建筑專業(yè)外文翻譯---用智能數(shù)據(jù)分析檢測建筑物中的能源異常消耗

1、<p><b>　　中文5120字</b></p><p>　　Energy and Buildings 39 (2007) 52–58</p><p>　　www.elsevier.com/locate/enbuild</p><p>　　Using intelligent data analysis to detect abnor

2、mal energy consumption in buildings</p><p>　　John E. Seem *</p><p>　　Johnson Controls, Inc., 507 East Michigan Street, Milwaukee, WI 53202, USA</p><p>　　Received 31 October 2005; re

3、ceived in revised form 11 March 2006; accepted 18 March 2006</p><p><b>　　Abstract</b></p><p>　　This paper describes a novel method for detecting abnormal energy consumption in buildi

4、ngs based on daily readings of energy consumption and peak energy consumption. The method uses outlier detection to determine if the energy consumption for a particular day is significantly different than previous energy

5、 consumption. For buildings with abnormal energy consumption, the amount of variation from normal is determined using robust estimates of the mean and standard deviation. This new data analysis</p><p>　　# 20

6、06 Elsevier B.V. All rights reserved.</p><p>　　Keywords: Energy consumption; Fault detection; Outlier analysis; Performance monitoring; Robust statistics</p><p>　　Introduction</p><p&g

7、t;　　Energy management and control systems can collect and store massive quantities of energy consumption data. Facility operators can be overwhelmed with the quantity of data. For many operators, it is not possible to de

8、tect equipment, design, or operation problems because of data overload. Modern building management systems have two systems to help the operators with this data overload: alarm and warning systems and data visualization

9、programs. Today, operators must select the thresholds for alarms</p><p>　　* Tel.: +1 414 524 4677; fax: +1 414 524 5810.</p><p>　　E-mail address: john.seem@gmail.com.</p><p>　　0378-

10、7788/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.enbuild.2006.03.033</p><p>　　The research community has developed a number of methods for detecting faults in buildings and h

11、eating, ventilating, and air-conditioning systems. Two major research efforts have been sponsored by the International Energy Agency: Annex 25 [1,2] and Annex 34 [3]. There are two basic approaches to fault detection and

12、 diagnostics in buildings: a component level (bottom-up) approach and a whole-building (top-down) approach. The component level approach looks for faults in individual systems such as </p><p>　　Claridge et a

13、l. [4] describe an energy consumption report method that helps building operators and facility managers identify if the building systems are working properly. The report contains scatter plots of daily chilled water ener

14、gy consumption versus average daily temperature and daily hot water consumption versus average daily temperature for a 3- month period. For the last month, the scatter plot uses letters (M, T, W, H, F, S, U) to identify

15、the days of the week. The letters helps buildin</p><p>　　Nomenclature</p><p>　　a 2 Ba is an element of set B</p><p>　　a2= Ba is not an element of set B</p><p>　　iind

16、ex used in for loop in Fig. 1 nnumber of elements in set X noutnumber of outliers in set X</p><p>　　pright tail area probability for t-distribution</p><p>　　Riextreme studentized deviate for

17、 ith extreme</p><p>　　sstandard deviation for elements in set X</p><p>　　srobustrobust estimate of standard deviation for ele- ments in set X</p><p>　　tn,p critical value (t

18、n,p) for the Student’s t-distribution with n degrees of freedom and a right tail area probability of p</p><p>　　xe,ivalue of ith extreme</p><p>　　xjvalue of jth observation in set X</p>

19、<p>　　x¯average of elements in set X</p><p>　　x¯robustrobust estimate of average of elements in set X Xset of observations that contain outliers and non-</p><p><b>

20、　　outliers</b></p><p>　　Xnon-out set of observations that contain no outliers</p><p>　　Xoutset of observations that contain outliers</p><p>　　zmmodified z-score (standard

21、score)</p><p>　　{ }set of observations or elements</p><p>　　jsuch that</p><p>　　Greek letters</p><p>　　aprobability of declaring a normal value an outlier</p>

