ASHRAE LO-09-062-2009 Evaluation of Typical Weather Year Selection Approaches for Energy Analysis of Buildings《建筑物能源分析用典型气象年选择方法的评估》.pdf

《ASHRAE LO-09-062-2009 Evaluation of Typical Weather Year Selection Approaches for Energy Analysis of Buildings《建筑物能源分析用典型气象年选择方法的评估》.pdf》由会员分享，可在线阅读，更多相关《ASHRAE LO-09-062-2009 Evaluation of Typical Weather Year Selection Approaches for Energy Analysis of Buildings《建筑物能源分析用典型气象年选择方法的评估》.pdf（14页珍藏版）》请在麦多课文档分享上搜索。

1、654 2009 ASHRAEThis paper is based on findings resulting from ASHRAE Research Project RP-1477.ABSTRACTIn this paper, the results a series of sensitivity analyses are presented and discussed to assess the impact of the selection methodology to generate a typical weather year suitable for energy analy

2、sis of building systems. In particular, the impact of weighting factors associated with various weather variables was investigated. Moreover, the effect of the length of recorded data used to generate the typical year was evaluated. The anal-ysis is carried out for 10 US sites for which measured wea

3、ther data are reported for at least 30 years. The results indicate that using the TMY2 selection approach it was better to assign more weight to global solar radiation than to direct normal radia-tion and that 15 years of recorded data would be sufficient to generate a typical weather year.INTRODUCT

4、IONAs detailed building energy simulation is becoming a common practice in the design and the evaluation of building energy projects, there is an increasing need to develop and format weather data suitable for whole-building energy anal-ysis tools. Weather data used for detailed simulation energy an

5、alysis include hourly values of dry-bulb temperature, dew-point temperature, solar radiation, and wind speed and direc-tion. Several selection approaches do exist to develop typical weather data using a single year of hourly data that are selected to represent the average weather patterns that can b

6、e found in a multi-year data set (Keeble, 1990). In the US, several approaches have been proposed to select a typical weather year for building energy analysis including the ASHRAE Test Reference Year or TRY (ASHRAE, 1976), Typical Meteorological Year or TMY (Hall et al., 1978), the Weather Year for

7、 Energy Calculations (Crow, 1981), and TMY2 (Marion and Urban, 1995), and more recently TMY3 (Wilcox and Marion, 2008).Other similar approaches and methods for developing typical weather data have been proposed in several other coun-tries (Lund and Eidorff, 1980; Pissimanis et al., 1988; Festa and R

8、atto, 1993; Mosalam and Tadros, 1994; Hui, 1996; Petra-kis et al., 1998; Gazela and Mathioulakis, 2001; ISO-15927-4, 2005). For instance, Pissimanis et al. (1988) generated TMY weather file for Athens, Greece using 17 years (1966 to 1982) standard meteorological data and measured global solar radiat

9、ion obtained by National Observatory of Athens. The authors used the TMY generation method described in Hall et al. (1978). They noted that the final TMY selection of the typi-cal year is rather subjective due to the large number of statis-tical parameters that need to be considered. They suggested

10、a more straightforward selection procedure using the monthly RMSE (Root Mean Square Error) values of mean hourly solar global horizontal irradiance (GHI) for each month associated with five candidate years. Hui (1996) generated a TRY weather file and a TMY2 weather file for Hong Kong using 16 years

11、(1977 to 1994) of standard meteorological data. He noted the challenges of using the TMY2 method to generate typical weather year due to the lack of a specific criteria set for the selection procedure. He attempted to utilize Kolmogorov-Smirnov statistics (Stuart et al., 1999) instead Finkelstein-Sc

12、hafer (1971) statistics using daily and hourly values instead of daily and monthly values.In this paper, a series of sensitivity analyses is presented to assess the impact of the selection methodology on generating a Evaluation of Typical Weather Year Selection Approaches for Energy Analysis of Buil

13、dingsDonghyun Seo Joe Huang Moncef Krarti, PhD, PEStudent Member ASHRAE Member ASHRAE Member ASHRAEDonghyun Seo is a graduate student and Moncef Krarti is a professor and associate chair in the Civil, Environmental, and Architectural Engi-neering Department, University of Colorado, Boulder, CO. Joe

14、Huang is a principal at White Box Technologies Inc., Moraga, CA.LO-09-062 (RP-1477) 2009, American Society of Heating, Refrigerating and Air-Conditioning Engineers, Inc. (www.ashrae.org). Published in ASHRAE Transactions 2009, vol. 115, part 2. For personal use only. Additional reproduction, distrib

15、ution, or transmission in either print or digital form is not permitted without ASHRAEs prior written permission.ASHRAE Transactions 655typical weather year suitable for energy analysis of building systems. In particular, the impact of weighting factors as well the impact of the length of historical

