ARTICLE IN PRESS Building and Environment 41 (2006) 184–194 www.elsevier.com/locate/buildenv Evaluation of simplified models for predicting CO2 concentrations in small commercial buildings Thomas M. Lawrencea,, James E. Braunb a Department of Biological and Agricultural Engineering, Driftmier Engineering Center, University of Georgia, Athens, GA 30602, USA b School of Mechanical Engineering, Purdue University, West Lafayette, IN 47907, USA Received 3 April 2004; accepted 4 January 2005 Abstract Evaluation of a building for application of demand-controlled ventilation (DCV) typically involves the use of computer simulations to predict energy use/costs for both fixed ventilation and ventilation adjusted to maintain fixed CO2 levels within the space. The simulation tools incorporate models for predicting CO2 concentrations in response to internal sources (people), infiltration/exfiltration, and ventilation. This paper presents a detailed evaluation of different modeling approaches for predicting levels of CO2 in occupied spaces for small, single-zone commercial buildings employing packaged air-conditioning equipment. Twozone and three-zone transient models were compared with a quasi-static equilibrium model applied to three distinctly different building types. Baseline data were derived from computational fluid dynamic models that were developed for field sites. A complete building system simulation model was then used to compare the impact of the different modeling approaches on the predicted energy cost savings associated with application of DCV in each building type. The use of a transient CO2 model did not have a significant impact on model prediction accuracy and energy cost savings predictions as compared with the quasi-static model. The difference in predicted annual energy costs between the various CO2 modeling types were small and less than might result from errors introduced by factors such as CO2 sensor uncertainty. Therefore, the use of an equilibrium model is sufficient for use in evaluating DCV for small commercial buildings. r 2005 Elsevier Ltd. All rights reserved. Keywords: Demand control ventilation; Energy simulation; Carbon dioxide; Modeling; CFD 1. Introduction Demand-controlled ventilation (DCV) is a method used to minimize the energy penalty associated with providing appropriate ventilation for removing odors and contaminants within a room. With DCV, the amount of ventilation air is adjusted according to the occupancy level. Typically, CO2 is used as a passive tracer gas to determine human occupancy in the space, as originally proposed by Kusuda [1]. The best potential applications for DCV are rooms with highly Corresponding author. Tel.: 1 706 5424322; fax: +1 706 5428806. E-mail address: [email protected] (T.M. Lawrence). 0360-1323/$ - see front matter r 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.buildenv.2005.01.003 variable occupancy, such as restaurants, stores, or auditoria [2]. Published studies going back to the 1970s (for example, [3]) have indicated the potential for energy savings with the application of ventilation control based on occupancy or CO2 levels. Emmerich and Persily [4,5] provided a thorough review of the published studies, summarizing the application of DCV in approximately 20 field studies, plus other published papers and standards on simulation studies and sensor locations. Alalawi and Krarti [6] presented a laboratory study comparing the effect of different feedback control strategies on the CO2 levels and HVAC equipment energy consumption. Schell and Smith [7] outlined the various control system decisions needed to retrofit ARTICLE IN PRESS T.M. Lawrence, J.E. Braun / Building and Environment 41 (2006) 184–194 Nomenclature mix C C_ p Np V V_ Zv oa occ ocst r s st v-st vent z concentration (ppmv) CO2 generation rate per person (l/min) number of persons in building zone volume (m3) volumetric flow rate (m3/s) ventilation effectiveness Subscripts inf 185 mixture flow between ventilation and occupied zone outdoor air occupied zone flow between occupied and storage zones return supply storage zone flow between ventilation and storage zones ventilation zone zone (occupied zone) infiltration CO2-based DCV in larger scale buildings and presented a case study that discussed the detailed modifications needed for a retrofit at one office building. Their discussion was based in part on earlier work [8] which described the use of the upper level CO2 set point in a proportional control strategy as an anchor point for evaluating zone ventilation rates. This work also outlined a method to estimate actual ventilation for an occupied space based on CO2 levels and known occupancy. Most of the published studies and proposed analysis methods have assumed perfect mixing in the room under consideration, for example [9–14]. The perfect mixing assumption simplifies the modeling of room CO2 concentrations, but at the expense of accuracy. Knoespel et al. [15] developed a pollutant transport model for analyzing the rate of change of contaminant concentrations in a building with multiple zones. O’Neill and Crawford [16] used the same model format to develop an inverse model for determining interzonal airflows and ventilation effectiveness from experimental data. Federspiel [13,14] used a similar model with the assumption of perfect room mixing and developed a method for recursively estimating the source strength when some of the transport parameters are unknown. Persily [17,18] noted that the assumption of equilibrium conditions for estimating ventilation rates using mass balance equations may not be valid. For