lable at ScienceDirect

Building and Environment 85 (2015) 104e113

Contents lists avai

Building and Environment

journal homepage: www.elsevier .com/locate/bui ldenv

Temperature sensor placement optimization for VAV control usingCFDeBES co-simulation strategy

Zhimin Du*, Peifan Xu, Xinqiao Jin, Qiaoling LiuSchool of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai, China

a r t i c l e i n f o

Article history:Received 24 September 2014Received in revised form17 November 2014Accepted 27 November 2014Available online 5 December 2014

Keywords:Co-simulationSensor placementOptimizationVAV controlHVAC

* Corresponding author. Tel.: þ86 21 34206533.E-mail address: duzhimin@sjtu.edu.cn (Z. Du).

http://dx.doi.org/10.1016/j.buildenv.2014.11.0330360-1323/© 2014 Elsevier Ltd. All rights reserved.

a b s t r a c t

The popular optimal control approaches in the heating, ventilation and air conditioning (HVAC) systemjust focus on the energy consumption mostly. It usually simplifies the thermal comfort issue throughusing an indoor average temperature, which may result in the improper indoor temperature distribution.The co-simulation technology, which integrates the building energy simulation (BES) and computationalfluid dynamics (CFD), can provide a possible solution to avoid the false optimization in the controlprocess. In this paper, a simple co-simulation strategy is presented to integrate the BES and CFD tech-niques for the HVAC system. The energy simulation and thermal comfort calculation are coupledtogether and the indoor temperature distribution is embedded into the VAV control process. The CFDeBES co-simulation method is validated in the HVAC simulator of an office building located in Shanghai.With the CFDeBES co-simulation strategy, the indoor temperature sensor placement is optimized thoughconsidering the energy consumption and predicted mean vote (PMV) simultaneously. The results showthat the commonly selected sensor position of indoor temperature is not always the best solution for theVAV terminal control.

© 2014 Elsevier Ltd. All rights reserved.

1. Introduction

Indoor thermal comfort and energy efficiency have alreadybecome the two main issues in the heating, ventilation and airconditioning (HVAC) system. To obtain the higher energy efficiency,some energy simulation tools such as DOE-2 [1], EnergyPlus [2] andTRNSYS [3] are widely used to analyze the energy consumingprocess and then develop the optimal control strategies for thedifferent buildings. Wang [4] developed a TRNSYS based simulatorfor the HVAC system in an office building. This HVAC simulator canbe used to test different optimal control strategies for the possibleenergy saving potentials. Fan [5] developed the simulator for anairport HVAC system using EnergyPlus. He presented severaloptimal control strategies in the simulator to validate the potentialenergy saving of the airport. Perez-Lombarda [6] employed DOE-2to evaluate the energy efficiency of HVAC system in an officebuilding. Yuan [7] developed the model predictive strategy torealize the temperature control of multiple zones. Chao [8] devel-oped a dual-mode demand control strategy to analyze the indoor

air quality and the possible energy saving. These energy simulationtools including the detailed component models usually carry outthe complicated control logic so as to obtain the better energyconserving results.

However, the capacities of these energy simulation tools arequite limited in the aspect of indoor thermal comfort analysis.During the energy simulation process, each room is usuallyconsidered to be one calculation node through assuming the indoorair to be well mixed. Since the indoor dynamic property is simpli-fied, no detailed room temperature distribution but indoor averagetemperature can be provided. As a result, it cannot supply thespatial variation of temperature during the control process.Without the spatial temperature distribution, it is difficult tooptimize the energy saving and thermal comfort comprehensively.Consequently, it is necessary to develop the optimal controlembedded with spatial temperature distribution for the higherenergy efficiency and better thermal comfort in the buildings.