16、 data used. The analysis was carried out for 10 US sites for which measured weather data for at least 30 years are reported.TMY-BASED SELECTION PROCEDUREThe TMY weather data are developed based on the Sandia selection method (Hall et al. 1978). This method uses 9 daily indices, i, associated to five

17、 weather parameters to select specific months over a multi-year set to form a typical year:Max/Min/Mean of dry-bulb temperature (DBT)Max/Min/Mean of dew-point temperature (DPT)Max/Mean Wind Speed (WSP)Global Horizontal Solar Irradiation (GHI)Direct Normal Irradiance (DNI)Seven steps are used to sele

18、ct typical months to form the TMY weather data. These steps are summarized below:1. Compare monthly CDFs with long-term CDFs obtained through Finkelstein-Schafer Statistics:For each index, i:(1)wherei,k= absolute difference between the long-term CDF and the candidate month CDFn = the number of daily

19、 readings in a month2. Calculate the weighted sum (WS) of the FS statistics:(2)wherewi= weighting for index iFSi= F-S statistic for index i3. Rank with respect to closeness of the month to the long-term mean and median to select 5 candidate months4. Check for persistence of mean DBT and daily GHI to

20、 determine the frequency and run length above and below fixed long-term percentile5. Select the typical meteorological month (TMM) using the highest ranked month from process 3 and persistence criteria6. Repeat the process for the other months comprising the year7. Assemble the TMMs and smooth the t

21、ransition data between all the TMMs using curve-fitting techniques.The TMY selection approach summarized above is utilized to generate other typical weather year data sets includ-ing TMY2 and International Weather for Energy Calculations or IWEC weather files (Thevenard and Brunger, 2002a and 2002b)

22、. The main difference between the various selection approaches is the weighting factors used in estimating the FS statistics as outlined in Equation (2). For the IWEC selection approach, the persistence criterion is not utilized. Moreover, for the IWEC weather data sets, both dry-bulb temperature (D

23、BT) and solar radiation (GHI) have an equal weighting factor of 40%. While IWEC uses the same nine indices used for TMY, TMY2 adds a 10th index: the Direct Normal Irra-diance (DNI). Table 1 summarizes the weighting factors utilized by each selection approach. While not common, the Equal weighting fa

24、ctor (EqWt) was included in this study for comparative analysis.STATISTICAL ANALYSIS METHODSIn this section, an overview is provided to describe some statistical indicators used to assess the performance of various selection approaches to generate a TMY2 weather file for a given site.t-StatisticsThe

25、 statistic indicator allows data sets to be compared and to determine whether or not a models estimates are statis-tically different from a reference data set at a desired confi-dence level (Stone, 1993). t-statistics can be computed using both the root mean square error (RMSE) and mean bias error (

26、MBE) and takes into account the dispersion of the results which is neglected when the root mean square error and mean bias error are considered separately. The t-statistic is given by Equation (3):(3)FSi1 n()ik,k=1n=WS wiFSii=110=Table 1. Weighting Factors with Respect toWeather Data TypeIWEC EqWt T

27、MY TMY2DBTmin 1/20 1/20 1/24 1/20DBTmax 1/20 1/20 1/24 1/20DBTmean 6/20 3/20 2/24 2/20DPTmin 0.5/20 1/20 1/24 1/20DPTmax 0.5/20 1/20 1/24 1/20DPTmean 1/20 3/20 2/24 2/20WSmax 1/20 2.5/20 2/24 1/20WSmean 1/20 2.5/20 2/24 1/20GHI 8/20 5/20 12/24 5/20DNI 5/20t1 n()dii=1nS / n-=656 ASHRAE Transactionswh

28、eren = degree of freedom (n = 12 months in this analysis),di= difference between calculated data mean and reference data mean,S = unbiased standard deviation of the differences between estimated and measured values.The unbiased standard deviation can be calculated from Equation (4):(4)RMSE and MBE a

29、re defined as follows:(5)(6)Combining Equation (6) and Equation (3) gives,(7)And using Equations (5) and (6) in conjunction with Equation (4) gives,(8)Substituting for S in Equation (7) gives,(9)The smaller the value of t, the better is the models perfor-mance. To determine whether estimates are sta

30、tistically signif-icant, one has to determine a critical t-statistic obtainable from standard statistical tables; i.e., t/2at the level of significance, , and n 1 degrees of freedom. For example, the critical t-statistic used in this study is typically 1.796 corresponding to 11 degree of freedom (n

31、= 12 months) and a 95% significant level. For the models estimates to be judged statistically significant at the 1 confidence level, the calculated t-statistic using Equation (9) must be less than the critical value.Weighted Sum t-statisticsThe comparison process of the two data sets in this study i