example, one particular study was described where air ventilation rates would be overestimated by a factor of two if equilibrium conditions were assumed to exist when in fact they did not. Under constant occupancy conditions, the time needed for a building or room to reach equilibrium is a function of the air exchange rate. A time period with stable occupancy and ventilation rate equal to three times the room time constant is required for room CO2 levels to reach 95% of their steady-state value [19]. Thus, in a room with a low air exchange rate, it may take up to 12 h of constant occupancy for equilibrium conditions to be reached [20]. Such Building Construction Data HVAC Equip’t Data Occupancy Information Weather Data Building System Simulation using Component Models DCV / Economizer Controller Building Thermal Load HVAC Equipment Performance Zone CO2 Model Output: Predicted Energy Use by HVAC Equipment, CO2 Levels Fig. 1. General information flow diagram for DCV analysis modeling. a condition would exist for offices set to older ventilation standards of around 170 l/min per person, or 0.25 air changes per hour in the building measured. Persily also specified a criterion for a building to be considered in equilibrium [18]. The evaluation of a site for DCV retrofit should ideally involve the use of a simulation model that estimates energy savings as compared with fixed ventilation rates. Fig. 1 depicts the simulation process, inputs, and models needed to perform this evaluation. The goal of the work described in this paper was to identify an appropriate model for predicting space CO2 concentrations to be used in analyzing DCV retrofits for small commercial buildings having packaged HVAC equipment. The existing approaches typically use quasistatic models and no previous study has evaluated the effect of modeling approach on energy cost savings predictions. ARTICLE IN PRESS 186 T.M. Lawrence, J.E. Braun / Building and Environment 41 (2006) 184–194 2. Methodology The process of identifying appropriate modeling approaches was carried out in several steps. Different field sites were identified that have a range of different occupancy patterns. Computational fluid dynamics (CFD) models were prepared for the sites in order to provide detailed benchmark data for evaluating the ability of simplified models to predict transient variations in space CO2 levels under realistic operating conditions. Transient and quasi-static models of the CO2 distribution were developed from the CFD results for each site. These models were then incorporated into a system simulation that allowed comparison of the impact of the different CO2 model types on predicted cost savings for DCV. Fig. 2. School site CFD physical model and layout. 2.1. Test sites Three field test sites were selected for study. These sites were instrumented and monitored as part of a larger research project investigating potential energy conservation technologies in commercial buildings, and are described in more detail in Braun et al. [21]. The sites selected include modular schoolrooms, children’s play areas located in fast food restaurants and retail stores. The buildings are all single rooms (i.e., no internal walls partition the rooms) and are conditioned by packaged HVAC equipment located on the rooftop or building sidewall. They range in size from fairly small, approximately 75 m2 (800 ft2) up to 835 m2 (9000 ft2) of floor area. The modular school site is a portable trailer type building with floor dimensions of 12.2 m 6.1 m (40 20 ft) and a ceiling height of 2.75 m (9 ft). The room is normally occupied by 32 students plus the teacher. The room layout provides desk spaces for the students lined up in several rows within the room, but can vary depending on the teacher’s personal preferences. Space conditioning is provided by a sidewall-mounted unit that delivers 0.52 m3/s (1100 cfm) of supply air. Conditioned air is ducted to two 0.6 m (2400 ) supply air diffusers located in the drop ceiling in line with the heat pump. A 0.6 m 0.8 m (2400 3000 ) grill opening in the wall aligned with the back of the HVAC unit allows return air back to the system for reconditioning. Additional items such as storage shelves and a file cabinet are included. Fig. 2 shows a cutaway view of the physical model used in creating the CFD model for the modular schoolroom site. The restaurant play area site consists of a more complicated room layout containing fixed seating areas and large play equipment located in the center of the room. The site has the basic room dimensions of 15.2 m 7.6 m (50 ft 25 ft) with a ceiling height of Fig. 3. Restaurant site CFD physical model and layout. 5.5 m (18 ft). The play structure contains two levels with a floor footprint of 4.6 m 7.6 m (15ft 25 ft). The specific site modeled contains a 1.2 m 2.4 m 3 m (4ft 8 ft with a height of 10 ft) entrance foyer section from the main dining room area, as shown in the physical model layout in Fig. 3. A maximum design occupancy of 80 people is assumed. Conditioned supply air (2.265 m3/s or 4800 cfm) is provided by three directional diffusers in the upper sidewall and a single return air grill in the upper corner. The room also has ceiling fans installed which are occasionally run. Provisions for the ceiling fans are included in the CFD model, but they are assumed not running in this study for simplicity. The retail store site is a rectangular floor plan with basic dimensions of 30.5 m 27.5 m (100 90 ft) and a ceiling height of 4.25 m (14 ft). The major room features are the product display shelves located on a regular pattern throughout the room. Four