1、4844 (RP- 1 05 1 ) Literature Review on Calibration of Building Energy Simulation Programs: Uses, Problems, Procedures, Uncertainty, and Tools T. Agami Reddy, PhD, PE Member ASHRAE ABSTRACT Calibrated simulation is the process of using an existing building simulation computer program and ?tuning or
2、cali- brating the various inputs to the program so that observed energy use matches closely with that predicted by the simula- tion program. The two primary reasons for adopting this approach is that it allows (1) more reliable identijcation of energy savings and demand-reduction measures (involving
3、 equipment, operation, and/or control changes) in an existing building and (2) increased confidence in the monitoring und verification process once these measures are implemented. Historically, the calibration process has been an art form that inevitably relies on user knowledge, past experience, st
4、atisti- cal expertise, engineeringjudgment, and an abundance of trial and error: Despite widespread interest in the professional community, unfortunately no consensus guidelines have been published on how to perform a simulated calibration using detailed simulation programs. ASHAE initiated a resear
5、ch project (?-1 051) intended to cull the best tools, techniques, approaches, andprocedures from the existing body of research and develop a coherent and systematic calibration methodol- ogy that includes both ?parameter estimation ? and determi- nation of the uncertainty in the calibrated simulatio
6、n. This paper provides a pertinent and detailed literature review of calibrated simulation techniques, describing their strengths, weaknesses, and applicability, thus serving as a precursor to reporting the results of the research project in subsequent papers. BACKGROUND Calibration as Part of ECM I
7、dentification and Monitoring and Verification The oil shock of 1973 triggered a flurry of energy conser- vation activities especially by federal and state agencies. This led to the widespread initiation of demand-side management (DSM) projects especially targeted to residential and small commercial
8、building stock. Subsequently, in the 1980s, build- ing professionals started becoming aware of the potential and magnitude of energy conservation savings in large buildings (office, commercial, hospitals, retail, etc.). DSM measures implemented included any retrofit or operations practice, usually s
9、ome sort of passive load curtailment measure during the peak hours such as installing thermal storage systems, retrofits to save energy (such as delamping, energy efficient lamping, changing constant air volume CV systems into variable air volume VAV, demand meters in certain equip- ment such as chi
10、llers, and energy management and control systems EMCS for lighting load management). The drastic spurt in activity by energy service companies (ESCOs) led to numerous papers being published in this area (such as that by Schuldt and Romberger 19981 and various other publications reviewed later in thi
11、s paper) and standard development efforts by organizations such as ASHRAE (2002) and USDOE (IPMVP 2001). During the last few years, electric market transformation and utility deregulation have led to a new thinking toward more proactive load management of single and multiple buildings (Reddy andNorf
12、ord 2004; Norford and Reddy 2004). Though several facilities do implement such proactive load management practices, these are yet to achieve a sufficient level of maturity at this time. T. Agami Reddy is a professor in the Civil, Architectural, and Environmental Engineering Department, Drexel Univer
13、sity, Philadelphia, Pa. 226 02006 ASHRAE. The proper implementation of DSM measures involved first the identification of the appropriate energy conservation measures (ECMs), and then assessing their impact or perfor- mance once implemented. This need resulted in monitoring and verification (M (b) th
14、at on an annual basis, these indices were 1 1 .O% and 19.0%, respectively, for gashel and 9.2% and 15.0%, respectively, for electricity use; and (c) that on a monthly basis, these indices were 16.7% and 24.0% for total energyuse, 26.3% and35.0% forgasfuel use, and 18.7% and 30.0% for electricity use
15、, These values are higher than those recommended in ASHRAE Guideline 14, but it must be realized that this study was conducted about 25 years ago at the dawn of the building simulation era and did not require any submetering whatsoever. The results of another USDOE-funded study a few years later are
16、 documented in a report by TRC (1984). Only one large office building was selected, detailed monitoring was done to various end-uses, and the accuracy of the DOE-2 program algorithms was evaluated at hourly time scales for each of the four seasonal periods for several components, such as chillers, w
17、ater pumps, cooling towers, boilers, and second- ary system components. Measurement errors of the sensors were also accounted for. It was found that monthly predicted total energy use was lower by about 5% on average over the year as compared to measured energy use (though individual months showed m
