Calibration of a building energy model using measured data.

BACKGROUND

Model calibration is essential to ensure that the architectural, mechanical, and electrical systems are properly modeled and integrated together for the purpose of estimating the building energy performance. Calibration of the energy model of a large building can be labor intensive even for an experienced modeler. It requires a thorough understanding of the architectural planning and mechanical systems, as well as of the assumptions, default values, mathematical models, and limitations of the energy analysis program.

During the calibration, the inputs for some variables with uncertain values are modified until the difference between the program prediction of energy use and measured data is acceptable. Utility bills are generally used for comparison with the predicted whole-building monthly and annual energy consumption since they contain, in a condensed format, the energy history of the building. This is also because they are available to the energy auditor at the beginning of the project, while monitored data require time and resources. Data monitored at the system or equipment level at smaller time intervals could give, however, more detailed and useful information than the whole-building energy consumption. For instance, Kaplan et al. (1990) calibrated the computer model of a small office building by using monitored energy end use for three months: a peak heating month, a peak cooling month, and a month with swing operation. The graphical representation of metered indoor air temperature was an additional help in understanding the room temperature setpoint schedule of operation and revealed differences from the design thermostat settings.

Kaplan and Canner (1992) have made recommendations for the maximum allowable difference between predicted and monitored data. For instance, the prediction of energy use for interior loads such as lighting, receptacles, or domestic hot water is satisfactory when the difference is within 5% on a monthly basis and 15% on a daily basis. However, the acceptable difference may increase up to 15%-25% (monthly) and 25%-35% (daily) for the simulation of HVAC systems. The annual simulated energy use should be within 10% of collected information, while a difference of less than 25% is acceptable on a seasonal basis.

Reddy (2006) presented a literature review of publications on calibration practices. The review showed that most studies were based on manual, iterative, and user-specific approaches. Several graphical comparative displays are also used to identify the calibration parameters. Other methods include short-term tests, measurements on test cells, energy signature, statistical screening techniques, and analytical models.

Carroll and Hitchcock (1993) proposed a method that automates the tuning of building parameters for matching the predicted energy use to utility data. Sun and Reddy (2006) presented a mathematical foundation to the calibration problem, with the goal of avoiding the "fudging process" of manipulating a large number of variables on a trial-and-error basis. Reddy et al. (2007a) reviewed literature on calibration from other scientific disciplines, such as environmental and structural engineering, and proposed a general methodology for the calibration of energy analysis programs against its utility bills. Reddy et al. (2007b) applied the new proposed methodology to the calibration of computer models of three office buildings, which were developed using the DOE-2 program. The calibration was performed using (1) synthetic data from two reference buildings, and (2) electricity use data from utility bills over 12 months from the third building.

The uncalibrated building energy models are still, however, largely used to estimate the energy savings from proposed energy conservation measures. This practice may lead to differences of 7%-73% when the annual predictions are compared with the measured whole-building electricity use, 25%-87% for chilled-water consumption, and 65%- 98% for hot-water use (Ahmad and Culp 2006). Other researchers concluded that a finely calibrated model does not ensure the accurate estimation of energy savings due to the building retrofit (Kaplan et al. 1990; Corson 1990). Nevertheless, the calibration of a computer model gives more confidence on the quality of work and the accuracy of predictions of the base case model.

A limited amount of information has been published thus far about the calibration of energy models for large buildings developed using the EnergyPlus program. Bellemare et al. (2002) modeled an institutional building with 54 interior zones and related variable-air-volume ( VAV ) systems. They presented the comparison between the predicted and monitored indoor air temperature and supply airflow rate of one classroom and found a similar pattern of variation. The weekly average difference between the room air temperature and airflow rate for the same classroom, as predicted by the EnergyPlus and DOE-2 programs, was 0.24[degrees]C and 9%, respectively. The daily average supply airflow rate of an air-handling unit was predicted within 8% of measured values. Ellis and Torcellini (2005) have simulated a tall building having an overall floor area of 240,000 [m.sup.2] (2,283,340 [ft.sup.2]). Their analysis was mainly focused on height-dependent properties and the use of floor multipliers, while HVAC systems were entered through the purchased air option that is offered by the EnergyPlus program. This approach reduces the computing time since it calculates the cooling and heating loads without taking into account the performance of HVAC equipment. Witte et al. (2001) used BESTEST guidelines to evaluate EnergyPlus for a base case building with mechanical systems. HVAC BEST-EST has been developed by the International Energy Agency and consists of a series of steady-state tests used to evaluate the ability of whole-building simulation programs (Hayer et al. 2001). The tests consist of analytical verifications of a specified mechanical system applied to a simplified near-adiabatic building envelope (Neymark et al. 2001). Results for eight different commonly used simulation programs are also included as a comparison tool for new software (Witte et al. 2001). The test helped to identify errors and documentation deficiencies. Most problems encountered were related to the system cycling mode and humidity. All issues encountered were investigated and fixed in later versions of the program.

OBJECTIVES

In this paper, the calibration of the case study is carried out separately over two periods with different operating conditions: period A, from May 4 to June 20, 2006, when the mechanical cooling system is in operation, and period B, from March 20 to May 3, 2006, which corresponds to the shoulder portions of the spring season, when the cooling coils are not in operation. The simulation of two distinct operation periods has the advantage of avoiding some compensating errors that can occur when the calibration process is performed over one year, with periods of different operating conditions.

