Model Predictive Control for Central Plant Optimization with Thermal Energy Storage

Transcription

1 Purdue Uiversity Purdue e-pubs Iteratioal Hig Performace Buildigs Coferece Scool of Mecaical Egieerig 2014 Model Predictive Cotrol for Cetral Plat Optimizatio wit ermal Eergy Storage Micael J. Wezel Joso Cotrols Ic. Uited States of America Robert D. urey Joso Cotrols Ic. Uited States of America Kirk H. Drees Joso Cotrols Ic. Uited States of America Follow tis ad additioal works at: ttp://docs.lib.purdue.edu/ipbc Wezel Micael J.; urey Robert D.; ad Drees Kirk H. "Model Predictive Cotrol for Cetral Plat Optimizatio wit ermal Eergy Storage" (2014). Iteratioal Hig Performace Buildigs Coferece. Paper 122. ttp://docs.lib.purdue.edu/ipbc/122 is documet as bee made available troug Purdue e-pubs a service of te Purdue Uiversity Libraries. Please cotact epubs@purdue.edu for additioal iformatio. Complete proceedigs may be acquired i prit ad o CD-ROM directly from te Ray W. Herrick Laboratories at ttps://egieerig.purdue.edu/ Herrick/Evets/orderlit.tml

2 3379 Page 1 Model Predictive Cotrol for Cetral Plat Optimizatio wit ermal Eergy Storage Micael J. WENZEL 1 * Robert D. URNEY 1 Kirk H. DREES 1 1 Joso Cotrols Ic.; ecology ad Advaced Developmet Buildig Efficiecy; Milwaukee WI Uited States of America mike.wezel@jci.com* robert.d.turey@jci.com kirk..drees@jci.com * Correspodig Autor ABSRAC Liear Programmig is used i order to determie ow to distribute bot ot ad cold water loads across a cetral eergy plat icludig eat pump cillers covetioal cillers water eaters ad ot ad cold water (termal eergy) storage. e objective of te optimizatio framework is to miimize cost i respose to bot real-time eergy prices ad demad carges. A plaig tool tat allows for te user to approimate a year s load distributio ad tus cost i a few miutes is demostrated. e optimizatio framework ca also be used i real-time plat operatio as a model predictive cotrol (MPC) problem. I simulatio te system as demostrated more ta 10% savigs over oter scedule based cotrol trajectories eve we te sub-plats are assumed to be ruig optimally i bot cases (i.e. optimal ciller stagig etc.) For large plats tis ca mea savigs of more ta US $1 millio per year. 1. INRODUCION Desig ad operatio of cetral plats is becomig a icreasigly difficult problem. May ig efficiecy products are available; owever te effectiveess of tese products i reducig te overall cost of operatig a plat is igly depedet o te cotrol tecology tat will be used to properly distribute te load across te may devices (Ma 2011) (Yu 2008). ermal eergy storage ca be used meet te desig day load durig te peak of te otter summer days. Additioally coupled wit real-time pricig for electricity ad demad carges termal eergy storage (ES) offers aoter degree of freedom tat ca be used to greatly decrease eergy costs by siftig productio to low cost times or we oter electrical loads are lower so tat a ew peak demad is ot set. Of course i order to get tese beefits of termal eergy storage optimized cotrol is ecessary a simple sceduled carge old ad discarge scedule will ot suffice. I fact to properly cotrol te ES system oe must predict te termal loads o te buildig or campus ad determie te load distributio across all cetral plat assets tat will result i te lowest cost. is optimizatio must be doe over a recedig orizo; te problem as most of te elemets of a traditioal model predictive cotrol (MPC) problem. is paper describes a model predictive cotrol tecique tat is capable of ruig a plat wit termal eergy storage optimally wile cosiderig real-time electrical eergy pricig demad carges as well as alterate metods of productio wic use differet fuels. e optimal cotrol is performed by splittig te optimizatio ito two cascaded sub-problems tat we solved produce a sub-optimal result but uder most coditios sould be very ear optimal. e lower level optimizatio determies for eac sub-plat (e.g. a assembly of eat pump cillers) te best devices to ru ad te optimal operatig setpoits for te cillers (e.g. flow temperature etc.) for ay give load request ad weater coditio. is optimizatio ca be doe offlie allowig for a optimal efficiecy curve of te sub-plat to be give to te ig level optimizatio. Hig level optimizatio is ru usig tese efficiecy curves o-lie wit a multiple day orizo over wic te best distributio of load across te sub-plats ad ay termal eergy storage is foud for eac our of te orizo usig liear programmig. I simulatio te system as demostrated more ta 10% savigs over oter scedule based cotrol trajectories eve we te sub-plats are assumed to be ruig optimally i bot cases (i.e. optimal ciller stagig etc.) For large plats tis ca mea savigs of more ta US $1 millio per year. 3 rd Iteratioal Hig Performace Buildigs Coferece at Purdue July