Computational fluid dynamics (CFD) technique is a widely usedtool for the thermal comfort calculation. Many applications havebeen applied in the building and HVAC system successfully [9e13].Zhai and Chen [14] compared several coupling methods betweenCFD and energy simulation tools. The static and dynamic couplingapproaches between CFD and energy simulation were presented

Nomenclature

T temperature (�C)G humidity (kg/kg)CC CO2 or pollutant concentration (ppm)C thermal capacitance (kJ/�C)R thermal resistance (�C/kW)Q heat (kW)S moisture or pollutant generation ratev volume flow rate (m3/s)V air volume (m3)M air mass (kg)A area (m2)F mass flow rate (kg/s)P pressure (Pa)t time (s)H control outputK gaine difference of controlled variable and its setpointA ~ G sensor positionPMV predicted mean voteHVAC heating, ventilation and air conditioningVAV variable air volumeCFD computational fluid dynamicsBES building energy simulation

Greek symbolsr air density (kg/m3)J sensible heat ratioq variable of sensort time constantD drifting bias

Subscripts and superscriptsSA supply airRA return airstr non-transparent layerwin transparent layerinf infiltrationexf exfiltrationint internalplt CO2 or pollutantin inletex outletwt waterC coils sensor measurementr real valueP proportional itemI integral itemD derivative itemi, i�1 time

Z. Du et al. / Building and Environment 85 (2015) 104e113 105

and tested in an office. Bartak [15] also integrated the energysimulation tool with the CFD models. This coupling strategy wasvalidated in the university building. Djunaedy [16] presented anexternal coupling strategy between the energy simulation and CFDsoftware. Through exchanging the necessary data, the energysimulation and CFD models were well coupled. Furthermore, Zhaiand Chen [17] made the sensitivity analysis for an office building.They compared several coupling approaches between CFD andenergy simulation tools through considering the building charac-teristic and calculation precision.

As to the comprehensive optimization, many control strategieshave been delicately designed [18e21] in the HVAC system. How-ever, the negligence of the key component such as the sensor maylead to the optimal target unreachable. Besides the accuracy ofmeasurements, selecting a better sensor position should be alsoconsidered in the optimal control process. Although the sensoractually plays essential roles in the control loop, its placementoptimization was not paid enough attention in the last few years.For the variable air volume (VAV) control loop in the HVAC system,the temperature sensor is usually placed at the return air inlet.Whether this placement method is always the best solution duringthe whole control process is an interesting but unsolved issue.

Recently, Fan and Ito [22] developed a BES-CFD integrationapproach to study various supply air location in an office building.With the co-simulation of CFD and energy analysis tools, the energyconsumption using the different ways was analyzed. Sun andWang[23] presented a CFD-based virtual testing method for the controlof indoor environment. The virtual sensor was used to compensatethe effect of nonuniform stratification on the temperature controlprocess. This virtual sensor improved the control reliability in amechanical ventilated room. Zhang [24] presented a contributionratio of indoor climate (CRI) method for the building energy system.With the CRI method, the temperature distribution of an officeusing CFD was combined with its thermal load simulation.

The co-simulation between CFD and energy consumption has notbeen well developed in the VAV control process. It is also necessaryto optimize the sensor placement in the HVAC system. In this paper,a simple CFDeBES (building energy simulation) coupling strategy ispresented to co-simulate the energy simulation together with thethermal comfort analysis. The indoor temperature distribution isembedded into the VAV terminal control process. The temperaturesensor placement is optimized through considering the indoorthermal comfort and energy consumption simultaneously.

2. System description

2.1. The HVAC system

The typical HVAC system in an office building is shown in Fig. 1,which can be partitioned into water side and air side. In the waterside, the supply chilled water coming from the chiller is transportedto the air handling unit by the 2nd level pump. The return chilledwater passing by the air handling unit is circulated back to thechiller by the 1st level pump. Actually, the air handling unit is theheat exchanging place between the air and water.

On the other hand, the supply air, which is the mixture of theoutdoor air and recycle air, is circulated to the air handling unit bythe supply fan and exchanges heat with the chilled water. Afterbeing cooled down (in the summer condition), the supply air iscirculated to the VAV terminals to meet the indoor requirement.With the return fan, the return air is divided into two parts: theexhaust air and the recycle air. The former is discharged into theoutside space and the latter is reused to another air circle.

