Model calibration for building energy efficiency simulation

Energy and Buildings 43 (2011) 3666–3679

Contents lists available at SciVerse ScienceDirect

Energy and Buildings journal homepage: www.elsevier.com/locate/enbuild

Calibrating whole building energy models: Detailed case study using hourly measured data Paul Raftery ∗ , Marcus Keane, Andrea Costa Informatics Research Unit for Sustainable Engineering, National University of Ireland, Galway, Ireland

a r t i c l e

i n f o

Article history: Received 28 June 2011 Accepted 30 September 2011 Keywords: Simulation Calibration EnergyPlus Whole building energy model Case study Hourly data Visualisation Version control

a b s t r a c t This paper demonstrates a systematic, evidence-based methodology for calibrating whole building energy models. The methodology uses version control software to store a complete history of the calibration process, including the evidence on which decisions were made. This paper details the calibration of a whole building energy model to hourly energy consumption data using the methodology. The case study building was a 30,000 m2 , four-floor office building located on Intel’s campus in Ireland. The final calibrated model represents the building to a high level of detail using a large number of zones and uses measured lighting and plug load data in the simulation at hourly intervals. The results show excellent correlation with the measured HVAC consumption data for the analysed year (2007), demonstrating the effectiveness of the methodology. Mean Bias Error (MBE) and Cumulative Variation of Root Mean Squared Error (CVRMSE(hourly) ) for HVAC electrical consumption were −4.16% and 7.8%, respectively for the final model. This model was then used to investigate Energy Conservation Measures (ECMs) for feasibility. The paper concludes with a discussion of discrepancies remaining in the model, the issues encountered related to the criteria used for determining when a model is calibrated, and recommendations for future calibration case studies. © 2011 Elsevier B.V. All rights reserved.

1. Introduction This paper describes the detailed, evidence-based calibration of a whole building energy simulation model of a large office building. This is a demonstration of the methodology described in a previous paper [1]. 1.1. Building description Intel Ireland manufactures microprocessors at a large production site in Leixlip, Co. Kildare, Ireland. The Intel office building was completed in 2003 to support the manufacturing facilities (Fig. 1). The four-floor building has a gross floor area of 30,000 m2 and an aspect ratio of 2.1:1. The size and deep floor plan nature of the building minimize the effects on calibration of several variables that are typically difficult to quantify. For example, the relative effect of variables such as infiltration, solar load and three-dimensional ground heat transfer are reduced when compared to smaller buildings. Measured total building and HVAC electricity consumption in

∗ Corresponding author at: National University of Ireland Galway, Civil Engineering Department, University Road, Galway, Ireland. Tel.: +353 91 49 3086. E-mail addresses: [email protected], [email protected], [email protected] (P. Raftery), [email protected] (M. Keane), [email protected] (A. Costa). 0378-7788/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.enbuild.2011.09.039

2007 was 4.581 GWh and 1.097 GWh, respectively. Figs. 2 and 3 present an overview of measured total building electricity consumption using box whisker mean (BWM) plots [2]. The thick black horizontal lines indicate the median. The outer limits of the boxes and whiskers show the 25th/75th and 5th/95th percentiles, respectively. The small black circles indicate outliers, which have been defined as any points outside the whiskers. Mean values are overlaid as blue interconnected circles. The building is primarily a daytime office, occupied from 8 am to 6 pm, Monday to Friday. However, the Intel facility operates on a 24-h basis and there are a significant number of staff who are directly involved in the operation of these production facilities. Thus, the building is at least partially occupied at all times. Night occupancy in the office areas is typically 10% of peak day occupancy based on multiple manual head count surveys and human resources interviews. The ground floor contains a kitchen, a canteen (which operates on a 24 h basis and provides services to staff from other buildings) and a facilities area; the first and second floors consist of open office space and conference rooms; and the third floor is currently unoccupied. 1.2. HVAC systems There are eight large Air Handling Units (AHU) on the roof that condition the majority of the building. Three identical AHUs supply

P. Raftery et al. / Energy and Buildings 43 (2011) 3666–3679

3667

the ground floor contains a ninth AHU that supplies the mechanical, electrical and battery rooms. This AHU has a constant speed supply fan, an outside air enthalpy economiser, a heating coil and a cooling coil. All of the air systems are single-duct systems. A range of VAV boxes, constant volume terminals, fan-powered terminals, both with and without reheat coils, condition the individual thermal zones within the building. In addition to these, hot water convective baseboards (radiators) heat the stairwells. At a plant level, site hot water and chilled water systems supply all the heating and cooling coils in the building. 1.3. Stock information and measured data

Fig. 1. Photograph of the Intel office building.

Fig. 2. Monthly BWM plot of total building electrical consumption in 2007.

a central duct in the east of the building. One of these three is uncommissioned additional capacity for the currently unoccupied third floor. Each AHU has Variable Air Volume (VAV) supply and return centrifugal fans with Variable Frequency Drives (VFD), an outside air dry-bulb economiser, a cooling coil, a preheating coil and a reheat coil. Three identical AHUs condition the west side of the building. The two remaining AHUs on the roof condition the canteen and kitchen areas on the ground floor. A dedicated outside air system (DOAS) with two-speed supply and exhaust fans and a large heating coil conditions the kitchen area. A DOAS with VAV supply and exhaust fans, a preheating coil, cooling coil, and reheat coil conditions the canteen area. The mechanical room on

The quality of stock information about the building is very high due to its recent construction and to the systematic approach to construction documentation undertaken by Intel. High quality as-built drawings and detailed information on materials and constructions are available. Detailed design information (such as as-built pipe and duct sizing information) and Operation & Maintenance manuals are available for most HVAC equipment. Furthermore, Intel has significant resources and expertise on-site for further spot and logged measurements. The Intel site has extremely high quality measurement systems when compared to other case studies examined by the authors to date [3]. A dedicated Energy Monitoring System (EMS) monitors electrical power consumption at every major electrical panel or Motor Control Center (MCC). The panels are generally organised into the following categories: lighting, general equipment, emergency power, uninterruptible power, and MCCs. An on-site weather station measures outdoor dry-bulb temperature, dew point temperature, relative humidity, barometric pressure, daily rainfall, wind speed and wind direction. Furthermore, the archived datasets are of extraordinarily high quality. Data-points are logged at 1-min intervals for weather data and 15-min intervals for EMS data. An archive stores the data in full for at least 5 years. Very few missing data-points or irregular logging periods were found in any of the data, and no spurious outliers were present. However, zone level measurement, such as zone air temperatures, is not available. Also, although the EMS data is extensive, and the panels are organised well, a more detailed look at the panel schedules show discrepancies that complicate the use of this data in calibrated building energy models. For example, although HVAC equipment is primarily supplied from MCCs, some HVAC equipment is supplied from lighting panels (e.g., VAV fan-powered boxes) and general equipment (e.g., unit heaters and air curtains) panels. Also, ‘heat’ meters (flow-rate and temperature sensors) for district hot and chilled water consumption were not in the original design. Thus, it was not possible to calibrate the hot and chilled water consumption for the building – this research analyses model error for electrical power consumption only. 2. Presentation and analysis of results 2.1. Data analysis This research presents results under three consumption categories: total building electrical consumption (total), HVAC electrical consumption (HVAC) and lighting (L), plug load (P) and emergency (E) power consumption (LPE). Mean Bias Error (MBE) and Cumulative Variation of Root Mean Squared Error (CVRMSE) values were calculated using the following formulae [4]:

NP (M − S ) i=1 NP i i MBE =

Fig. 3. Hourly BWM plot of total building electrical consumption in 2007.

i=1

Mi

(1)

3668

P. Raftery et al. / Energy and Buildings 43 (2011) 3666–3679

NP MP =

i=1

Mi

NP

(2)



NP ((Mi i=1

CVRMSE(P) =

− Si )2 /NP )

(3)

MP

where Mi and Si are measured and simulated data at instance i, respectively; p is the interval (e.g., monthly, weekly, daily & hourly); Np is the number of values at interval p (i.e., Nmonth = 12, Nday = 365, Nhour = 8760) and MP is the average of the measured data. Several other data analysis methods were used in this research. Biased percentage error was calculated at each hour using the following formula: biased percentage error =

Mi − Si Mi

(4)

Thus, positive values illustrate periods when the model under predicts actual consumption and vice versa. However, average calculations using biased percentage error values will not account for offsetting errors. Thus, absolute percentage error is also an important variable: absolute percentage error =



Mi − Si Mi

2 (5)

Furthermore, characteristics of subsets of the data provide immediate insight into the circumstances under which the model least accurately represents the building. This powerful analysis technique, commonly known as ‘bin’ analysis, consists of collating data into subsets (or bins) based on the value of another variable. Typical variables on which consumption is likely to be dependent are: hour of day; day of week; month of year; outdoor dry-bulb temperature; outdoor wet-bulb temperature; and solar irradiance. The analysis in this paper does not include solar radiation and wet-bulb temperature bins because energy consumption is not highly dependent on these variables for the Intel building. An example form of the formula to calculate the value (ˇh,v ) of a variable (v) for a specific hour of the day (h) averaged over all the days in the year (d) is:

Nday

ˇh,v =

d

vh,d

Nday

(6)

where vh,d is the value of the variable at a specific hour of the day (h) on a specific day of the year (d). A specific example of bin use could be to determine the average absolute error between the simulated and measured data for total building electrical consumption (total) for a specific hour of the day. Thus, ˇ5,AbsErr(total) is the absolute error for total building electrical consumption averaged at each 5 am period over the whole annual simulation period. 2.2. Notes on visualisation techniques Whole building energy simulation models output significant amounts of data, typically at hourly (or 15 min) intervals. This is 8760 (or 35,040) data-points per data-stream for each annual run. There are inherent difficulties in presenting this amount of data in a coherent and readily understandable manner. Threedimensional plots are essential, as they significantly increase the number of data-points that can be visualised when compared to two-dimensional techniques. Furthermore, even if surface plots are used, the sheer number of data points in an annual simulation makes it difficult to visualise the data in a meaningful manner. Consider that the output from an annual simulation plotted hourly would require a surface plot 365 data-points long and only 24 wide. Such plots are unwieldy to display and require significant time to review and identify patterns. Combining surface plots with bin analysis allows an analyst to efficiently assess patterns present

in the data and how well the simulation predicts actual energy consumption. Figs. 7–9 illustrate the absolute model error using surface plots by bin for hour of day, day of week, month of year and outdoor dry-bulb air temperature (in 2 ◦ C bins). It should be noted that only independent variables are plotted on any given surface plot. For example, outdoor dry-bulb temperature is plotted with respect to day of week (of which temperature should be relatively independent) instead of hour of day or month of year (of which temperature is clearly dependent). Furthermore, given a fixed total number of colours (20) the smallest possible integer step size is used for each image (e.g., 1%, 2%, 3%, etc.). This illustrates each image to the highest level of detail, with the disadvantage of making it more difficult to compare plots that use different colour ranges. Numerous other visualisation tools were used over the course of this research, such as time-series graphs and scatter plots. 3. Overview of calibration process A previously published paper [1] describes the calibration methodology in detail. The number of revisions at each stage of the process (can be seen in Table 5) relates specifically to the Intel building case study. A higher or lower number of revisions could be required in other projects depending on the size and complexity of the modelled building and systems, the amount of information available at each stage of the calibration process, and the number of iterative process steps required to meet the acceptance criteria and satisfy the analyst(s). 3.1. Version control TortoiseSVN [5], an open-source software tool, was used create a version control repository of the calibration process. This repository stores each revision of the model, a description of modifications made between revisions and the evidence based on which each change was made. This improves the reproducibility and reliability of the calibration process by allowing future users of the model to review the entire history of the calibration process and the evidence on which the current version is based. 3.2. Preparation 3.2.1. Initial model As building energy simulation is still a relatively rare practice in Ireland, no initial model was available and hence one was created for this project. The initial model was created according to design documentation and program defaults where design information was unavailable. A standard core and 4 perimeter zone strategy was applied to each floor. Zones were defined at higher resolution where multiple air conditioning systems served areas on the same floor. EnergyPlus v3.1 was selected as the simulation engine based on a comprehensive review of program capabilities [6]. The EnergyPlus program has been validated against experimental measurements and through comparative testing with the BESTEST suite [7]. In addition, EnergyPlus input files are in humanreadable text format, which allows analysts to use version control software capabilities to immediately and precisely highlight the differences between two model revisions. An Industry Foundation Class (IFC) [8] based Building Information Model (BIM) of the geometry and constructions was created using architectural drawings. An EnergyPlus input data file (IDF) of the building geometry was created from the BIM automatically using GST/IDF Generator; a newly developed data transformation tool at the Lawrence Berkeley National Laboratory [9]. This is an example of partial automation in the calibration process through