The goal of the calibration presented in this paper is the development of a model of a large institutional building that predicts well the supply airflow rate and the supply and return air temperatures, which are both directly measured variables. The whole-building cooling loads, which are calculated from measured data, are also compared with the program predictions.

SHORT DESCRIPTION OF THE CONCORDIA SCIENCES BUILDING

The Concordia Sciences Building is located on the Loyola campus in Montreal and has a total floor area of 32,000 [m.sup.2] (344,445 [ft.sup.2]). The building is divided in three main sectors: A, B, and C (Figure 1).

[FIGURE 1 OMITTED]

According to the design specifications, the building has walls with the overall thermal resistance varying between 2.6 and 3.1 [m.sup.2]*[degrees]C/W (14.8 and 17.6 [ft.sup.2]*[degrees]F*h/Btu) and roofs between 2.8 and 4.2 [m.sup.2]*[degrees]C/W (15.9 and 23.8 [ft.sup.2].[degrees]F*h/Btu). Most walls consist of insulated brick completed with an air space, a vapor barrier, and one or two layers of gypsum board. The roofs are composed of a bitumen membrane, a concrete layer, two types of insulation, a plywood panel, a vapor barrier, and another concrete layer. Two types of glazing are present: (1) double low-e clear with film 6/6 mm (0.24/0.24 in), and (2) double low-e clear 6/13mm (0.24/0.51 in). The glazing accounts for about 32% of the total area of exterior walls (PMA 2003). The fenestration assemblies have aluminium framing with thermal break.

The lighting installed load is between 7 and 10 W/[m.sup.2] (0.7 and 0.9 W/[ft.sup.2]), and the equipment load is between 2 and 10 W/[m.sup.2] (0.2 and 0.9 W/[ft.sup.2]). To reduce the zone loads, motion detectors are installed in all rooms of the Sciences Building. The motion detectors shut off lights after an adjustable delay of no activity. When lights shut off, a signal is also sent to the building automated system to reduce the amount of air sent to the room.

The supply airflow rate for laboratories changes from 10 air changes per hour (ach) during occupied hours to 6 ach while unoccupied. This is further reduced to 3 ach at night. The ventilation is brought back to 10 ach whenever occupants are present (Lemire and Charneux 2005). The minimum supply airflow rate to other room types is 3 ach if the room is located on the perimeter and 1.5 ach if it is an interior zone (PMA 2003).

The VAV system of sector B&C is served by two identical air-handling units (AHU-7 and AHU-8). The design airflow rate of the VAV system is 75.5 [m.sup.3]/s (160,000 cfm), the cooling coil capacity is 1655 kW (5650 mbh) and the heating coil capacity is 2340 kW (7990 mbh). The VAV system of sector A is served by four identical air-handling units (AHU-1 to AHU-4). The design airflow rate of the VAV system is 151.0 [m.sup.3]/s (320,000 cfm). The overall cooling and heating coil capacity for this sector is 3310 kW (11,300 mbh) and 4580 kW (15,600 mbh), respectively. The installed cooling and heating coils capacity includes built-in capacity for future building additions.

The total supply airflow rate of each system is composed of the amount of air required for cooling/heating purposes plus the additional amount of air that must be supplied to laboratories to compensate for the air exhausted by the laboratory hoods. Therefore, a large amount of energy is required to heat and cool the outdoor air introduced into the building. To reduce the energy burden, a run around heat recovery glycol loop is installed between the exhaust airstream and the outdoor airstream. Variable frequency drives are also installed on fans to improve efficiency at part-load operation.

A thermal central plant serves all sectors of the building. During the summer season, the hot-water system, which is used for the re-heating coils in the VAV boxes, operates at 35[degrees]C (95[degrees]F) supply and 29.4[degrees]C (85[degrees]F) return water temperatures. The water temperature is increased to 51.7[degrees]C (125[degrees]F) supply and 29.4[degrees]C (85[degrees]F) return during the winter season. The hot water is also used, through plate heat exchangers, to warm up a 50% glycol solution from 26.7[degrees]C to 48.9[degrees]C (80[degrees]F to 120[degrees]F), which is supplied to the glycol heating coils installed in the air-handling units. Plate heat exchangers are used to recover the heat rejected from chillers and from the exhaust gases of two existing boilers, with 70%-75% efficiency, to preheat the hot water. If the recovered heat flow rate is insufficient to achieve the required water temperature, a tube-and-shell heat exchanger is used to further heat up the water using steam produced by an 85% efficient natural gas boiler having a capacity of 815 kW (2785 mbh). The steam generated in the central plant is also used for the humidifiers installed in the air-handling units. The steam production efficiency is 75%; however, the heat recovery system installed on the two existing boilers gives a total production efficiency of 96%.

Two chillers are installed with a cooling capacity of 3165 kW (900 tons) each, and a nominal coefficient of performance (COP) of 5.76. They operate during the summer and shoulder months to supply chilled water of 5.6[degrees]C/13.3[degrees]C (42[degrees]F/56[degrees]F) to the cooling coils. Two additional chillers are installed in the building to serve the fan-coil units during the winter and part of the shoulder seasons. The combination of energy-efficient measures and operating strategies has led to a 50% reduction in energy consumption compared to the same building that would comply with Model National Energy Code of Canada for Buildings (MNECCB 1997).

