ASHRAE 4738-2004 Proposed Tools and Capabilities for Proactive Multi-Building Load Management Part 2 - Aggregated Operation《为积极主动的多建设负荷管理的建议工具和能力 第2部分-汇总运作RP-1146》.pdf

《ASHRAE 4738-2004 Proposed Tools and Capabilities for Proactive Multi-Building Load Management Part 2 - Aggregated Operation《为积极主动的多建设负荷管理的建议工具和能力 第2部分-汇总运作RP-1146》.pdf》由会员分享，可在线阅读，更多相关《ASHRAE 4738-2004 Proposed Tools and Capabilities for Proactive Multi-Building Load Management Part 2 - Aggregated Operation《为积极主动的多建设负荷管理的建议工具和能力 第2部分-汇总运作RP-1146》.pdf（14页珍藏版）》请在麦多课文档分享上搜索。

1、4738 (RP-1146) Proposed Tools and Capabilities for Proactive Multi-Building Load Management: Part 2-Aggregated Operation Leslie K. Norford, Ph.D. Member ASHRAE ABSTRACT ASHRAE Research Project 1146, “Building Operation and Dynamics Within an Aggregated Load,” was meant to (a) identzfi situations und

2、er which aggregating individual build- ing loads is attractive for managing total, multi-building loads and (b) identi to simultaneously forecast the - Les Norford is an affiliate of Tabors Caramanis and Associates and professor of Building Technology in the Department of Architecture, MIT, Cambridg

3、e, Mass. Agami Reddy is an associate professor in the Civil and Architectural Engineering Department, Drexel University, Phila- delphia, Fenn. 02004 ASHRAE. c 457 Tod I: Customer Pre- screening Tool9: Tool 1 O: Interaction with Monitoring and to estimate the impact of measures that have the potentia

4、l to reduce the controllable loads. The term “building load” focuses on electrical loads in this study, but thermal loads directly impact electrical loads and influence thermal comfort, which constrains load reduc- tions. A detailed literature review of building load models was compiled by Reddy et

5、al. (1998a). Load-forecasting methods can be classified as follows: 1. Semi-empirical. The aggregator has an empirical estimate of demand increase, for example a kWJC metric for each building and day type. Whether such rules of thumb allow end-use to be predicted with the required degree of accu- ra

6、cy is uncertain. 2. Statistical/adaptive models fim historic data. If measured demand data are available for a year, which is usually not the case for end-use loads, one could develop statistical models for different types of days. Studies in the past (Fels 1986; Kissock et al. 1998; Katipamula et a

7、l. 1998) indicated that the outdoor dry-bulb temperature is the most important regressor variable, at monthly and even at daily time scales. Classical linear functions are not appropriate for describing energy use in many buildings because of the presence of functional discontinuities, called “chang

8、e points.” These change-points are caused by HVAC operating and control algorithms and schedules, including economizer cycles (Reddy et al. 1998b). 3. Simulation-based. The building simulation approach adopts an engineering simulation model and “tunes” the inputs of the program so that simulated out

9、put and measured values of building energy use match closely. A simulation program thus calibrated could then serve as a more reliable means of predicting the energy use of the building when operated under different climatic or different pre-specified operating conditions. One can distinguish betwee

10、n two different types of engineering simulation models: “detailed,” general-purpose, fixed-schematic models such as DOE-2 (Norford et al. 1989; Bronson et al. 1992; Bou-Saada 1994), and BLAST (Manke et al. 1996) or “simplified,” fixed-schematic HVAC system models based on the air-side models develop

11、ed by ASHRAE TC 4.7 (Knebel 1983) and adopted in slightly different forms by many workers (Katipamula and Claridge 1993; Liu and Claridge 1995). Both the detailed and the simplified cali- brated model approaches have yet to reach a stage of matu- rity in methodology development where they can be use

12、d routinely and with confidence by people other than skilled analysts. Whole-Building Thermal and Electrical Load Models Few papers describe on-line models for building thermal loads. MacArthur et ai. (1 989) presented results for a recursive time series model. Kawashima et al. (1995) evaluated auto

13、re- gressive integrated moving average (ARIMA), exponentially weighted moving average (EWMA), ordinary regression, and artificial neural network (ANN) models and found ANN to be the most accurate. Henze et al. (1997) considered various mathematical forms to predict thermal loads and assess their imp

14、act on the performance of a controller for thermal storage systems. Their load models included an unbiased random walk, a bin predictor model, a harmonic predictor model, and an autoregressive network predictor model. Katipamula and Brambley (2003) switched from a neural net to a set of time series,

