Вы находитесь на странице: 1из 11

'r

ELSEVIER

0306-2619(95)

Applied Energy52 (1995) 35-45 Copyright 1995 Elsevier Science Limited Printed in Great Britain. All rights reserved 0306-2619/95/$9.50 00013-5

Error Estimation in Bin Method Energy Calculations V. I. Hanby


Department of Civil and BuildingEngineering,Universityof Technology, Loughborough, UK

ABSTRA CT The use of the bin method for energy calculations in heated and cooled buildings entails a compromise between the calculation time taken and the increased accuracy of using a larger numbers of bins. A deterministic expression has been developed which can be used to predict the error attributable to different degrees of discretization. The expression was tested using two typical calculation problems, which were run using both binned input data and the parent time-series data files. The expression was found to give a satisfactory prediction of the error due to discretization.

INTRODUCTION The bin method ~ has been in use for many years as a simple hand-calculation procedure for calculating energy consumptions in heated and air conditioned buildings. The method integrates a steady-state thermal model of the building envelope with the capability of including casual and solar gains, together with the part-load characteristics of heating and cooling plant. 2 Although it has been eclipsed as an analytical tool by dynamic computer simulation programs, the virtues of simplicity and transparency have ensured the continuing use of the bin method in many teaching and practical applications. In addition to its use in hand calculations, the advantages of using binned data have been described 3 in steady-state HVAC system simulations. This is particularly useful where repeated simulation runs are required as part of the analytical stages of design, or when the simulation is part of a design optimization procedure. 4 35

36

v. I. Hanby

USE OF D I S C R E T I Z E D DATA In any calculation procedure based on the use of binned data, there will be a compromise between the higher accuracy obtainable by choosing smaller intervals for the bins and the calculation or execution time which will increase in proportion to the number of times the model function has to be evaluated. In a practical sense, this balance has been determined by the form of the available binned data. For example ASHRAE recommends using outside-air dry-bulb data in 5F wide bins, whereas the current 'banded' weather data in the CIBSE Guide 5 group all data into 10 bins for each month. The objective of this work was to derive and test an expression which could estimate the error attributable to the discretization of the input data in the bin method. This expression could then be used to quantify the relationship between the degree of discretization and the accuracy of the result. The error estimate was to be tested against the output from typical building/plant models. It was important in this respect to remove from consideration any limitations in the integrity of the model itself. Hence two prototype models (an air-handling plant and a building/heating system model) were run firstly using hourly time-series data for an example year, then with different numbers of bins for the primary driving variable. In both cases, the variable used was the outside-air dry-bulb temperature. There are two ways of handling secondary input variables. If there exists a statistically valid relationship, the secondary variable can be calculated from the value of the primary driving variable. An example of this is given by the relationship between wet bulb and dry bulb temperatures for the CIBSE Example Year for Kew. 6 An alternative approach is to use a single value for the secondary variable averaged over the time that the primary variable was in each of its bins. This is the basis of the data published in the CIBSE Guide. Both of these techniques have been incorporated into the example calculations reported here. The basis of the bin method is that a single representative value of the primary driving variable is taken for each group of binned data. This value is frequently the centre line value of each bin, but it is rather more accurate to take the mean value of the data within each bin. The model function is evaluated for these representative values and this is combined with the probability of occurrence of each bin to give an overall estimate of the long-term model output (i.e. the energy consumption). The following factors could be expected to influence the accuracy of this procedure: the number of bins chosen;

Error estimation in bin method energy calculations

37

the sensitivity of the model output to changes in the value of the binned variable; the distribution of values of the variables within each bin.

E R R O R ESTIMATION In the course of a calculation, it is the absolute value of the system's energy consumption that is generally required. In developing this error analysis, it was more convenient to refer to the mean value of the calculated energy consumption per time period. The total value is, of course, easily obtained by multiplying the mean value by the number of time periods. Consider a benchmark calculation which is driven by a time-series (v~, rE, V3,...,Vk), where each item vi represents the value of the variable (usually the outside-air dry-bulb temperature) at time i. The actual energy consumption calculation procedure can be represented by a single function g(vi), which gives the desired model output for vi. The mean value of the calculated energy consumption over the entire period of the timeseries is
k

1
j=l

If the input to the discretized model is divided into N bins, then it can exist in N states, each of which is denoted by M~. In this case g(M~) is the model output for state M~. The expected mean output of the bin calculation is then
N

~.~g(Mi) P (Mi)
i=1

where P(Mi) denotes the probability of the system being in state M~. The difference between the mean value predicted by the bin calculation and the true mean is thus
N 1 iv

i=1

/=1

In order to combine these two summations, the time-series data can be divided up into N sets, according to which state M~ the variable vi falls into:
N
i=1

1 N vx~M i g(Mi) P (Mi) - k

~
i=1

~g(Vx)