DEVELOPMENT OF THE COMPUTER MODEL

This paper uses, as a case study, a building submitted for the Commercial Building Incentive Program (CBIP) to Natural Resources Canada to obtain additional funding for integrated design, with the goal of reducing the annual energy use. For this purpose, the EE4 program, which uses the DOE-2 program as the calculation engine, is used to simulate the building thermal loads and the energy performance of secondary and primary HVAC systems. The estimated annual energy consumption of the proposed design is compared with the performance of a reference building of an equal size that complies with the MNECCB (1997). The EE4 input file provided by the mechanical consultants was helpful to create the initial input file in this study. The DOE-2 input file, which contains the building description, is generated by EE4 and then translated into an "*.idf" input file compatible with the EnergyPlus program. This conversion is achieved using a utility program named DOE2Translator, provided as a pre-process program by EnergyPlus. The translation program provides incomplete design object information and, therefore, additional work and several modifications are required to obtain a working input file for the EnergyPlus program.

The computer model was developed with the EnergyPlus program, Version 1.3.0, released on April 28th, 2006. The complexity of the Concordia Sciences Building has led to several modeling issues related to the determination of space loads. Given the size of the building and its vocation, the model is developed using 97 thermally controlled zones (58 for sector A and 39 for sector B&C), 26 plenums and 1773 surfaces (walls, roofs, partitions, and floors). In order to obtain accurate results, it is recommended to include every surface within a zone. In the case of an interior partition that separates two zones, the vertices of this partition must be defined in both zones. This condition helps in obtaining the same surface area on both sides of the partition, and therefore ensures the heat transfer through the partition complies with the conservation of energy principle.

To ease the entry process for HVAC systems in Energy-Plus, compact HVAC systems were initially used. Compact HVAC objects provide a shorthand way of describing standard HVAC system configurations. They allow for the specification of simple zone thermostats and HVAC systems with automatically generated node names. The models include built-in default data and user input data entry for basic system options. EnergyPlus automatically sets up loops, branches, and node names for the specified objects. Each object can be expanded in future runs to further detail each component (DOE 2007). This approach abbreviates and simplifies the initial modeling.

For simplification, the two identical AHUs that are installed in sector B&C are combined in the model as one unit of equal capacity. The four identical AHUs of sector A are combined in one unit of equal capacity.

The large number of air branches required to simulate the building as one entity exceeded the maximum number supported by EnergyPlus. Consequently, sector B&C and sector A are modeled separately, which adds to difficulties encountered in the simulation of the central plant. The modeling of the primary system itself is also challenging. In the existing central plant, there are several heat recovery systems, heat exchangers, a steam boiler, and related controls that cannot be directly simulated by the EnergyPlus program. They are simulated using equivalent and simpler primary equipment and plant configurations. Consequently, the calibration process and system analysis is limited to the secondary HVAC systems and building air-side thermal loads.

The size and complexity of the building also has a direct impact on the computing time required to perform the simulations. For comparison purposes, two different computers were used for simulations: (1) a laptop with Dell Latitude D600 Pentium M of 1.4 GHz, and 512 MB of RAM, and (2) a desktop with Dell Precision 360 Pentium 4 of 2.8 GHz, and 2.0 GB of RAM. Computation times are presented in Table 1 for the calibration period (spring season) and the annual simulation. In both cases, the simulation was carried out using a one-hour time step.

Table 1. Computing Time

                Sector A            Sector B&C

Computer    Spring     Annual    Spring    Annual
Type

Laptop    3 h16 min  8 h 38 min  31 min  1 h 56 min
Desktop   2 h18 min  7 h 22 min  21 min  1 h 09 min

MONITORED DATA

Information about the as-built and as-operated thermal performance of the Sciences Building is obtained from the Monitoring and Data Acquisition System (MDAS) with the collaboration of the Physical Plant of Concordia University. Data from 49 sensors are collected every 30 minutes. The following data are used in this study: the airflow rate of each supply fan, the supply and return air temperatures for each AHU, and the supply and return temperatures of hot, chilled, and glycol water. The weighted average values of each sector are calculated from data collected at each AHU.

The following conclusions are drawn from the analysis of monitored data on the secondary systems during the specified spring period:

1. For sector B&C, the average supply airflow rate during the occupied period is 30.0 [m.sup.3]/s (63,570 cfm), and is 14.0 [m.sup.3]/s (29,665 cfm) during the unoccupied period. The supply fans work on an average 40% of total capacity during the occupied period, and at 18.5% capacity during the unoccupied period. During the spring season, the maximum load on supply fans is 50% of total capacity.

2. For sector A, the average supply airflow rate during the occupied period is 73.5 [m.sup.3]/s (155,740 cfm) and is 52.0 [m.sup.3]/s (110,180 cfm) during the unoccupied period. During the occupied period, the supply fans work an average 50% of total capacity and, during the unoccupied period, 35% of capacity. During the spring season, the maximum load on supply fans is 60% of total capacity.

3. The supply airflow rate does not significantly vary with the variation of outdoor air temperature; hence, one can conclude that the cooling/heating loads due to heat gains/losses through the building envelope are much smaller than the internal gains.