18、uch greater variability: underprediction in July by 6% and overprediction in March by 12%). Values of such differences for each of the equipment are also specified. It must be noted that these results are building- and location- specific. The authors conclude that the ability of the DOE-2 simulation
19、 program to predict building energy use is within the accuracy of empirical measurements of building energy use. The approach adopted by Kaplan et al. (1990a, 1990b) in the framework of small ofice buildings within the Energy Edge program is to monitor several end-uses during short peri- ods and to
20、perform the calibration for these periods only as against a whole year. The short tuningperiods recommended are one month during a hot period, one month during a cold period, and one month in between. The authors studied shorter periods of a week but concluded that one-month intervals ASHRAE Transac
21、tions: Research 229 tended to better smooth variability. In general, the tuning process first corrected obvious simulation errors highlighted by the discrepancies, then adjusted for internal loads, and, finally, adjusted other inputs such as HVAC end-uses. Tuning tolerances of about 10% for whole-bu
22、ilding energy use on an annual basis were achieved. Hunn et al. (1992) describe a study using as-built draw- ings, site interviews, and whole-building electric use data for the Texas Capitol Building to calibrate DOE-2 to the pre-reno- vation energy use status. Energy savings due to the extensive re
23、novations could then be identified using post-renovation data. A similar methodology was adopted by Reddy et al. (1994) for a 250,000 fi2 university building in Austin using two months of pre-retrofit monitoring and site visits to identify the most significant end-use components. They were able to a
24、chieve calibration tolerances of 4.5% (monitored data were lower) in whole-building energy use for the entire seven months of pre-retrofit period and 2.8% for heating energy use. However, large differences in cooling energy during certain months (up to 10%) were detected. Norford et al. (1994) prese
25、nt results of a case study in a large ofice building in New Jersey with a rich variety of architectural and engineering features designed to minimize energy use. The individual effects of a large number of variables, such as power density and schedule of lights and office equipment, HVAC schedules a
26、nd thermostat settings, and HVAC and building shell perfor- mance, were studied. The study reiterates the need for prop- erly accounting for internal loads and recommends that one- time measurements be adopted for reliable calibration. Finally, Lunneberg (1999) also points out the critical impor- ta
27、nce of monitoring key short-term end-use internal loads so that more realistic schedules of the building can be obtained. The results of simulated calibration to a 49,000 ft2 commercial office building in San Diego are presented. Pedrini ct al. (2002) describe a methodology involving three steps: us
28、ing as-built drawings, walk-through visits, and electric and thermal measurements. The methodology has been used to calibrate the DOE-2 simulation program with monitored data from about a dozen office buildings. They found that the first step of calibration usually results in large unacceptable diff
29、erences (up to 20% at monthly time scales). Next, schedule adjustments are made by monitoring certain key end-uses for a few days. They find that this usually reduces the errors to about 5% to 6%. Next a walk-through audit is conducted where spot measurements of quantities such as lighting levels, a
30、irflow, air temperature, and ordoff status of power circuits are made with handheld instruments. The mean differences can thus be brought down to less than 1%. The final calibration step involves using end-use data measure- ments for more accurate calibration. The paper presents monthly and diurnal
31、plots to illustrate how the calibration process is gradually improved. Yoon et al. (2003) present a seven-step methodology for calibrated simulation: (a) base case modeling that involves gathering building data, utility bills, weather data, and as-built drawings as well as carefully considering buil
32、ding zoning; (b) using monitored data of several end-uses during a week, analyzing differences between simulations and measurements in the base load (i.e., weather-independent gas and electric use) using scatter plots with outdoor temperature; (c) fine tuning a simulation model during the swing seas
33、on period when heating and cooling energy use is low; (d) performing additional site visits and interviews to refine lighting power densities, equipment quan- tities, schedules, number of occupants; (e) calibrating the heat- ing/cooling season; (f) evaluating accuracy of the calibration by calculati