2.2. Pressure-independent VAV control

In the office building, pressure-independent VAV terminalcontrol (Fig. 1) is employed to maintain the required indoor

Fig. 1. The diagram of HVAC system.

Z. Du et al. / Building and Environment 85 (2015) 104e113106

temperature of each room. In the pressure-independent VAVcontrol loop, the temperature controller and air flow ratecontroller are integrated to realize the cascade control process.The room temperature controller, which is the inner one of thecascade controllers, outputs the setpoint of air flow rate to theouter controller according to the current temperature measure-ment. As the outer one of the cascade controllers, the air flowrate controller modulates the VAV damper according to the realflow rate measurement and the setpoint from the innercontroller.

Fig. 2. The grid partition of the CFD model.

3. CFDeBES co-simulation methodology

3.1. CFD models

The CFD model is used to calculate the indoor spatial temper-ature distribution. It will be combined with TRNSYS to co-simulatethe operation and control process of HVAC system. A RNG keεmodel is used in this paper that is appropriate to characterize theflow field in an air conditioning room [25]. After selecting the formand coefficients of RNG keε model [26], the SIMPLE algorithm isused to solve the equations of continuity, momentum and energyconservation [27]. In the CFD model of room 1, 12 occupants,computers, tables and chairs are all considered as the simplerectangular solid.

As to the boundary conditions, there is no penetration and non-slip condition imposing at the solid wall boundaries. The supply airvent is defined as the velocity flow inlet. The east wall is set withthe radiation heat flux. The ceiling is set with the lighting heat flux.The west, south and north walls are considered as adiabatic. Therelative humidity in the room is not considered.

The grid generation influences the accuracy of the numericalsolver in modeling the air flow profile. Too intensive cells maydecrease the calculation speed and cost the relative long

calculation time. Since the calculation results from CFD are used toassist the simulation with seconds-level time step and its control,the grid with 54,242 is selected for the CFD calculation shown inFig. 2.

3.2. Simulation of the HVAC system

3.2.1. The models of building and HVAC systemA simplified dynamic building model [4,28,29] is used to

simulate dynamic process of the office building including the bal-ances of energy, moisture, CO2 and the pollutants [30,31]. Someassumptions are made during the modeling process:

1) Only vertical direction heat transfer is considered for the wall;2) The equivalent sol-air temperature is used for the radiation heat

of sun;3) The indoor air is well mixed. The indoor air temperature dis-

tribution is not considered and an average indoor temperature isused.

The models of multiple zones are expressed as Equ. (1)e(3).

Z. Du et al. / Building and Environment 85 (2015) 104e113 107

Cair;idTair;i ¼�

Tstr;i � Tair;i� XN 1 þ �

Twin;i � Tair;i�

dtm¼1

Rstr;eq2;i;m

�XNm¼1

1Rwin;eq2;i;m

þ Qint;i þ rSA;ivSA;icp�TSA � Tair;i

�

� Tair;i � Tfur;iRfur;i

þXk1

k¼1

Tik � Tair;iRik;eq2

þXk2

k¼1

ui;kAi;kcp�rair;kTair;k � rair;iTair;i

�

þminf ;icpTout �mexf ;icpTair;i(1)

Mair;idGidt

¼ SG;i þ rSA;ivSA;iðGSA � GiÞ þminf ;iGout �mexf ;iGi

þXk2

k¼1

ui;kAi;k�rair;kGk � rair;iGi

�

(2)

Vair;idCCidt

¼ Splt;i þ vSA;iðCCSA � CCiÞ þminf ;i

rair;outCCout �

mexf ;i

rair;iCCi

þXk2

k¼1

ui;kAi;kðCCk � CCiÞ

(3)

where T and G are the temperature and humidity, CC is the CO2 orother pollutant concentration. C is thermal capacitance, R is thethermal resistance, N is the number of external walls. The subscriptstr and win means the non-transparent (heavy) and transparent(light) layer of the wall accordingly. Qint is the internal heatincluding the equipment, lighting and occupants, SG and Splt are themoisture and CO2 (or pollutant) generation rate, respectively. rSA isthe density of supply air, vSA is the supply air volume flow rate, ui;kis the air exchange velocity between zone i and k, A is area,minf andmexf are the infiltration and exfiltration air flow rate, Mair and Vairare the total air mass and volume.