P. Raftery et al. / Energy and Buildings 43 (2011) 3666–3679

the use of BIM technology mentioned in the methodology paper [1]. The HVAC systems and miscellaneous other objects were added to this IDF file using a separate interface tool developed by the authors – EnergyPlus HVAC Generator. The model uses a 15 min zone time-step and a 1 min system time-step for HVAC and plant calculations. Decreasing the zone time-step has very little effect on the results (e.g., using a 5-min time-step affects MBE and CVRMSE by less than 0.005%). 3.2.2. Historical weather file A historical weather file was created from measured 2007 data obtained from the on-site weather station. Typical Mean Year 2 (TMY2) data for Dublin was used for solar radiation as this measurement was unavailable. This does incur a certain (unavoidable) amount of error. However, this is not significant due to the buildings’ deep floor plan, highly insulated exterior constructions and low-emissivity double-glazed windows. 3.2.3. Calibration data Hourly measured electrical consumption data for 2007 was obtained for the building and was used for comparison with the model output. As discussed in Section 1.3, measured data for district chilled and hot water energy consumption was unavailable and thus, this research focuses only on electrical energy consumption. 3.2.4. Documentation The companion paper (reference to be added upon publication) defines the need for acceptance criteria and significance criteria in the calibration process. ASHRAE Guideline 14 defines the acceptable limits for calibration to hourly data as within ±10% MBE and ≤30% CVRMSE(hourly) measured at a utilities level [4]. These were selected as the acceptance criteria for this project. In the absence of guidelines for significance criteria, for the purpose of this demonstrator any modification to the model that results in a change of greater than 1% MBE, 1% CV RMSE(monthly) , or 2% CV RMSE(hourly) was considered significant. The source hierarchy was defined according to the methodology. A complete record of all the sources of information used in the model was kept at every stage of the calibration, starting at the preparation stage (revision 1). Table 1 illustrates the source hierarchy for the final revision under the categories described in the methodology paper [1]. Table 2 illustrates the model parameters affected by each source of information. These two tables are populated as follows: • A new source of information is reviewed for use in the model. This source of information is added the relevant category in Table 1 (i.e., short term electrical load measurements are added to Category 2: spot measurement); • The parameters affected by this new information are added to Table 2. Table 2 also highlights an occurrence where a source of information overrides one that is lower down in the source hierarchy. In this case, power consumption per light fixture was updated based on asbuilt electrical panel schedules at revision 15 (entry 29). However, this parameter was later updated based on measured data (entry 32) as this source is higher in the source hierarchy and is expected to be a more accurate estimate for this parameter. The version control repository also contains a detailed change log and the actual evidence used to make a change to the model (e.g., in this case: a spreadsheet containing the lighting load measurements, calculations and assumptions). This level of documentation allows future users to review the evidence on which the model is based at any stage of the calibration process. It should be noted that these tables

3669

have been shortened for this publication. For example, Table 1 excludes links to the associated evidence locations in the version control repository and Table 2 contains brief synopses rather than a detailed description due to space limitations. 3.3. Update model 3.3.1. Zone-typing This section describes the zone-typing process as applied to the Intel building. Fig. 4 illustrates the initial model floor plan described in Section 3.2.1. This model uses the traditional zoning strategy in which a core zone and a zone at each perimeter face of the building represent the actual thermal zones in the building. Note however, that in this case the floor plan is additionally split in two halves (East and West) because separate conditioning systems supply each side of the building. This yields 8 zones per floor (1 core zone and 3 perimeter zones per conditioning system) instead of the usual 5 (1 core zone and 4 perimeter zones). This model also includes the assumption that all zones are separated by a physical internal partition (in this case, a simple un-insulated gypsum plasterboard construction), even in open office spaces where there is no physical separation between zones. This practice is a result of limitations in older simulation engines regarding zone boundaries. The zonetyped model couples long-wave radiation between zones that are not separated by a physical wall (e.g., in open plan offices) using the Material: InfraRed Transparent objects in EnergyPlus [13]. Fig. 5 illustrates the thermal zones on one floor of the model after the zone-typing process has been completed. Zone-types were defined based on the criteria discussed in part the methodology paper [1]. For example, the final zone-typed model: • Describes communications rooms, conference rooms, stairwells and toilets as separate zone-types because these zones will have significantly different internal load profiles and conditioning methods (e.g., Variable Air Volume terminal boxes with return air plenum, constant air volume with exhaust air, hot water baseboard heaters, etc.). • Describes large ducts (e.g., ‘risers’) as separate zone-types in order to ensure accurate floor areas for other zones and so that the simulation can estimate heat transfer between these ducts and the other zones in the building. • Describes return plenums as individual zones instead of incorporating them into floor constructions as an increased thermal resistance (e.g., as an additional ‘air gap’ resistance under the floor slab). Actual zones in the building of the same zone-type were agglomerated (i.e., represented as a single zone in the model) when: • The inter-zone surface-length to perimeter ratio was greater than 0.05%. • The resultant agglomerated zone had a maximum distance in any dimension of 75 m and a maximum floor area of 1500 m2 . The zone-typed model has 106 thermal zones and although it is a more accurate representation of the Intel building, it has an associated cost both in terms of the additional time required to create the detailed model, and in model run-time. Run time increased from 46 min to 3 h and 38 min (an increase of 370%) between the initial model and the zone-typing strategy, respectively for the Intel building. 3.3.2. Constructions Wall constructions and materials properties were updated in revision 3 based on commissioned as-built material datasheets (Table 1, Category 5) and as-built drawings (Table 1, Category 7).

3670

P. Raftery et al. / Energy and Buildings 43 (2011) 3666–3679

Table 1 Source hierarchy of information used in the final model (revision 23). Category

Source type

Unique ref.

Description

Revision

1

Logged measured data

o p

EMS lighting and general equipment electrical load data EMS emergency power electrical load data

6 7

2

Spot measured data

h v w aa ae

Conference room booking database Measured electrical load data Measured electrical load data Spot airflow measurements Spot check of air set-points at VAV boxes

4 16 17 21 23

3

Surveys and physical verification

a i x y z

Geometry verified by physical inspection Office occupancy counts (4 surveys – 3 days, 1 night) Photographs Site visit and walk through Photographs

2 4 18 18 19

4

Interviews

j s

HR Interviews Building operator interview

4 11,22

5

Material data-sheets

d

Material properties datasheets

3

6

Operation & Maintenance manuals

r ab

Air handling unit Operation & Maintenance manuals Air distribution unit Operation & Maintenance manuals

9,10 21

7

As-built documentation

b c g k m n q ac

Architectural and mechanical as-built drawings Architectural elevation as-built drawings Wall materials and constructions taken from as-built drawings Official construction project scope document Electrical as-built drawings Electrical panel schedules Further mechanical as-built drawings As-built VAV box design sizing data

2,14 3 3 4,8 6,7,9 6,15,16,21 9 21

8

Benchmark or best practice models

f l

Glass properties – Window 6 datasets [10] US DOE benchmark models [11]

3 5,11

9

Standards and guidelines

e t ad

ASHRAE 2005 handbook of Fundamentals [12] EnergyPlus input/output reference v3.1 [13] Advanced Variable Air Volume System Design Guide [14]

3,14 13 22

10

Initial model

Table 2 Excerpt of the sections of the model affected by each source of information. Entry no.