separate packaged rooftop units, each providing 1.7 m3/s (3600 cfm), are at regularly spaced locations. Each unit supplies air via four diffusers with a central return air grill, as seen in Fig. 4. For this study, a normal full occupancy of 90 people was assumed. ARTICLE IN PRESS T.M. Lawrence, J.E. Braun / Building and Environment 41 (2006) 184–194 III. IV. Fig. 4. Retail store site CFD physical model and layout. V. 2.2. CFD models CFD models were built representing the three field test site types to generate benchmark data for evaluating the CO2 modeling approaches. The goal was not to have models that provided exact predictions for the specific field sites, but rather to have models that gave representative predictions and allowed investigation of the primary factors that affect internal CO2 concentrations. Each site has specific configuration issues that can affect the overall airflow and/or occupancy distribution. For example, the restaurant play area has seating located just under the play equipment. This region is both likely to be occupied, and hence have CO2 generated within, and is blocked from much of the direct interaction with the HVAC supply airflow. Careful attention was given to developing reasonable physical representations for each site. Several key features were common to the CFD models and are described below. I. CO2 sources were located in areas expected to be occupied: for example, the seating areas in both the restaurant play areas and the modular schoolrooms. The retail store represents a different situation where the occupants are more mobile and the CO2 sources were evenly distributed. II. CO2 source generation was represented by a low velocity, low volumetric air–CO2 mixture flow from the top or side surface of solid blocks. For the school and restaurant sites, the blocks are 0.2 m 0.2 m square and 1 m high. The retail store site model used an air–CO2 mixture flow from the sides of several of the product display shelves. The solid blocks represent occupied seating areas, for the restaurant and school sites, or the product shelves for the retail store configuration. These blocks also develop a more representative flow field in the zones since these represent areas of generally VI. 187 no or very low airflow due to blockage. The volumetric flow from the blocks introduces only a minor error to the overall flow field and air balance in the CFD simulation results, with the mixture flow being between 1.4% and 3% of the total HVAC supply flow for all building types modeled. The CO2 source generation rate was varied by adjusting the relative concentration of CO2 in the air–CO2 mixture flowing from the source blocks. A source generation rate of 0.31 l/min per person was assumed. The directional character of the HVAC supply airflow introduced into the zone was modeled as close as possible based on the outlet diffuser design at each site. The total supply airflow to the zone was taken from the manufacturer’s rating information for the HVAC equipment and from field site evaluations. Infiltration flow for each site was assumed to enter the occupied zone via the door openings at the specified volumetric flow rate. The basic room physical descriptions were transformed into models using a commercially available mesh generation package and a commercial CFD simulation program. Cutaway views of the physical models created for each site are given in Figs. 2–4 for the school, restaurant and retail store sites, respectively. Features to note in the school physical model shown in Fig. 2 are the circular diffusers in the ceiling, the student desk locations and the return air grill in the right-hand sidewall. The restaurant play area model shown in Fig. 3 contains the table seating areas, two directional supply air diffusers in the upper center region, a single return grill in the upper left corner and the play equipment floor surface located approximately 2.5 m above the floor in the room center. In Fig. 4, the 13 product display shelves make up the bulk of the retail store floor space. Also shown are the locations of the four supply outlets and central return grill for the four HVAC rooftop units. The schoolroom CFD model was meshed with the edge nodes, representing the room volume, the student desks, file cabinet and storage desk, on a spacing of 0.2 m. This mesh size criterion matched the size of the CO2 mixture flow blocks. The inlet and outlet airflow areas were meshed to a finer spacing to get better resolution of the flow. The doors and return grill had edges meshed to 0.15 m spacing. The supply air openings were meshed with a total of 12 nodes around the circumference of the circles, for a node spacing of approximately 0.15 m also. The supply air registers are designed with louvers to provide air in a mostly radial direction so as not to direct airflow down on the students below. Although no hard data were available to indicate the resulting flow angle from the supply vent, ARTICLE IN PRESS 188 T.M. Lawrence, J.E. Braun / Building and Environment 41 (2006) 184–194 based on the louver construction the air was assumed to be leaving the vent at a 301 angle from the horizontal. The room volume was meshed using a hybrid tetrahedron scheme, as was done for all the site models. The final meshed room had a total of 159,477 mesh volumes. The impact of mesh size was studied and finer mesh volumes had minimal impact on the results. A comparison of the effect of using a courser grid based on 0.4 m spacing indicated minimal impact (o3% difference) on the predicted return CO2 levels when tested during the first occupancy transient of the school