Measuring Industrial Energy Savings by yst42447

VIEWS: 39 PAGES: 38

									            Measuring Industrial Energy Savings
                                                             *
                                 J. Kelly Kissock and Carl Eger
          University of Dayton, Department of Mechanical and Aerospace Engineering,
                        300 College Park, Dayton, Ohio 45469-0238 USA


ABSTRACT

Accurate measurement of energy savings from industrial energy efficiency projects

can reduce uncertainty about the efficacy of the projects, guide the selection of

future projects, improve future estimates of expected savings, promote financing of

energy efficiency projects through shared-savings agreements, and improve

utilization of capital resources. Many efforts to measure industrial energy savings,

or simply track progress toward efficiency goals, have had difficulty incorporating

changing weather and production, which are frequently major drivers of plant

energy use. This paper presents a general method for measuring plant-wide

industrial energy savings that takes into account changing weather and production

between the pre and post-retrofit periods. In addition, the method can

disaggregate savings into components, which provides additional resolution for

understanding the effectiveness of individual projects when several projects are

implemented together. The method uses multivariable piece-wise regression

models to characterize baseline energy use, and disaggregates savings by taking

the total derivative of the energy use equation. Although the method incorporates

search techniques, multi-variable least-squares regression and calculus, it is easily


   Corresponding Author. Tel.: +1-937-229-2852; fax: +1-937-229-4766

   E-mail address: Kelly.kissock@notes.udayton.edu (J.K. Kissock)



                                                                                      1
implemented using data analysis software, and can use readily available

temperature, production and utility billing data. This is important, since more

complicated methods may be too complex for widespread use. The method is

demonstrated using case studies of actual energy assessments. The case studies

demonstrate the importance of adjusting for weather and production between the

pre- and post-retrofit periods, how plant-wide savings can be disaggregated to

evaluate the effectiveness of individual retrofits, how the method can identify the

time-dependence of savings, and limitations of engineering models when used to

estimate future savings.


Keywords: Measuring, Manufacturing, Industrial, Retrofit, Energy, Savings

1. INTRODUCTION

The decision to spend money to reduce energy expenditures frequently depends

on the expected savings. Decision makers must then weigh the expected savings

with several other issues. These issues include the availability of capital,

competing investments, the synergy of the proposed retrofit with other strategic

initiatives, and, not insignificantly, the certainty that the expected savings will be

realized.


This uncertainty about whether the expected savings will be realized depends

largely on the type of retrofit. In some cases, it is relatively easy to verify expected

savings; for example, expected energy savings from a lighting upgrade can be

easily verified by measuring the power draw of lighting fixtures before and after a

lighting upgrade. A history of verified savings reduces the uncertainty about future



                                                                                         2
lighting recommendations and encourages this type of energy efficiency retrofit. In

other cases, however, the retrofit may occur on a component of a larger system,

and the energy use of the component may be difficult or impossible to meter.

Moreover, the energy use may also be a function of weather and/or production,

which frequently changes between the pre- and post retrofit periods. In these

cases, it is more difficult to measure energy savings and, as a consequence,

savings are seldom verified.


This lack of verification hurts the effort to maximize industrial energy efficiency. In

some cases, retrofit measures which would realize the expected savings are not

implemented since there is no history of successful verification. In other cases,

retrofits that do not achieve the expected savings get implemented, which wastes

resources that may have been directed to more effective measures. Both of these

problems could be minimized by systematically measuring savings, and comparing

expected and measured savings. The information could guide the selection of

future retrofits, improve methods to calculate expected savings, promote financing

of energy efficiency through shared-savings agreements and improve utilization of

capital resources.


Several studies confirm that uncertainty in expected savings reduces

implementation of energy reduction measures. [1, 2] Moreover, the lack of reliable

estimates of savings increases the subjectivity of the evaluation of savings

initiatives, which weakens arguments for addition initiatives. For example,

Ramesohl et al. [3] report ―in quite some cases the quantitative saving effect of a



                                                                                          3
conservation measure is not exactly known by the companies due to insufficient

monitoring and measuring. Even in cases where cost savings are perceived to be

significant and relevant to the decision, reliable data is rarely obtainable, leaving

the assessment open to personal judgment. This lack of objective decision

parameters underlines the subjective character of profitability assessments.‖


Because of the importance of measuring savings, numerous efforts have been

made to develop standard protocols for measuring savings. For example, the

National Association of Energy Service Contractors developed protocols for the

measurement of retrofit savings in 1992. In 1994, the US Department of Energy

initiated an effort that resulted in publication of the North American Energy

Measurement and Verification Protocols [4] and, later, the International

Performance, Measurement and Verification Protocols [5]. The U.S. Federal

Energy Management Program developed their own set of Measurement and

Verification Guidelines for Federal Energy Projects. [6] ASHRAE published its

guideline, Measurement of Energy and Demand Savings, in 2002. [7]


A principle method for measuring savings included in all of the aforementioned

protocols relies on regression modeling. The regression method of measuring

savings has been widely used in the residential and commercial building sectors.

[8,9] This paper describes an extension of the method to measure savings in the

industrial sector, in which both weather and production are frequently strong

drivers of energy use, and synergisms between manufacturing operations negate

the efficacy of measuring savings by submetering individual pieces of equipment.



                                                                                        4
The paper begins with a brief review of the regression method for measuring

savings, discusses the extension of the method to measure industrial energy

savings, and demonstrates the method with case studies.



2. REGRESSION METHOD FOR MEASURING SAVINGS

Perhaps the simplest method of measuring retrofit energy savings is to directly

compare energy consumption in the pre- and post-retrofit periods. This method

implicitly assumes that the change in energy consumption between the pre-retrofit

and post-retrofit periods is caused solely by the retrofit. However, energy

consumption in most industrial facilities is frequently influenced by weather

conditions and the quantity of production—both of which may change between the

pre- and post-retrofit periods. If these changes are not accounted for, savings

determined by this simple method will be erroneous. Because direct comparison

of pre- and post-retrofit energy consumption does not attempt to adjust the pre-

retrofit model to account for these changes, savings measured using this method

are called ―unadjusted‖ savings.


One way to account for these changes is to develop a weather and production-

dependent regression model of pre-retrofit energy use. The savings can then be

calculated as the difference between the post-retrofit energy consumption

                                    ˆ
predicted by the pre-retrofit model EPr e and measured energy consumption during

the post-retrofit period E Meas . The procedure to calculate savings is summarized

by:



                                                                                     5
            m
      S   ( E Pr e , j  E Meas, j )
              ˆ                                                                   (1)
            j 1




where m is the number of post-retrofit measurements.