15、 binned by temperature, to predict whole-building load as part of a diagnostics tool. Daryanian et al. (1994) developed a two-step online procedure for forecasting the day-ahead hourly cooling load. First, the total load for the next day was estimated on the basis of the forecasted average outdoor t

16、emperature, the total load for the previous day, and the day type (weekday or weekend). Second, the total load was distrib- uted among the 24 hours on the basis of historical load-distri- bution percentages. Regression analyses showed that outdoor temperature accounted for about 80% of the variation

17、 in load and that the use of three independent variables (temperature, previous load, and day type) produced a correlation coefficient (R2) of 0.95. Forty days of data were used to establish the hourly load shapes. Electric utilities monitor the whole building loads of most of their larger customers

18、 at 15-minute or 30-minute intervals. It would be advantageous for proactive load aggregators to make use of this rich source of information. Akbari (1995) showed that such data could be used to understand customer 458 ASHRAE Transactions: Research patterns as well as separate the effects of weather

19、-dependent and weather-independent effects, both on an individual customer as well as customer-class basis. Forrester and Wepfer (1 984) used multiple linear regression to develop a load prediction algorithm for the whole-building electricity use of a large commercial building. The algorithm allowed

20、 summer energy and peak use to be predicted up to four hours in advance with an accuracy of2.5%. Seem and Braun (1991) reviewed deterministic (including polynomial, exponential, and sinusoidal functions) and stochastic (including autore- gressive and autoregressive moving average) time-series models

21、. They described a Cerebellar Model Articulation Controller (CMAC) to forecast electricity demand, relying on the EWMA method to update a lookup table to map system inputs and outputs. They noted that stochastic time series methods could be used to model the difference between a time series and a de

22、terministic model forthat time series. They then combined a deterministic and a stochastic model and adap- tively determined the three autoregression parameters used in the stochastic model. Electricity data gathered from a grocery store and a restaurant were used to demonstrate the accuracy and rob

23、ustness of the algorithm. Efforts have been made to estimate end-use loads from whole-building measurements: 1. Econometric modeling (Usoro and Schick 1986). The objective was to develop and demonstrate new methods for estimating load shapes for residential end uses by disaggre- gating metered whole

24、-house data. Hourly data for a year were obtained from 125 utility customers. At the first of two analysis levels, 60-70 parameters characterizing daily, weekly, seasonal, and weather-sensitive patterns of the load were extracted. At the second level, cross-sectional regres- sions measured the influ

25、ence of household demographics and appliance ownership. Algorithmic approaches such as the statistically adjusted engineering (SAE) method and the end-use disaggregation algorithm (EDA). EDA (Akbari 1995) used hourly whole- building data along with audit information about the build- ing and certain

26、physical constraints to produce hourly load profiles for air conditioning, lighting, fans and pumps, and miscellaneous loads. This approach was applied to two buildings (office and retail) with an average error of less than 5% during daytime operation. 3. The signal processing approach developed by

27、MIT researchers (Norford and Leeb 1996; Lu0 et al. 2002) where rapid sensing of whole-building electrical use along with sophisticated processing techiques allow individual loads to be detected with reasonable accuracy. 2. Component Models Measurement techniques and calculations are detailed in ASHR

28、AE, AMCA, and ASME standards for air-handling- unit fans and in ASME and the Hydraulics Institute standards for pumps, as described by Phelan et al. (1 997a). For constant- volume airflow systems, a one-time power measurement and a knowledge of the operating schedule are sufficient. For VAV systems,

29、 Phelan et al. studied the ability of linear and quadratic models to predict electricity use as a function of mass flow rate and concluded that although quadratic models are superior to linear models in predicting energy use, the linear model seemed to be the better overall predictor of both energy

30、and demand (i.e., maximum monthly power consumed by the fan). For variable-flow water systems, achieved with either a throt- tling valve or a variable-speed drive, Phelan et al. concluded that quadratic models were superior to linear models in predicting electricity use as a function of mass flow ra

31、te. Polynomial and thermodynamic models can describe in- situ chiller performance. Polynomial models correlate chiller (or evaporator) thermal cooling capacity or load and the elec- trical power consumed by the chiller (or compressor). For example, based on functional forms in the DOE-2 building sim

32、ulation software (LBL 1980), electricity use can be modeled as a tri-quadratic polynomial model. This model has 1 1 parameters to identify. All of them are unlikely to be statis- tically significant and a step-wise regression with the sample data set yields the optimal set of parameters to retain. B