4. For all sectors, the supply air temperature is maintained constant at around 16[degrees]C (61[degrees]F) when the cooling coils are in operation, while the airflow rate varies depending on the cooling/heating loads and the level of occupancy in the building.

5. From March 20 to May 4, 2006, the cooling coils installed in the air-handling units are not in operation; thus, the supply air temperature is above the supply air setpoint when the outdoor air temperature is above 16[degrees]C (61[degrees]F).

6. For all sectors, the return air temperature is around 22[degrees]C (72[degrees]F) for all hours of the day. This information revealed that there is no change in the room thermostat setpoint at night or during the unoccupied hours.

Conclusions drawn from the analysis of monitored data helped to improve the initial input file, generated from the DOE-2 file, and to develop a model of the Concordia Sciences Building that better simulates the performance of secondary HVAC systems over the calibration period.

CALIBRATION FOR PERIOD A

The Canadian Weather for Energy Calculations (CWEC) weather data file is modified to reflect the on-site conditions for the calibrated period. The dry-bulb air temperature in the original weather file is replaced with the outdoor air temperature measured and recorded onsite by the MDAS, while the relative humidity is kept as per the original weather file. However, when the psychometric calculations give the state of moist air with more than 100% relative humidity, the program calculations do not converge. Hence, for those particular hours, the relative humidity and the atmospheric pressure are modified based on hourly data collected by Environment Canada at the Montreal Pierre-Elliott-Trudeau airport.

Initial Simulation

In the initial input file for the EE4 program, the zone setpoint temperature is set at 24[degrees]C (75[degrees]F) in cooling mode during occupancy and 35[degrees]C (95[degrees]F) at night, while in heating mode the zone setpoint temperature is 22[degrees]C (72[degrees]F) during occupancy and 18[degrees]C (64[degrees]F) at night. The temperature setpoint for hot, chilled, condenser, and glycol water is entered based on design specifications. Figure 2 shows the variation of predicted and measured supply airflow rate for sector B&C during period A. The predicted supply airflow rate does not significantly vary, and the minimum supply airflow rate is higher than measured minimum values. This difference may be explained by the fact that the supply air required to compensate for the laboratories' exhaust is neither included in the input file nor calculated by the program. The supply airflow rate calculated by the EnergyPlus program corresponds only to space cooling loads.

[FIGURE 2 OMITTED]

The predicted and measured supply and return air temperatures are also compared. For sector B&C, the supply air temperature is set to 16[degrees]C (61[degrees]F) in EnergyPlus, which agrees well with the measured value of 15.6[degrees]C[+ or -]1.0[degrees]C (60.1[degrees]F[+ or -]1.8[degrees]F). The measured return air temperature does not vary throughout the day, which reveals that the zone thermostat setpoint is not changed during unoccupied periods. The initial input file of EnergyPlus uses different values during the occupied and unoccupied periods. For this reason, the return air temperature, which is calculated as the weighted average of air temperature of all zones, is overestimated by about 2[degrees]C (3.6[degrees]F).

Revised Simulation

Sector B&C. The input file is modified to reflect the actual operation. A constant zone thermostat setpoint of 22[degrees]C (72[degrees]F) for heating and 24[degrees]C (75[degrees]F) for cooling is used. The setpoint temperatures for hot, chilled, condenser, and glycol water are adjusted according to measured data. The minimum supply airflow rate is reduced to match the minimum supply airflow rate for sector B&C, which is approximately 20% of the maximum supply airflow rate measured over the spring season. Since the compensating airflow rate for the laboratories' exhaust is not included in the EnergyPlus model, the additional airflow rate is added to the simulation results. By considering the average simultaneous usage of ventilation hoods in each room (Table 2), the compensating airflow rate is estimated at 18.5 [m.sup.3]/s (39,200 cfm), and is added to the airflow rate that is calculated by EnergyPlus to satisfy the cooling loads. The predicted and measured airflow rates are now in better agreement (Figure 3).

[FIGURE 3 OMITTED]

Table 2. Average Simultaneous Usage of Ventilation Hoods (PMA 2003)

Number of Hoods Per Room  Percentage of Hoods
                          Running at Full Capacity

1                                100%
2-3                               90%
4-5                               80%
6-7                               70%
8-9                               60%
10+                               50%

The revised simulated air temperatures are now closer to measured data. The average predicted return air temperature is 23.1[degrees]C (73.6[degrees]F), while the average measured value is 22.1[degrees]C (71.8[degrees]F). The average measured and predicted temperatures of supply air are both close to 15.6[degrees]C (60.1[degrees]F).

In order to compare the predicted and measured building cooling loads, an indirect approach similar to the principle of a calorimeter is used. The measured cooling loads of sector

B&C, [Q.sub.m] (kW), are estimated from the measured air-side data as follows:

[Q.sub.m] = [V.sub.a]*[[rho].sub.a]*[C.sub.p,a]*([T.sub.R/A] - [T.sub.S/A]) (1)

where

[V.sub.a] = airflow rate, [m.sup.3]/s

[[rho].sub.a] = density of air, [[rho].sub.a] = 1.169 kg/[m.sup.3] (0.072978 lb/[ft.sup.3])

[C.sub.p,a] = specific heat of air, [C.sub.p,a] = 1.004 kJ/kg*K (0.2398 Btu/lb*[degrees]F)

[T.sub.R/A] = return air temperature, [degrees]C

[T.sub.S/A] = supply air temperature, [degrees]C

The predicted cooling loads have two components: (1) the sector cooling loads, [Q.sub.Sector], due t*o internal and external heat gains, and (2) the cooling load, [Q*.sub.Sector], of the compensating air, which is required by l*aboratory hoods. [Q.sub.Sector] is calculated with Equation 1, and [Q.sub.Sector] is estimated with Equation 2, both using data predicted by EnergyPlus.