The models of air handling unit are expressed as Equ. (4)e(6).

CCdTCdt

¼ TSA;in � TCRair

� TC � Twt;in

Rwt(4)

TSA;ex ¼ TSA;in �J$ðTSA;in � TCÞmaircPair$Rair

(5)

Twt;ex ¼ Twt;in þ TC � Twt;in

mwtcPwt$Rwt(6)

where CC is the heat capacitance of coil, Rair and Rwt are the heatresistance at air and water sides, respectively. TC is the mean tem-perature of coil. TSA,ex and TSA,in are the outlet and inlet supply airtemperature. Twt,ex and Twt,in are the outlet and inlet supply watertemperature in the cooling coil. J is the sensible heat ratio.

Besides the above models, the component models of chiller,pump, fan, air duct, air damper, water pipe and water valves con-structed [28,29,31e33] are used in this paper. The flow-pressurebalance model [34] is employed to calculate the mass balance,resistance and flow balances, which connects the related compo-nents in the HVAC system.

3.3. The models of control system

To ensure the system operate on the required condition, thecontrol system should be integrated into the HVAC simulator thatincludes the sensor, controller and actuator. The dynamics of sen-sors including temperature (T), pressure (P) and flow rate (F) aresimulated using the following equation.

qis ¼�qir � Dq0

��h�

qir � Dq0

�� qi�1

s

i$e�

Dtt (7)

Where qs represents the output of the sensor, qr represents the realvalue, Dq0 is the drifting bias, t is the time constant, i and i�1represent the current and last time step, respectively.

The typical PID controller model shown in Equ. 8 is used torealize the cascade control of VAV terminal in the HVAC system.

HPID ¼ KP$

�eðtÞ þ 1

tI

ZeðtÞdtþ tD

deðtÞdt

�(8)

Through connecting the simulation models of physical compo-nents and control system, the complete simulator of HVAC systemcan be constructed [28,29]. With this simulator, the further optimalcontrol strategies and energy efficiency evaluation can be testedeffectively [31e33].

3.4. Co-simulation strategy for CFD and HVAC simulator

A simple co-simulation strategy is presented to combine theCFD and HVAC simulator shown in Fig. 3. One interface module isdeveloped to share and exchange the simulation data betweenenergy simulation and thermal comfort analysis (Fig. 3(a)). At eachtime step, the HVAC simulator calculates the physical process andrelated control response. Then the CFD part calculates the indoortemperature distribution and velocity distribution according to thereceived information such as the load, supply air temperature andsupply air flow rate. Once the CFD calculation is convergent atcurrent time step, the CFD results are transferred back to the HVACsimulator. Receiving the related information from CFD, thepressure-independent VAV controller modulates the terminaldamper to maintain the indoor temperature at the desired value.The spatial temperature distribution is embedded into the VAVcontrol process as the reference. At the same time, the responses ofbuilding to the VAV cascade controllers are calculated using theHVAC simulator. And the building energy simulation is also carriedout according to the return air temperature from the CFD outputs.

Although the time step of HVAC simulator is 1 s, the indoor loadand thermal distribution don't change greatly in the short time. Tosave the simulation time and improve the discordance betweenCFD and HVAC simulator, the indoor air flow distribution is sup-posed to be constant in five minutes. Consequently, the CFD carriesout the spatial temperature calculation and exchange data withHVAC simulator every five minutes. Fig. 3(b) illustrates the com-plete co-simulation process between thermal comfort calculation(5 min time step) and energy simulation (1 s time step).