32、nvolves 5 weather parameters. Generally, some parameters may satisfy the degree-of-fit criteria using the t-statics provided by Equation (9) while some parameters may not. To make an objective decision on the goodness of fit between two data sets involving the 5 parameters, a weighted sum t-statisti

33、c was developed. This value was calculated by multiplying the five parameters with a set of weighting factors, the same factors that are used for the selection the typical meteorolog-ical months (TMMs). The weighted t-statistic was obtained using the same principle with the weighted sum F-S value ob

34、tained in Equation (2):WStstatistics= w1 DBTt+ w2 DPTt+ w3 GHIt+ w4 DHIt+ w5 WSPt(10)Where, w1 through w5: weights of each variable, sum of all weights should be 1 (refer to Table 1).ANALYSIS RESULTSIn order to verify whether or not the TMY2 weather files are representative of long-term weather patt

35、erns, a 30-year average weather data is computed and is utilized as a reference long-term average data set throughout the analysis performed in this study. The 30-year average data were estimated by taking the 30-year means of all hourly parameters. The t-statistics were used for the comparative ana

36、lysis to assess the differences between any two weather files. In the analysis presented in this paper, 10 weather stations representing vari-ous US climates were used.First, the data used in the analysis are presented. Then, the results of the analysis are summarized and discussed in the following

37、sections.Data Used in the AnalysisIn this study, 10 US stations (referred to as primary stations in the National Solar Radiation Data Base or NSDB) that have measured solar radiation data were selected. The 10 sites represent a wide range of climatic zones in US and are summarized in Table 2. The NS

38、RDB data have 30 years (1961-1990) of quality controlled weather and solar data and cover many sites throughout the US. The climatic zone classification provided in Table 2 is based on the definition provided in ASHRAE Standard 90.1 (2004).Comparison Between Various Typical Weather Year Types Table

39、3 summarizes the selected years associated with the 12 months for four typical weather year data sets (IWEC, TMY, TMY2, and EqWT) and for the 10 US sites. The selected typical months are generally different for the four weather data types (which have different weighting factors) on 1 to 6 months out

40、 of 12 months except for Madison where 9 months have been selected differently depending on the weather data format type used.Table 4 shows the determination coefficient, R2, values of monthly mean of weather parameters against 30-year average data obtained from NSRDB. The coefficient R2is defined a

41、s R2= 1-SSE/SST where SSE is the squared sum of errors and SST is squared sum of variances (square of differences between sample mean and sample values). S2ndi2i=1ndii=1n2nn 1()-=nRMSE2di2i=1n=nMBE dii=1n=tMBES / n-=SnRMSE2MBE2()n 1()-=tn 1()MBE2RMSE2MBE2-=ASHRAE Transactions 657Regardless of typica

42、l weather selection methodology, the results of Table 4 show good correlation with the 30-year aver-age reference data for all parameters except for the direct normal irradiance (DNI) for which the correlation between the monthly means obtained for the four weather format types and the 30-year month

43、ly means is generally weak with R2values as low as 0.322.It should be noted that only TMY2 provides a weighting factor for DNI. However, based on the results of Table 4, the R2values associated with DNI for TMY2 against 30-year average year are generally low. This result is most likely attributed to

44、 the fact that there is already a strong corre-lation between the global solar radiations with direct normal. Thus, there is no need generally to assign an additional weight factor associated with DNI.For a comparative evaluation between various weather data format types, the weighted sum of t-stati

45、stics were calcu-lated for each typical weather data type using the 30-year aver-age as reference. The t-statistics obtained for four weather parameters (DBT, DPT, GHI, and DNI) are shown in Figure 1 for all 10 US sites and for the four weather data format types(IWEC, EqWt, TMY, and TMY2). All typic

46、al weather format types show good agreement with the reference data. It should be noted that the TMY2 data set does not exhibit consistently better performance compared to the other weather format types even though it has an additional weighting factor asso-ciated with DNI.In Figure 1, the t-statist

47、ics values obtained for some parameters such as dry bulb temperature for the IWEC data in Atlanta, dew point temperature for the TMY2 data in Las Vegas, and dew point temperature for the EqWt data in Madi-son were uncharacteristically high compared to the t-statistics for the other parameters. These

48、 results were attributed to the fact that high mean bias errors (MBE) or low root mean square errors (RMSE) are obtained causing high t-statistic values. There is always some probability that the t-statistics of one parameter was high even when t-statistics of the other param-eters were small. For i

49、nstance, for the dew point temperature of TMY2 data in Las Vegas, year 1976 is selected for typical weather year for the month of April due to the lowest F-Svalue. The F-S value for the dew point temperature (DPT) however for 1976 is 0.01212 as compared to 0.0070 for April of 1992. Impact of Weighting Factors on TMY2 DataHall et al. (1978) indicated that the selection of weighti