CDD60, CDD65,..., billg dys usg pproprite weights ech cycle. 4. Compute CDD Sple HDD Sple vribles from CDD60, CDD65,... 5. Divide CDDSple HDDSple by number billg dys. Th will look like dily dt. For exmple, vlue 10 dictes tht verge dy month hd 10 degree dys. 6. If pproprite do th ll THI DD coolg where THI cludes dily temture humidity. 2006, Itron Inc. 18
How Regression Sees Monthly Dt UPC Per Dy (KWh) Residentil Residentil Billg Billg Dt. Dt. Ech Ech pot pot one one month. month. Dt Dt 1991 1991 2006. 2006. Cycle Averge Temture UPC Per Dy (KWh) CDD Sple CDD Sple Reltionships Reltionships less less cler cler monthly monthly dt. dt. Mny Mny months months hve hve both both HDD HDD CDD CDD over over cycle cycle dys. dys. UPC Per Dy (KWh) Ech Ech pot pot billg billg month. month. Color Color codg codg by by seson seson Blue Blue = = Wter, Wter, Green Green = = Sprg Sprg Red Red = = Summer, Summer, Ornge Ornge = = Fll Fll HDD Sple CDD Sple HDD Sple 2006, Itron Inc. 19
Models Monthly Dt Dependent Dependent vrible vrible monthly monthly use use cusmer cusmer billg billg dy. dy. Addg Addg Monthly Monthly bries bries ccounts ccounts lightg lightg or or sesonl sesonl effects. effects. 2006, Itron Inc. 20
Models Sesonl CDD Slopes Dependent Dependent vrible vrible monthly monthly use use cusmer cusmer billg billg dy. dy. Sesonl Sesonl CDD CDD Sple Sple slopes slopes constent constent expecttions. expecttions. CDD CDD effects effects lower lower Sprg Sprg months. months. 2006, Itron Inc. 21
Wer Effects Chnge Over Time Wer slopes chnge over time Cusmer growth implies more hetg/coolg equipment Chnges hetg/coolg sturtion levels chnge CDD HDD slopes cusmer Chnges hetg/coolg efficiency chnge CDD HDD slopes cusmer Dt Dt from from 2004 2004 2005 2005 Dt Dt from from erly erly 1990 s 1990 s For For residentil residentil cusmers, cusmers, response response hot hot wer wer hs hs cresed cresed significntly significntly between between erly erly 1990 s 1990 s 2005 2005 on on cusmer cusmer bs. bs. Th Th constent constent ntionwide ntionwide trend trend wrd wrd lrger, lrger, more more completely completely cooled cooled homes. homes. Models Models tht tht use use hricl hricl dt dt must must ccount ccount th th growth growth wer wer sensitivity, sensitivity, project project contued contued chnges chnges future. future. 2006, Itron Inc. 22
Models Sesonl Slope Trends Dependent Dependent vrible vrible monthly monthly use use cusmer cusmer billg billg dy. dy. Trend Trend vrible vrible creses creses by by 1 1 ech ech yer yer 0 0 2005. 2005. So So estimted estimted slopes slopes 2005 2005 slopes. slopes. Slopes Slopes erlier erlier yers yers reduced reduced by by.066.066 KWh/degree KWh/degree yer. yer. For For exmple, exmple, July July slope slope 1.64 1.64 1993 1993 2.43 2.43 1995. 1995. 2006, Itron Inc. 23
Models AR1 Residul Addg Addg n n AR1 AR1 term term improves improves overll overll fit. fit. coefficients coefficients not not ltered ltered significntly. significntly. Coefficients Coefficients out out AR1 AR1 2006, Itron Inc. 24
Actul & Predicted Vlues Fl Model Monthly Use Per Cusmer Per Billg Dy (KWh) Actul Actul vlues vlues Red Red Predicted Predicted vlues vlues Blue Blue Vlues Vlues Monthly Monthly Use Use Per Per Cusmer Cusmer Per Per Billg Billg Dy Dy 2006, Itron Inc. 25
Usg Structured HDD, CDD SAE Approch With th pproch, structured vribles constructed hetg coolg uses. CDDSple HDDSple vribles cn be used usge equtions. For exmple: XCool = CoolIndex CoolUse CoolIndex t y t = rmleff y Weight Type t Type St St Type y Type by Eff Eff Type y Type by CoolUse t = P P t 98 Inc Inc t 98 b HHSize HHSize t 98 c CDDSple CDDSple t 98 2006, Itron Inc. 26
Buildg Models on Per Dy Bs 1. By buildg models on dy bs, y cn esily be used : -- Budget ecstg on clendr month or cycle bs -- Wer normliztion booked billed sles -- Clendriztion billed sles -- Estimtion unbilled energy 2. Estimte models use billg dy s Y vrible (or use cusmer billg dy). 3. Estimte models CDD billg dy HDD billg dy s X vribles. 4. Evlute models pproprite wer -- Clendr month CDD/HDD clendr dy -- Unbilled corner CDD or HDD unbilled dy n multiply by number dys 5. Usg dy models mkes billg dt look like dily lod reserch dt... except it still just one number month. 2006, Itron Inc. 27
Conclusions Lod reserch dt cn be used refe improve models wer response. Improved models will do better job > Forecstg > Wer normliztion > Estimtion clendr month energy > Estimtion unbilled energy > Vrince nlys 2006, Itron Inc. 28