ASHRAE IJHVAC 17-5-2011 HVAC&R RESEARCH An International Journal of Heating Ventilating Air-Conditioning and Refrigerating Research.pdf

1、HVAC Ei(Engineering Information, Inc.) Compendex and EngineeringIndex; ISI (Institute for Scientific Information) Web Scienceand Research Alert; BSRIA (Building Services Research ACS (American Chemical Society) Chem-ical Abstracts Service and Scientific and Technical Informa-tion Network; CSA: Guide

2、 to Discovery CSA Materials Re-search Database with METADEX, CSA Engineering ResearchDatabase, and CSA High Technology Research Database withAerospace;IIR(InternationalInstituteofRefrigeration)Bulletinof the IIR and Fridoc; and Thomson Gale. Current contents arein ISI Engineering, Computing Online I

3、SSN: 1938-5587Institutional Subscribers: $270, 150, 216.Personal Subscribers: $175, 97, 140.Production and Advertising Office: Taylor European Commission 2006). Smart grid cannot be successful without:information and communication technologies, sensing and measuring technologies, advanced electriceq

4、uipment and systems with new technologies, decision support systems and automatic control strategies.Managing the loads in demand side to match the generation in supply side and the means for establishingthe real-time interaction between both sides are important and essential.Problems and issues in

5、power supply and distributionInmanysmartgridplans,nuclearpowerplantsorthermalpowerplantsaffordthemainelectricalloadofthe power network, while plants located in different locations provide supplemental power using renewableenergysources.Thiskindofarrangementiscalleddistributedgeneration.Therearesever

6、alissuesthatneedto be addressed at in supply side, such as low generation efficiency of thermal power plants, compatibilityof distributed generation power in cases of “plug-in” smart grid and energy storage for surplus electricity.Inelectricitytransformation/distribution,energylossishugeforlongdista

7、nceelectricitydeliveryunderexisting framework. Safety and reliability are important factors that indicate quality of power services.In many cases, an incident in a certain part of the power network may cause a blackout of the wholeregion. Furthermore, low electricity quality (e.g., voltage sags) wil

8、l lead to elevator malfunction that willput passengers in danger. Existing distributed automation systems only process events automatically in aunidirectional way without collecting information from end-users.A combined heat and power (CHP) system is often used to improve the energy efficiency of th

9、ermalpower plants. Distributed generation can be coordinated to control and balance the electricity generationaccording to the predicted load demand and power generation potential. By adopting AC/DC converters,voltage/frequency regulators and reactive power compensators, better electricity quality c

10、an be achieved615HVACa114effective load prediction of (PCM-integrated) buildings for accurate load prediction of the grid;a114effective demand response control methods and strategies;a114response time of demand side management, which is a crucial issue under real time management;a114optimization of

11、PCM integration with buildings; anda114supervisory and optimal control strategies for maximizing the potential of PCM-integrated buildings inload management.Downloaded by T accepted March 9, 2011Haorong Li, Member ASHRAE, is Associate Professor. Daihong Yu, Student Member ASHRAE, PhD student. James

12、E. Braun,PhD, PE, Fellow ASHRAE, is Professor.ical sensors, on average. Today, about 40 relativelylow-cost embedded physical sensors are employedalong with virtual sensors to optimize driving per-formance, safety, functionality, and reliability of ve-hicles (Healy 2010).Incontrast,buildingsystemsrar

13、elyprovide feed-back on energy efficiency or the need for ser-vice and generally do not provide optimized con-trols. In fact, typical information provided to abuilding owner and occupants, even with a directdigital control (DDC) energy management controlsystem (EMCS), is not significantly better tha

14、nwhat was provided 50 years ago. Although theenergy efficiency of individual building compo-nents has improved significantly (e.g., the ratedefficiency of new residential cooling equipmenthas nearly tripled), the operating efficiency is typ-ically degraded by 20% to 30% due to improperinstallation/c

15、ommissioning and inadequate mainte-nance/repair (CEC 2008).619HVAC for example, it is currently notpossible to measure directly the amount of refriger-ant charge within an air-conditioner or heat pump.Virtual sensors have been developed in otherfields to obtain measurements indirectly in a cost-effe

16、ctive, noninvasive, or/and practical manner, butthey are only recently the subject of developmentfor building systems. There is no widely accepteddefinition of virtual sensing. In the context of thisarticle, virtual sensing is considered to include anyindirectmethodofdeterminingameasureablequan-tity

17、 that utilizes outputs from other physical and/orDownloaded by T Hardy and Ahmad 1999; Hardyand Maroof 1999; Oosterom and Babuska 2000;Kestell et al. 2001; Oza et al. 2005; Srivastava et al.2005; Gawthrop 2005; Kabadayi et al. 2006; Boseet al. 2007; Ibarguengoytia et al. 2008; Said et al.2009; Ravee