4. Validation

An office building located in Shanghai is selected to validate theco-simulation strategy presented in this paper. The building has thegross area 11,000 m2 and each storey is 2332 m2. The façade of thisoffice building is 95 mm double layer glass. The height of eachstorey is 3.9 m and the thickness of the floor is 0.4 m. In each storey,two same air handling units are used to serve for each half part. Thedesign supply air flow rate of VAV is 7.5 m3/s. The setpoint of supplyair static pressure can be varied from 300 Pa to 700 Pa. The north

Fig. 3. The diagram of co-simulation logic.

Z. Du et al. / Building and Environment 85 (2015) 104e113108

half floor shown in Fig. 1 served by one air handling unit is simu-lated as the target air side for the convenience. Based on the orig-inal chilled water system, an equivalent chilled water system isdesigned to match the half storey and its AHU air side. The heattransfer between two half floors is neglected. The room 1 studied inthis paper has the structure size of 12.8m� 8.6 m (length�width).The weather data is selected according to the Shanghai weatherdata records. The main component models of HVAC system havebeen validated using the real structure parameters of equipment.And the TRNSYS based simulator for the building and HVAC systemwas calibrated and tested using various control strategies[28,29,31e33]. The VAV terminal control loops are well tuned toensure the proper P/I/D gains, which can promptly respond to theload changing of each air conditioning room.

4.1. Simulation test condition

One typical summer day of Shanghai is selected as the simula-tion test outdoor condition. Fig. 4 illustrates the load conditionsincluding the sun radiation load, occupants load, lighting andequipment load.

The top view of the test room is illustrated in Fig. 5. Twelveoccupants and their computers, tables and chairs are selected in theoffice. Besides the return air inlet, the indoor air temperature

Fig. 4. The room cooling load profiles.

sensor may be placed at various positions (A … G) respectively(Fig. 5). The sensor placements from A to G are in the 1.1 m heightlevel of the room. To consider the thermal comfort of the head andfeet levels, two sets of observing points are selected to analyze thespatial temperature distribution shown in Fig. 5. The points from 1to 9 are used to analyze the thermal comfort of occupant's headlevel. And the points from 11 to 19 are used to analyze that of oc-cupant's feet level.

4.2. Flow field under various supply air flow rates

The CFDeBES co-simulation platform developed for HVAC sys-tem is validated first in this paper. To validate the impact of varioussupply air flow rates to the indoor temperature and flow distribu-tion, three cases with various supply air flow rates are tested as the

Fig. 5. Top view of the sensor placement and observing points.

Fig. 6. Indoor temperature distribution in case 1.

Z. Du et al. / Building and Environment 85 (2015) 104e113 109

following. The indoor cooling load is not changed and the supply airtemperature is set at the fixed value.

CASE 1: the supply air flow rate is 2400 m3/h, the supply airtemperature is 13 �C;

CASE 2: the supply air flow rate is 1200 m3/h, the supply airtemperature is 13 �C;

CASE 3: the supply air flow rate is 600 m3/h, the supply airtemperature is 13 �C.

The comparisons of temperature distribution and air velocitydistribution for Case 1, 2 and 3 are illustrated in Figs. 6e8 andFigs. 9e11, respectively. No matter in the 0.1 m or 1.1 m height level,the indoor temperatures of Case 1 are the smallest values (21e22 �C)shown in Fig. 6. And the air flow velocities are the largest ones inFig. 9 (0.2e0.4 m/s). Consequently, the occupants would have thecooler feeling about the indoor thermal environment in Case 1.

As for the Case 2 shown in Figs. 7 and 10, the temperatures andair flow velocities in 0.1 m and 1.1 m levels are appropriate thatindicates the indoor thermal comfort is the most satisfied. For the

Fig. 7. Indoor temperature

Fig. 8. Indoor temperature

Case 3 shown in Figs. 8 and 11, in addition, the temperatures in both0.1 m and 1.1 m levels are largest (about 26 �C) while the air flowvelocities are smallest (0.1e0.15 m/s). The indoor occupantsconsequently would have the warmer feeling.

Obviously, the temperature and velocity distribution may bedifferent under various conditions. Changing the supply air flowrate may result in quite different indoor temperature distribution.Then the indoor thermal comfort will be affected inevitably.Consequently, it is necessary to modulate the supply air flow ratetimely to satisfy the indoor requirement during the controlprocess.