38

V./. Hanby

If the number of items (k) in the time-series is large, then P(M) is given by the ratio of the number of members (n) of each bin divided by the total membership of the time-series

P(M) =

n(Vx I Vx eMi) k

The two summations i = 1...N can now be combined to give

.i~l -~

n(Mi) g (Mi) - - " ~ g ( v x )

Further combination of the two summations and noting the difference between the two function values as Agx gives
N 1 vxeMi

E
i=l

~ agx
i=1

(1)

Omitting higher-order terms, a bound can be placed on Agx using a first-degree Taylor expansion in terms of the binned variable v:

Agx

8g iMiAvi Ov

where Av t represents the difference between one of the time-series values of the variable and its representative value in the ith bin. Equation (1) can now be simplified further if Agx is replaced by a bound max (IAgl):
N

1 txeMi N 1 -~ ~,Agx -<~ -~ n(Mi)max (]Ag)l


i=l

i=1

giving
N

~ P (Mi)max(lag[)
i=l

This summation is equal to unity, since it is the sum of the probabilities of being in all possible states. The value of max (IAgl) then provides a bound on the size of the difference between the energy consumption as found by the bin calculation and that found from the time-series data. Substituting the first-degree Taylor approximation for Agx in eqn (1) gives N
i=1

1 .(u) 8g

E -;E
x=l

lg, a, Vx

Error estimation in bin method energy calculations


A further simplification leads to

39

N
i=1

1 Og 1 .(Mr) --k n(Mi) -~v [ Mi n(Mi) ~ A , vx


x=l

where again

n(Mi) k
Noting that

e(Mi)

1 n(M~) n(Mi~- ~4 Ai Vx ~
x=l

is the mean deviation of the time-series values from the representative value for v in state i, this can be abbreviated to A~,v;. It is appropriate to use absolute values for both the partial derivatives and for the differences, so the expression becomes
N

P(M~) abs
i=1

Ov

I M~ ) abs (A. v~)

(2)

A P P L I C A T I O N TO E X A M P L E M O D E L S In order to test the applicability of the error estimation expression (2), results from two example models were obtained. In both cases, the primary driving variable was the outside-air dry-bulb temperature, and the energy consumption of the plant in each case was reported. The weather data were taken from the CIBSE Example Year for Kew, UK, which ran from 1 October 1964 to 30 September 1965.

Air handling unit model


This example modelled a heating and cooling coil in series (Fig. 1). A constant flow-rate of outside air was maintained over the coils and the controller maintained an off-coil dry-bulb set-point temperature with the coils sequenced and a dead zone of lC between the operation of the heating and cooling coils. The wet bulb temperature of the entering air was calculated from twet = 0.455 + 0.848td~y, a relationship which has a correlation coefficient of 0.97 for this Example Year. 7 The individual coil models were an implementation of the IEA Annex X Specifications and used the three-line method for calculating dehumidification.

40

V. I. Hanby

!"i

!"'!

Fig. 1. Schematic of air-handling plant model.

In order to estimate the error in the case of the results from the binned data, the mean deviation of the data points from the representative value (the mean) in each bin was determined directly from the original timeseries file. The partial derivative of the energy consumption with respect to the dry-bulb temperature was obtained numerically from the calculated energy consumption for each bin representative value. The three-point formula 9 was used for this purpose, using only forward differences for the lowest representative temperature and only backward differences for the highest. For a central point:
f(xo)

1 --:-- [t'(xo + h) - f ( X o -

2h

h)]

and for an end point:


1
f(xo) ~" ~

2h

[3f(Xo) + 4f (Xo + h)

-flXo

2h)],

where h is the bin width. The model was run with the input dry-bulb temperatures divided into 5, 10, 15, 20 and 30 bins. The distribution of the representative temperatures (the means in each bin) for 10 bins is shown in Fig. 2 and the energy consumption distribution for this case in Fig. 3. The shape of this distribution has two peaks, attributable to the dead zone between the operation of the coils. The comparison of the results of the bin method calculations with the time-series benchmark is shown in Fig. 4, where the absolute value of the error is shown as a percentage of the benchmark figure as a function of

Error estimation in bin method energy calculations

41

2000 1800 1600 1400 1200


t~

1000 800 600 40O 200 0 -10


I

-5

Bin Temperature(degC)

10

15

20

25

30

Fig. 2. D i s t r i b u t i o n o f dry-bulb temperatures for 10 bins.


|0000 i i i

9000 80O0 7000

tU

-/
60O0 5000 3000 2000 1000
0 I I I I I I

-5

Bin Temperature(dagC)

10

15

20

25

30

Fig. 3. D i s t r i b u t i o n o f energy consumption in the A H U model.

the number of bins used. It can be seen that the results from the binned data converge quickly onto the benchmark figure up to about 15 bins of dry-bulb temperature. This represents a considerable saving in execution time over that of the original 8760 time-series input values. The error expression gave a very conservative estimate at low numbers of bins - this was found to be typical behaviour over a wide range o f steady-state simulation problems.