                        ˆ
The pre-retrofit model, EPr e , is called the baseline model. Savings measured using

a baseline model, are called ―adjusted‖ savings when the baseline model is

adjusted to account for the weather and production conditions in the post retrofit

period. Adjusted savings are more accurate than unadjusted savings, and should

be used whenever the energy use data used to measure savings is weather and/or

production dependent.                Two types of baseline regression models that are

appropriate for measuring industrial energy savings are described below.



3. MULTI-VARIABLE CHANGE-POINT MODELS

In most industrial facilities, the weather dependence of energy use can be

accurately described using a three-parameter change-point model.               Three-

parameter change-point models describe the common situation when cooling

(heating) begins when the air temperature is more (less) than some balance-point

temperature.       For example, consider the common situation where electricity is

used for both air conditioning and production-related tasks such as lighting and air

compression. During cold weather, no air conditioning is necessary, but electricity

is still used for process purposes. As the air temperature increases above some

balance-point temperature, air conditioning electricity use increases as the outside

air temperature increases (Figure 1a). The regression coefficient 1 describes



                                                                                    6
non-weather dependent electricity use, and the regression coefficient 2 describes

the rate of increase of electricity use with increasing temperature, and the

regression coefficient 3 describes the change-point temperature where weather-

dependent electricity use begins. This type of model is called a three-parameter

cooling (3PC) change point model.         Similarly, when fuel is used from space

heating and production-related tasks, fuel use can be modeled by a three-

parameter heating (3PH) change point model (Figure 1b). In the statistical

literature, these types of models are known as piecewise linear or spline models.



In buildings, the largest components of weather-induced heating and cooling loads

are conduction through the building envelope and air infiltration/ventilation, both of

which vary linearly with outdoor air temperature.       This linearity is even more

pronounced in industrial facilities and processes, where high ventilation,

combustion air and process infiltration loads, which are completely linear with air

temperature, are dominant.     Thus, the choice of a linear relationship between

heating and cooling energy and outdoor temperature is mandated by the physics

of building and process energy use. Similarly, the choice of a spline fit over a

polynomial to describe the onset of heating and cooling derives from the type of

control used in virtually all buildings, a thermostat. Because thermostats initiate

heating and cooling at fixed temperatures, and because heating and cooling loads

are essentially linear with outdoor air temperature, piecewise fits such as the 3PC

and 3PH models described here, provide a better representation than polynomials




                                                                                     7
for the relationship between heating and cooling energy use and outdoor air

temperature. [10]



These basic change-point models can be easily extended to include the

dependence of energy use on the quantity of production by adding an additional

regression coefficient.      The functional forms for best-fit multi-variable three-

parameter change-point models for cooling energy use, Ec, (3PC-MVR) and

heating energy use, Eh, (3PH-MVR), respectively, are:



      EC  1   2 T   3    4  P
                             
                                                                                  (2)


      EH  1   2  3  T    4  P
                             
                                                                                  (3)


where 1 is the constant term, 2 is the temperature-dependent slope term, 3 is

the temperature change-point, and 4 is the production dependent term.          T is
                                                                                    +
outdoor air temperature and P is the quantity of production. The superscript

notation indicates that the value of the parenthetic term is zero when the value of

the term enclosed by the parenthesis is negative.


The use of a single regression coefficient,4, and a single metric of production, P,

is arbitrary; additional terms can be added to account for multiple products. The

number of production variables needed to characterize plant energy use depends

on the plant and process.          In many plants, such as auto assembly plants or

foundries, the relationship between energy use and production is accurately



                                                                                   8
characterized by a singe variable. In other plants with a heterogeneous product

mix, multiple variables for the most energy-intensive products may be needed. In

this paper, the method is demonstrated using one production variable; however,

the methodology is unchanged with addition production variables.


In Equations 2 and 3, the 1 term represents energy use that is independent of

both weather and production, such as lighting energy use in plants with limited

daylighting. The 2·(T-3)+ or –2·(3-T)+ term represents outdoor air temperature-

dependent energy use.      Because several studies have shown that outdoor air

temperature is the single most important weather variable for influencing energy

use in most buildings, we refer to this as weather-dependent energy use. [8,9] In

cases for which the weather dependent term represents space-conditioning energy

use, the coefficient, 2, represents the overall building load coefficient, UA, divided

by the efficiency of the space conditioning equipment, . When considering cases

where space cooling is present, this efficiency term is the efficiency of the space

cooling equipment (i.e. chillers, cooling towers, etc.).          Alternatively, when

considering cases where spacing heating is present, this efficiency term

represents the efficiency of the space heating equipment (i.e. boiler system, direct-

fire make-up air unit, unit heat, etc.) The coefficient,3, represents the balance-

point temperature, which is the outdoor air temperature below which heating

energy is used or above which cooling energy is used. The 3·P term represents

production-dependent energy use. Using these terms, these simple regression

equations can statistically disaggregate whole-plant energy use into independent,



                                                                                     9
weather-dependent and production-dependent components.             The interpretation

and use of this disaggregation technique is called Lean Energy Analysis, and is

useful for identifying energy saving opportunities, measuring energy effects of

productivity changes, and developing energy budgets. [11,12,13]


Several algorithms have been proposed for determining the best-fit coefficients in

piecewise regressions such as Equations 2 and 3. The simple and robust method

proposed here uses a two-stage grid search. The first step is to identify minimum

and maximum values of T, and to divide the interval defined by these values into

ten increments of width dx. Next, the minimum value of T is selected as the initial

value of 3 and the model is regressed against the data to find 1, 2, 3 and

RMSE. The value of 3 is then incremented by dx and the regression is repeated

until 3 has traversed the entire range of possible T values. The value of 3 that

results in the lowest RMSE is selected as the initial best-fit change-point. This

method is then repeated using a finer grid of width 2dx, centered about the initial

best-fit value of 3. [14] For discussions of the uncertainty of savings determined

using regression models see Kissock et al. [15] and Reddy et al. [16].


This method has been incorporated in several software tools for measuring

savings. One tool is the ASHRAE Inverse Modeling Toolkit [17,18], which supports

ASHRAE Guideline 2002-14.         Another tool is ETracker [19,20], which is free

software used to support the EPA Energy Star Buildings program [21]. Another

tool is Energy Explorer, which is used in the analysis that follows. [22]




                                                                                   10
3.1.   MULTI-VARIABLE VARIABLE-BASE DEGREE-DAY MODELS

Multi-variable variable-base degree-day (VBDD-MVR) models can also be

developed that yield similar results. The use of VBDD models to measure savings

traces its origin to the PRInceton Scorekeeping Method (PRISM), which has been

widely used in the evaluation of residential energy conservation programs.