22、<p>　　licritical value for Rosner’s generalized ESD many-outlier procedure</p><p>　　through the tedious process of manually inspecting graphs to detect abnormal energy consumption. Instead, the operat

23、or or maintenance operator can investigate only buildings with abnormal energy consumption. The method accounts for weekly variation in energy consumption by grouping days of the week with similar power consumption. A ro

24、bust outlier detection method is used to determine if the energy consumption is significantly different than previous energy consumption. For time periods with abno</p><p>　　Overview of data analysis method&

25、lt;/p><p>　　Fig. 1 shows the major steps required to identify abnormal energy consumption in buildings. The feature extraction block determines features such as the average daily consumption or peak demand for

26、a day from energy data such as the whole- building electrical consumption. The features are then sorted into groups based on days of the week with similar energy consumption profiles. (In this paper, the term ‘‘day type’

27、’ refers to days of the week with similar consumption profiles.) After the data is</p><p>　　whole-building electric consumption. By inspecting these plots, building operators can identify days of abnormal en

28、ergy consumption. Haberl and Abbas [5,6] review several new graphical displays for viewing building energy data.</p><p>　　Dodier and Kreider [7] present a method for detecting whole-building energy problems

29、for the following energy uses: whole-building total electric energy, whole-building total thermal energy, HVAC-other-than-chiller electric energy, and chiller energy usage. They used an Energy Consumption Index (ECI) to

30、determine if the energy consumption was higher than normal, normal, or lower than normal. The ECI is the ratio of actual energy consumption to expected energy consumption as determined from a neu</p><p>　　Th

31、is paper presents an intelligent data analysis method [8] for automatically detecting abnormal energy consumption in buildings. With this method, operators will not have to go</p><p>　　Fig. 1. Block diagram

32、for detecting abnormal energy consumption.</p><p>　　Outlier identification: GESD many-outlier procedure</p><p>　　An outlier is an observation that appears to be inconsistent with the majority of

33、 observations in a data set. For example, in</p><p>　　Block 1: Set nout = 0. This step is used to initialize the number of outliers to zero.</p><p>　　Block 2: Compute average (x¯) of elemen

34、ts in set X. The average is determined from</p><p>　　the data set {1, 2, —1, 0, 3, 2, 101, —2}, the observation 101</p><p>　　appears to be an outlier. Data sets may contain more than one outlier

35、. For example in the data set {1, 2, —1, 0, 3, 2, 101, —2,</p><p><b>　　x¯ ¼</b></p><p><b>　　j¼1 x j</b></p><p><b>　　n</b></p>

36、;<p><b>　　(1)</b></p><p>　　96, 2, 0, —209}, the observations 101, 96, and —209 appear to be outliers.</p><p>　　Barnet and Lewis [11] provide details on several common outlier

37、identification methods. After comparing several popular outlier identification methods, Iglewicz and Hoaglin [9] highly recommend the generalized extreme studentized deviate (ESD)</p><p>　　where xj is a mem

38、ber of set X and n equals the number of elements in set X.</p><p>　　Block 3: Compute standard deviation (s) of elements in set X.</p><p>　　The standard deviation is determined from</p>&l

39、t;p>　　s?????????????????????????????</p><p>　　j¼1ðx j — x¯Þ</p><p>　　many-outlier procedure that was proposed by Rosner [12]s ¼</p><p>　　because

40、it works well under a variety of conditions.</p><p><b>　　n — 1</b></p><p><b>　　(2)</b></p><p>　　The generalized ESD many-outlier procedure can identity the e

41、lements in a set that are outliers. Fig. 2 is a flow chart for determining one or more outliers from a set of n observations X 2 {x1, x2, x3, .. ., xn}. The user needs to specify the probability, a, of incorrectly declar

42、ing one or more outliers when no outliers exist and an upper bound, nu, on the number of potential outliers. Carey et al. [13] said the upper bound (nu)</p><p>　　Block 4: s = 0. This block checks if the stan

