Download Comparing Industrial Energy Savings: Absolute, Intensity, and Regression and more Essays (university) Energy and Environment in PDF only on Docsity!
ECEEE INDUSTRIAL SUMMER STUDY PROCEEDINGS 389
The value of regression models in
determining industrial energy savings
Peter Therkelsen, Prakash Rao, Darren Sholes & Aimee McKane Lawrence Berkeley National Laboratory One Cyclotron Road Berkeley, CA 94720 USA
Bill Meffert & Randy Green Georgia Institute of Technology Enterprise Innovation Institute 75 5th^ Street NW Atlanta, GA 303308 USA
Sachin Nimbalkar Oak Ridge National Laboratory One Bethel Valley Road Oak Ridge, TN 37831 USA
Keywords
energy savings, modelling, energy management system, energy model, ISO 50001
Abstract
Determination of industrial facility energy savings is common- ly undertaken to comply with legal requirements, meet sus- tainability goals, and quantify greenhouse gas emissions, cost savings, and as one way to demonstrate energy performance improvement per the requirements of an energy management system standard, such as ISO 50001. At their most basic level, energy savings are determined through a process of account- ing for baseline and reporting period energy consumption fol- lowed by a calculation of energy savings as the difference in the levels of energy consumption between the two periods. The determination of industrial facility-wide energy sav- ings can be undertaken in a wide variety of ways and in rela- tion to energy performance improvement; energy savings can be expressed as any number of energy performance indicators (EnPIs). Energy savings values will vary depending upon if an absolute, intensity, or regression basis of calculation is selected for use. The use of one of the three basis of energy savings cal- culation will alter the relevance, meaning, and comparability of the resulting energy savings values To increase consistency in the determination of energy savings values, standards such as the International Performance Measurement and Verification Pro- tocol (IPMVP), ISO 17747 – Determination of Energy Savings in Organization, and the U.S. Department of Energy (U.S. DOE) Superior Energy Performance (SEP) program Measurement and Verification Protocol (SEP M&V Protocol) have been published. Industrial facilities in the United States, Mexico, and Cana- da are certified to the U.S. DOE SEP program after becoming
ISO 50001 certified and verification by a third party of calcu- lated energy savings per the SEP M&V Protocol. This paper discusses some of the different meanings energy savings values will have when calculated by using the three different bases. Additionally, data from five SEP certified industrial facilities are used to calculate energy savings values using the three bases and compared as energy performance improvement percentage (EPI %) values. Discussion of why a large degree in variation in EPI % values are seen for any given set of data is provided. Ultimately, while each energy saving calculation basis has merit and can be used appropriately in its own context, the regres- sion basis is shown to best translates energy savings values into contextualized energy performance improvement values by conveying comparable information regarding operational and behavioural improvements not otherwise observable with the absolute or intensity bases.
Introduction
The determination of energy savings is conducted for a wide variety of reasons, including to comply with legal requirements, meet sustainability goals, quantify greenhouse gas emissions and cost savings, and as one way to demonstrate energy perfor- mance improvement per the requirements of an energy man- agement system standard, such as ISO 50001 – Energy manage- ment system standard – Requirement with guidance for use. This wide variety of reasons to determine energy savings is matched with an equally wide variety of methods detailing processes to conduct energy savings determination. Different energy cal- culation methods will be expressed as different energy perfor- mance indicators (EnPIs): quantitative values or measures of energy performance.
3-079-16 THERKELSEN ET AL
390 INDUSTRIAL EFFICIENCY 2016
- ENERGY MANAGEMENT: THE NUTS AND BOLTS
At their most basic level, energy savings are determined through a process of calculating the difference between energy consumption accounted for during an established baseline and reporting period. Adjustments to energy consumption values prior to calculation of energy savings are sometimes performed in order to impart additional meaning to calculated energy sav- ings values. These adjustments create different EnPIs that can convey context regarding uses and independent variables that affect energy consumption and thus energy performance levels. Rarely can the entirety of energy performance improvement be understood by a singular EnPI and thus a variety of methods of calculating energy savings may be of use (Goldstein and Al- maguer, 2013). A variety of internationally relevant standards present meth- ods to transparently determine energy savings values. These standards include International Performance Measurement and Verification Protocol (IPMVP), ISO 17747 – Determina- tion of Energy Savings in Organizations, and the U.S. Depart- ment of Energy (U.S. DOE) Superior Energy Performance (SEP) program Measurement and Verification Protocol (SEP M&V Protocol). The SEP M&V Protocol is used across North America and was drawn from in the formation of ISO 17747. In turn, the recent revised version of the SEP M&V Protocol has benefitted from the international development of ISO 17747. Methods to determine facility-wide energy savings at indus- trial facilities are typically based upon either a bottom-up or top-down energy accounting approach. These approaches can be used in isolation or together such that one approach can reconcile the other. A bottom-up approach to determining facility-wide energy savings conducts accounting of energy consumption before and after the implementation of one or more energy perfor- mance improvement actions. The bottom-up approach can be used to determine facility-wide energy savings by aggregating the saving from multiple energy performance improvement actions. The additional expense of sub-metering may be re- quired to be placed on individual equipment, systems, or pro- cesses. Bottom-up energy savings are useful for focusing on a specific or a set of energy performance improvement actions. Complications due to interactive effects and difficulty in quan- tifying savings from operational or behaviour