[Q*.sub.Sector] = [V.sub.Sector,hoods]*[[rho].sub.a]*[C.sub.p,a]*([T.sub.R/A - [T.sub.S/A]) (2)

where

[V.sub.Sector,hoods] = airflow rate of laboratory hoods, [m.sup.3]/s

Both predicted and measured building cooling loads show a similar trend of variation (Figure 4), and a similar increase in cooling loads with the increase in outdoor air temperature. For sector B&C, the measured cooling load is 155 [+ or -] 67 kW (529 [+ or -] 229 mbh), while the predicted cooling load is 188 [+ or -] 97 kW (642 [+ or -] 331 mbh). The average percentage difference of 21.1% is below the recommended maximum difference of 25% between measured and predicted results (Kaplan and Canner 1992). It is important to note that the prediction of cooling loads is based on the assumption that the air temperature in each zone is uniform and equal to the return air temperature; this assumption implies that the ventilation efficiency is 100%. The measured data correspond to the real ventilation efficiency.

[FIGURE 4 OMITTED]

Sector A. The minimum airflow rate is modified by adding the airflow rate of continuous exhaust during the unoccupied periods, that is, 17.0 [m.sup.3]/s (36,000 cfm). This value is calculated based on the design specifications of the VAV units installed on the ventilation hoods. Moreover, based on design specifications of laboratory hoods (about 170 hoods are installed in sector A), and by considering the average simultaneous usage of hoods in each room (Table 2), the airflow required by the hoods is 45.5 [m.sup.3]/s (96,400 cfm). Since the continuous minimum exhaust airflow rate of 17.0 [m.sup.3]/s (36,000 cfm) has already been included in the input file, only the additional airflow rate of (45.5 - 17.0) = 28.5 [m.sup.3]/s (60,400 cfm) is added to the results obtained using EnergyPlus during the occupied period. By including the additional flow for the laboratories exhaust, the difference between the simulated results and the monitored data is reduced.

The comparison of measured and predicted cooling loads of sector A is also performed (Figure 5). For the unoccupied period, the measured and estimated loads are close. However, during the occupied hours, the EnergyPlus program underestimates the load. The average predicted cooling loads by EnergyPlus are 370 [+ or -] 117 kW (1264 [+ or -] 400 mbh), while the cooling loads calculated from measured data are 421 [+ or -] 150 kW (1438 [+ or -] 512 mbh), which gives an average relative difference of 12.2%.

[FIGURE 5 OMITTED]

Table 3 presents an overview of average results from period A for sectors A and B&C. The predicted and measured data are in good agreement, since the relative difference is less than 6% for directly measured variables and less than 25% for the derived variable (the building cooling loads).

Table 3. Overview of Period A

                                        Sector A

Variable            Measured           EnergyPlus           Diff

Airflow Rate,  59.0 [+ or -] 14.9  62.2 [+ or -] 14.4       5.4%
[m.sup.3]/S

[T.sup.s/a],   15.9 [+ or -] 1.4   16.2 [+ or -] 0.4    0.3[degrees]C
[degrees]C

[T.sup.s/a],   22.0 [+ or -] 0.5   21.2 [+ or -] 0.8   -0.8[degrees]C
[degrees]C

Thermal load,   421 [+ or -] 150    370 [+ or -] 117       -12.2%
kW

                                      Sector B&c

Variable           Measured           EnergyPlus         Diff.

Airflow Rate,  20.2 [+ or -] 7.8  20.9 [+ or -] 9.9      3.5%
[m.sup.3]/S

[T.sup.s/a],   15.6 [+ or -] 1.0  15.7 [+ or -] 0.1  0.1[degrees]C
[degrees]C

[T.sup.s/a],   22.1 [+ or -] 0.4  23.1 [+ or -] 0.4  1.0[degrees]C
[degrees]C

Thermal load,   155 [+ or -] 67    188 [+ or -] 97       21.1%
kW

CALIBRATION FOR PERIOD B

Sector B&C

The approach used for period A is also used for this period, when the cooling coils are not in operation. Figure 6 shows that, during the unoccupied hours, there is good agreement between measured and predicted supply airflow rate for sector B&C. However, during occupied hours, the predicted supply airflow rate, which is calculated by adding the EnergyPlus prediction and the compensating air, is greater than the measured value.