[8,23,24] Sonderegger extended the method to include additional variables, such

as production. [25,26]


The forms of multi-variable VBDD models of cooling energy use, Ec, and heating

energy use, Eh, are shown below:


       EC   1   2  CDD  3    4  P                                 (4)


       E H   1   2  HDD  3    4  P                                (5)


where 1 is the constant term, 2 is the slope term, HDD(3) and CDD(3) are the

number of heating and cooling degree-days, respectively, in each energy data

period calculated with base temperature 3, and 4 is the production-dependent

term. P is the quantity of production. The number of cooling and heating degree-

days in each energy data period of n days is:


                      n
       CDD  3    (Ti   3 )                                           (6)
                     i 1




                      n
       HDD  3    (  3  Ti )                                          (7)
                     i 1




                                                                              11
where Ti is the average daily temperature. To use this method, the balance point

temperature which gives the best-fit to the data must be estimated or determined

by a search algorithm.


A simpler method uses degree days with a fixed 18 °C base temperature.


       EC   1   2  CDD T  18C    4  P                                        (8)


       E H  1   2  HDD T  18C    4  P                                        (9)


The loss in accuracy of the 18 °C degree-day method compared to the variable-

base degree-day method depends on the deviation between the assumed 18 °C

balance-point temperature and the actual balance-point temperature of the facility.


3.2.   OTHER MODELS

Other models and statistical techniques have also been used to describe facility

energy use. For example, neural network models have been shown to accurately

capture non-linear relationships and cross correlation among multiple independent

variables. [27,28,29]      Principal component analysis has been used to handle

multicollinearity associated with time series data. [30,31]           Other examples of

empirical modeling of industrial energy use include the application of a productivity

index to the container glass sector to understand productivity, efficiency and

environmental performance. [32]           In addition, Boyd applied a variation of best-fit

multivariable regression modeling techniques to identify best practices in industrial

sectors. [33] Although these methods all have appropriate applications, they were




                                                                                        12
not selected for this approach for two reasons. First, the goal of this research is to

describe a transparent method for measuring industrial energy savings, which still

accounts for major sources of error associated with unadjusted savings, and can

be applied by the industrial community. Unfortunately, the complexity of applying

and interpreting many of the models described above generally inhibits their

widespread use. In addition, the model coefficients in the proposed method have

physically meaningful interpretations, which enhance the usefulness of the

method, and enables savings to be disaggregated into meaningful components.

Because of these reasons, these more complicated methods will not be further

discussed.


3.3.   CALCULATING ADJUSTED SAVINGS

To calculate adjusted savings, the appropriate baseline model (Equation 2, 3, 8 or

                                     ˆ
9) is used as the pre-retrofit model EPr e in Equation 1. The total adjusted savings

during the post-retrofit period is the summation of the difference between the

values of energy use predicted by the pre-retrofit model, which has been adjusted

to account for weather and production during the post-retrofit period, and the

actual energy use in the post-retrofit period.



4. DISAGGREGATING SAVINGS INTO COMPONENTS

One of the strengths of this method is that the 3PC-MVR or 3PH-MVR pre and

post-retrofit models lend insight into how energy is used in a facility. As indicated

above, regression coefficients 1, 2, 3 and 4 correspond to energy use




                                                                                   13
independent of weather and production, the weather-dependent energy use

coefficient, the facility balance-point temperature, and the production-dependent

energy use coefficient, respectively. Thus, graphical comparison of the pre and

post retrofit models, or direct comparison of the pre and post-retrofit coefficients,

yields much insight into the nature of the savings. For example, an energy efficient

lighting retrofit should decrease in 1, weather and production-independent energy

use, between the pre and post retrofit periods. Adding insulation to the building

envelope should decrease 2, the weather-dependent energy use coefficient.

Decreasing the thermostat set-point temperature in winter should decrease 3, the

facility balance point temperature.         And improving the energy efficiency of

production related equipment should decrease 4, the production-dependent

energy use coefficient.


These insights can be quantified by noting that savings, S, is the change in energy

use, dE, and can be estimated by taking the total derivative of the energy use

equation (Equation 2 or 3).


         ˆ                      E        E         E         E 
     S  EPr e  E Post  dE      d1  
                                            d 2  
                                                         d 3  
                                                                   d 4
                                                                                 (10)
                                1         2          3          4


Following Equation 10, the total savings can be disaggregated into independent,

weather-dependent, production-dependent, and interior-temperature dependent

components (Table 1).




                                                                                   14
The multiple equations required for balance point temperature–dependent savings

are due to the discontinuous nature of the change-point equations.



5. CASE STUDY 1

The first case study demonstrates the use of the method to measure adjusted

savings and disaggregate savings into components. The savings opportunities

were identified during an energy assessment by the University of Dayton Industrial

Assessment Center (UD-IAC).        The UD-IAC is one of twenty-six Industrial

Assessment Centers at universities throughout the United States. [34]       Each

center is funded by the United States Department of Energy Industrial

Technologies Program to perform about 25 energy assessments per year for mid-

sized industries, at no cost to the industrial client. Each assessment identifies

energy, waste, and productivity cost saving opportunities, and quantifies the

expected savings, implementation cost and simple payback of each opportunity.

This information is delivered to the client in a report summarizing current energy

and production practices and the savings opportunities identified during the

assessment. About one year after each assessment, the client is contacted to

collect implementation results.


The use of this method to measure savings is demonstrated by analyzing fuel data

before and after an energy assessment of the Staco Energy Products Company in

Dayton, Ohio on February 2, 2004. [35]     Staco employed about 80 people and

occupied an 11,334 m2 (122,000 ft2) facility. The facility operated about 2,000




                                                                                15
hours per year and produced variable transformers, industrial voltage regulators,

uninterruptible power supply systems and other power management equipment.

During the year from July, 2002 to June, 2003, the facility used 967,061 kWh of

electricity, 6,209 GJ of diesel fuel and 3,067 GJ of natural gas. The diesel fuel was

used in a 6.7 GJ/hr hot-water boiler dedicated solely to space heating. Natural gas

was used in three drying and curing ovens, rated at 0.52 GJ/hr, 1.1 GJ/hr and 0.58

GJ/hr.    Total annual energy expenditures were $140,702.


The assessment generated 17 recommendations addressing electricity, fuel, waste

and productivity savings opportunities. These recommendations identified a total

of about $97,629 per year in potential savings with a total implementation cost of

about $21,121. This total includes non-energy related savings opportunities from

improving productivity and reducing waste. The estimated simple payback for all

recommendations was about 3 months.