based energy performance improvement actions can cause uncertainty in energy savings values. Bottom-up approaches are widely used throughout industry, they often do not require modelling and rely on pre and post engineering estimates or metered data for determining the energy savings. A top-down approach to determining facility-wide energy savings conducts accounting on a facility basis, that is, most commonly at the physical boundaries of a facility. Utility meters are typically the source of data and the contribution of individual energy performance actions within the facility boundaries is not determined. This approach provides the overall savings level for the facility, thus avoiding issues related to the implementation of multiple energy performance improvement actions such as in- teractive effects or double counting. This paper will focus solely on energy savings calculated with the top-down approach. The calculation of energy savings with a top-down approach can conducted on one of three bases; absolute, intensity, or nor- malization. The absolute energy savings basis utilizes unadjusted energy consumption values while both the intensity and normal-
ization bases make adjustments to energy consumption values. Adjustments to energy consumption values are done to impart context and additional meaning to calculated energy savings values. Industrial organizations as well as national and regional energy efficiency and performance improvement programs use all three of these calculation bases for various reasons. Relevant variables are independent variables such as pro- duction or weather that directly affect the amount of energy consumed within the facility boundaries. These variables can be used to normalize energy consumption as part of an adjust- ment model. As part of the energy accounting, relevant vari- ables and energy consumption data are both recorded. Most energy savings standards require relevant variables to be physi- cal quantities, characteristics, or conditions. Financial metrics such as product price or energy costs are not allowed because they lack a physical relationship to energy consumption. Whether simple or complex expressions of performance, en- ergy savings values can be misinterpreted or misrepresented as a measure of energy performance improvement (Goldstein and Almaguer 2013, Goldstein et. al., 2011) In most cases, this can be linked to selecting the wrong energy savings calculation method and basis to convey accurate meaning regarding ener- gy performance improvement for the user. In order to be used effectively, EnPIs must be designed and used for the intended and specific purpose. This paper discusses the various reasons why one of the three bases of energy savings calculation may be selected for use over another. Additionally, variation in cal- culated energy savings values resulting from the different bases is demonstrated by applying each of the three bases to facility- wide industrial energy consumption and relevant variable data from five SEP certified facilities. A discussion of why a variation in the calculated energy savings values occurs is presented.
Absolute, Intensity, and Normalization Bases
A top-down approach to energy savings is applicable when considering energy savings across multiple energy end-uses, such as a process or facility. The top-down approach simul- taneously captures the effects of multiple energy performance improvement actions, including those that may be difficult to enumerate, such as behavioural actions. However, these ap- proaches may rely on developing proxies or relationships to ad- just energy consumption to reflect energy productivity. While these adjustments are intended to isolate the effect of energy performance improvement actions, they may introduce errors. Three bases, absolute, intensity, and normalization, used in top-down facility-wide energy savings determination are eval- uated in this paper. The selection of a basis will often depend on the desired use of the energy savings estimate. Each basis can be applied to a single energy type or to multiple energy types in aggregate.
ABSOLUTE BASIS The absolute savings basis involves comparing unadjusted ener- gy consumption data from two time periods, with the difference being the energy savings. Energy savings are reported in units of energy (e.g. kWh, MJ, MMBTU), or as an energy performance improvement percentage. This basis can be beneficial to organi- zations interested in reporting progress towards absolute goals to reduce greenhouse gas emissions or for those that are unable
3-079-16 THERKELSEN ET AL
392 INDUSTRIAL EFFICIENCY 2016
- ENERGY MANAGEMENT: THE NUTS AND BOLTS
timate of energy savings resulting from energy performance improvement actions by accounting for the effects of relevant variables such as production variation and weather. Four types of normalization are commonly used and are described below. Analysis in this paper utilizes only forecast normalisation.
Forecast Normalization Forecast normalization results in a model of baseline period energy consumption that is applied to the reporting period relevant variable values to calculate adjusted baseline period energy consumption for comparison with observed (actual) reporting period energy consumption. The adjusted baseline period energy consumption is an estimate of the energy con- sumption that would have been expected at reporting period relevant variable values, if the baseline operating systems and practices were still in place during the reporting period.
Retrospective Normalization Retrospective normalization results in a model of reporting pe- riod energy consumption that is applied to the baseline relevant variable values to calculate adjusted reporting period energy consumption for comparison to observed (actual) baseline pe- riod energy consumption. The adjusted reporting period ener- gy consumption is an estimate of the energy consumption that would have been expected at baseline period relevant variable values, if the reporting period operating systems and practices were in place during the baseline period.
Standard Condition Normalization Standard condition normalization results in two adjustment models, one of baseline period energy consumption and one for reporting period energy consumption. Standard conditions are applied to each of the models to calculate adjusted energy consumption values. The adjusted energy consumption for each period is the estimated energy consumption that would have been expected at a standard set of conditions (relevant variable values).