[FIGURE 6 OMITTED]

For further investigation, the hourly predicted airflow rate is plotted against the hourly outdoor air temperature (Figure 7). Two different levels of airflow rate are noticed-one for occupied hours and another for unoccupied hours. At outdoor air temperatures above 16[degrees]C (61[degrees]F), the predicted airflow rate is increased significantly, especially during the occupied hours. The cooling coil is not operational for that period, and if the outdoor air temperature is higher than the design supply air temperature, the program might calculate a higher supply airflow rate to keep the indoor air temperature around the thermostat setpoint. At outdoor air temperatures below 10[degrees]C (50[degrees]F), the systems work in heating mode, and the supply airflow rate predicted by EnergyPlus is also increased. The predicted and measured return air temperatures are in good agreement, while the predicted supply air temperature varies between 16[degrees]C and 30[degrees]C (61[degrees]F and 86[degrees]C) (Figure 8). During period B, the outdoor air is not treated by the cooling coils, and it is directly introduced into the building, thus explaining some high values in supply air temperature. At outdoor air temperatures between 4[degrees]C and 16[degrees]C (39[degrees]F and 61[degrees]F), the predicted supply air temperature is maintained around the design setpoint of 16[degrees]C (61[degrees]F), which corresponds to the supply air flow rate of 30-40 [m.sup.3]/s (63,570-84,755 cfm) (Figure 7). At outdoor air temperatures above 16[degrees]C (61[degrees]F), the predicted supply air temperature is equal to the outdoor air temperature, since the systems operates with 100% outdoor air.

[FIGURE 7 OMITTED]

[FIGURE 8 OMITTED]

At outdoor air temperatures below 4[degrees]C (39[degrees]F), the supply air temperature is between 20[degrees]C and 25[degrees]C (68[degrees]F and 77[degrees]F). The analysis of the input file and outputs generated by EnergyPlus reveals that the minimum outdoor air level of the economizer control was initially set to proportional minimum with the minimum outdoor airflow rate set to autosize. The zone outdoor airflow rate is therefore set proportional to the total system airflow rate. Also, the economizer lower temperature limit was initially set to 4[degrees]C (39[degrees]F). Under this set of conditions, the outdoor airflow rate becomes lower than the calculated minimum coincident outdoor airflow rates of all zones.

To eliminate the error in the input file, the minimum coincident outdoor airflow rate of all zones is calculated in terms of the number of occupants in each zone during unoccupied hours. For sector B&C, the minimum outdoor airflow rate is estimated at 3.46 [m.sup.3]/s (7330 cfm). The lower economizer cutoff temperature limit is also modified from 4[degrees]C to -10[degrees]C (39[degrees]F to 14[degrees]F) to ensure that the outdoor air level decreases linearly with the decrease in outdoor air temperature until reaching the minimum predetermined outdoor airflow rate.

After this revision, the predicted average supply airflow rate and air temperatures are in good agreement with measured data, with a relative difference of less than 4% (Table 4). There are, however, a few significant differences from measured data (Figures 9 and 10). The differences occur at outdoor air temperatures above 16[degrees]C (61[degrees]F) when the outdoor air is directly introduced into the building and the cooling coil is not in operation (Figure 11); thus, the program increases the airflow rate in order to compensate for the decrease of air temperature difference between return and supply.

Table 4. Overview of Period B

                                       Sector A

Variable             Measured         Energy Plus            Diff.

Airflow Rate,  61.9 [+ or -] 11.2  62.0 [+ or -] 14.2        0.1%
[m.sup.3]/s

[T.sub.S/A],   16.8 [+ or -] 1.5   16.9 [+ or -] 2.0    0.1 [degrees]C
[degrees]C

[T.sub.R/A],   21.5 [+ or -] 0.6   20.7 [+ or -] 1.2   -0.8 [degrees]C
[degrees]C

Thermal load,   337 [+ or -] 111    274 [+ or -] 116       -18.8%
kW

                                      Sector B &C

Variable           Measured           Energy Plus          Diff.

Airflow Rate,  20.4 [+ or -] 8.2  22.0 [+ or -] 12.4      -7.8%
[m.sup.3]/s

[T.sub.S/A],   16.3 [+ or -] 1.8  16.9 [+ or -] 2.8   0.6[degrees]C
[degrees]C

[T.sub.R/A],   22.0 [+ or -] 0.4  22.8 [+ or -] 1.2   0.8[degrees]C
[degrees]C

Thermal load,   129 [+ or -] 57    143 [+ or -] 89        10.4%
kW

[FIGURE 9 OMITTED]

[FIGURE 10 OMITTED]

[FIGURE 11 OMITTED]

To complete the analysis of period B, the cooling loads of sector B&C predicted by EnergyPlus are compared with those estimated from measured data. The average cooling loads from measured data are 129 [+ or -] 57 kW (440 [+ or -] 195 mbh), while the average estimated cooling loads are 143 [+ or -] 89 kW (488 [+ or -] 304 mbh).

Sector A

The economizer settings are modified to the fixed minimum outdoor airflow rate of 5.8 [m.sup.3]/s (12,290 cfm) and a lower cut-off outdoor air temperature limit of -10[degrees]C (14[degrees]F). Consequently, the measured and simulated airflow rate variations show a similar trend (Figure 12), with a relative difference between the average values of 0.1%. The average cooling loads of sector A as predicted by EnergyPlus are 274 [+ or -] 116 kW (936 [+ or -] 396 mbh), while the average cooling loads estimated from measured data are 337 [+ or -] 111 kW (1151 [+ or -] 379 mbh), which gives an average difference of less than 20%.