On July 25, 2005, Staco was contacted to find out which recommendations had

been implemented, and to collect recent utility billing data for measuring savings.

According to management, 13 of the 17 recommendations had been implemented.

Of the 13 implemented recommendations, six were specific to fuel consumption.

The estimated savings and implementation cost of each recommendation are

shown in Table 2. Total expected fuel savings were 2,845 GJ per year.


5.1.     CASE STUDY 1: UNADJUSTED SAVINGS




                                                                                      16
Data used for this case study are monthly fuel use, average outdoor air

temperature, and monthly sales. The fuel use data were compiled from utility bills,

and represent the total energy from both natural gas and diesel fuels. Note that

the boiler was operated using diesel fuel in the pre-retrofit period and natural gas

in the post-retrofit period. No energy using equipment was added or removed

between the pre and post-retrofit periods. Due to the variety of products produced,

sales data were the best metric of production available. The sales data were

lagged by one month, since sales in one month influenced production during the

next month when production restocked depleted inventory. The average outdoor

air temperature for each period was calculated using average daily temperatures

from the UD/EPA Average Daily Temperature Archive, which posts average daily

temperatures for 324 cities around the world from 1995 to present. [33]


Unadjusted savings are calculated as the difference between pre- and post-retrofit

energy use. The time-trends of monthly fuel use from the pre- and post-retrofit

periods are shown in Figure 2. The time trends clearly show decreased fuel use

during the winter. The mean fuel consumption during the pre-retrofit period was

25.52 GJ per day, and is indicated by the top horizontal line. The mean fuel

consumption during the post-retrofit period was 21.07 GJ per day, and is indicated

by the lower horizontal line. Using the mean energy use from the pre-retrofit

period as a baseline model, the unadjusted fuel savings are calculated from

Equation 1 to be about 4.45 GJ per day. Annual unadjusted savings are 1,624 GJ

per year.




                                                                                       17
5.2.   CASE STUDY 1: WEATHER-ADJUSTED SAVINGS


A quick inspection of Figure 2 shows that fuel use peaks in the winter months,

which indicates the strong weather dependence of fuel use. Thus, the effect of

changing weather must be accounted for to accurately measure savings. To do so,

a weather-dependent model of pre-retrofit fuel use is developed.


Figure 3 shows three-parameter heating (3PH) models of fuel use as functions of

outdoor air temperature. The top (blue) model shows pre-retrofit fuel use and the

bottom (red) model shows post-retrofit fuel use. Both models show that space

heating fuel use increases linearly as outdoor air temperature decreases. The

outdoor air temperature at which space heating begins is 18 °C in the pre-retrofit

period and 16.5 °C in the post-retrofit period. Both models have good fits to the
           2
data; the R and CV-RMSE of the pre-retrofit model are 0.93 and 21.3%, and the

R2 and CV-RMSE of the post-retrofit model are 0.99 and 7.2%.


To adjust the baseline model for possible changes in production, as well as

weather, an additional regression coefficient for production can be added to the

pre-retrofit model (Equation 3). In this case, lagged sales data were the best

indicator of production available. The R2 and CV-RMSE of the weather and

production-dependent pre-retrofit model are 0.93 and 20.6% and the R2 and CV-

RMSE of the post-retrofit model are 0.99 and 7.2% (Figure 3). Thus, the addition

of lagged sales data as an independent variable added almost no information to

the models. Hence, for simplicity and clarity, the weather-dependent model




                                                                                     18
(Equation 17), rather than the weather and production-dependent model (Equation

3), will be used as the basis for calculating savings. The best-fit coefficients of the

pre and post-retrofit models are shown in Table 3.


            day   1  GJ day    2  GJ day  C    3 C   Toa C 
       Fuel GJ
                                                                                  
                                                                                 (17)
                                                  

The fuel use savings are the sum of the differences between the actual energy use

in post-retrofit period and the energy use predicted by the pre-retrofit period for the

same weather conditions. Fuel use savings can be visualized as the difference

between the model lines in Figure 3. Fuel use savings can also be visualized by

projecting the weather-adjusted baseline model onto the post-retrofit period (Figure

4). The weather-adjusted baseline model shows the energy use that would have

occurred if the retrofits had not taken place given the actual weather conditions in

the post-retrofit period. The savings are calculated using Equation 1 to be about

3.0 GJ per day. Annual weather-adjusted savings are 1,095 GJ per year.


5.3.     CASE STUDY 1: COMPARISON OF EXPECTED, UNADJUSTED AND

   ADJUSTED SAVINGS


Total expected savings from implementing all six recommendations was 2,845 GJ

per year (Table 2). Unadjusted savings were 1,624 GJ per year, and weather-

adjusted savings were 1,095 GJ per year. Many important lessons can be learned

from comparing these results.




                                                                                       19
First, the dramatic difference between expected and measured savings shows the

importance of measuring savings. In this case, measured savings were only 34%

of expected savings. This difference illustrates the limitations of engineering

modeling to predict the long term behavior of complicated systems. Some of these

limitations are functions of the assumptions and simplifications used to create

workable engineering models. The limitations may also include actual errors in the

engineering models. Finally, the recommendations may not be implemented in

exact accordance with recommendation specifications.


Second, the importance of weather adjustment when measuring savings is also

clear. Weather-adjusted savings were only 67% of unadjusted savings. The large

difference between unadjusted and weather-adjusted savings may not be apparent

from a casual inspection of the weather data. For example, the average annual

temperatures were 10.6 °C and 10.9 °C during the pre- and post-retrofit periods,

and average temperatures during the heating season (October through May), were

1.6 °C and 3.6 °C during the pre- and post-retrofit periods. Intuition may not

conclude that such small temperature differences could lead to such a large

change in measured savings. This suggests that the best way to account for

changes in weather is to employ weather-adjusted models, rather than by simple

inspection of weather data.


5.4.   CASE STUDY1: DISAGGREGATING SAVINGS INTO COMPONENTS




                                                                                   20
Inspection of the pre and post-retrofit models lends insight into the nature of the

savings and how much energy was saved by each type of retrofit. For example,

visual inspection of Figure 3 shows that:


   Weather-independent fuel use increased

   The balance-point temperature of the facility decreased

   The slope of the fuel use versus temperature line decreased.



This indicates that:


   Negative savings resulted from non-weather dependent retrofits.