Section of model affected

Unique ref.

Description

Revision

Overwrite entry no.

23

Systems

All AHU Coils

r

10

N/A

Systems

AHU set-points

s

11

N/A

Zones Zones Zones Lighting Emergency power Emergency lighting

Ground floor zones All zones All zones N/A N/A All zones

t e b v n w

Updated AHU coil characteristics (design air/water on/off temperatures, flow-rates and nominal capacities) Updated AHU mixing box, cooling, heating and preheat coil set-points Added ground surface temperatures Added internal mass objects to represent furniture Updated building orientation

25 26 27 28 30 31 32

13 14 14 16 16 17

N/A N/A N/A N/A N/A 29

Added exterior lighting loads Update of constant emergency lighting load

Fig. 4. Typical zoning strategy – first floor plan of the initial model (revision 1).

P. Raftery et al. / Energy and Buildings 43 (2011) 3666–3679

3671

Fig. 5. First floor plan of the model (revision 2).

Benchmark model practices and standards were used where these were unavailable (Table 1, Categories 8 and 9). The sources of the information used were also entered as comments under the name of each material object in the input file for the EnergyPlus model.

3.3.3. Internal loads The internal loads were first updated in revisions 4–7. Occupancy data for each zone-type was updated based on human resources interviews, personnel counts, and multiple day/night occupancy surveys (Table 1, Categories 3 and 4). A centralised room-booking database was used to determine conference room occupancy (Table 1, Category 2). Infiltration parameters were added to the model based on best practice models (Table 1, Category 8) as no other source of information was available. Measured data from the EMS was used to update lighting and equipment loads (Table 1, Category 1) at revisions 6 and 7. Average values were used for measurement streams that were constant within a maximum error band of ±5% over any given hour. Modelling of the remaining (the majority) lighting and general plug electrical measured data using typical schedules (i.e., day-typing [15]) could have been employed. However, this introduces error into the simulation. This is unnecessary if complete annual measured data is available and the simulation tool is capable of using this data directly. Thus, instead of day-typed schedules, the actual measured values for these loads were used in the model at each hour of the simulation. Each of the individual EMS data-streams (measured at multiple panels in the building) were added to the model on a floor-area-weighted basis. This was deemed a better approach because the primary purpose of the calibrated model is to investigate changes in HVAC consumption due to a proposed ECM, not to explicitly model lighting and plug loads. This is especially the case when this would incur significant (>5%) unnecessary error in the hourly results. Plug loads are typically occupant controlled and building operators do not have a significant amount of control over these aspects of building consumption. Furthermore, cost effectiveness analyses of retrofit options related to lighting can be performed easily when measured data is available. Thus, there is no need to model these aspects of building consumption using schedules when accurate measured data is available. However, lighting and plug load consumption data was measured at a multiple zone level. Therefore, assumptions were made in order to assign these loads to specific zones: • As-built electrical drawings and panel schedules were used to allocate these loads on a floor area weighted basis. • Additional spot measurement, as-built panel schedules and asbuilt electrical drawings (Table 1, Categories 2 and 7) were used

to further refine the floor area weighting of this measured data at later stages of the calibration process (revisions 15–17). • Large constant loads were assigned to the specific zone in which the equipment is located. • Exterior equipment (external lighting) was added to the model based on additional short term measurements.

As always, the version control repository stores a complete record of these assumptions including the measured data and calculations on which the loads in the model are based.

3.3.4. HVAC and plant The model inputs relating to HVAC and plant were updated from revisions 8 to 11 based on the readily available information: the official design scoping document, commissioned as-built mechanical drawings, operator interviews and O&M manuals including spot air hood measurements (Table 1, Categories 4, 6 and 7). Further refinements were made at later stages of the calibration process. For example: • Pump characteristics were verified by site visit at revision 19 (Table 1, Category 3). • Air Distribution Units O&Ms (including spot measurements of airflow at each terminal) were obtained and used to update the maximum airflows and heating coil capacities on a zone by zone basis at revision 21 (Table 1, Category 2). • Fan part load curves were updated to reflect operating static pressure set-points at revision 22 (Table 1, Category 4). • Manual spot checks of VAV box set-points were used to update the model at revision 23 (Table 1, Category 2).

All HVAC characteristics in the final model were either updated or verified based on more reliable information in the source hierarchy than the initial model. For example, • Fan type, maximum airflow, pressure & operating efficiency were updated based on the O&M manuals (Table 1, Category 6). Part load curves were selected from the Advanced Variable Air Volume Systems Design Guide [14] based on fan static pressure set-points (Table 1, Category 8). • Coil types; air and water on and off design temperatures; maximum air and water flow-rates; and heating/cooling capacity were updated based on heating and cooling coil O&M manuals (Table 1, Category 6).