43、dard deviation of</p><p>　　the elements in set X is zero. If the standard deviation equals zero, then the elements in set X all have the same value and there are no outliers in the remaining elements in set

44、X. (During field-testing of this method, several data sets had a standard deviation of zero.) To prevent a divide by zero in Block 6, execution goes to Block 10 when the standard</p><p>　　deviation determine

45、d in Block 3 equals zero.</p><p>　　could be determined by finding the largest integer that satisfies</p><p>　　the following inequality: nu Ç 0.5(n — 1). Following are details</p><

46、;p>　　Block 5: Find ith extreme (xe,i</p><p>　　) in set X. The extreme element,</p><p>　　on the numbered blocks in Fig. 2.</p><p>　　xe,i, is the element in set X that is furthest

47、from x¯. Of all the</p><p>　　elements in set X, the extreme element xe,i maximizes the</p><p>　　function jx j — x¯j where xi is an element of set X.</p><p>　　Block 6: Co

48、mpute ith extreme studentized deviate Ri. The</p><p>　　extreme studentized deviate is determined from</p><p><b>　　xe;ix¯ j</b></p><p><b>　　Ri ¼ j—</

49、b></p><p><b>　　(3)</b></p><p>　　where Ri is a normalized measure of how far the ith extreme is from the average value (x¯) determined in Block 2.</p><p>　　Block

50、7: Compute ith critical value li. Rosner [12] developed the following equation for determining the critical value:</p><p>　　n — iÞtn—i—1; p </p><p><b>　　li ¼ q</b></p&g

51、t;<p><b>　　ð</b></p><p>　　???????????????????????????????????????????????????????????????</p><p><b>　　(4)</b></p><p>　　ðn — i þ 1Þ

52、ðn — i — 1 þ t2Þ</p><p>　　where tn—i—1,p is the Student’s t-distribution with (n—i—1) degrees and the tail area probability p is determined from</p><p><b>　　a</b><

53、/p><p><b>　　p ¼ 2 n</b></p><p><b>　　i þ 1Þ</b></p><p><b>　　(5)</b></p><p>　　Fig. 2. Flow chart for implementing Rosner’s

54、 generalized many-outlier pro- cedure.</p><p>　　Abramowitz and Stegun [14] review equations for estimating the Student’s t-distribution.</p><p>　　Block 8: Ri > li. This block determines

55、 if the ith extreme studentized deviate, Ri, determined in Block 6 is greater than the ith critical value, li, determined in Block 7.</p><p>　　Block 9: Set nout = i. This block sets the number of outliers,&l

56、t;/p><p>　　nout, equal to i.</p><p>　　Block 10: Remove extreme element xe,i from set X. The extreme element xe,i is removed from set X and after removing the extreme element xe,i, the number of ele

57、ments in Set X is n — i. If i equals nu, then execution goes to Block 11; otherwise, return to the for loop on i.</p><p>　　Block 11: Outliers ¼ fxe;1; xe;2; ... ; xe;nout g. This block determines the ex

58、treme values that are outliers. The first nout extremes identified in Block 5 are considered outliers. Note that all the extreme values determined in Block 5 are not outliers.</p><p>　　Block 5: Compute modif

59、ied z-score(s). The modified z-score is a measure of the number of robust standard deviations an outlier is from the robust estimate of the mean. In equation form, we determine the modified z-score with</p><p&

60、gt;　　xoutlier — x¯ robust</p><p>　　Modified z-score</p><p><b>　　zm ¼</b></p><p><b>　　srobust</b></p><p><b>　　(9)</b><

61、/p><p>　　To help facility operators rank the severity of an outlier, a modified z-score [9] is used to quantify how far and in which direction an outlier is from the mean value of typical (i.e., non- outlier) o

62、bservations. Fig. 3 is a flow chart for determining a modified z-score from robust estimates of the mean and standard deviation [15,16] for a set of n observations X 2 {x1, x2, x3, .. ., xn}. Following are details on the