   Some savings resulted from decreasing the set point temperature

   Some savings resulted from increasing the efficiency of the boiler




The observations can be quantified using Equation 10. Applying Equations 11, 13

and 15 to this case study, gives the disaggregated savings shown in Table 4. The

savings expected from each type of retrofit can then be compared to the

disaggregated savings to show the measured savings associated with each type of

recommendation.


For example, the recommendation ―Reduce Air Flow Through Dispatch and

Jensen Ovens‖ should reduce fuel use in these production-related ovens. Fuel

use in these ovens is relatively insensitive to outdoor air temperature since the

ovens use indoor air for ventilation and combustion, and all parts enter the ovens



                                                                                      21
at the indoor air temperature of the plant. Thus, the energy savings from this

recommendation should reduce the weather-independent energy use, as

measured by 1, in the regression models. The total expected fuel savings from

this recommendation was 345 GJ/year (Table 2). However, applying Equation 12

to the data from the post-retrofit period indicates that weather-independent fuel

use actually increased by 181 GJ/year. Thus, no savings from this

recommendation could be measured. In part, the lack of measurable savings

results from the lack of production data in the model. In addition, the standard

error of the 1 regression coefficients is greater than the value of the coefficient.

Thus, in this case, this data set does not have enough resolution to identify

savings from these measures.


In comparison, the impact of the recommendation ―Reduce Night Setback

Temperature from 65 F to 60 F‖ is clearly apparent in the models, as the shift in

the change-point temperature from 18 °C (64.4 °F) to 16.5 °C (61.7 °F) (Figure 3).

Applying Equation 13 to the data from the post-retrofit period indicates that the

measured savings from reducing the night setback temperature are 779 GJ/year,

compared to the expected savings of 950 GJ/yr. Thus, this retrofit produced

significant and measurable savings.


The recommendations, ―Run Boiler in Modulation Mode‖ and ―Reduce Excess

Combustion Air in Boiler‖ recommend measures that improve the efficiency of the

boiler. The recommendation, ―Reduce Air Flow Through Dispatch and Jensen

Ovens‖ would reduce space heating use by reducing infiltration into the plant. The



                                                                                        22
impact of these recommendations is apparent in the models as the reduced slope

of the post-retrofit model (Figure 3). The slope in these models represents the

building load coefficient divided by the efficiency of the heating source. Thus, the

effect of increasing the efficiency of the boilers and reducing infiltration is

measurable as reduction in slopes, 2, from -2.108 to -1.921 GJ/day-°C (Equation

10). Applying Equation 14 to the data from the post-retrofit period indicates the

measured savings from improving boiler efficiency are 550 GJ/year, compared to

the expected savings of 1,444 GJ/yr. In subsequent work, the model for estimating

savings by switching to modulation mode was refined after field calibration showed

that stack losses with on/off control were less than originally predicted. Thus, the

expected savings used in the refined model are about 75% of the previous

estimate. [37]


In summary, the recommendations to reduce the building set point temperature at

night and to improve the efficiency of the boilers produced significant and

measurable savings. The effect of the recommendations to improve the efficiency

of the production ovens could not be measured with the available data.



6. CASE STUDY 2


The second case study demonstrates measuring adjusted savings and how the

method provides insight into the time dependence of savings. In this case study, a

tile manufacturer began to turn excess kiln burners off in February 2005 to reduce

energy use, but was forced to turn some burners back on in February 2006 due to



                                                                                    23
product quality issues. A time trend of normalized plant fuel use, with mean

models of pre and post-retrofit fuel use, is shown in Figure 5 The unadjusted fuel

savings, calculated by comparing the mean fuel use in the pre and post-retrofit

periods, is 2.2 units/month.


However, fuel use is correlated with both outdoor air temperature and production.

To account for these effects, 3PH-MVR models of normalized fuel use as a

function of outdoor air temperature and normalized production during the pre and

post retrofit periods are constructed (Equation 3). The model coefficients and

standard errors are shown in Table 5.


Pre and post-retrofit normalized fuel use, with a projection of the pre-retrofit model

during the post-retrofit period, is shown in Figure 6. The adjusted savings, which

accounts for changes in outdoor air temperature and production between the pre

and post-retrofit period, is 2.8 units/month. This is an increase of about 27% over

the unadjusted savings. In addition, close inspection of Figure 6 shows that

savings were not measurable at the start of the post-retrofit period, but gradually

increased as more burners were turned off. Further, Figure 6 shows that no

savings were apparent beginning in about February 2006 after burners were

turned back on for production quality purposes. This type of time-resolution can

help energy managers measure the persistence of savings over time, and possibly

act in a timely manner to correct problems.




                                                                                      24
Comparison of the coefficients in Table 5 shows that weather and production-

independent fuel use actually increased slightly during the post retrofit period while

production-dependent fuel use decreased as expected. Applying Equations 11,

13, 15 and 16 to this case study gives the disaggregated savings, which are shown

in Table 6. The results indicate that, as expected, the greatest savings are from

reducing production-dependent fuel use. The results also show that weather and

production-independent fuel use increased during the post-retrofit period, which

decreased the total savings. The increase in independent fuel use may be

attributable to other fuel using equipment in the plant or may be caused during

periods of reduced production when kiln burners are left on even when no tile is

being fired in the kiln. If so, the production-related savings associated with

decreasing the number of burners, is even greater than the total savings suggest.



7. CONCLUSIONS, LIMITATIONS AND FUTURE WORK


This paper presents a general method for measuring industrial energy savings and

demonstrates the method using case studies from actual industrial energy

assessments and energy efficiency projects. The method takes into account

changes in weather and production, and can use sub-metered data or whole plant

utility billing data. In addition to calculating overall savings, the method is able to

disaggregate savings into weather-dependent, production-dependent and

independent components. This disaggregation provides additional insight into the

nature and effectiveness of the individual savings measures. Although the method




                                                                                          25
incorporates search techniques and multi-variable least-squares regression, it is

easily implemented using data analysis software and readily available energy,

production and outdoor air temperature data. Use of the method to measure

savings can lead to greater industrial energy efficiency by identifying energy

conservation retrofits which do not perform up to expectations, providing data to

refine engineering methods for estimating savings, and redirecting resources to

retrofits that consistently produce the best results.