3672

P. Raftery et al. / Energy and Buildings 43 (2011) 3666–3679

3.4. Error check Several inconsistencies were identified and corrected at this stage of the calibration process. Revision 12 corrected a case in which the outputs from the simulation were clearly unreasonable. The minimum airflow through the VAV Air Distribution Units (ADUs) was zero, despite an entered value of 0.3. This was due to a bug in EnergyPlus HVAC Generator v1.08 where two lines of code were reversed, yielding a blank entry for the minimum airflow parameter (i.e., interpreted as zero by the simulation software). Revision 14 illustrates an instance where previously obtained information was added to the model incorrectly. In this case, infiltration was accidentally omitted for Unconditioned and Stairs zone-types. These issues highlight the need for detailed review of simulation outputs by the analyst(s) performing the calibration in order to verify that the outputs are reasonable. 3.5. Iterative process The model met the defined acceptance criteria (Section 3.2.4) for calibration to hourly data at the first iteration (revision 15). However, even though the revision 15 model meets the acceptance criteria, model error was still quite high under certain conditions, such as peak occupancy and high outdoor dry bulb temperature (see Section 6.2). Thus, the authors continued the calibration in order to improve the model and to thoroughly demonstrate all stages of the iterative process. Fig. 6 illustrates all paths through the iterative process for each revision using letters A to F. Table 3 gives a detailed description of the changes at one revision for each step (A to F) through the iterative process. Revision 16 describes a case where it was possible to obtain new information but significant resources were required: logged measurement of electrical panels. Thus, a significance test was performed and two runs were made to evaluate the significance criteria as described in the companion paper [1]. The parameter in question (exterior lighting) was modified between estimated maximum and minimum values based on the commissioned as-built panel schedules. The results (Table 4) indicate that this parameter did not meet the significance criteria, however, the analyst deemed it necessary to obtain the new information and add it to the model (Table 3). 4. Results and discussion 4.1. Results over the history of the calibration process Table 5 presents model error by revision for total building electrical consumption (total) and the maximum/minimum model error at any hour over the entire annual simulation period. The stages of the calibration process refer to the methodology described in a previous paper [1]. Section 3.3.3 describes that the model uses lighting, plug and emergency (LPE) power data on an hourly basis from revision 7 onwards, with assumptions and modifications made to assign this data on a floor-area-weighted basis. Thus, MBE for this consumption (LPE) in the final model is zero and CVRMSE(hourly) is very low, at 0.02%, as shown in Table 6, which shows the changes in LPE error over the course of the calibration process. The use of internal load data in this manner means that error in total building electrical consumption (total) is effectively a diluted version of the HVAC consumption error (as total consumption is equal to the sum of HVAC and LPE consumption). However, it should be noted that the final model would meet the acceptance criteria even if they were also applied to HVAC energy consumption. It should be noted that the significant decrease in error between the initial model (revision 1) and revision 2 is due primarily to an

Fig. 6. Overview of the path through the iterative process as occurred in this demonstrator.

upgrade from EnergyPlus HVAC Generator v1.0 to v1.07. This later version included basic automated error checking of many parameter values. This feature identified an unreasonably low value for AHU fan efficiency and corrected it to a default value. As expected, when the zone-typing change is examined in isolation (i.e., using the same version of EnergyPlus HVAC Generator as the initial model) it shows a less significant effect on model error: MBE, CVRMSE(monthly) and CVRMSE(hourly) change by +7.63%, −7.73%, and −6.94%, respectively.

P. Raftery et al. / Energy and Buildings 43 (2011) 3666–3679

3673

Table 3 An example of each stage of the iterative process. Example of point

Revision Description

A

15

The simulated lighting consumption was greater than the installed lighting power in some zones – a clear modelling error. It was discovered that the consumption of several high load devices was spread out across many zones in the building. This flawed assumption in emergency power distribution was remedied and a new approach was taken based on an as-built electrical panel schedules document.

B

19

There were no clear discrepancy sources in the model. Starting at the lowest level in the source hierarchy, parameter(s) were identified for which more reliable information or further measurement was possible. It was discovered that parameters related to the pumps were based on the initial model and that further information was available (physical nameplate inspection).

C

18

A discrepancy was identified between spot measured data for wattage/light fitting and previous revision value (which was based on as-built electrical panel schedules). There were little further resources required as an Intel technician had previously measured that panel. Thus, no significance test was required.

D

21

Maximum air flows and heating coil max capacities related to some of the air distribution units (VAV boxes) were based on the initial model and it was possible to obtain further information. However, significant resources were required to obtain this information and enter it into the model. A significance test was performed and the parameter met the criteria. Thus the information was obtained and the model updated accordingly.

E

16

Exterior lighting fed from internal lighting panels was discovered through an in-depth review of the as-built panel schedules. The heat generated by these lights does not affect HVAC consumption in the real building, but due to the floor area weighting of EMS data, simulated HVAC consumption was affected. Analysis showed that the parameter did not meet significance criteria, but the analyst deemed it necessary. Thus, spot measured data was obtained to identify the power consumption of these lights and the floor area weighting of the EMS data was corrected to account for these loads.

F

18

The floor area weighting of the EMS data from one of the equipment panels did not account for a relatively high load contained in one zone on the ground floor (vending machines). However, the parameter did not meet the significance criteria required to justify additional measurement.

Table 4 Results of significance test at revision 16.

Run 1: constant maximum design load Run 2: photocell operation at 75% of design load Significance criteria Difference between the two significance runs

MBE

CVRMSE (monthly)

CVRMSE (hourly)

7.80% 8.50% 1.00% 0.70%

7.85% 8.55% 1.00% 0.70%

8.14% 8.96% 2.00% 0.82%

Table 5 Analysis of error in total building electrical consumption by revision. Revision number

Stage of calibration process

MBE

CVRMSE (monthly)

CVRMSE (hourly)

Max error

Min error

Meets hourly acceptance criteria?

Meets monthly acceptance criteria?

1a 2 3 4 5 6 7b 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

Preparation Zone-typing Constructions Internal loads Internal loads Internal loads Internal loads HVAC and plant HVAC and plant HVAC and plant HVAC and plant Error check Error check Error check Iteration 1 Iteration 2 Iteration 3 Iteration 4 Iteration 5 Iteration 6 Iteration 7 Iteration 8 Iteration 9

−78.89% −6.34% −4.61% −7.21% −7.21% 28.31% −4.40% −1.40% −31.44% −32.81% −10.68% −11.87% −12.45% −13.20% −8.75% −8.45% −8.25% −5.86% 0.08% 0.39% 1.41% −1.32% −1.00%

79.26% 6.63% 4.97% 7.43% 7.44% 28.33% 4.57% 1.79% 31.59% 33.03% 10.78% 11.94% 12.51% 13.26% 8.80% 8.51% 8.31% 5.93% 0.81% 0.89% 1.68% 1.63% 1.35%

94.02% 38.40% 38.44% 38.05% 38.04% 28.45% 5.03% 2.36% 31.90% 33.40% 12.31% 12.99% 13.67% 14.44% 9.21% 8.91% 8.68% 6.11% 1.53% 1.47% 2.16% 2.19% 1.87%

−21.52% 35.42% 37.15% 34.33% 34.35% 62.19% 1.68% 4.38% −21.34% −21.23% −0.32% −3.27% −3.54% −4.20% −1.40% −0.87% −0.76% 1.57% 5.62% 5.82% 7.20% 4.44% 4.58%

−263.30% −99.49% −98.15% −98.38% −98.32% 15.18% −14.24% −9.07% −53.36% −56.14% −47.79% −45.86% −47.72% −51.56% −24.50% −23.38% −22.26% −12.30% −6.76% −5.43% −4.39% −7.52% −6.54%

No No No No No No Yes Yes No No No No No No Yes Yes Yes Yes Yes Yes Yes Yes Yes

No No Yes No No No Yes Yes No No No No No No No No No No Yes Yes Yes Yes Yes

a b

This revision includes the fan efficiency error discussed in Section 4.1. Measured lighting and plug load data used in the model from this revision onwards.