18、ndranathan et al. 2009; among others).In particular, virtual sensing has found widespreadapplication in process controls and automobiles, sothis article focuses on these two fields in providinga brief history of notable developments.Virtual sensing in process controls“Virtual sensors,” also termed “

19、soft sensors,”have found widespread application in process con-trol engineering since the early 1980s. Researchersin process control engineering (Venkatasubrama-nian et al. 2003a, 2003b, 2003c; Fortuna et al. 2007;Kadlec et al. 2009) use the term virtual sensors tocharacterize software that includes

20、 several interact-ing measurements of characteristics and dynamicsthat are processed (fused) together to calculate newquantities that need not be measured directly. Underthis definition, well-known software algorithms thatare considered to be soft sensors include Kalmanfilters and state observers or

21、 estimators, such aselectric motor velocity estimators. In process con-trol engineering, virtual sensing is focused on esti-mating system dynamics or state variables throughconstruction of state observers. Accordingly, vir-tual sensor development involves representing thewhole control system using a

22、 mathematical tran-sient model through ordinary or partial differen-tial equations, and then constructing state observersor estimators to estimate nonmeasured states usingmathematical transformation techniques. The trans-formed estimators or observers are considered to bevirtual sensors.Virtual sens

23、ing in automobilesThere have been a large number of virtual sens-ing developments for automobiles during the pastdecade. Unlike applications in process control engi-neering, where system dynamics or transient statesare of primary interest, the focus for virtual sensingin automobiles has primarily be

24、en on determina-tion of steady-state variables. The methods for con-structingvirtualsensorshavebeenmorefragmentedand component oriented. Steady-state models rep-resented by algebraic equations are often used torelate the quantity that is not measured directly toone or more quantities that are direct

25、ly measuredusing physical sensors. These steady-state modelscan be considered to be virtual sensors.Figure1depictsavehiclethatemploystenvirtualsensors that have been developed in the past decadeto provide increased functionality, safety, and relia-bility. Table 1 describes the corresponding physical

26、sensorandreferencesforeachvirtualsensor.Forex-ample, the virtual sideslip angle and velocity of thecenter of gravity sensors play important roles in re-ducing the potential dangers associated with loss ofcontrol of a vehicle. These quantities would be dif-ficult and expensive to measure directly and

27、 are es-timated using a hyperbolic tangent switching func-tion (Zhang et al. 2009; Shraim et al. 2006; amongothers) that combines available vehicle information(e.g., mass of vehicle, friction coefficient). Anotherexample is the virtual tire pressure sensor. Conven-tionally, tire air pressure is meas

28、ured directly usinga pressure responsive element located within thetire. However, this construction is complicated andcostly. There have been a number of different devel-opments related to tire pressure indication, reflectedin more than 40 patents (Gustafsson et al. 2001;Yoshihiro 1998; Takeyasu 199

29、7; among others). Awidely studied approach utilizes a Kalman filter toestimate tire pressure in a simple model that useswheel speed and road friction that are also sensedusing virtual observers (Zhang et al. 2009; Shraimet al. 2006; Gustafsson 1997, 1998, 2001; amongothers). On-vehicle sensors alone

30、 do not provideDownloaded by T more than 40 patentsVirtual vehicle tire-roadforces sensorForce sensor, such as straingaugeDoumiati et al. (2009), etc.Virtual vehicle transversalforces sensorForce sensor, such as straingaugeStephant et al. (2004b)Virtual vehicle roadfriction sensorForce sensor, such

31、as straingaugeGustafsson (1997, 1998),Gustafsson et al. (2001), etc.Virtual vehicle sideslipangle sensorwithout a correspondingphysical sensorZhang et al. (2009), Milaneseet al. (2007), Stephant et al.(2004a, 2004b, 2007), etc.Virtual vehicle velocitysensorSpeed sensor, originallydeveloped as virtua

32、lsensorsZhang et al. (2009), Shraimet al. (2006), etc.Virtual vehicle car-2-carcommunication sensorwithout a correspondingphysical sensorRockl et al. (2008)Virtual motor combustiontiming sensorHolmberg and Hellring (2003)Virtual vehicle enhancedlighting preview controlsensorLauffenburger et al. (200

33、7)Virtual vehicle stabilitycontrol sensorCanale et al. (2008a, 2008b),Wenzel et al. (2007), etc.Downloaded by T Kestell et al. 2001; Said et al. 2009). Simi-lartoapplicationsinautomobiles,theprimaryfocusfor virtual sensing developments in these emergingfields has been on estimating steady-state vari