Although this method seeks to extract as much information about savings as

possible from easily obtainable utility billing, production and temperature data, the

extractable information is limited by the information in the data set, which is sparse

in the both the system and time domains. The most important limitation is in the

system domain, where the method attempts to determine savings from individual

subsystems using whole-plant energy use. One of the advantages of the use of

whole-plant energy use data is that the method may be able to capture the net

effect of synergisms, if present, between sub-subsystems. However, use of whole-

plant energy use necessitates the assumption that energy use from the non-

retrofitted systems is unchanged between the pre and post-retrofit periods. While

the method accounts for changes in overall production and weather, which are two

of the biggest factors influencing energy use, it cannot account for non-production

and non-weather related changes in other subsystems. Thus, the user must

determine if the energy use of non-retrofitted equipment changes between the pre

and post retrofit periods, and must adjust for these changes if substantial.




                                                                                    26
The sparseness of data in the time domain is less problematic for those interested

in the long term energy savings, say, on the order of the retrofit‘s payback period.

However, use of monthly data makes changes with time-intervals of less than one-

month invisible. For example, this method could provide no information about

whether energy savings were occurring on both weekdays and weekends. The

fact that multiple retrofits rarely are completed at the same time is more a problem

of lack of system resolution than time resolution; with greater subsystem

resolution, specific pre and post retrofit periods could be applied to each retrofit.

While the limitations noted above are real, they are attributable to the data set and

not inherent to the method. When the method is used with short time-interval or

sub-metered data, the limitations mentioned above are reduced or negated.


A second caution for the use of this method is also important; the conclusions

drawn from this analysis should be measured against the statistical uncertainty of

the results. The uncertainty of savings measured using regression modeling is

discussed in detail in Kissock et al, 1993; Reddy et al., 1998; and Kissock et al.,

1998. [15,16,9] Although the savings estimated by this method are based on

‗best-fit‘ models, and are thus the ‗best‘ estimate of savings available using this

method and data set, the uncertainty of the savings is often large. In these cases,

it is recommended that decisions derived from this estimate of savings include a

sensitivity analysis that reflects the reported uncertainty. In addition, as in all

regression, the uncertainty with which the individual model coefficients are known

is greater than the uncertainty of the overall model prediction. Thus, estimates of




                                                                                        27
disaggregated savings, which are based on the regression coefficients, are

inherently large. In cases where the standard error of the regression coefficient is

larger than the absolute value of the coefficient, no inferences based on this

coefficient should be attempted.


With these limitations in mind, future work will focus on improving understanding of

inter-correlations between the components of disaggregated savings, and on

improving understanding of interpretation of the uncertainty of the results.



ACKNOWLEDGEMENTS


We are grateful for support of this work from the U.S. Department of Energy

Industrial Technology Program, through the Industrial Assessment Center

program. We would like to thank Staco Energy Products Company, and especially

to Dan Sweda and Ed Kwiatkowski, for allowing us to publish these results, and for

the help and assistance provided throughout the assessment and analysis

processes. We are also grateful to Daltile Corporation and Jesse Hamiltion for

providing the data and information used in the second case study.



REFERENCES
 [1]   Sandberg, P. and Söderström, M., 2003, ―Industrial Energy Efficiency: The
       Need for Investment Decision Support from a Manager Perspective‖,
       Energy Policy, Vol. 31, No. 15, pp. 1623-1634.

 [2]   De Groot, H.L.F., Verhoef, E., and Nijkamp, P., 2001, ―Energy Savings by
       Firms: Decision-making, Barriers, and Policies‖, Energy Economics, Vol.
       23, Issue 6, pp. 717-740.




                                                                                   28
[3]   Ramestohl, S., Clases, C., and Prose, F., 1997, ―Duplicating the Success –
      From Positive Examples to Socio-economic Marketing Strategies for
      Greater Energy Efficiency in Industry.‖, Proceedings of the European
      Council for an Energy Efficient Economy (ECEEE) Summer Study,
      Spindleruv Mlýn, Czechia, June 9-14.

[4]   United States Department of Energy, 1996a. "North American Energy
      Measurement and Verification Protocol", DOE/EE-0081, U.S. Department of
      Energy, Washington, D.C.

[5]   Efficiency Valuation Organization, 2002, ―International Performance
      Measurement and Verification Protocol: Concepts and Options for
      Determining Energy and Water Savings: Volume 1‖, www.ipmvp.org.

[6]   United States Department of Energy, 1996b. "Measurement and
      Verification Guidelines for Federal Energy Projects ", DOE/GO-10096-248,
      U.S. Department of Energy, Washington, D.C.

[7]   American Society of Heating, Refrigeration and Air Conditioning Engineers,
      2002. ―Measurement of Energy and Demand Savings‖, Guideline 14-2002.
      Atlanta, Georgia.

[8]   Fels, M. 1986. "PRISM: An Introduction", Energy and Buildings, Vol. 9, pp.
      5-18.

[9]   Kissock, K., Reddy, A. and Claridge, D., 1998. "Ambient-Temperature
      Regression Analysis for Estimating Retrofit Savings in Commercial
      Buildings", ASME Journal of Solar Energy Engineering, Vol. 120, No. 3, pp.
      168-176.

[10] Kissock, J.K., 1993. "A Methodology to Measure Energy Savings in
     Commercial Buildings", Ph.D. Dissertation, Mechanical Engineering
     Department, Texas A&M University, College Station, TX, December.

[11] Kissock, K. and Seryak, J., 2004a, ―Understanding Manufacturing Energy
     Use Through Statistical Analysis‖, National Industrial Energy Technology
     Conference, Houston, TX, April 21-22.

[12] Kissock, K. and Seryak, J., 2004b, ―Lean Energy Analysis: Identifying,
     Discovering And Tracking Energy Savings Potential‖, Advanced Energy and
     Fuel Cell Technologies Conference, Society of Manufacturing Engineers,
     Livonia, MI, October 11-13.

[13] Seryak, J. and Kissock, K., 2005, ―Lean Energy Analysis: Guiding Energy
     Reduction Efforts to Theoretical Minimum Energy Use‖, ACEEE Summer
     Study on Energy in Industry, West Point, NY, July 19-22.

[14] Kissock, J.K., Haberl J. and Claridge, D.E., 2003. ―Inverse Modeling Toolkit
     (1050RP): Numerical Algorithms‖, ASHRAE Transactions, Vol. 109, Part 2.




                                                                               29
[15] Kissock, K., T. Agami, D. Fletcher and D. Claridge. 1993. "The Effect of
     Short Data Periods on the Annual Prediction Accuracy of Temperature-
     Dependent Regression International Solar Engineering Conference, pp. 455
     - 463.

[16] Reddy, T., J. Kissock and D. Ruch. 1998. ―Uncertainty In Baseline
     Regression Modeling And In Determination Of Retrofit Savings‖, ASME
     Journal of Solar Energy Engineering, Vol. 120, No. 3, pp. 185-192.