3 3379 Page 2 2. PROBLEM DESCRIPION Cetral plat optimizatio is cocered wit cotrollig ay umber of sub-plats feedig ay umber of loads i te most cost efficiet maer possible. Figure 1 sows a illustrative view of te resource flow i a cetral plat tat serves bot ot ad cold water loads of a buildig. e plat cotais a ciller sub-plat eater sub-plat ad a eat recovery ciller sub-plat ad is served by electricity atural gas ad water utilities. e goal is to serve te loads i a way tat as te least ecoomic cost. I real-time pricig scearios or we tere is a electrical demad carge. o perform te optimizatio te termal loads (ad electrical loads) of te buildig must be predicted for some orizo (a umber of days). For tis reaso te problem as all te elemets of a model predictive cotrol problem. It ca be broke ito two parts: predictio ad optimizatio. e predictio problem is posed as: give weater forecast ˆ w te day type day te time of day t ad te past measured load data Y k-1 determie te best estimate of te future weater data. at is fid te best estimate of te loads for legt of te orizo. w Y k 1 l ˆ ˆ day t (1) k e optimizatio problem is ot as simple. If properly desiged te termal eergy storage ca ave a very log time costat eergy ca be stored for a fairly log time before it is lost to te eviromet or eat trasfer across te termoclie makes te eergy uusable. Because te dyamics of te tak are log tere may be some advatages i usig a log orizo. However te equipmet performace curves (power used vs. equipmet load) are i geeral o-cove ad tere are several device o/off decisios to be made. e optimizatio problem is a oliear mied iteger program (NLMIP). is may be itractable i a sort computatioal time. For tis reaso te optimizatio problem is broke ito subproblems. e equipmet (low) level optimizatio determies wic equipmet witi a give subplat to ru give a load Figure 1: Illustrative view of resource use ad assets of a cetral plat. 3 rd Iteratioal Hig Performace Buildigs Coferece at Purdue July

4 ad evirometal coditios. is is described i Q arg mi J Q * LL w LL LL LL w 3379 Page 3 (2) were θ * LL cotais te optimal low level decisios (i.e. biary equipmet o/off decisios flow setpoits ad temperature setpoits) based o te Q te subplat load ad w all pertiet weater coditios. o fid te optimal set of low level decisios te low level cost fuctio J LL is miimized. e low level cost fuctio is te sum of te cost of all utility use per device summed over all equipmet i te subplat. is is give by J LL e u Q c u Q LL w j ji LL w (3) i 1 j 1 were e ad u are te umber of devices i te subplat ad te umber of utilities servig te plat respectively c j is te ecoomic cost of utility j at te curret time ad u ji is te rate of use of utility j by device i. Similar problems ave bee solved o optimal ciller selectio (Deg 2013). At te equipmet level tere is little i te form of system dyamics. e optimizatio is ru slow eoug tat oe ca assume tat te equipmet cotrol as reaced its steady-state. erefore all te parameters ad decisios eed to be made oly at a istace of time rater ta over a log orizo. e subplat (ig) level optimizatio o te oter ad requires a log orizo due to te time costat of te storage taks. Its goal is to miimize te cost ruig meetig te load over te etire orizo by properly distributig te load across te subplats ad storage taks * HL J HL HL HL arg mi (4) * were HL are te optimal ig level decisios (i.e. wat load sould eac of te subplats ad storage taks provide) for te etire orizo. J HL is te ig level cost fuctio te sum of te ecoomic cost of eac utility used by eac subplat at every time i te orizo J HL (5) k 1 i 1 j 1 s u HL c jku jik HL were c jk is te ecoomic cost of utility j at time k ito te orizo ad u jik is te rate of use of utility j by subplat i at time k ito te orizo. e solutio sould be desiged i suc a way tat it provides for two distict use cases. e optimizatio may eiter be used operatioally to determie optimal plat operatio (ad eiter sed te results directly to te buildig automatio system or preset te results to a buildig operator for implemetatio) or as a plaig tool i order to determie te cost of ruig suc a optimized system. e plaig tool sould allow for te user to cage cetral plat cofiguratios ad recalculate cost for a etire year. e plaig tool as muc stricter computatio time requiremets as it must calculate a etire year of plat load distributios i a time frame tat leds itself to iteractive desig. 3. SOLUION DESIGN 3.1 Cascaded Subproblem Descriptio Figure 2 sows te cascaded approac to cetral plat optimizatio. e cascade as two advatages over solvig te wole optimizatio problem: 1) Differeces i te dyamics allow te equipmet level optimizatio to be ru wit a very sort or o orizo; wereas te subplat level optimizatio must look far ito te future to properly make use of te termal eergy storage. 3 rd Iteratioal Hig Performace Buildigs Coferece at Purdue July