63、 numbered blocks in Fig. 3.</p><p>　　Block 1: Identify outliers in set X. Outliers ðxe;1; xe;2; ... ; xe;nout Þ in set X are identified with the generalized ESD many-outlier proce

64、dure described in the previous section. The outliers are members of set Xout (i.e., Xout ¼ fxe;1; xe;2; ... ; xe;nout g). If there are no outliers in set X, then Xout is an empty set.</p><p>　　Block 2

65、: Determine set of non-outlier observations.</p><p>　　Determine the set of observations Xnon-out in set X that do not include the outliers. In equation form, the set of non- outlier observations is

66、 determined from</p><p>　　Xnon-out ¼ fxjx 2 X and x2= Xoutg (6)</p><p>　　Block 3: Compute mean of non-outlier observations. The robust estimate of the

67、 mean, x¯robust, is the average value of the elements in set Xnon-out.</p><p><b>　　Pn—nout</b></p><p><b>　　j¼1</b></p><p>　　where xoutlier is the

68、value for an outlier.</p><p>　　Field test results</p><p>　　We used the data analysis method presented in the previous section to analyze electrical consumption data for 97 buildings. When analyz

69、ing energy consumption for a particular day, the energy consumption was compared to previous energy consumption for days of the same day type. The maximum number of previous days used in the analysis was limited to 30. M

70、any of the 97 buildings did not have abnormal high-energy consumption. This section contains field test results for three buildings that had abno</p><p>　　Many buildings had abnormally low energy consumption

71、</p><p>　　during holidays and special events such as offsite company meetings. The method described in this paper is robust to these unusual conditions. For example, following periods of low energy consumpti

72、on during the winter holiday season, the method did not detect false outliers in the beginning of January. Also, for this study, we did not find situations where the method failed by incorrectly detecting high outliers.&

73、lt;/p><p>　　The method for detecting abnormal energy consumption</p><p>　　requires knowledge of day type (i.e., days of the week with similar energy consumption profiles). There are three basic app

74、roaches to determining the day type: (1) an operator can select</p><p>　　x¯robust ¼</p><p><b>　　n — n</b></p><p><b>　　out</b></p><p>

75、<b>　　(7)</b></p><p>　　the day type based on knowledge of the building consumption,</p><p>　　(2) an operator can use time series or box plots [17,18] to</p><p>　　where xj

76、 2 Xnon-out and nout is the number of outliers in set X.</p><p>　　Block 4: Compute standard deviation of non-outlier</p><p>　　observations. The robust estimate of the standard deviat

77、ion, srobust, is the sample standard deviation of the elements in set Xnon-out</p><p>　　s??????? ??????????????????????????????????</p><p>　　j¼1 ðx j — x¯robustÞ&

78、lt;/p><p>　　determine the day type, or (3) a pattern recognition algorithm can automatically determine day type [19]. The operator can select the days based on knowledge of building use, or the operator can use

79、 schematic or time series plots to determine the days with similar power consumption. Fig. 4 is a two-panel box plot (trellis plot [20,21]) of the average daily consumption and peak daily</p><p>　　demand for

80、 a telecommunications building 1. From this plot, we</p><p>　　srobust ¼</p><p><b>　　n — n</b></p><p><b>　　out — 1</b></p><p><b>　　(8

81、)</b></p><p>　　can conclude there are two day types: weekdays (Monday through Friday) and weekends (Saturday and Sunday).</p><p>　　Fig. 5 is a time series graph of the peak daily electri

82、c consumption for a telecommunications building. On 7 August, the outlier analysis method determined that the energy consumption was abnormally high. Robust estimates of the mean and standard deviation determined that th

83、e energy consumption was 10 standard deviations above the mean on 7 August. Investigation into the high-energy consumption revealed a controls problem that was caused by failure of the primary chiller. After the primary

84、chill</p><p>　　Fig. 3. Flow chart for determining modified z-scores of outliers.</p><p>　　7 August cost the customer around 12,000 United States</p><p>　　Fig. 4. Two-panel b