[FIGURE 12 OMITTED]

For period B, the predicted and measured data for sectors A and B&C are in good agreement, since the relative difference is less than 5% for directly measured variables and less than 20% for the derived variable (the building cooling loads) (Table 4).

SENSITIVITY ANALYSIS

When developing a computer model, a number of assumptions and approximations are made about the architectural, mechanical, and electrical characteristics of the building. Sensitivity analysis is often used to assess the impact of changes of input values of some selected variables on the variation of performance variables used for the calibration process. If the model is not, or is less, sensitive to a given variable, any input value can be selected, within the range of study, for that variable. If the model is sensitive, however, the input data must be selected with care, and this may even require additional measurements. The conclusions of the sensitivity analysis normally increase the level of confidence of the user in the developed model.

So far, most studies presented the sensitivity analysis on an annual or monthly basis. In this study, the sensitivity analysis is performed separately for two different weeks, one for each calibration period: (1) the week of June 12 to June 18, with mechanical cooling, and (2) the week of March 20 to March 26, without mechanical cooling. The analysis is performed for sector B&C only.

Six categories of parameters are considered by Sun and Reddy (2006) for the sensitivity analysis, which are related to: (1) the envelope loads, (2) the systems schedules, (3) the load schedules, (4) the auxiliary electrical loads, (5) the internal loads, and (6) the systems variables.

The building under study has been in operation for three years, and, thus, information related to the installed equipment and operating conditions is available for developing the EnergyPlus input file. Also, several conclusions were drawn from the calibration process and from the analysis of collected data, which can help in the selection of variables for the sensitivity analysis. First, it is noticed that the supply airflow rate does not significantly vary with the variation in outdoor air temperature; hence, the space cooling loads due to heat gains/losses through the building envelope are smaller compared to the building internal gains. Therefore, parameters influencing the envelope assemblies are not considered in the sensitivity analysis. Since the computer model is developed based on monitored data, no modifications are required for the system operating schedule and load schedules. The auxiliary non-HVAC electrical loads are not included in the model; hence, this parameter is not considered in the sensitivity analysis. Therefore, the sensitivity analysis is limited to the air infiltration and internal loads.

In the base case, the air infiltration is assumed to occur only when the HVAC systems is OFF. When the system is ON, no infiltration occurs due to building pressurization. For this building, the systems are always ON and, thus, for the base case, the air infiltration rate is set to zero in the calibrated model. For the sensitivity analysis, the infiltration rate is changed from zero to the constant value of 0.15 ach over 24 hours for sector B&C. This value is calculated from the recommended value of 0.25 (L/s)/[m.sup.2] (0.05 cfm/[ft.sup.2]) of gross exterior wall area for natural infiltration rate, for above ground perimeter zones only (MNECCB 1997).

The variation of the average supply airflow rate due to changes in the selected variables is used to assess the sensitivity of the model. The impact of the natural air infiltration rate on the supply airflow rate is negligible for both weeks. The sensitivity coefficient is about -3.3 [m.sup.3]/s per ach for the week of June with mechanical cooling, and it is equal to -13.1 [m.sup.3]/s per ach for the week of March without mechanical cooling. These two values correspond to a variation of the supply airflow rate of less than 1%. The assumption that there is constant infiltration throughout the day probably overestimates the actual infiltration rate in the building. However, since it has a limited impact on the simulation results, one can conclude that any air infiltration rate between 0 and 0.15 ach does not influence the predicted results from the calibrated EnergyPlus model.

When the lighting load is modified by [+ or -]20%, the sensitivity coefficient is 0.35 [m.sup.3]/s per W/[m.sup.2] for the week of June when cooling is required, and it is 0.19 [m.sup.3]/s per W/[m.sup.2] for the week of March when cooling is not needed. In the first case the supply airflow rate varies, on average, by 16%, while in the second case it varies by 9%. Therefore, the initial assumption made about the significant impact of internal loads compared to the envelope loads is proven to be correct.

CONCLUSIONS

This paper presents the pathway used for the calibration of a specific computer model and shows other aspects of the calibration that are beyond the adjustment of some inputs on a trial-and-error basis. The path followed to calibrate the model was based on revisions of the assumptions made for two different sets of operating conditions. The simulation of two distinct operation periods has the advantage of avoiding some compensating errors that can occur when the calibration process is performed over one year, with periods of different operating conditions. The initial input file was created using design information. The monitored data analysis helped to improve the initial input file by adjusting its operating conditions to match the actual building conditions. In this study, the comparison between simulation results and measurements is made with respect to the supply airflow rate and the supply and return air temperatures. The verification of accuracy of those predictions is essential before estimating the energy use by primary equipment such as chillers. This approach is not common in the calibration of energy analysis programs. Moreover, the principle of a calorimeter is used to estimate the building cooling loads from both predictions and measurements. The graphical representation of the variation of the supply airflow rate and temperature with time and outdoor air temperature reveal problems in the building operation and in the computer model. Without those graphical representations, it would be difficult-if not impossible-to correctly calibrate the model in a reasonable time.