base case. Although the majority of the room airflow for each site was expected to be nearly laminar flow, in the regions near the supply air vents and the return grill turbulent flow would be expected. Therefore, the flow field was determined using a viscous flow solver with the k-epsilon two-equation model. For the restaurant play area model, a total of 20 blocks for low volumetric air–CO2 mixture flow were located in the seating areas and within the play equipment space. The supply air flow was modeled to include the directional characteristic that matched the vane pattern on each diffuser outlet. A similar mesh size philosophy was followed as with the modular schoolroom in that a standard mesh node spacing of 0.2 m was applied to all surface, and a finer grid of 0.15 m applied to the air inlet and outlet points. The final meshed room had a total of 214,315 volume cells. The retail store model is a simpler layout and therefore a somewhat courser grid spacing could be employed. This included a standard node spacing of 0.5 m applied on all wall and block surfaces and a finer node spacing of 0.25 m was used for all flow (supply and return) surfaces. The final meshed room had a total of 192,926 volume cells. The three field sites have very different occupancy, ventilation requirements and air flow patterns. Compared to the other sites the school site has a relatively small ventilation time constant with respect to the length of time for each different occupancy period. The restaurant and retail store sites are larger in volume and take longer to reach steady-state conditions. The restaurant site in particular has a complicated internal air flow pattern, with some seating areas located under the children’s play areas which are effectively blocked from good airflow distribution. For each site, a base case was defined to represent a ‘‘typical’’ operating day. Changes in occupancy and infiltration flow conditions were made hourly or as necessary for the two commercial sites, while the basic class schedule for the school site was used to define its occupancy. Table 1 summarizes the occupancy levels and infiltration flow rates used in the base case simulation runs. The occupancy loading listed under the ‘‘% Occupied’’ column represents the occupancy expressed as percentage of the listed design occupancy Table 1 CFD simulation base case test plan Building: school Design occupancy ¼ 32 Start time End time % Occupied Infiltration (l/min) Status 8:30 10:50 11:05 12:30 13:15 14:20 14:30 10:50 11:05 12:30 13:15 14:20 14:30 15:00 100% 10% 100% 0% 100% 10% 100% 1416 8496 1416 2832 1416 8496 1416 Class Recess Class Lunch Class Recess Class Building: restaurant 9:00 11:00 12:00 13:00 14:00 15:00 17:00 19:00 20:00 21:00 11:00 12:00 13:00 14:00 15:00 17:00 19:00 20:00 21:00 22:00 5% 25% 75% 100% 75% 50% 60% 70% 50% 15% 2832 2832 4248 4248 4248 4248 4248 4248 4248 2832 Design occupancy ¼ 80 Morning Initial Lunch Lunch Lunch Intermittent Dinner Dinner Intermittent Cleanup Building: retail 8:00 12:00 13:00 17:00 19:00 12:00 13:00 17:00 19:00 22:00 10% 20% 10% 25% 10% 5664 5664 5664 5664 5664 Design occupancy ¼ 90 Morning Noon shoppers Afternoon Evening rush Evening ARTICLE IN PRESS T.M. Lawrence, J.E. Braun / Building and Environment 41 (2006) 184–194 for each site. Changes in occupancy were implemented as a step function at the times noted in Table 1. Variations on the base case model conditions were also run and evaluated. These included variations in the infiltration flow rates and patterns, changing the occupancy patterns on a more frequent basis than hourly (such as every 15 min), and changing the source locations at the restaurant play area site by turning off the sources located in the play area. Although several different variations to the base case conditions were run, these were found to have no impact on the overall conclusions discussed in this paper. Therefore, only the results from CO2 modeling investigations for the base case are presented here, since they provide sufficient information to evaluate the modeling types. The use of hourly step changes in occupancy is conservative in terms of evaluation of the simplified models and should exaggerate the importance of considering transient effects. Only the school site experiences large step changes in occupancy similar to those given in Table 1. The occupancies within the restaurant and retail store sites change more continuously. All models were run to simulate a complete daily occupancy period. Before beginning the simulated day, the CFD models were first run to obtain a steady-state flow field solution using the flow and turbulence models. The transient simulations used an implicit first-order solver and the species transport solution used a full multi-component diffusion model. The occupancy and flow schedules listed in Table 1 were then implemented using a time step of 1 min. The impact of time step on model predictions was considered and a smaller step had a relatively minor effect on results. The solution convergence criterion was when the residual in the overall mass balance being less than 5 108. 