Table 6 MBE and CVRMSEs for LPE electrical consumption by model revision. Revision number

MBE

CVRMSE (monthly)

CVRMSE (hourly)

1 2–5 6 7–21

4.04% 9.89% 36.91% 0.00%

4.41% 10.03% 36.94% 0.01%

50.10% 47.53% 36.98% 0.02%

3674

P. Raftery et al. / Energy and Buildings 43 (2011) 3666–3679

Table 5 shows that the acceptance criteria (the ASHRAE hourly calibration criteria) for this project are met at every stage of the iterative process from revision 15 onwards. The ASHRAE monthly acceptance criteria (±5% MBE and ≤10% CVRMSE(monthly) ) are met from revision 19 onwards due to more stringent requirements on Mean Bias Error. The authors continued the calibration process despite the results meeting the acceptance criteria because significant discrepancies remained in the model (see Section 6.2).

4.2. Overall results This section compares the model results with the measured consumption data (for 2007) in detail using techniques described in Section 2. Only HVAC absolute error plots are shown due to space constraints. Figs. 7 and 8 show a large improvement in correlation between simulation output and measured data. HVAC MBE drops from −342.49 to −4.16% between the initial model and final calibrated model, respectively. Likewise, CVRMSE(hourly) drops from 349.1% to 7.8% between the initial model and final calibrated model, respectively.

4.3. Comparison to further measured data A historical weather file and measured energy consumption data were obtained for the following year (2008) as a means to further verify that the model is an accurate representation of the building. EMS data for lighting, plug and emergency power was added to the model. No further changes were made to any other parameters. Table 7 shows that the results of this simulation comfortably meet the acceptance criteria. Fig. 9 illustrates HVAC absolute error in the same manner as for the 2007 analysis in Fig. 8.

5. Energy Conservation Measures Numerous Energy Conservation Measures (ECMs) were identified over the course of the demonstrator, as can be expected from an in depth audit of any building. Table 8 gives a brief description of some of these ECMs. Although a complete analysis of all of these ECMs is outside the scope of this paper, two specific ECMs were analysed as examples.

5.1. ECM 1: static pressure reset Implementing a static pressure reset on the office AHUs based on the damper positions of a representative sample of VAV boxes would yield significant savings. Implementing this change in the model showed a reduction of 181 MWh of electricity and 28 MWh of chilled water, offset by an increase of 71 MWh of hot water, for 2007. These values correspond to 3.9%, 8.4% and −2.3% of simulated total building electrical, chilled water, and hot water consumption, respectively. This estimate was obtained using the calibrated model and static pressure reset fan curve from the Advanced Variable Air Volume System Design Guide [14].

5.2. ECM 2: conference room demand controlled ventilation (DCV) The conference rooms in the building are sporadically occupied and have very low lighting and equipment loads during these periods. However, these zones have fixed minimum ventilation airflow rates sized to supply ventilation air for the maximum design number of occupants. The unnecessary airflow overcools these zones during the unoccupied periods. Thus, this situation wastes reheat energy and electricity (fan power). This is an example of a situation in which heating and cooling loads in adjacent zones (in the conference rooms and open office space directly outside these rooms, respectively) oppose each other. Core-and-four-perimeter zoning strategies do not capture these effects. Also, this ECM analysis is an example of a situation in which a detailed model yielded an estimate with fewer assumptions than would be required in a less detailed model (such as the core-and-four perimeter approach). Implementing rudimentary demand controlled ventilation (DCV) in these zones would have saved 19.5 MWh (0.4%), 5.09 (1.5%) and 40.9 MWh (1.3%) of electricity, chilled and hot water, respectively in 2007. This estimate was obtained by reducing the VAV box minimum airflow fraction in these zones to the design leakage rate (taken from VAV box as-built documentation) during unoccupied periods. Implementing DCV using occupancy sensors (e.g., CO2 or infrared) would yield further savings, however at additional initial cost. 6. Discussion 6.1. Discrepancies remaining in the final model The results still show a significant amount of discrepancy between measured and simulated values despite the level of calibration effort. This is in part caused by poor estimates for some parameter values. However, some of this is due to the unpredicted operation of HVAC equipment. For example, in 2007, the model under predicts HVAC consumption in November 2007. Fig. 8 shows the magnitude of this error. The fact that it is an underestimate (i.e., a negative error) was identified using separate biased error plots (not shown due to space constraints). Measured data showed that this was primarily due to a large reduction in airflow at one of the AHUs. The second AHU that supplies the central duct in the west of the building compensated for this by increasing airflow to maintain the pressure set-point in the duct. However, power consumption is proportional to the cube of airflow at each fan and thus, this change caused increased power consumption overall. Conversely, the model over-predicts energy consumption in January 2007 (Fig. 8 shows the magnitude of this error) due to a reduction in power consumption. These issues were identified from the data in 2010 and unfortunately, their causes are unknown. Also, many discrepancies in the model are due to the assumptions and simplifications made by the simulation engine. A thorough discussion of the simplifications and assumptions made by whole building energy simulation tools can be found in recent research [17]. For example, the model overestimates HVAC consumption during the summer months at peak occupancy (Fig. 8 shows the magnitude of this error). This is due, at least in part, to the fact that the simulation engine models VAV terminal boxes with

Table 7 Results of model error analysis using 2008 data. Consumption

MBE

CVRMSE (monthly)

CVRMSE (hourly)

Max

Min

Total HVAC LPE

−1.11% −4.61% 0.00%

1.34% 5.53% 0.02%

2.11% 8.72% 0.02%

13.89% 40.99% 0.04%

−7.54% −37.39% −0.08%

P. Raftery et al. / Energy and Buildings 43 (2011) 3666–3679

3675

Fig. 7. Initial model (revision 1) average absolute percentage error for HVAC electrical consumption.