5 3379 Page 4 Figure 2: Illustrative view of resource use ad assets of a cetral plat. 2) e subplat level optimizatio is performed witout kowledge of te flow etwork. e equipmet level optimizatio is commuicatig wit BAS ad eeds to be tailored i some way to te plat. e subplat level optimizatio is more geeral ad tus oly as to depeds o te subplats preset. I order to perform te optimizatio te subplat power curve (i.e. te rate of utility use by te subplat as a fuctio of load produced) will be calculated. is is performed by ruig te equipmet level optimizatio for several differet loads ad weater coditios. A curve is te fit to te data ad te subplat curve is give to te subplat level optimizatio for its use. After obtaiig te subplat power curve for eac subplat te cotrol is ready. A predictio is made ad adjusted for feedback. Wit te predictios te subplat level optimizatio is able to use te power curves ad utility rate data ad fid te distributio of te predicted loads across all subplats for te et (orizo) samples. e load distributio for te first time period of te orizo is give to te equipmet level optimizatio. e equipmet level optimizatio is te resposible for determiig wic devices to use te temperature setpoits ad te flow setpoits tat will optimally deliver te requested load from eac subplat. e buildig automatio system troug closed loop cotrol will te modulate te actuators i order to maitai te desired setpoits. e wole process of predictig ad optimizig te subplat ad equipmet level is repeated every sample period. 3.2 Plaig ool Mode of Operatio e plaig tool uses te same optimizatio algoritm; owever tere is o eed to predict te loads i real-time. e data etered ito te plaig tool will cotai all loads for te year. A orizo of te give eatig coolig ad electrical loads alog wit utility pricig is take ad te plat load distributio tat results i te lowest ecoomic cost is foud usig te subplat level optimizatio algoritm. A block of resultat load distributio is take (a legt of time tat is less ta or equal to te orizo) ad accepted to be te true plat dispatc. e orizo is te sifted forward by te block size ad te process is repeated as sow i Figure 3. is allows for te plaig tool to be ru i sorter periods of time ad scale to yet scale to ig fidelity overigt rus. I te plaig tool tere is o reaso tat te optimizatio must be ru for every sample period as is doe i te operatioal tool. Because predictio is essetially perfect i te plaig tool (data is just take from te load time series) te oly data tat ca cage te optimizatio results is te ew block of data tat is obtaied we te orizo is sifted. If te block size is a small percetage of te orizo tis sould ave very little affect o corol. 3 rd Iteratioal Hig Performace Buildigs Coferece at Purdue July

6 3379 Page 5 It ca be see i Figure 3 tat ours 7 troug 12 i te first optimizatio are early idetical to ours 1 troug 6 i te secod optimizatio suggestig tat eve a 12 our orizo would ave ad similar results i tis case. Oce te optimal subplat load distributio is foud for te etet of te plaig tool ru te results of te equipmet level optimizatio are used to calculate te productio ad utility use of eac device witi a subplat. e fuctios wic perform tis calculatio are determied at te begiig of te plaig tool ru i te by sedig various loads ad weater coditios to te equipmet level optimizatio ad fittig te curves i te same way it is doe i te operatioal tool. 4. OPIMIZAION FRAMEWORK 4.1 Liear Programmig Liear programmig was cose as te optimizatio framework for te subplat level optimizatio. A liear programmig problem as te form give by arg mi c ; subject to A b H g (6) Figure 3: I plaig mode te algoritm will optimize load distributio over a orizo ad te accept a block of tose (b) as te actual plat dispatc. is is repeated as te orizo is slid forward i time. 3 rd Iteratioal Hig Performace Buildigs Coferece at Purdue July