For each room, on the other hand, the supply air flow rate iscontrolled automatically. The indoor temperature PID controllermodulates the VAV terminal according to the measurement oftemperature sensor and its setpoint. The installing position oftemperature sensor may influence the supply air flow directly,which affects the indoor temperature distribution and energyconsumption.

distribution in case 2.

distribution in case 3.

Fig. 9. Indoor velocity distribution in case 1.

Fig. 10. Indoor velocity distribution in case 2.

Fig. 11. Indoor velocity distribution in case 3.

Fig. 12. Supply air flow rate under different sensor placement.

Z. Du et al. / Building and Environment 85 (2015) 104e113110

4.3. Thermal comfort under various sensor placement

In this office building, the HVAC system operates from 8 AM to8 PM on the working days. The temperature setpoint of room 1 is24 �C. The supply air temperature setpoint of air handling unit is13 �C. The temperature sensor is placed at the different positionincluding A, B, C, D, E, F, G and return air inlet. The co-simulationillustrates that the supply air flow rates of room 1 are quitedifferent when the temperature sensor is placed at different posi-tion. It can be seen in Fig. 12 that the supply air flow rate is thesmallest one when the sensor is placed at position C. And the flowrate is the largest onewhen the sensor is placed at E or F. The reasonis that the position C is close to the supply air diffuser with thelower temperature, while position E or F is close to the occupantsand computers with the higher temperature. It should be notedthat the supply air flow rate is also larger when the sensor is placedat the return air inlet.

On the other hand, different supply air flow rate changes theindoor temperature distribution and velocity distribution. Conse-quently, the indoor thermal comfort is also influenced. The indoor

Fig. 13. PMV via time under different sensor placement.

Z. Du et al. / Building and Environment 85 (2015) 104e113 111

Table 1Energy consumption under different sensor placement.

Sensor isplaced at

Energy consumption (kWh)

Supply fans Return fans Pumps Chillers Total Relativevariation

Return air inlet 153.32 85.89 115.91 434.1 789.22 0A 151.65 84.68 116.21 434.92 787.46 �0.22%B 151.07 84.24 116.24 434.95 786.50 �0.34%C 149.14 83.03 117.24 438.95 788.36 �0.11%D 151.29 84.42 116.24 434.91 786.86 �0.30%E 154.01 86.44 115.89 433.96 790.30 0.14%F 153.87 86.31 115.82 433.76 789.76 0.07%G 151.73 84.72 116.09 434.52 787.06 �0.27%

Z. Du et al. / Building and Environment 85 (2015) 104e113112

temperatures and velocities of the observing points are obtainedusing the co-simulation of CFD and HVAC simulator. The predictedmean votes (PMV) of these points with various placing ways arecalculated to evaluate the indoor thermal comfort. The PMV resultsvia time in the head level are illustrated in Fig. 13.

Generally, the PMV value is recommended to be maintainedbetween �0.5 and þ0.5 to get the most satisfied indoor thermalcomfort. If the sensor is placed at the position of E, F or return airinlet, the PMV values are lower indicating the cooler feelings for thethermal comfort. If the sensor is placed at position of A, B, D or G,the PMV values are appropriate indicating the satisfied thermalcomfort. Furthermore, the occupants may feel warmer if the sensoris placed at position C because the PMV values mostlyexceeded þ0.5. Therefore, the temperature sensor placement in-fluences the indoor thermal comfort. When the temperature sensoris placed at the positions of return air inlet, supply air outlet or closeto the heat source, the indoor thermal comfort is not good as theexpectation. When the sensor is placed far away from these posi-tions, the thermal comfort is satisfied.

4.4. Energy consumption under various sensor placement

Besides the thermal comfort during VAV control process, theenergy consumption of HVAC system is also compared when thesensor is placed at different position. The energy consumptionunder each case is illustrated in Table 1.