ideal, instantaneous control. Fig. 10 compares the response of a VAV box (without reheat) in the simulation software and the idealized response in a typical VAV box (such as those in the Intel building). When cooling is required, the software calculates the exact required airflow to maintain the cooling set-point (23 ◦ C) based on the total load in the zone (up to a fixed maximum airflow, after which zone air temperature begins to increase) [18]. However, in reality, VAV box airflow is proportional to the error between current zone air temperature and the cooling set-point temperature over a fixed proportional band (2 ◦ C) (up to a fixed maximum airflow). Thus, simulated airflow is much more responsive to cooling loads than the actual building, causing fan power overestimation during high cooling load periods. 6.2. Issues with the acceptance criteria Several revisions illustrate that monthly acceptance criteria do not adequately capture how well the model matches the measured data. For example, Table 5 shows that the revision 3 model has a MBE and CVRMSE(monthly) of −4.61% and 4.97%, respectively, and thus it clearly meets the monthly acceptance criteria. However,

the revision 3 model has a CVRMSE(hourly) of 38.44% and thus only serves to highlight the fact that it is possible for a model to meet even the most stringent monthly acceptance criteria without accurately representing the building at more frequent intervals. When analysed in detail, this revision also illustrates the need for hourly measured data at sub-utilities level, as lighting and plug underestimation (by 245 MWh) partially offsets HVAC overestimation (by 555 MWh). Table 5 also shows that the model met both the monthly and hourly acceptance criteria at revisions 7 and 8. This was before the HVAC equipment parameters had been updated from the initial model, which included major discrepancies such as automatically sized parameter values for large equipment and inlet vane part load curves (instead of Variable Frequency Drive curves). These major discrepancies between the model and the real building clearly cannot be captured by current acceptance criteria. Furthermore, the model met the acceptance criteria at many stages of the final iterative process (revision 15 onwards), even though other significant errors remained. For example, the model comfortably meets the acceptance criteria at the beginning of the first iteration (revision 15). However, detailed analysis using the

3676

P. Raftery et al. / Energy and Buildings 43 (2011) 3666–3679

Fig. 8. Final model (revision 23) average absolute percentage error for HVAC electrical consumption.

visualisation technique described in Section 2.2 shows that this revision still contains significant errors. For example, the model overestimates HVAC electrical consumption by 108.25% during the most extreme hour of the year. Visualisation of HVAC model error showed that significant, systematic errors remained: biased HVAC error during peak occupancy (10 am to 5 pm, Monday to Friday) was an average of more than 48% over the whole year. Lastly, the most stringent currently available acceptance criteria focus only on comparing energy consumption between the model and the real building. Given the size of the parameter space and the number of possible solutions for the currently available acceptance criteria, there is no guarantee that any particular solution is a fair representation of how the building actually operates. For example, energy consumption data may closely correlate but there could be large differences between zone air temperatures in the model and those in the building. This would not be captured by current available acceptance criteria. 6.3. Recommendations It is clear that current acceptance criteria did not identify significant errors in the model (Section 6.2). More stringent acceptance

criteria are required to ensure that the model closely represents the operation of the real building: 1. Clear issues were encountered when using only monthly data (Section 6.2) for comparison. Thus, a requirement for hourly measured data over the calibration period is recommended. 2. Given the number of solutions yielded by the current acceptance criteria, a reduction in the acceptable MBE and CVRMSE(hourly) from 10% to 5% and 30% to 20%, respectively is recommended. 3. Overestimated end-use consumption in one area can offset underestimates in other areas to yield a reasonable correlation at the utilities level (Section 6.2). Thus, a requirement to explicitly measure energy consumption by end-use is recommended. The acceptance criteria should be applied to each of these end-use measurements. As an example, a typical office building would require separate measurement of lighting & plug loads, HVAC, chilled water, and hot water consumption. Each of the simulated versus measured data end-use comparisons would need to meet the acceptance criteria. 4. Current acceptance criteria evaluate energy consumption only. They do not account for how well the model represents other key aspects of the building’s operation, such as zone temperature

P. Raftery et al. / Energy and Buildings 43 (2011) 3666–3679

3677

Fig. 9. Model (revision 23) average absolute percentage error for HVAC electrical consumption using 2008 data.

conditions. Thus, a requirement to compare zone temperatures in the model and the actual building is recommended. This could be implemented by comparing zone temperatures on a zone volume weighted basis for the whole building. 5. Combination of criteria 3 and 4 into a single goodness of fit equation (using suitable weighting factors for each individual comparison criterion) to give a single overall indicator of how well the model captures the real operation of the building. These recommendations are based only on this case study. A numbers of detailed calibration case studies are necessary in order to further develop and formalise new acceptance criteria. Specifically, these criteria may be unreasonably stringent for buildings with HVAC systems that are more difficult to model (or require more assumptions and simplifications) using currently available software, such as buildings that use natural ventilation or Thermally Activated Building Systems (TABS). Furthermore, while the methodology applied in this case study yielded a solution under the defined acceptance criteria, there is no guarantee that it would do so for all calibration case studies. For

example, in cases with poorer evidence (limited building information, lack of spot measurement resources, etc.), with more difficult to model HVAC systems, or with more stringent acceptance criteria (such as those outlined above), the analyst may find that no more

Fig. 10. VAV box airflow.

3678

P. Raftery et al. / Energy and Buildings 43 (2011) 3666–3679

Table 8 Examples of ECMs identified during the calibration process. Description

Primary initial costs

Initial cost category

AHU static pressure reset (ECM 1): implement AHU pressure reset based on a representative selection of VAV boxes damper positions [14]. Conference room DCV (ECM 2): implement DCV in conference rooms. Canteen DCV (ECM 3): control canteen exhaust fan speed based on measured ventilation requirements. Canteen ERV (ECM 4): implement energy recovery ventilation on canteen AHU. Pump differential pressure reset (ECM 5): implement and test a pump (secondary) differential pressure reset based on AHU cooling coil valve positions. AHU split signal damper control (ECM 6): implement and test an AHU split signal damper control strategy [16]. Night-time fresh air reduction (ECM 7): reduce AHU outdoor air flow rate during low occupancy periods (currently fixed at the minimum fresh air rate based on maximum occupancy). AHU preheat coils (ECM 8): lower office AHU preheat coil set-points. Reduce AHU supply air temperature (ECM 9): reduce office AHU set-points for mixing box, cooling coil and heating coil to 15 ◦ C, 16 ◦ C and 14 ◦ C, respectively. VAV box re-evaluation (ECM 10): examine design calculations, damper min/max positions, and control settings for VAV boxes. Monitor and investigate reheat coil usage on a floor by floor basis. Computer standby (ECM 11): use network wide program to automatically synchronize computer standby settings. Internal loads (ECM 12): inform staff about shutting down unnecessary equipment and implement an A-rated equipment purchasing policy to address excessively high plug loads. Investigate high night-time plug loads. Perform a full lighting control schedule review and inform night staff how to operate the web-based control system.