7 3379 Page 6 were c is te cost vector is te decisio matri A ad b are te matri ad vector wic describe te iequality costraits ad H ad g are te matri ad vector wic describe te equality costraits. is framework appears igly restrictive; owever witi tis framework it is possible to determie te subplat load distributio for a log orizo i a very sort time frame complete wit load cage pealties demad carges ad plat performace curves. 4.2 Cetral Plat Optimizatio as a Liear Programmig Problem First te problem is formulated for te simple case were oly eergy cost ad equipmet costraits are cosidered. ake te eample plat give i sectio 2. e plat assets across wic te loads are to be distributed are a ciller subplat a eat recovery ciller subplat a eater subplat cold water storage ad ot water storage. e loads across eac oe of tese subplats are te 5 decisio variables tat te optimizatio must determie for eac sample period of te orizo i.e. Q ciller 1... Q rciller 1... Q eater1... Q otstorage 1... Q coldstorage (7) I te simplest form it is possible to assume tat eac subplat as a specific cost per load. is costat COP (efficiecy) ca cage for ay give elemet of te orizo but for tis simple case is ot a fuctio of te loadig. c is give by c u j1 c u j u u j ciller c ju j rciller c ju j eater 0 0 (8) j1 j u were c ju j rciller is used to represet a vector of sums oe for every elemet of te orizo. e last 2 j elemets are 0 to idicate tat cargig or discargig te storage tak as o cost (pumpig power is eglected). It is also ecessary to defie te costraits o te decisio variables. Eac subplat as two capacity costraits Q Q cillerk cillerk Q cillerma 0 k orizo. k orizo (9) ese iequality costraits ca be placed i te form of (6) by eterig A I I Q b 0 0 ciller ma 1 (10) ito te rows of te iequality costrait matri ad vector. Here [I ] used as te by idetity matri [0 ] is used as eiter a by zero matri or by 1 zero vector ad [1 ] is te by 1 oes vector. e storage taks ave similar costraits for teir maimum carge ad discarge rate (i tis case we cosider discargig as a positive load i te vector ). e costraits are give by A I I 0 Q b Q discarge ma carge ma 1 1 (11) ad similarily for te cold water tak. A total demad costrait P elecma ca be implemeted by addig te electrical usage of all te subplats ad te buildig/campus itself P eleccampus. e rows of costraits are A u electrical cilleri uelectricalrciller I uelectricaleater I 0 0 b Pelec ma1 Pelec campus k (12) 3 rd Iteratioal Hig Performace Buildigs Coferece at Purdue July