42

I~. I. Hanby

12 10
4\ \

bin m o d e l
bound

-*-.....

\,

2 0

10
No. of bins

15

20

Fig. 4. Actual and estimated errors for the A H U model.

Building/heating plant model


The second example chosen was a more conventional use of the bin method - - a dwelling with a design heat loss of around 10 kW. Casual gains were taken as averaging out at 600 W and solar-gain was accounted for on two facades of the building. A constant solar-gain factor was assumed and the values of solar intensity were taken from the original weather data file, averaged out over the respective periods for the dry-bulb temperature bins. The CIBSE Example Year was again used, but in this case only the conventional heating season (October to May) was used. The house was heated by a boiler with known part-load characteristics, given in Table 1. The distribution of temperatures was the same as in the first example, but in the heated house calculation, the distribution of energy consumption follows the more normal bell-shaped curve (Fig. 5).
TABLE 1 Part-load boiler characteristics

Load factor

(%)
100 80 60 40 20

Efficiency

("~)
79 77 65 57 35

Error estimation in bin method energy calculations

43

12000

I0000

8000

60OO

6
4OOO

2000

IO

-I0

-5

5 I0 Bin Temperature (degC)

15

20

25

Fig. 5. Energy-consumption distribution for the heated house.

The predicted and actual errors for this bin calculation are shown in Fig. 6. Once again eqn (2) gives a conservative estimate at low numbers of bins, but thereafter predicts the error satisfactorily. It would seem that the convention of using 10 bins gives, in this case, a good compromise between accuracy and calculation time, but halving this number to five still gives an answer to within 4% of the benchmark value. It is of particular interest to note that, in the case of this model, increasing the number of bins to large values gives an error which
12
\,\

I0

bin model

bound . . . .

o=

0 0

I0 No. of bins

15

20

Fig. 6. Actual and estimated errors for the heated house model.

44

1I. L Hanby

converges on a value around 0.5%, rather than zero. This is caused by the way in which the secondary input variable (solar radiation) is allowed for. As the number of bins allocated to the primary variable increases, its distribution decomposes towards the original time-series data, but this does not happen in the case of the solar radiation data.

CONCLUSIONS An expression has been derived which can predict the error due to discretization of the input variables in the bin energy calculation procedure. The analysis takes into account the number of bins used, the sensitivity of the model output to changes in the input variable and the distribution of the variable within each bin. The model was tested by running two representative calculation problems with binned data and with the original time-series data, which formed a benchmark figure. The calculated errors were found to be in satisfactory agreement with the errors obtained from the example calculations, but at low numbers of bins there was a consistent overestimate of the error. In some cases, it would be possible to use less bins than the currently recommended figure without a significant loss of accuracy. In the case where there is more than one input variable (such as drybulb temperature and solar radiation) to the calculation, increasing the number of bins for the primary-input variable causes the calculation error to converge onto a non-zero value. However, this residual error is of no practical significance.

ACKNOWLEDGEMENTS The error prediction equation used was developed from an approach devised by A. Dil. ~ The air-handling plant simulation runs were carried out by L. F. E. Yeo. The author would like to acknowledge the Science and Engineering Research Council (UK) for financial support of part of this work. REFERENCES
1. ASHRAE Handbook, Fundamentals, 28.7-28.15. American Society of Heating, Refrigerating and Air Conditioning Engineers, Atlanta, 1981. Letherman, K. M. & Dewsbury, J., The bin method - - a procedure for predicting seasonal energy requirements in buildings. Building Services Engineering Research and Technology, 7(2) (1986) 55-64.

2.

Error estimation in bin method energy calculations

45

Hanby, V. I. & Round, M., HVAC system simulation: use of condensed input data. Building Services Research and Technology, 13(1) (1992) 13-17. 4. Hanby, V. I. & Wright, J. A., HVAC optimisation studies: component modelling methodology. Building Services Engineering Research and Technology, 10(1) (1989) 35-40. 5. CISBE Guide, Book A, Banded Weather Data, A2-21-A2-27. Chartered Institution of Building Services Engineers, London, 1986. 6. Holmes, M. J. & Hitchin, E. R., An example year for the calculation of energy demand in buildings. Building Services Engineer, 45 (1978) 186-9. 7. Yeo, L. F. E., Preparation of condensed weather data for HVAC simulation. Internal report, Department of Civil Engineering, Loughborough University, UK, 1988. 8. Holmes, M. J., Coils. International Energy Agency Annex X Specification An10-880212-1, University of Liege, 1988. 9. Burden, R. L. & Faires, J. D., Numerical Analysis. Prindle, Weber and Schmidt, Boston, 1985. 10. Dil, A. J., Markov modelling of HVAV systems. PhD thesis, Loughborough University, UK, 1993.

3.

Вам также может понравиться