189 on the stratification model originally defined by Janssen et al. [22] and included in Appendix F of the ASHRAE Standard 62-2001 [23], with the addition of a term for infiltration. The ventilation effectiveness is a measure of how well the supply airflow mixes with the occupied zone for removal of CO or other pollutants and is 2 defined as Zv ¼ Cr Cs , Cz Cs (2) where Cr is the CO2 concentration in the return air. A two-zone transient model which relies on interzonal airflow is shown in Fig. 5. This model is similar to the quasi-static stratification model, with the difference being that two distinct volumes for the occupied and ventilation zones and an interzonal airflow term are used instead of the ventilation effectiveness term. An alternative three-zone model, which assumes that a portion of the room is partially isolated from the main breathing and ventilation zones, is illustrated in Fig. 6. The three-zone model may be better suited to situations like the restaurant play area or modular school room sites in that there are regions within the rooms at these sites that will have limited interaction with the ventilation air paths (such as under and between the desks or under the play area equipment) or are not normally occupied. The corresponding coupled equations for each model are given in Eqs. (3)–(7). where Return Supply Both ventilation zone and occupied zone are considered fully mixed. Ventilation Zone 2.3. Simplified CO2 models The following model formats and variations were evaluated as compared with transient CO2 predictions from CFD models: Occupied Zone Fig. 5. Two-zone room model with interzonal airflow. (a) a quasi-static (equilibrium) model; (b) a two-zone transient model with interzonal airflow; (c) a three-zone transient model with interzonal airflow. where Supply Return A generalized quasi-static model for predicting zone CO2 concentrations can be represented as follows: (1) Zv V_ s ðC z C s Þ ¼ N p C_ p þ V_ inf ðC oa C z Þ, where V_ is volumetric flow rate, C is CO2 concentration, C_ p is CO2 generation rate per person, Np is number of people in the zone, ZV is ventilation effectiveness, while the subscripts inf, oa, s, and z refer to conditions associated with air from the infiltration stream, outdoor ambient, supply to zone, and zone. The model is based Both ventilation zone and occupied zone are considered fully mixed. Storage Zone Occupied Zone Fig. 6. Three-zone room model with interzonal airflow. ARTICLE IN PRESS T.M. Lawrence, J.E. Braun / Building and Environment 41 (2006) 184–194 190 Two-zone model: V vent dC r ¼ V_ mix ðC z C r Þ þ V_ s ðC s C r Þ þ V_ inf ðC z C r Þ, dt (3) V occ dC z ¼ V_ mix ðC r C z Þ þ N p C_ p þ V_ inf ðC oa C z Þ. dt (4) Three-zone model: V vent V occ V st dC r ¼ V_ mix ðC z C r Þ þ V_ s ðC s C r Þ þ V_ inf ðC z C r Þ dt ð5Þ þ V_ v-st ðC st C r Þ, dC z ¼ V_ mix ðC r C z Þ þ N p C_ p þ V_ inf ðC oa C z Þ dt þ V_ oc-st ðC st C z Þ, ð6Þ dC st ¼ V_ oc-st ðC z C st Þ þ V_ v-st ðC r C st Þ. dt on the overall average room concentration. The cost function was the root-mean-square error (RMSE) of the predicted versus actual CFD model results of the bulk room average CO2 concentration. During the model training process, the room concentration was computed based on a volume-weighted average of the two or three zones. The RMSE cost function was computed using simulation results from the entire occupied day as a training data set. The parameter estimation process used a commercially available optimization routine to obtain estimates of the learned parameters. The optimization procedure is based on the gradient of change in the cost function with respect to the parameters of interest. A Nelder–Mead simplex (direct search) method determines the step taken along the steepest gradient. The only constraints placed on the flow and volume parameters during the estimation process were that they remain positive values. (7) In these equations, the volume subscripts vent, occ and st refer to the ventilation, occupied and storage zones, respectively. For the flow rates, the subscripts mix, vst, and ocst refer to flows between the ventilation and mixing zones or between the occupied and storage or ventilation and storage volumes. Note that in all models considered, the infiltration rate could be negative and represent exfiltration. In the application of the different CO2 models, the supply concentration is solved for based on a mass balance in the mixing of room and outdoor air. The room, zone and storage (in the three-zone model) concentration are obtained through numerical integration of Eqs. (3)–(7). The quasi-static model in Eq. (1) is based on an empirical room ventilation effectiveness. Separate CFD test cases were run with constant CO2 source and flow rates until steady-state conditions were reached. The steady-state return and bulk room CO2 concentrations were then computed and the ventilation effectiveness determined using Eq. (2). A parameter estimation process was applied to learn unknown values in the two-zone and three-zone models. In the two-zone model, the two zone volumes (V vent and V occ ) and the inter-zonal flow (V_ mix ) are unknown, with the other values either known from the CFD model or calculated as part of the equation solution and parameter estimation process. For the three-zone model, the additional storage zone volume (V st ) and inter-zonal airflows associated with the storage zone (V_ ocst and V_ vst ) are also learned parameters. The building system simulation program used for DCV evaluations bases the amount of outdoor ventilation air required on the bulk room CO2 concentration. Therefore, training of the zone CO2 models was based 3. CO2 model training results Tables 2–4 summarize model training results for the various formats studied and applied to the base case at each site. Based on the RMSE, the trained models all performed well. This conclusion is also supported by the plots of Figs. 7–9 for the school, restaurant and retail store sites, respectively. These plots show the ability of each model form to predict CO2 concentration transients for the base case simulated day as