8 3379 Page 7 to implemet a demad costrait. e fial iequality costraits deal wit tak capacities. e tak must ever carge above its capacity or be discarged below zero. is leads to a series of costraits tat esure tat te tak level at te begiig of te orizo Q 0 Hot plus all te cargig from 1 to k elemets ito te orizo (wit discarge from te tak take as positive tis will be a subtractio) is less ta te capacity. A similar Q ma Hot costrait prevets over discargig te tak. ese etries ito te costrait matri ave triagular matrices. For te ot storage tak s Q0 Hot 1 A b 1 (13) s Qma Hot Q0 Hot were is a lower triagular matri of oes ad s is te legt of time of a elemet of te orizo. Fially te loads must be satisfied wic leads to two sets of equality costraits oe for te ot water load ad oe for te cold water load. o implemet te load costraits H I I 0 0 I 0 1 u I I I 0 lˆ 1... ˆ 1... Cold k g l Hot k electricalrciller (14) For tis eample problem (assumig a orizo of 72 oe our samples) te liear program as 360 decisio variables ad 1224 costraits. However i te liear programmig framework tis ca be solved i less ta 200ms so a plaig problem wit 12 our blocks ca be solved i oly 2 miutes. 4.3 Demad Carge Optimizatio Proper iclusio of te demad carge ito te optimizatio framework is oe way to greatly improve te performace of cetral plat optimizatio. Iclusio of demad optimizatio as bee sow to save as muc as 5% of plat operatio cost o top of te already 8 10% eergy optimizatio aloe will save. o iclude te demad carge it is ecessary to modify te cost fuctio. e first equatio i (6) must be caged to c c map ( ) arg mi (15) demad were c demad is te period s demad carge. wo tigs make te iclusio of te demad carge complicated: first te cost fuctio is o loger liear due to te iclusio of te ma fuctio; secod te c is te eergy cost over te orizo wereas te demad carge is over te demad period. ese two periods migt ot be te same. o cast te ew cost fuctio ito te liear frame work a ew decisio variable peak (te peak demad) is required. e c ca simply be augmeted wit c demad ad wit peak ew elec k c c c. (16) demad Costraits are used to isure tat peak is greater ta te greatest of all te demads over te orizo. peak would ever be greater ta tis as it would be suboptimal. e costraits required are give by A u electrical cilleri uelectricalrciller I uelectricaleater I b Pelec campus k ew peak (17) Additioally te peak decisio variable must be greater ta it as bee at aytime i te past durig tis demad period. o properly make te trade-off betwee icreasig te demad carge versus icreasig eergy cost it is ecessary to weigt te demad carge. e cost fuctio i (16) as compoets tat are over differet periods ad caot be directly compared. e eergy cost is over te orizo wereas te demad carge is over te demad period. o 3 rd Iteratioal Hig Performace Buildigs Coferece at Purdue July

9 3379 Page 8 reweigt te objective fuctio it is ecessary to fid te average eergy cost per day over te orizo tis ca te be multiplied by te umber of days left i te demad period (d demad ) so tat te etire cost fuctio is over te demad period. e ew optimizatio fuctio would be give by wic is equivalet to d arg mi demad c c demad peak (18) arg mic cdemad peak. (19) d demad Eq. (19) simply as te advatage of adjustig oly oe elemet of te cost vector rater ta several. 4.4 Performace Curves ad Cage of Load Pealty Performace curve ca be easily added i a maer similar to te metod demostrated i addig te demad carge. Ay cove performace curve ca be added by te additio of a decisio variable for eac utility for wic its usage vs. productio curve is oliear but cove. I tis case te cost associated wit te variables actual productio is zero wile te ew variable for eac utility is a give a cost equal to te utility s cost at tat time. Liear iequality costraits are te used to costrai te utility use state to be i a piecewise liear approimatio of te epigrap of te performace curve. Of course te utility use will lie o te curve (boudary of te epigrap) because to move above te curve would be suboptimal. Ofte times te optimizatio algoritm will take a subplat from off to full load ad back to off agai i a matter of 3 elemets of te orizo. e optimizatio is fidig areas were tere are small fluctuatios i te utility cost cause tis beavior to ave te least ecoomic cost. is beavior is certaily ot optimal especially if te cost saved is o te order of few cets or dollars. is problem ca also be attacked by augmetig wit additioal decisio variables. I tis case a load cage amout is added at every step i te orizo. e cost of tis decisio variable is give a adjustable pealty (wic ca be specified i dollars per percet cage). e load cage decisio is te costraied to te epigrap of te absolute value of differece betwee te two previous load decisios usig te iequality costraits. 5. PRELIMINARY SIMULAION RESULS o demostrate te cetral plat optimizatio algoritm a eample plat was costructed usig 42.1 MW (12000 to) of ciller capacity 26.3 MW (7500 to) of eat pump ciller capacity ad 53.2 MW (162 mmbu/r) of water eater capacity. e cold termal eergy storage ad 316 MW (90000 to r) of capacity ad could carge or discarge at a maimum rate of 20% per our. e ot termal eergy storage ad 176 MW (600 mmbu) of capacity ad could also carge ad discarge at a rate of 20% per our. All ciller were assumed idetical wit a COP tat depeded o te wetbulb temperature all water eaters were assumed idetical wit a efficiecy of 0.85 ad eac of te tree eat pump cillers ad a capacity of 8.78 MW (2500 to) ad COP (defied as coolig output over electrical iput) of ad o perform a simulatio te data was ru i plaig mode wit te epected ot cold ad electrical loads of te campus served by te cetral plat. e simulatio results for various orizos ad block sizes are sow i table 1. ese results iclude te electricity gas ad water required to ru te cetral plat alog wit te correspodig costs (demad is sow for te etire buildig). As sow i te table te cost decreases as te orizo icreases ad block size decreases. However te icrease i savigs from a orizo above 72 ours is less ta $10k. e optimizatio provides approimately $910k i savigs compared to a sceduled termal eergy storage solutio. Iclusio of te demad optimizatio is wort aoter $400k i savigs. 3 rd Iteratioal Hig Performace Buildigs Coferece at Purdue July