34、ng statistical indices such as normalized mean bias error and coefficient of variation values as well as using graph- ical plots such as scatter plot and box and whisker plots; and (g) evaluating the effect of promising ECMs. The results of applying this methodology to a large 26-story commercial bu
35、ilding of 83,000 m2 in downtown Seoul, South Korea, are also presented. Calibration Based on a Suite of Informative Graphical Comparative Displays This approach is described in several papers by Haberl and colleagues (Bronson et al. 1992; Haberl et al. 1993a, 1993b, 1993c; Bou-SaadaandHaberl1995a, 1
36、995b; Haberlet al. 1996; Haberl and Abbas 1998a, 1998b; Haberl and Bou- Saada 1998). The issue of determining where errors exist between measured and simulated performance is simple when viewed at monthly or diurnal plots. When calibration is being performed with hourly data, the analyst is overwhel
37、med with too many data points and is at a loss to determine exactly where the differences exist. Hence, along with the manual, iterative, and pragmatic approach to calibration described in the previous section, certain types of visual graphics can high- light differences and be useful aids to the an
38、alyst in deciding which parameters to calibrate or “tweak“ for the next iteration. Such graphical plots cannot be generated with spreadsheet programs and, hence, there is a need for special toolkits. Common plots include carpet plots, three-dimensional time- series plots of energy use and residuals,
39、 superposed and juxta- posed binned box, and whisker and mean (BWM) plots in addition to the standard two-dimensional plots, such as scatter plots and time-series plots. Other practitioners have also developed similar graphics capability (e.g., McCray et al. 1995a, 1995b1). Calibration Based on Spec
40、ial Tests and Analytical Procedures In this category, we include a compendium of specialized approaches as described below. Intrusive Blink-Tests. In case resources are limited, “blink“ tests or ordoff tests can be used to determine snapshot end-use measurements of lighting and motor control center
41、electricity use. Blink tests can be performed over a weekend using a data recorder to record whole-building electricity use. A series of tests are then performed whereby groups of end- use loads are turned on and off in a controlled sequence, and their incremental power readings provide the necessar
42、y end- 230 ASHRAE Transactions: Research use information. Data are usually recorded at one- to five- minute intervals. Soebarto (1997) presents the results of two case studies (one a university campus building consisting of laboratories, offices, and classrooms and the other a municipal building) us
43、ing monthly utility bills in conjunction with site visits, building drawings, and hourly monitored whole-building electricity use and odoff “blink tests.” The paper showed that only two-to-four weeks of monitoring at any period of the year was enough to provide adequate accuracy: coefficient of vari
44、- ation (CV) of 6.7% for hourly whole-building electricity and within 1% for chilled water use for the university building. Shonder et al. (1 998) suggest the use of detailed calibra- tion to estimate the effect of various ECMs in types of campus buildings such as large military housing, federally s
45、ubsidized low-income housing, and planned communities such as condo- miniums, townhouses, and senior centers. The approach has been illustrated for a large army base in Louisiana consisting of 46 buildings and 200 individual apartments. Pre-retrofit data were used to calibrate TRNSYS (SEL 2000), whi
46、ch was then used to predict savings, subsequently found to be within 5% of the measured post-retrofit energy use. The type of ECMs modeled involved replacement of air-source heat pumps with geothermaj heat pumps, installation of low-flow shower heads, and lighting retrofits (delamping and compact li
47、ght fixtures). The authors also found that only six months of data were adequate to calibrate the TRNSYS model satisfactorily. STEM Tests. These include a protocol of short-term energy monitoring (STEM) tests along with a systematic approach of reconciling differences between measured and simulated
48、data. The origin of this method was that developed by Subbarao (1988) for residential buildings that was extended to small-medium commercial buildings by Manke and Hittle (1 996). The basic approach in both methods is simi- lar in that it involves intrusive and controlled heating and cool- ing tests
49、 during a three- to five-day period. For smaller buildings portable heaters could be used, while for larger buildings the building heating and cooling system could be used. The test protocol consists of a period during which temperature is controlled to be constant over time and uniform over space and another period during which temperatures were allowed to float. The coheating period is intended to provide a good estimate of the building loss coefficient during nighttime when heat flows other than through the shell are negligible. The cooldown period is intended to provid