The predicted supply airflow rate is in agreement with data monitored, over two periods of the spring season with different operating conditions, as the average difference is between 0.1% and 5.4%. The correct estimation of the supply airflow rate of the AHUs is essential for the calculation of the loads of the cooling coils, and of the water-side loads of the chiller. This information is then used to calculate the electricity demand for the chiller and circulating pumps by using additional information about the coefficient of performance of the chiller, its nominal capacity, and setpoint temperature of chilled water and condenser water. Therefore, before estimating the chiller electric demand it is important to accurately simulate the supply airflow rate.

The supply and return air temperatures are well estimated since the difference between the estimated values and the measured data is between 0.1[degrees]C and 1.0[degrees]C.

The building cooling loads calculated by the program are 10.4%-21.1% higher than those calculated from measured data for sector B&C, and 12.2%-18.8% lower for sector A. This difference does not exceed the maximum acceptable value of 25% for HVAC systems (Kaplan and Canner 1992).

Overall, the calibration exercise indicates that the computer model developed with the Energy Plus program gives good estimations of variables used in this study.

The central plant is simulated by using equivalent and simpler primary equipment and plant configurations. This is because the current version of the program cannot model the existing configuration with several heat recovery systems, heat exchangers, a steam boiler, and the related controls. Future work will focus on the monitoring of primary HVAC systems and the exploration of possibilities of modeling those systems with the Energy Plus program.

REFERENCES

Ahmad, M., and C. Culp. H. 2006. Uncalibrated building energy simulation modeling results. HVAC&R Research 12(4):1141-56.

Bellemare, R., S. Kajl, and M.-A. Roberge. 2002. Modelisation de systemes CVCA a l'aide du logiciel EnergyPlus. Proceedings of the e-Sim Conference. pp. 64-70.

Carroll, W.L., and R.J. Hitchcock. 1993. Tuning simulated building descriptions to match actual utility data: Methods and implementation. ASHRAE Transactions 99(2):928-34.

Corson, G.C. 1990. A comparative evaluation of commercial building energy simulation software. Prepared for Bonneville Power Administration. Contract No. DE-AC79-89BP96543.

DOE. 2007. Energy Plus Manual 2007: Input/Output Reference. US Department of Energy. http://www.eere.energy.gov/buildings/energyplus/pdfs/input-outputreference.pdf (latest access 05/02/2008).

Ellis, P.G., and P.A. Torcellini. 2005. Simulating tall buildings using Energy Plus. Proceedings of the 9th IBPSA Conference. pp. 279-86.

Hayer, S., C. Dalibart, G. Guyon, and J. Feburie. 2001. HVAC BESTEST: Clim 2000 and CA-SIS results. Proceedings of the 7th IBPSA Conference. pp. 1127-34.

Kaplan, M.B., J. McFerran, J. Jansen, and R. Pratt. 1990. Reconciliation of a DOE 2.1C model with monitored end-use data for a small office building. ASHRAE Transactions 96(1)981-93.

Kaplan, M., and P. Canner. 1992. Guidelines for energy simulation of commercial buildings. Portland: Bonneville Power Administration.

Lemire, N., and R. Charneux, R. 2005. Energy-efficient laboratory design. ASHRAE Journal 47(5):58-64.

MNECCB. 1997. Model National Energy Code of Canada for Buildings. http://oee.nrcan.ge.ca/newbuildings/mnecb-emneb/index_e.cfm? Text = N & Print View = N (latest access 28/10/2006).

Neymark, J., R. Judkoff, G. Knabe, H.-T. Le, M. Durig, A. Glass, and G. Zweifel. 2001. HVAC BESTEST: A procedure for testing the ability of whole building energy simulation programs to model space conditioning equipment. Proceedings of the 7th IBPSA Conference. pp. 369-76.

PMA. 2003. Pageau, Morel & Associes. Soumission PEBI -simulation EE4. Universite Concordia, Pavillon des Sciences, Montreal, Quebec.

Reddy, T.A. 2006. Literature review on calibration of building energy simulation programs: Uses, problems, procedures, uncertainties and tools. ASHRAE Transactions 112(1):226-40.

Reddy, T.A., I. Maor, and C. Panjapornpon. 2007a. Calibrating detailed building energy simulation programs with measured data-Part I: General methodology (RP-1051). HVAC&ssR Research 13(2):221-41.

Reddy, T.A., I. Maor, and C. Panjapornpon. 2007b. Calibrating detailed building energy simulation programs with measured data-Part II: Application to three case study office buildings (RP-1051). HVAC&R Research 13(2):243-65.

Sun, J., and T.A. Reddy. 2006. Calibration of building energy simulation programs using the analytic optimization approach (RP-1051). HVAC&R Research 12(1):177-96.

Witte, M.J., R.H. Henninger, and J. Glazer. 2001. Testing and validation of a new building energy simulation program. Proceedings of the 7th IBPSA Conference. pp. 353-60.

Danielle Monfet

Student Member ASHRAE

Roland Charneux, PEng

Fellow ASHRAE

Radu Zmeureanu, PhD, PEng

Member ASHRAE

Nicolas Lemire, PEng

Member ASHRAE

Danielle Monfet is a doctoral student at Concordia University, Montreal, Quebec, Canada. Radu Zmeureanu is a professor in the Department of Building, Civil, & Environmental Engineering, Concordia University. Roland Charneux is an associate, senior design engineer, chief operations officer, and operations vice president and Nicolas Lemire is an associate and design engineer at Pageau Morel and Associates, Montreal, Quebec, Canada.