33、raun (1 992) used a bi-quadratic model with two regressor variables containing six empirical coefficients, namely, cooling load on the chiller and the difference between the ambient wet-bulb temperature and the fluid temperature leaving the evaporator (or the supply temperature to the building). Oth

34、ers (for exam- ple, Hydeman 1997) have proposed slightly different variants of such polynomial models. Thermodynamic models, recommended by Phelan et al. (1997b), are preferred because they generally have fewer model parameters and those parameters, based on physical principles, tend to be more robu

35、st. The model proposed by Gordon and Ng (2000) correlates the chiller COP (the ratio of chiller thermal cooling capacity to the electrical power consumed by the compressor) with the easily measurable fluid inlet temperature to the condenser, the fluid temperature leav- ing the evaporator, and the th

36、ermal cooling capacity of the evaporator. The complete model has three parameters that are identified by multiple linear regression. This model and poly- nomial models have been studied in detail by Reddy and Andersen (2002) and Jiang and Reddy (2003). Available Software Private load consultants, lo

37、ad aggregators, and utilities have developed and have been using software tools that pertain to the load-forecasting tool. Although many of these tools are proprietary, a report by EPRI (Ried 1987) summarizes the capabilities of numerous tools by functionality. Software pertinent to Tool 5 can be di

38、vided into three groups: load data analysis software that allows processing and analyzing large amounts of monitored data from a particular facility or build- ings; tools that evaluate numerous load management technol- ogies along with their long-term financial implications for a specific customer;

39、and probabilistic and scenario-based fore- ASHRAE Transactions: Research 459 casting tools that assess and predict risks associated with fore- casting uncertainties and decomposing sources of errors. PROPOSED TOOL 6: COMFORT PENALTY FRAMING The objective of this proposed tool is to provide a proce-

40、dure to modi the indoor environment (specifically by allow- ing the indoor dry-bulb temperature to increase in a predetermined and controlled manner) so as to reduce electric demand at the expense of predefined overall occupant comfort. This differs from the approach described in numer- ous studies

41、(for example, Braun et al. 2001; Braun and Chaturvedi 2002), which involves pre-cooling the building and making use of the building thermal heat capacity to shave demand peaks without compromising occupant comfort. The basic issues involve: (1) formulating the strategy in terms of being able to math

42、ematically quanti9 varying degrees of occupant discomfort associated with specific indoor environmental changes given the fuzziness surround- ing human comfort modeling; (2) assessing whether the tech- nical capability of present day HVAC systems and their associated controls permits such a strategy

43、 to be implemented practically; and (3) determining whether the building-specific data needed for this tool can be collected for specific circum- stances. The procedure may seem drastic and unacceptable when electricity prices are moderately low but may be more attractive when demand charges are ver

44、y high. The semi-empirical model based on the Predicted Mean Vote (PMV) index proposed by Fanger (1972) and described in ASHRAE (2001) represents an average value of a large group of people. Dissatisfied occupants are likely to complain, and Fanger (1972) suggested another index called the PPD (Pred

45、icted Percentage of Dissatisfied occupants), which he related to PMV. In an office setting, air temperature and humidity are the two main variables for the physical comfort model. Gagge et al. (1986) proposed an extension of the PMV method that is more accurate under sweating conditions. Berglund (1

46、989) found that temperature had an order of magnitude greater effect on human comfort than did humidity in determining human comfort. He also found that subjects indicated that equal changes in humidity are more perceptible at higher humidity levels than at lower humidities. Load management is likel

47、y to be implemented during hot summer days, when high humidity levels and their adverse impacts on human comfort will be a major issue. An excellent review and results of recent climate chamber tests of subjects exposed to high humidity levels have been provided by Fountain et al. (1 999). Suggested

48、 Strategy and Data Requirements The portfolio of the aggregator includes several buildings. Within each building, there are several control panels and circuit loops in each control panel. A sample chart is shown in Figure 2. The practical implementation of this tool involves first creating a knowled

49、ge table like that shown in Table 1. 460 a1 N 81. 03 r &;i 4 . . . . , . . . . . . . I l Figure 2 Schematic illustrating the type of disaggregation needed for the Load Optimization Tool. Each circuit loop is designated by a code, an index number, and the type of load. The load-reduction potential (in kW) of each loop is determined by either measurement or best-guess estimates. Note that ordoff (O/O) equipment has one number, while the ramped (R) loads associated with HVAC equipment and the dynamic (D) loads associated with thermal storage can take on numerical values from O to a ma