10 rz blck elec. eergy (GW) elec. eergy cost (M$) elec. demad (MW) elec. demad cost (M$) atural gas eergy (GW) 3379 Page 9 It sould be oted tat we icorporatig te demad carge tere are sigificat gais to be made by etedig te orizo to 96 ours. We icludig te demad carge optimizatio te eergy cost is averaged to a per day basis ad te etrapolated to fit te wole demad period. is etrapolatio gets icreasigly better as te legt of te orizo is icreased. Also te results sow tat icludig te cage pealty decreases cost. I geeral tis would ot be te case. Here te load cage pealty ad a secodary effect of reducig te demad. If te load cage were ru wit demad optimizatio te solutio wit te load cage pealty would defiately be greater ta te US $9.34 millio cost witout te cage of load pealty. Figure 4 sows te results of te simulatio. e top two plots are zoomed i to sow te effect of te load cage pealty o te cold water load. Wit te load cage pealty it ca be clearly see tat te ciller load is sigificatly smooter. e secod two plots sow te effect of demad optimizatio. O te first plot demad peaks are clearly over 50 MW i several locatios. However demad optimizatio effectively trims tose peaks to a target tat is establised eac mot. 6. CONCLUSION A cascaded approac to cetral plat optimizatio as bee sow. e subplat level determies ow to distribute te loads betwee differet asset classes witi a cetral plat wereas te equipmet level determies ow to best ru te subplat at tat load. Additioally liear programmig was sow to optimize subplat level distributio ad able to icorporate demad carge load cage pealty ad performace curves. e cascaded approac allows oe make optimal use of computatioal time. e cascaded approac uses o orizo at te equipmet level we dyamics are fast compared to te time to re-optimize plat loads ad use a log orizo we te dyamics ad capacity of te termal eergy storage allow oe to defer loads for log time periods. Additioally liear programmig appears to be a good optimizatio framework for te subplat level optimizatio. It is capable of icorporatig real-world problems like demad carges load cage pealties ad performace curves ito its framework. e liear program ca be solved i a time frame tat makes possible a plaig tool capable of ruig all te ours of a year i a time tat facilitates iteractive plat desig possibly plat desig optimizatio. Simulatios ave sow tat te optimizatio framework is capable of savig over 10% of plat operatio cost over a sceduled termal eergy storage system. REFERENCES atural gas cost (M$) water (km 3 ) Deg K. Yu S. Cakraborty A. Lu Y. Brouwer J. Meta P.G. Optimal scedulig of ciller plat wit termal eergy storage usig mied iteger liear programmig America Cotrol Coferece Jue 2013 pp water cost (M$) otal Cost (M$) Comp. ime (s) INCLUDE DEMAND CHARGE OPIMIZAION INCLUDE LOAD CHANGE PENALY $1 PER PERCEN PER HOUR NO OPIMIZAION N/A N/A N/A able 1: Simulatio results 3 rd Iteratioal Hig Performace Buildigs Coferece at Purdue July

11 3379 Page 10 Figure 4: Simulatio results Ma Z. Wag S Supervisory ad optimal cotrol of cetral ciller plats usig simplified adaptive models ad geetic algoritm Applied Eergy vol. 88 o. 1 pp Yu F.W.. Ca K Optimizatio of water-cooled ciller system wit load-based speed cotro Applied Eergy vol. 85 o. 10 pp ACKNOWLEDGEMEN Special taks are due to Joe Stager Eecutive Director of Sustaiability ad Eergy Maagemet Departmet Staford Uiversity for is collaboratio i defiig te problem statemet ad use cases. Special taks are also due to Dr. Moammad ElBsat of Joso Cotrols ecology ad Advaced Developmet Buildig Efficiecy for providig te data for te o optimizatio case. 3 rd Iteratioal Hig Performace Buildigs Coferece at Purdue July