[17] Kissock, K., Haberl, J. and Claridge, D., 2002, ―Development of a Toolkit for
     Calculating Linear, Change-point Linear and Multiple-Linear Inverse
     Building Energy Analysis Models‖, Final Report, ASHRAE 1050-RP,
     November.

[18]        Haberl. J., Sreshthaputra, A., Claridge, D.E. and Kissock, J.K., 2003.
   ―Inverse Modeling      Toolkit (1050RP): Application and Testing‖, ASHRAE
   Transactions, Vol. 109, Part 2.

[19] Kissock, J.K., 1997. "Tracking Energy Use and Measuring Chiller Retrofit
     Savings Using WWW Weather Data and New ETracker Software", Cool
     Sense National Forum on Integrated Chiller Retrofits, San Francisco, CA,
     June 23-24.

[20] Kissock, J.K., 1999, ―ETracker Software and Users Manual‖, University of
     Dayton, Dayton, Ohio, http://www.engr.udayton.edu/weather/.

[21] EPA, 2006, United States Environmental Protection Agency, Energy Star
     Buildings Program,
     http://www.energystar.gov/index.cfm?fuseaction=labeled_buildings.showBui
     ldingSearch

[22] Kissock, J.K., 2005, ―Energy Explorer Software and User‘s Guide‖,
     University          of          Dayton,             Dayton,        Ohio,
     http://www.engr.udayton.edu/faculty/jkissock/http/RESEARCH/informaticn.ht
     m.

[23] Fels, M. and Keating, K., 1993, ―Measurement of Energy Savings from
     Demand-Side Management Programs in US Electric Utilities‖, Annual
     Review of Energy and Environment, 18:57-88.

[24] Fels, M., Kissock, J.K., Marean, M. and Reynolds, C., 1995. "PRISM
     (Advanced Version 1.0) Users Guide", Center for Energy and Environmental
     Studies, Princeton University, Princeton, NJ, January.

[25] Sonderregger, R. 1997. "Energy Retrofits in Performance Contracts: Linking
     Modeling and Tracking". Cool Sense National Forum on Integrated Chiller
     Retrofits, San Francisco, September 23-24.

[26] Sonderegger, R. A., 1998. "Baseline Model for Utility Bill Analysis Using
     Both   Weather and         Non-Weather-Related  Variables",     ASHRAE
     Transactions, Vol. 104, No. 2, pp. 859-870.



                                                                               30
[27] Anstett, M. and Kreider, J. F., 1993, “Application of neural networking
     models to predict energy use”, ASHRAE Transactions, Part 1; pp. 505-517.

[28] Kalogirou S.A., 2000, ―Applications of artificial neural-networks for energy
     systems‖, Applied Energy, Vol. 67, No. 1, September, pp. 17-35(19).

[29] Kissock, J.K., 1994. "Modeling Commercial Building Energy Use with
     Artificial Neural Networks", Intersociety Energy Conversion Engineering
     Conference, Vol. 3, pp. 1290-1295, Monterey, CA, August.

[30] Katipamula, S., Reddy, T.A., Claridge, D.E., 1998, ―Multivariate regression
     modeling‖, Journal of Solar Energy Engineering, Vol. 120, Issue: 3.

[31] Reddy, T.A. and Claridge, D.E., 1994, ―Using synthetic data to evaluate
     multiple regression and principal component analyses for statistical
     modeling of daily building energy consumption‖, Energy and Buildings. Vol.
     21, no. 1, pp. 35-44.

[32] Boyd, G., Tolley, G., Pang, J., 2002, ―Plant Level Productivity, Efficiency,
     and Environmental Performance of the Container Glass Industry‖,
     Environmental and Resource Economics, Volume 23, Number 1,
     September, Pages: 29 – 43.

[33] Boyd, G.A., 2005, ―A Method for Measuring the Efficiency Gap between
     Average and Best Practice Energy Use‖, Journal of Industrial Ecology,
     Volume 9.

[34] United States Department of Energy, 2006, Industrial Assessment Center
     Program,            Industrial           Technologies        Program,
     http://www.eere.energy.gov/industry/bestpractices/iacs.html

[35] University of Dayton Industrial Assessment Center, 2004.      ―Assessment
     Report 695‖, University of Dayton, Dayton, Ohio.

[36] Kissock, J.K., 1999b. ―UD EPA Average Daily Temperature Archive‖,
     (http://www.engr.udayton.edu.weather).

[37] Carpenter, K. and Kissock, K., 2005, ―Quantifying Energy Savings From
     Improved Boiler Operation‖, Industrial Energy Technology Conference,
     2005 New Orleans, Louisiana, May 11-12.




                                                                              31
Table 1. Disaggregation of Savings Into Components

Table 2. Estimated savings and implementation cost of the six implemented fuel-
related recommendations.

Table 3. Regression coefficients and standard errors for 3PH models of pre- and
post retrofit fuel use.

Table 4. Disaggregated savings from Case Study 1.

Table 5. Model coefficients and standard errors for 3PH-MVR models of pre and
post-retrofit fuel use.

Table 6. Disaggregated savings from Case Study 2.


Figure 1. a) 3P-cooing and b) 3P-heating regression models.

Figure 2. Time trend of pre- and post-retrofit fuel use with mean models of energy
use during both periods. The bold solid line represent pre-retrofit fuel use and the
bold dashed line represents post-retrofit fuel use.


Figure 3. Pre- and post-retrofit fuel use plotted against outdoor air temperature
with 3PH models through each data set. The solid line (upper) model represents
pre-retrofit fuel use and the dashed line (lower) model represents post-retrofit fuel
use.


Figure 4. Time trends of pre- and post-retrofit energy use, with a projection of the
weather-adjusted baseline model (light dashed line) during the post-retrofit period.
Savings are the difference between the adjusted baseline (light dashed line) and
actual post retrofit energy use (bold dashed line).


Figure 5. Time trend of pre- and post-retrofit fuel use with mean models of energy
use during both periods. The bold solid line represents pre-retrofit fuel use and the
bold dashed line represents post retrofit fuel use.


Figure 6. Time trends of pre- and post-retrofit energy use, with a projection of the
adjusted baseline model (bold dashed line) during the post-retrofit period. Savings
are the difference between the adjusted baseline (bold dashed line) and actual
post retrofit energy use (light dashed line).