34、ables.However, transient-state variables are also utilizedin many virtual sensing schemes, notably in activenoise control, which uses mechanistic models simi-lar in nature to the process control field. Data-drivenmodeling methods are more frequently used in thefields of wireless communication, senso

35、r networks,and data fusion because the amount of data andinformation in these fields is very rich. Also, re-searchers in these fields are more accustomed to ap-plying data processing techniques and less skilledin developing physical models.Another emerging area for adoption of virtualsensingisthebui

36、ldingindustry.Thedevelopmentofvirtual sensors for building components has laggedbehind other fields, probably because of the frag-mented nature of the industry and the emphasis oninitial costs. In fact, the concept and potential forvirtual sensing has only recently been consideredfor building applic

37、ations, leading to some initial de-velopments (Li and Braun 2007a, 2007b, 2009a,2009b).Development methodology forvirtual sensingVirtual sensors are the embodiment of virtualsensing techniques. For the sake of simplicity, theterm virtual sensor is used interchangeably with vir-tual sensing to presen

38、t development methodology.Although many different types of virtual sensorshave been developed, there is no widely accepteddefinition and no systematic virtual sensing devel-opment methodology. It is useful to categorize vir-tual sensors before attempting to describe generalapproaches for their devel

39、opment.Categorization of virtual sensorsVirtual sensors can be categorized, as shown inFigure 2, according to three interrelated criteria thataffect development approaches: (1) measurementcharacteristics, (2) modeling methods, and (3) ap-plication.The measurement characteristic category refersto whe

40、ther the desired virtual sensor outputs aretransient or steady-state variables. A transient vir-tual sensor incorporates a transient model to predictthe transient behavior of an unmeasured variable inresponse to measured transient inputs. This type ofFigure 2. Categorization scheme for virtual senso

41、rs.Downloaded by TKampjarvietal.2008).Transient-statevirtualsensorsarealsoverycommoninthespecialtychem-istry field (Bonne and Jorgensen 2004). However,steady-state virtual sensors represent the majorityof the applications in different fields (Qin 1997;Casali et al. 1998; Park and Han 2000; Jos de As

42、-sis and Maciel Filho 2000; Meleiro and Finho 2000;Radhakrishnan and Mohamed 2000; Devogelaere etal. 2002; James et al. 2002; among others).With respect to modeling methods, virtual sen-sors can be divided into three types: first-principle(model-driven), black-box (data-driven), and gray-box virtual

43、 sensors. First-principle (physical orwhite-box) virtual sensors are most commonly de-rived from fundamental physical laws and have pa-rameterswithsomephysicalsignificance.Forexam-ple, DeWolf et al. (1996) developed a virtual slurrypolymerization reactor sensor based on a Kalmanfilter, and Prasad et

44、 al. (2002) applied a multi-rateKalman filter to the control of a polymerization pro-cess.Forthesameapplication,Doyle(1998)utilizeda nonlinear observer method. In contrast to first-principle virtual sensors, black-box (data-driven)approaches utilize empirical correlations withoutany knowledge of the

45、 physical process. Examplesinclude multivariate principle component analysis(PCA) (Gonzalez 1999; Warne et al. 2004), par-tial least squares (Frank and Friedman 1993; Kourti2002), artificial neural networks (ANNs) (Poggioand Girosi 1990; Bishop 1995) and so on. A gray-box virtual sensor utilizes aco

46、mbination ofphysicaland empirical models in estimating the output of anunmeasuredprocess(Casalietal.1998;MeleiroandFinho 2000; James et al. 2002).According to application, virtual sensors can bedivided into backup/replacement and observing vir-tualsensors.Backup/replacementvirtualsensorsareused eith

47、er to back up or replace existing physicalsensors.Abackupvirtualsensorcanprovideacheckon the accuracy of an installed sensor and even en-able virtual calibration. For example, the reliabilityof temperature sensors is affected by incorrect in-stallation, hostile environmental conditions, or nat-ural

48、drift (ASHRAE 2009). A replacement applica-tionisdictatedbycostandreliabilityconsiderations,such as the virtual tire air pressure sensor studiedfor automobiles (Gustafsson et al. 2001; Yoshihiro1998; Takeyasu 1997; among others.). A physicalpressure sensor located within the tire is expensiveand exp

49、osed to high rotational speeds and vibration,which, over time, can lead to unreliable measure-ments and failure. Within automotive applications,the majority of the existing virtual sensors eitherbackuporreplacephysicalcounterparts.Incontrast,observingvirtualsensorsestimatequantitiesthatarenot directly observable (or measurable) using exist-ing physical sensors. For example, typically there isno physical sensor to directly determine engine per-formance. Mihelc and Citron (1984) proposed a vir-tual engine-performance monitor for determiningthe relative combustion efficiencies of each