compared with the CFD predictions. For simplicity, the CFD runs were set up assuming 100% outdoor air for the supply flow ðC s ¼ C oa Þ and the values given in Figs. 7–9 are differences between the average room and supply concentrations. Table 2 CO2 model evaluation for school site base case SCHOOL Volume: 206 m3 Model format Values Quasi-static Room CO2 RMSE (ppmv) 48.3 Two-zone V v ðm3 Þ V occ ðm3 Þ V_ mix ðm3 =sÞ 75.7 35.7 0.571 Three-zone V v ðm3 Þ V occ ðm3 Þ V st ðm3 Þ V_ mix ðm3 =sÞ V_ oc-st ðm3 =sÞ V_ v-st ðm3 =sÞ 75.8 24.7 10.8 0.374 0.248 0.072 8.2 8.1 ARTICLE IN PRESS Table 3 CO2 model evaluation for restaurant site base case Restaurant Volume: 608 m Model format Values 3 Quasi-static Room CO2 RMSE (ppmv) 7.8 Two-zone V v ðm3 Þ V occ ðm3 Þ V_ mix ðm3 =sÞ 111.6 353.5 1.498 Three-zone V v ðm3 Þ V occ ðm3 Þ V st ðm3 Þ V_ mix ðm3 =sÞ V_ oc-st ðm3 =sÞ V_ v-st ðm3 =sÞ 124.4 264.3 7.5 0.485 0.104 0.039 4.1 Bulk Room Concentration Above Supply (ppmv) T.M. Lawrence, J.E. Braun / Building and Environment 41 (2006) 184–194 450 400 350 Quasi-static 250 Values Two-Zone and Three-Zone Models Give Nearly Identical Results 150 100 50 0 60 120 180 240 300 360 Minutes from start of day Fig. 7. Predicted room CO2 levels compared to CFD: school base case. Room CO2 RMSE (ppmv) Quasi-static 5.0 Two-zone V v ðm3 Þ V occ ðm3 Þ V_ mix ðm3 =sÞ 2.5 408.7 1952.7 6.516 Three-zone V v ðm3 Þ V occ ðm3 Þ V st ðm3 Þ V_ mix ðm3 =sÞ V_ oc-st ðm3 =sÞ V_ v-st ðm3 =sÞ 279.0 470.7 1008.7 2.035 0.692 0.011 Bulk Room Concentration Above Supply (ppmv) Model format Two-Zone 200 0 Table 4 CO2 Model evaluation for retail site base case Volume: 3360 m3 CFD 300 3.9 Retail store 191 200 150 100 CFD Quasi-Steady State 50 Two-Zone Three-Zone 0 0 60 120 180 240 300 360 420 480 540 600 660 720 780 Minutes from start of day Fig. 8. Predicted room CO2 levels compared to CFD: restaurant base case. Certain anomalies appear in the model training results shown in Figs. 8 and 9 for the restaurant and retail store sites. For some of the periods, the trained CO2 models did not match as well as the others, for example, during the middle of the day when the occupancy decreased after the noon-time rush. This is because there was not sufficient time for the CO2 levels to stabilize between the occupancy step changes. Only minor differences were seen between the twozone and three-zone transient models, and they both resulted in good predictions of the transient room concentration. Even with the quasi-static model, the maximum RMSE computed using 1-min time step data, 48 ppmv at the school site, was less than the manufacturer’s rated accuracy of 50 ppm for the sensors used at the field test sites. The RMSE values listed in Tables 2–4 Bulk Room Concentration AboveSupply (ppmv) 1.7 70 60 CFD Room Quasi-Static Two-Zone Three-Zone 50 40 30 20 10 0 0 60 120 180 240 300 360 420 480 540 600 660 720 780 840 Minutes from start of day Fig. 9. Predicted room CO2 levels compared to CFD: retail store base case. are based on the 1-min time scales. If the errors in the quasi-static predictions are based on 5-min and 1-h averages of the data, then the maximum RMSE values for the school site decrease to 41 and 15 ppmv, respectively. The CFD simulations used step changes in the occupancy, while actual field sites generally would ARTICLE IN PRESS 192 T.M. Lawrence, J.E. Braun / Building and Environment 41 (2006) 184–194 have more gradual changes. Therefore, although the transient simulations do provide somewhat better model predictions, the differences are exaggerated by the assumption of large step changes in occupancy and the quasi-static model may be adequate. This was evaluated formally by comparing predictions of DCV cost savings using a building system simulation model. 4. Evaluation of CO2 model formats using a building system simulation Even though different CO2 modeling approaches may give different CO2 levels, especially during periods of rapid or significant changes in occupancy, the impact of these differences on the overall cost savings estimates when evaluating DCV control may be relatively small. If this were the case, then the simplest modeling approach would be the most appropriate choice. The practical significance of using a transient CO2 model or the more simple quasi-equilibrium approach was studied by comparing results of yearly simulations. The program chosen for this analysis was based on the model developed by Brandemuehl and Braun [2] for evaluation of DCV cost savings. The program was modified to use the quasi-equilibrium and transient CO2 model formats. The simulation model performs calculations with an hourly timestep and incorporates separate models for transient heat gains from the building envelop, mass and energy balances on the air within the zone and distribution system, and HVAC equipment energy requirements. The overall input and output information flow is outlined in Fig. 1, using hourly input of the local weather and building occupancy. Heat gains/losses from external walls, roofs and floors are modeled using onedimensional transfer functions. The cooling and heating equipment are modeled using routines from the ASHRAE toolkit [24]. The program compares DCV with a base case for fixed ventilation associated with satisfying ASHRAE Standard 62-2001. Energy costs are computed using a three-tiered utility rate structure consisting of on-peak, mid-peak and off-peak pricing. For each site, the local utility rate structure information in place in late 2002 was obtained and used for the analyses in this study. The simulation program was validated against other commercial and public domain