85、oxplot of average daily consumption and peak demand.</p><p>　　Dollars (USD). To prevent this problem in the future, the customer revised the control strategy to prevent the two chillers from operating simul

86、taneously.</p><p>　　Fig. 6 shows the peak and average daily power usage for a second telecommunications building that uses district heating. The customer had an agreement to not use (curtail) district-heat d

87、uring times requested by the supplier. On 16 and 17 December, the average and peak electric consumption were abnormally high because the supplier asked the customer to not use district- heat. The customer then used elect

88、ric boilers to heat water. This caused the electric bill to increase by approximately 4300 US</p><p>　　Fig. 5. High outlier caused by chiller failure at a telecommunications building.</p><p>　　F

89、ig. 7 shows the peak and daily energy usage for the second telecommunications building. On 8 February 2000 a new electric distribution panel was installed. To test the panel, a number of electric devices were turned on.

90、This caused the peak electric consumption to rise significantly, and increased the electric bill for the customer by about 10,000 USD. The customer could have reduced the costs by performing tests during periods of low e

91、lectric costs.</p><p>　　Fig. 8 shows the peak and daily power consumption for a third telecommunications building that had a duct system that was not sized to allow free cooling during time periods when the

92、outdoor temperature is low. The control strategy of using cool outside air to reduce or eliminate mechanical cooling is commonly called economizer cycle control [22]. The high</p><p>　　Fig. 6. High outliers

93、caused by district-heating curtailment at a data center for a telecommunications building.</p><p>　　Fig. 7. High outliers caused by equipment testing at a telecommunications building. Next to the high outlie

94、rs are the corresponding modified z-scores.</p><p>　　Fig. 8. High outliers caused by HVAC system problem at a telecommunications building. Next to the high outliers are the corresponding modified z-scores.&l

95、t;/p><p>　　outliers shown in Fig. 8 could have been avoided if the duct system was sized to allow economizer cycle control. The higher electric consumption on 9 and 10 November cost the building owner approxima

96、tely 600 USD.</p><p>　　Conclusions</p><p>　　This paper presented a new method for converting energy consumption data into information. The method is computa- tionally efficient and thus can be i

97、mplemented in today’s building energy management and control systems. The method uses robust statistical methods to determine if the energy consumption is significantly different than previous energy consumption. If the

98、energy consumption increases significantly, then building operators or maintenance staff can investigate and correct the problem t</p><p>　　faulty operation. Also, building energy costs will decrease because

99、 control, system, or operation problems will be detected and corrected.</p><p>　　Acknowledgements</p><p>　　I would like to thank Jim Kummer, Bill Huth, and the field organization of Johnson Cont

100、rols, Inc. for their valuable contribution in collecting data and working with our customers to identify the causes and associated costs of abnormal energy consumption.</p><p>　　References</p><p&g

101、t;　　[1] J. Hyva¨rinen, S. Ka¨rki (Eds.), Building Optimization and Fault Diagnosis Source Book, International Energy Agency: Energy Conservation in Buildings and Community Systems Annex 25, Real Time Simulation

102、 of HVAC Systems for Building Optimization, Fault Detection and Diag- nosis, VTT Building Technology, Espoo, Finland, 1996.</p><p>　　[2] J. Hyva¨rinen (Ed.), Technical Papers of the IEA Annex 25, Intern

103、ational Energy Agency: Energy Conservation in Buildings and Community Systems Annex 25, Real Time Simulation of HVAC Systems for Building Optimization, Fault Detection and Diagnosis, VTT Building Technology, Espoo, Finla

104、nd, 1996.</p><p>　　[3] A. Dexter, J. Pakanen (Eds.), Demonstrating Automated Fault Detection</p><p>　　and Diagnosis Methods in Real Buildings, International Energy Agency: Energy Conservation in

105、 Buildings and Community Systems Annex 34, Computer Aided Evaluation of HVAC System Performance, VTT Tech- nical Research Centre of Finland, Espoo, 2001.</p><p>　　[4] D.E. Claridge, J.S. Haberl, R.J. Sparks,