VAV damper position logging

3 (med)

CO2 sensors CO2 sensors; exhaust fan VFD

3 (med) 4 (high)

ERV; ductwork; design cost Pressure logging and control

4 (high) 3 (med)

CO2 sensors; pressure sensors

2 (low)

CO2 sensors

2 (low)

N/A N/A

1 (nil) 1 (nil)

Personnel hours; flow & temp sensors

4 (high)

N/A

1 (nil)

N/A

1 (nil)

evidence is available and that the model still does not meet the acceptance criteria. Given this issue, the authors propose a further development of the methodology: to combine the evidence-based, version control approach with a sensitivity analysis approach. The analysts could develop the model according to the existing methodology. When the readily available sources of evidence are exhausted, the next step would be to assign expected ranges of variation to parameter values based on the (assumed) quality of the available evidence. The resultant parameter space can then be explored using a sensitivity analysis approach in a manner similar to that described in other case studies [19]. Such a methodology is currently under development and will be applied to a second case study building (in which hourly energy consumption and individual zone temperatures are logged) using more stringent acceptance criteria, such as those outlined above. 7. Conclusions This research details the calibration of a whole building energy model of the Intel building in Ireland using hourly, sub-utilities measured data. The methodology uses version control software to track the calibration process. This yields a complete repository of the model at each stage of the calibration process and the evidence on which the model is based. The final model represents the building to a high level of detail using a large number of zones (which were defined according to the zone-typing method) and uses internal load data measured at hourly intervals instead of day-typed schedules. The results show excellent correlation with the measured hourly electrical consumption data both for the analysed year (2007) and for the subsequent year (2008), demonstrating the effectiveness of the methodology. The paper concludes with a discussion on discrepancies remaining in the model, issues encountered related to the acceptance criteria, and recommendations for future calibration case studies. Overall, due to the amount of manual information transformation currently involved in the calibration of a model to this level of detail, the process takes considerable time and resources. The commercial use of high quality, detailed calibrated building

models is infeasible without a systematic approach to measurement in buildings and a measure of automation in the calibration process. A complete Industry Foundation Class (IFC) [8] Building Information Model (BIM) that includes accurate geometry, constructions and HVAC information linked to a structured database of measured building data may well provide this in the future. In the mean-time however, calibration case studies such as this one are a valuable way to evaluate simulation engine capabilities, to investigate modelling assumptions against real building operation, to develop best-practice modelling approaches, and to evaluate ECMs. Acknowledgements The authors thank Intel Ireland Corporation, who granted access to the building; Luke Fenner and Kevin Geoghegan for their time and invaluable assistance with data collection and Intel building audits; and Tobias Maile at Lawrence Berkeley National Laboratory for developing GST/IDFGenerator and for his advice throughout this research. The Irish Research Council for Science, Engineering and Technology (IRCSET) Embark Initiative, the Fulbright Commission in Ireland, and Enterprise Ireland funded this research. References [1] P. Raftery, M. Keane, J. O’Donnell, Calibrating whole building energy models: An evidence-based methodology, Energy and Buildings 43 (2011) 2356–2364. [2] J.S. Haberl, M. Abbas, Development of graphical indices for viewing building energy data: part 1, Journal of Solar Energy Engineering 120 (1998) 156–161. [3] P. Raftery, M. Keane, J. O’Donnell, A. Costa, Energy Monitoring Systems: value, issues and recommendations based on five case studies, in: Proceedings of the 10th REHVA World Congress, Antalya, Turkey, 2010. [4] ASHRAE, ASHRAE Guideline 14-2002: Measurement of Energy and Demand Savings, 2002. [5] TortoiseSVN Core Development Team, TortoiseSVN, 2010, http://tortoisesvn. tigris.org/. [6] D.B. Crawley, J.W. Hand, M. Kummert, B.T. Griffith, Contrasting the capabilities of building energy performance simulation programs, Building and Environment 43 (2008) 661–673. [7] R. Henninger, M. Witte, D. Crawley, Analytical and comparative testing of EnergyPlus using IEA HVAC BESTEST E100–E200 test suite, Energy and Buildings 36 (2004) 855–863. [8] International Alliance for Interoperability, IAI Tech, 2010, http://www.iaitech.org/.

P. Raftery et al. / Energy and Buildings 43 (2011) 3666–3679 [9] T. Maile, M. Fischer, V. Bazjanac, Building Energy Performance Simulation Tools – A Life-Cycle and Interoperable Perspective, Center for Integrated Facility Engineering, Stanford, CA, 2007. [10] LBNL Window& Daylighting Software – WINDOW 6 Research Version, 2010, http://windows.lbl.gov/software/window/6/index.html. [11] US DOE, Commercial Building Reference Models, 2010, http://www1.eere. energy.gov/buildings/commercial initiative/new construction.html. [12] ASHRAE, ASHRAE Handbook of Fundamentals, 2005, ISBN 13-978-1931862738. [13] US DOE, EnergyPlus Input/Output Reference v3.1, 2009, http://apps1.eere. energy.gov/buildings/energyplus/. [14] M. Hydeman, S. Taylor, J. Stein, E. Kolderup, T. Hong, Advanced Variable Air Volume System Design Guide, Technical Report: P500-03-082-A-11, Alameda, CA, Taylor Engineering, 2003.

3679

[15] M.B. Kaplan, J. McFerran, J. Jansen, R. Pratt, Reconciliation of a DOE2.1C model with monitored end-use data for a small office building, ASHRAE Transactions 96 (1990) 981–992. [16] N. Nassif, S. Moujaes, A new operating strategy for economizer dampers of VAV system, Energy and Buildings 40 (2008) 289–299. [17] T. Maile, Comparing Measured and Simulated Building Performance Data, Doctorate Thesis, Stanford University, 2010. [18] US DOE, EnergyPlus Engineering Reference v3.1, 2009, http://apps1.eere. energy.gov/buildings/energyplus/. [19] T. Reddy, I. Maor, C. Panjapornpon, Calibrating detailed building energy simulation programs with measured data—part 1: general methodology, HVAC&R Research 13 (2007) 221–241.