building simulation models, as documented in Mercer [25]. The original simulation program employed a quasistatic model for room CO2 concentrations and mass and energy balance calculations were performed on an hourly basis. For the transient CO2 models, a smaller internal timestep was used. Calculations of room and return CO2 concentrations and ventilation adjustments for DCV were performed using 1-min timesteps. For each hour, average outdoor ventilation air requirements were computed and used along with building envelope heat gains determined with 1-h time steps in order to estimate total hourly equipment loads and energy requirements. For the comparisons of the different modeling approaches, a bulk room CO2 concentration of 950 ppmv was used as the setpoint for DCV. This value resulted in a return air concentration Table 5 Key building simulation model parameters for DCV analysis Parameter Building thermal Window net transmissivity Window resistance Floor overall resistance Floor overall capacitance Wall overall resistance Wall overall capacitance Ceiling overall resistance Ceiling overall capacitance a; exterior wall a; roof Interior mass Equipment and ventilation Ventilation effectiveness Supply fan power Make up air Infiltration air Sensible gain lights & equip’t Energy efficiency ratio—air conditioning (EERAC) Furnace efficiency or heat pump COP Supply air flow Units M2 K/W M2 K/W KJ/m2 K m2 K/W KJ/m2 K m2 K/W KJ/m2 K kg/m2 W/(m3/min) (l/min)/m2 (l/min)/m2 W/m2 M3/min School Restaurant Retail store 0.52 0.27 2.65 39.4 1.25 4.4 2.84 12.1 0.7 0.6 122 0.63 0.27 0.19 238.3 1.26 7.5 3.19 12.1 0.7 0.7 73 0.39 0.27 0.41 24.8 1.84 83.2 3.80 19.8 0.7 0.7 366 0.79 12.4 15 30 27 9.5 3 31.2 0.57 10.6 30 60 9.7 9.5 75% 136 0.6 10.6 15 10 27 8.9 3 408 ARTICLE IN PRESS T.M. Lawrence, J.E. Braun / Building and Environment 41 (2006) 184–194 of approximately 800 ppmv, which was the setpoint for DCV studies at the field test sites. The evaluation of a site for potential application of DCV requires the determination of the ventilation flow rate with both DCV control active and with the existing control strategy. For CO2 modeling evaluations in this study, the reference baseline assumed a fixed outdoor ventilation level as prescribed in ASHRAE Standard 622001 for each specific site. The reference baseline for each site also assumed enthalpy economizer cooling control. Site-specific values for the building construction, plus the window area, shading and orientation, were used for each building type studied, and these are summarized in Table 5. Table 5 also includes sitespecific HVAC equipment parameters. The system simulation tool was used to study the sensitivity of DCV energy cost savings predictions to the three different CO2 modeling approaches. The simulations evaluated the difference in HVAC heating and cooling energy costs for DCV plus economizer control (DCV On) compared to economizer cooling only (DCV Off). These comparisons were done using estimates of CO2 Generation Rate (liters/min) 10 School (Weekdays Only) Retail Store (Weekday& Weekend) 9 8 7 6 5 4 3 2 1 0 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Hour of Day Fig. 10. Assumed CO2 source generation rate pattern for school and retail store sites. 193 building occupancy (and hence the CO2 source generation rate) that were derived from preliminary investigations at the field sites. Separate average hourly CO2 source generation rates were used during the weekday or weekend periods for the restaurant site. The school site was assumed occupied only during weekdays. For the retail store site, the same occupancy pattern was assumed for both weekday and weekend periods. Fig. 10 shows the variation in assumed occupancy (CO2 generation rate) for the school and retail store sites, while Fig. 11 provides this information for the weekend and weekday periods at the restaurant site. The annual energy cost savings for including DCV plus economizer cooling control compared to the baseline economizer cooling alone ranged from $260 (US) for the school site to over $3100 for the retail store site. Fig. 12 shows a comparison of the annual energy cost savings percentage associated with adding DCV. The resulting differences in predicted energy cost savings between the three different CO2 model types were very small, ranging from 0.3% for the school and retail store to around 1% for the restaurant site. In terms of absolute costs, the net annual energy cost difference between the equilibrium and transient models ranged from a low of $8 for the school site to a high of $70 at the restaurant site. Compared to the total energy cost savings, the differences are only about 3% of the total predicted savings. The difference in total energy cost savings between the equilibrium and transient CO2 models for the retail store site, expressed as percentage of the total savings, is only about 1.5%. The differences in cost savings for DCV due to the choice of CO2 model are less than any differences that would be expected due to other uncertainties, such as the occupancy schedule or errors in the CO2 sensor readings. For example, Fig. 13 shows the estimated energy cost savings for the restaurant site when the bulk room CO2 setpoint was adjusted750 ppmv from the baseline 950 ppmv setpoint. For this site, the difference 30% Restaurant Weekday Restaurant Weekend 20 Restaurant, Sacramento Area Retail Store, RiversideArea % Energy Cost Savings with DCV Source Generation Rate (Liters/min) 25 15 10 5 0 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Hour of Day Fig. 11. Assumed CO2 source generation rate pattern for restaurant site. 25% ModularSchool, Sacramento Area 20% 15% 10% 5% 0% Equilibrium Model Two-zone with control to bulk room Three-zone with control to bulk room Fig. 12. Comparison of energy cost savings for different CO2 model types. ARTICLE IN PRESS T.M. Lawrence, J.E. Braun / Building and Environment 41 (2006) 184–194 194 30% Restaurant building type % Savings with DCV 25% 20% 15% 10% 5% 0% Equilibrium Model, 900 ppm Equilibrium Model, 950 ppm Equilibrium Model, 1000 ppm bulk room setpoint bulk room setpoint bulk room setpoint Fig. 13. Variation in energy cost savings for different CO2 setpoints. in annual energy costs estimated between a 900 ppmv and a 1000 ppmv setpoint was nearly $300, or four times the difference between using the equilibrium and transient CO2 models. 