When the temperature sensor is placed at position C, the energyconsumption of fans is the smallest because it has the lowest supplyair flow rate. When the sensor is placed at return air inlet, E or F, theenergy consumption of fans is larger because of the higher supplyair flow rate. In addition, the total energy consumption are largerwhen the sensor is placed at return air inlet, C, E and F, while that is

Fig. 14. Return air temperature under different sensor placement.

smaller when placed at position A, B, D and G. Although the energyconsumption of fans is the smallest for the placement of C, theenergy consumption of pump and chiller is the largest. The reasonis that placing the sensor at position C means the highest return airtemperature shown in Fig. 14 indicating the largest cooling loadrequirement. Table 1 also shows the relative variation of energyconsumption if the sensor placed at the return air inlet is used asthe comparison baseline.

Consequently, the sensor installing position as well as sensoraccuracy can affect the energy consumption of HVAC system.Compared with the sensor measuring bias [35], the impact of sensorplacement shown in Table 1 seems to be smaller. The reason is onlyone room's sensor position is changed in this paper. The temperaturedistribution of other seven rooms is not considered so as to avoid toomuch calculation time of CFD. If all of the rooms change the sensorpositions, the impact to system energy consumptionwould be larger.

5. Conclusions

A simple coupling strategy for the building energy simulationand computational fluid dynamics is presented for building andHVAC system. The co-simulation method between the energysimulation of TRNSYS and thermal comfort analysis of CFD isdeveloped in this paper.

The CFDeBES co-simulation strategy is used to assist the VAVoptimal control in HVAC system through optimizing the energyconsumption and thermal comfort simultaneously. The placementof indoor temperature sensor in the VAV terminal control loop isoptimized using the CFDeBES co-simulation approach. The PMVand energy consumption are evaluated when the temperaturesensor is placed at different position. The results show that thepopular temperature sensor placement being close to the return airinlet may be not always the most optimal solution. The sensorplaced at the position B, which is at the head level close to theinternal north wall, is the best placement considering both PMVand energy consumption.

With the application of wireless sensors in the buildings, mul-tiple temperature sampling information representing the differentindoor position can be used to improve the VAV control capacity.With the multiple indoor temperature sensors, the VAV controlstrategy can be designed more flexible and robust. Optimalselecting themost suitable sensor as the controller input, accordingto the actual changing requirement, may be the optimization so-lution for the HVAC system.

Acknowledgment

This work was supported by National Natural Science Founda-tion of China under the contract No. 51376125.

References

[1] http://doe2.com/download/DOE-21E/DOE-2BasicsVersion21E.pdf.[2] EnergyPlus documentation. http://apps1.eere.energy.gov/buildings/

energyplus.[3] TRNSYS17. http://sel.me.wisc.edu/trnsys/features/features.html.[4] Wang SW. Dynamic simulation of building VAV air conditioning system and

evaluation of EMCS online control strategies. Build Environ 1999;34:681e705.[5] Fan B, Jin XQ, Du ZM. Optimal control strategies for multi-chiller system based

on probability density distribution of cooling load ratio. Energy Build2011;43(10):2813e21.

[6] Perez-Lombarda Luis, Ortizb Jose, Maestrec Ismael R, Coronela Juan F. Con-structing HVAC energy efficiency indicators. Energy Build 2012;47:619e29.

[7] YuanS,PerezR.Multiple-zoneventilationand temperaturecontrolofa singleductVAV system using model predictive strategy. Energy Build 2006;38:1248e61.

[8] Chao CYH, Hu JS. Development of a dual-mode demand control ventilationstrategy for indoor air quality control and energy saving. Build Environ2004;39(4):385e97.

Z. Du et al. / Building and Environment 85 (2015) 104e113 113

[9] Aganda AA, Coney JER, Sheppard CGW. Airflow malt distribution and theperformance of a packaged air-conditioning unit evaporator. Appl Therm Eng2000;20:515e28.

[10] Fletcher CAJ, Mayer IF, Eghlimi A. CFD as a building services engineering tool.Int J Archit Sci 2001;2:67e82.