                                                                                  32
Table 1
Disaggregation of Savings Into Components

Weather and
production-      E
                      d 1   1,Pr e   1, Post                                               (11)
independent       1
energy use
Weather-
                 E
                      d 2  TPost   3,Pr e   1,Pr e   1, Post 
dependent                                        
                                                                                                (12)
cooling energy    2
use
Weather-
                 E
                      d 2   3,Pr e  TPost   1,Pr e   1, Post 
dependent                                        
                                                                                                (13)
heating energy    2
use
                 E
                      d 3    2,Pr e TPost   3,Pr e 
Cooling                                                                        when
balance           3                                                       TPost   3, Post
temperature-                                                                                    (14)
                 E
                      d 3    2,Pr e  3,Pr e   3, Post 
                                                                                when
dependent
energy use        3                                                       TPost   3, Post
                 E
                      d 3    2,Pr e  3,Pr e   3, Post 
Heating                                                                         when
balance           3                                                       TPost   3, Post
temperature-                                                                                    (15)
                 E
                      d 3    2,Pr e  3,Pr e  TPost 
                                                                               when
dependent
energy use        3                                                       TPost   3, Post
Production-      E
dependent             d 4  PPost  4,Pr e   4, Post                                       (16)
                  4
energy use




                                                                                                  33
       Table 2

       Estimated savings and implementation cost of the six implemented fuel-related
       recommendations.

                                                 Expected Annual         Project      Simple
      Assessment Recommendation                       Savings             Cost       Payback
                                                Fuel (GJ) Dollars
AR 1: Run Boiler in Modulation Rather Than
On/Off Mode                                       1,039       $7,720      None       Immediate
AR 2: Reduce Thermostat Setpoint from 65 F
to 60 F on Nights and Weekends                     950        $7,056      None       Immediate
AR 3: Reduce Excess Air Flow Through
Dispatch and Jensen Ovens                          345        $1,575      None       Immediate
AR 4: Turn Off Exhaust Fans on IR Oven
When Not in Use                                    307         $2,281     $175        1 month
AR 5: Shut Off Boiler at the Beginning of May      106          $784      None       Immediate
AR 6: Reduce Excess Combustion Air in Boiler        98          $729      None       Immediate
Total                                             2,845       $20,145     None       Immediate


       Table 3
       Regression coefficients and standard errors for 3PH models of pre- and post
       retrofit fuel use.

             Coefficient       Units          Pre-retrofit        Post-retrofit
                 1           GJ/day        6.9954 ± 2.2544     7.492 ± 0.6302
                 2          GJ/day-°C      -2.108 ± 0.1841     -1.921 ± 0.0637
                 3             °C           18.01 ± 0.0061      16.49 ± 0.005




                                                                                       34
     Table 4
     Disaggregated savings from Case Study 1.
           Weather-independent savings                               -181 GJ/year

           Temperature set point or internal load savings            779 GJ/year

           Heating efficiency or building loss coefficient savings   550 GJ/year




     Table 5
     Model coefficients and standard errors for 3PH-MVR models of pre and post-
     retrofit fuel use.
                                                      Pre-retrofit      Post-retrofit

                                                            model                 model

R2                                                           0.72                  0.84

CV-RMSE (%)                                                  8.2                    5.2

1 (independent fuel-use)             Units/day         11.37 + 13.70       13.98 + 8.567

2 (temperature-dependence)          Units/day-°C     -0.2786 + 0.2515     -0.3928 + 0.1562

3 (balance-point temperature)            °C           27.38 + 0.0042      29.84 + 0.0042

4 (production-dependence)          Units/product     0.8270 + 0.1574      0.7452 + 0.0978


     Table 6.
     Disaggregated savings from Case Study 2.
             Weather and production-independent savings       -2.61 units/month

            Balance temperature dependent savings             -0.82 units/month

            Weather-dependent savings                         -0.69 units/month

            Production-dependent savings                      6.57 units/month




                                                                                          35
                            Figure 1. a) 3P-cooing and b) 3P-heating regression models.


                      60

                      50
  Fuel Use (GJ/day)




                      40                                                                            Pre-retrofit
                                                                                                    Post-retrofit
                      30
                                                                                                    Pre-retrofit Average
                      20                                                                            Post-retrofit Average

                      10

                        0
                      8/20/2002   3/8/2003   9/24/2003   4/11/2004 10/28/2004 5/16/2005 12/2/2005
                                                          Date

Figure 2. Time trend of pre- and post-retrofit fuel use with mean models of energy
use during both periods. The bold solid line represent pre-retrofit fuel use and the
bold dashed line represents post-retrofit fuel use.




                                                                                                                           36
                      70


                      60


                      50
  Fuel Use (GJ/day)




                      40


                      30


                      20


                      10


                      0
                               -10                     -5          0         5           10          15        20         25          30
                                                                                   Temperature (C)

 Figure 3. Pre- and post-retrofit fuel use plotted against outdoor air temperature
with 3PH models through each data set. The solid line (upper) model represents
pre-retrofit fuel use and the dashed line (lower) model represents post-retrofit fuel
                                         use.

                                           60

                                           50
                       Fuel Use (GJ/day)




                                           40

                                           30

                                           20

                                           10

                                             0
                                           8/20/2002        3/8/2003   9/24/2003     4/11/2004   10/28/2004   5/16/2005   12/2/2005
                                                                                       Date

Figure 4. Time trends of pre- and post-retrofit energy use, with a projection of the
weather-adjusted baseline model (light dashed line) during the post-retrofit period.
Savings are the difference between the adjusted baseline (light dashed line) and
actual post retrofit energy use (bold dashed line).



                                                                                                                                           37
                         120

                         100
  Fuel Use (units/day)




                         80

                         60
                                                                                                                                Pre-retrofit
                         40                                                                                                     Post-retrofit
                                                                                                 .
                                                                                                                                Pre-retrofit Average
                         20
                                                                                                                                Post-retrofit Average

                          0
                          1/4/2004                                7/22/2004           2/7/2005          8/26/2005        3/14/2006

                                                                                          Date



Figure 5. Time trend of pre- and post-retrofit fuel use with mean models of energy
use during both periods. The bold solid line represents pre-retrofit fuel use and the
bold dashed line represents post retrofit fuel use.

                                                      120

                                                      100
                               Fuel Use (units/day)




                                                       80

                                                       60

                                                       40
                                                                                                 .

                                                       20

                                                        0
                                                       1/4/2004           7/22/2004         2/7/2005         8/26/2005       3/14/2006

                                                                                                 Date



Figure 6. Time trends of pre- and post-retrofit energy use, with a projection of the
adjusted baseline model (bold dashed line) during the post-retrofit period. Savings
are the difference between the adjusted baseline (bold dashed line) and actual
post retrofit energy use (light dashed line).




                                                                                                                                                       38

								
To top