5. Conclusion The use of a transient CO2 model is not necessary for evaluating cost savings associated with DCV for small commercial buildings. The errors are less than those that would occur due to other uncertainties such as CO2 sensor errors. One of the most important factors impacting DCV savings is the occupancy schedule. A companion paper addresses the issue of identifying sitespecific occupancy schedules from field measurements using a quasi-equilibrium model and parameter estimation [26]. Acknowledgments This research was supported in part by the California Energy Commission Pubic Interest Energy Research (PIER) Building Energy Efficiency Program. References [1] Kusuda T. Control of ventilation to conserve energy while maintaining acceptable indoor air quality. ASHRAE Transactions 1976;82(1). [2] Brandemuehl MJ, Braun JE. Impact of demand-controlled and economizer ventilation strategies on energy use in buildings. ASHRAE Transactions 1999;105(2):39–44. [3] Ogasawara S, Taniguchi H, Sukehira C. Effect of energy conservation by controlled ventilation: case study in a department store. Energy and Buildings 1979;2:3–8. [4] Emmerich SJ, Persily AK. Literature review on CO2 based demand controlled ventilation. ASHRAE Transactions 1997;103(2):229–43. [5] Emmerich SJ, Persily AK. State-of-the-art review of CO2 demand controlled ventilation technology and application. National Institute of Standards Technical Report NISTIR 6729, July 2001. [6] Alalawi MA, Krarti M. Experimental evaluation of CO2-based demand-controlled ventilation strategies. ASHRAE Transactions 2002;108(2). [7] Schell M, Smith D. Assessing CO2 control in retrofits. ASHRAE Journal 2002;44(11):34–43. [8] Schell MB, Turner SC, Shim RO. Application of CO2 based demand controlled ventilation using ASHRAE Standard 62: optimizing energy use and ventilation. ASHRAE Transactions 1998;104(2):1213–25. [9] Drees KH, Wenger JD, Janu G. Ventilation air flow measurement for ASHRAE Standard 62-1989. ASHRAE Journal 1992:40–5. [10] Janu GJ, Wenger JD, Nesler CG. Strategies for outdoor airflow control from a systems perspective. ASHRAE Transactions 1995;101:631–43. [11] Sørensen B. Simulation of a small VAV plant. Proceedings of Indoor Air ’96, vol. 2, 1996. p. 199–204. [12] Ke Y, Mumma SA, Starke D. Simulation results and analysis of eight ventilation control strategies in VAV systems. ASHRAE Transactions 1997;103:381–91. [13] Federspiel CC. Estimating the inputs of gas transport processes in buildings. IEEE Transactions on Control Systems Technology 1997;5(5):480–9. [14] Federspiel CC. Conditions for the input–output relation of perfect mixing processes to be first order with application to building ventilation systems. Transactions of ASME, Journal of Dynamic Systems, Measurement and Control 1998;120:170–6. [15] Knoespel PD, Mitchell JW, Beckman WA. Macroscopic model of indoor air quality and automatic control of ventilation airflow. ASHRAE Transactions 1991;97(2):1020–30. [16] O’Neill PJ, Crawford RR. Identification of flow and volume parameters in multi-zone systems using a single-gas tracer technique. ASHRAE Transactions 1991;97(1):49–54. [17] Persily AK. Ventilation, carbon dioxide and ASHRAE Standard 62-1989. ASHRAE Journal 1993;35(7):40–4. [18] Persily AK. Evaluating building IAQ and ventilation with indoor carbon dioxide. ASHRAE Transactions 1997;103(2):193–204. [19] Seppänen OA, Fisk WJ, Mendell MJ. Associations of ventilation rates and CO2 concentrations with health and other responses in commercial and institutional buildings. Indoor Air 1999;9:226–52. [20] Nabinger SJ, Persily AK, Dols WS. A study of ventilation and carbon dioxide in an office building. ASHRAE Transactions 1994:1264–73. [21] Braun JE, Lawrence TM, Mercer K, Li H. Description of Field Test Sites. Report submitted to Architectural Energy Corporation for the California Energy Commission Building Energy Efficiency Program, Deliverables 2.1.1a, 2.1.1b, and 3.1.1a, December 2001. [22] Janssen JE, Hill TJ, Woods JE, Maldonado EAB. Ventilation for control of indoor air quality: a case study. Environment International 1982;8:487–96. [23] ASHRAE. ANSI/ASHRAE Standard 62-2001: Ventilation for acceptable indoor air quality. Atlanta, GA: American Society of Heating, Refrigerating and Air-Conditioning Engineers; 2001. [24] Brandemuehl MJ, Gabel S, Andresen I. HVAC2 Toolkit: Algorithms and Subroutines for Secondary HVAC System Energy Calculations. ASHRAE, 1791 Tullie Circle, NE, Atlanta, GA 30329, 2000. [25] Mercer K. Modeling and testing strategies for evaluating ventilation load reduction technologies. Master’s Thesis, Purdue University, West Lafayette, IN, 2003. [26] Lawrence TM, Braun JE. A methodology for estimating occupant CO2 source generation rates from measurements in small commercial buildings. Building and Environment, Submitted, 2004.
© Copyright 2026 Paperzz