[11] Xu W, Chen Q. A two-layer turbulence model for simulating indoor airflow e

part I: model development. Energy Build 2001;33:613e25.[12] Xu W, Chen Q. A two-layer turbulence model for simulating indoor airflow e

part II: applications. Energy Build 2001;33:627e39.[13] Lam JC, Chan ALS. CFD analysis and energy simulation of a gymnasium. Build

Environ 2001;36:351e8.[14] Zhai Zhaiqiang, Chen Qingyan, Haves Philip, Klems Joseph H. On approaches

to couple energy simulation and computational fluid dynamics programs.Build Environ 2002;37(8e9):857e64.

[15] Bartak M, Beausoleil-Mottison I, Clarke JA. Integrating CFD and buildingsimulation. Build Environ 2002;37(8e9):865e71.

[16] Djunaedy E, Hensen JLM, Loomans MGLC. External coupling between CFD andenergy simulation: implementation and validation. ASHRAE Trans2005;111(1):612e24.

[17] Zhai Zhiqiang, Chen Qingyan. Sensitivity analysis and application guides for in-tegrated building energy and CFD simulation. Energy Build 2006;38(9):1060e8.

[18] Wang SW, Jin XQ. Model based optimal control of VAV air conditioning sys-tem using genetic algorithm. Build Environ 2000;35:471e87.

[19] Engdahl F, Dennis J. Optimal supply air temperature with respect to energyuse in a variable air volume system. Energy Build 2004;36:205e18.

[20] Ning Min, Zaheeruddin M. Neuro-optimal operation of a variable air volumeHVAC&R system. Appl Therm Eng 2010;30(5):385e99.

[21] Huang GS, Wang SW, Xu XH. A robust model predictive control strategy forimproving the control performance of air-conditioning systems. EnergyConvers Manag 2009;50(10):2650e8.

[22] Fan Yunqing, Ito Kazuhide. Energy consumption analysis intended for realoffice space with energy recovery ventilator by integrating BES and CFD ap-proaches. Build Environ 2012;52:57e67.

[23] Sun ZW, Wang SW. A CFD-based test method for control of indoor environ-ment and space ventilation. Build Environ 2010;45:1441e7.

[24] Zhang Weirong, Hiyama Kyosuke, Kato Shinsuke, Ishida Yoshihiro. Buildingenergy simulation considering spatial temperature distribution for nonuni-form indoor environment. Build Environ 2013;63:89e96.

[25] Chen Q. Comparison of different keεmodels for indoor air flow computations.Numer Heat Transf 1995;28:353e69.

[26] FLUENT user's guide, version 6.2. Fluent Inc.; 2005.[27] Clarke JA. Energy simulation in building design. Bristol: Adam Hilger; 1985.[28] Jin XQ. Study on simulation of VAV air-conditioning system and online

optimal control [D]. PhD thesis. Shanghai Jiao Tong University; 1999.[29] Jin XQ, Xia J, Wang SW. Simulation of multi-zone VAV system and local DDC

controllers. J Refrig 1999;1:31e5.[30] Wang SW, Jin XQ. CO2-based occupancy detection for on-line outdoor air flow

control. Indoor Built Environ 1998;7(3):165e81.[31] Jin XQ, Ren HG, Xiao XK. Prediction-based online optimal control of outdoor

air of multi-zone VAV air conditioning systems. Energy Build 2005;37(9):939e44.

[32] Jin XQ, Du ZM. Energy evaluation of optimal control strategies for centralVWV chiller systems. Appl Therm Eng 2007;27(5e6):934e41.

[33] Yang XB, Jin XQ, Du ZM. Evaluation of four control strategies for building VAVair-conditioning system. Energy Build 2011;43(2e3):414e22.

[34] Du ZM, Jin XQ. Detection and diagnosis for sensor fault in HVAC systems.Energy Convers Manag 2007;48(3):693e702.

[35] Du ZM, Jin XQ. Fault tolerant control of outdoor air and AHU supply airtemperature in VAV air conditioning systems using PCA method. Appl ThermEng 2006;26:1226e37.