Chaining Normalization Chaining normalization is a composite of the forecast and ret- rospective normalization methods. The chaining method may be used if there is a 12-month (1 year) intermediate period be- tween the baseline period and the reporting period for which a model can be developed In this case, the intermediate period model is used to backcast to the baseline period year and fore- cast to the reporting period year. Chaining normalization may be useful if the reporting period conditions for energy con- sumption and relevant variables are outside the range of con- ditions of the baseline period model, and the baseline period conditions are outside the range of conditions of the reporting period, or a model cannot be developed for either the reporting or baseline periods.
ISO 50001 and the U.S. DOE SEP M&V Protocol
Published in June 2011, ISO 50001 – Energy management sys- tem standard – requirements with guidance for use, is an in- ternational standard that provides a framework for the imple- mentation of an energy management system (EnMS) for the purpose of continuously improving energy performance (ISO
2011). ISO 50001 provides guidance to industrial and commer- cial facilities for integrating energy efficiency into their daily management practices. For industrial facilities this includes fine-tuning production processes and improving the energy efficiency of industrial systems. The standard gives organiza- tions management strategies that can be used to reduce energy consumption, carbon emissions, and costs, and to improve environmental performance. Organizations implementing ISO 50001 conduct an energy review, develop an energy policy, establish objectives, targets and action plans related to its sig- nificant energy uses, and engage top management in decision making. Though quantitatively unspecified, ISO 50001 does re- quire that an organization demonstrate improvement in energy performance. Many organizations seek to demonstrate energy performance improvement through the calculation of energy savings. The basis on which energy savings values are calcu- lated for ISO 50001 will vastly change the contextualization of the demonstrated energy performance improvement. A large number of countries have nationally adopted ISO 50001. These countries are encouraging uptake of ISO 50001 within their industrial sector in a wide variety of ways. To facilitate clear understanding of energy performance improvement resulting from ISO 50001, countries have de- veloped programs requiring facility-wide top-down energy savings determined using either the absolute, intensity, or normalization basis. The U.S. DOE developed the Superior Energy Performance (SEP) program in collaboration with industry and other stakeholders. The SEP program requires participating indus- trial facilities to achieve ANAB third-party verification of its ISO 50001 EnMS and meet pre-established energy perfor- mance improvement targets, of at least 5 % over 3 years (US DOE 2016). The SEP certification program provides industrial facilities and companies a transparent, globally accepted sys- tem for verifying improvements in energy performance and management practices through the SEP M&V Protocol. As of May 2016, 35 facilities representing a diverse range of sectors, sizes, and locations are certified to SEP, improving their energy performance on average by 12 % over 3 years (US DOE 2016). The success of the program’s initial demonstration has led to a growing interest in SEP.
The SEP M&V Protocol sets forth the verifiable method-
ology for determining and demonstrating achievement of the energy performance improvement level claimed by an organi- zation for a defined facility. The SEP M&V Protocol contains requirements necessary to demonstrate, through the SEP pro- gram specific energy performance indicator (EnPI), SEnPI, the achieved energy performance improvement for the SEP program. The SEP M&V Protocol was initially developed and released in 2012 and based upon stakeholder feedback, the SEP program has been updated, requiring modification of the 2012 version of the SEP M&V Protocol. The structure and language of the SEP M&V Protocol was updated to reflect common ele- ments between ISO 50001 and supporting standards such as ISO 50015 – Measurement and verification of organizational energy performance – General principles and guidance and ISO 17747 – Determination of energy savings in organization. The EnPI Tool is a regression analysis based Excel Add-in developed by the U.S. Department of Energy. The Tool was created to help plant and corporate managers establish a nor-
- ENERGY MANAGEMENT: THE NUTS AND BOLTS
ECEEE INDUSTRIAL SUMMER STUDY PROCEEDINGS 393
3-079-16 THERKELSEN ET AL
malized baseline of energy consumption and track annual progress of intensity improvements, energy savings, and SEP EnPIs, while accounting for variations due to weather, produc- tion, and other relevant variables. The EnPI Tool is available as a free download from the U.S. DOE Advanced Manufacturing Office Energy Resources Centre. The EnPI Tool will accept any relevant variables that are input by the user. However, it is up to the user to ensure SEP allows for all relevant variables included in the model. For example, although the EnPI Tool would ac- cept them, SEP does not allow for the inclusion of financial based relevant variables because they do not have a direct effect on energy consumption. The Tool requires monthly or weekly energy consumption data. The energy consumption can be adjusted for multiple variables such as heating degree days (HDD), cooling degree days (CDD), dew point temperature, product output, moisture content of the product, shift sched- ule adjustments, etc. Regression analysis can be performed for multiple energy types separately, or for aggregated energy con- sumption. It is also capable of modelling multiple parts of a facility independently, and then aggregating final results.
Methodology
The three different bases; absolute, intensity and normalization, were used to determine facility-wide top-down energy savings for five industrial facilities using real data collected for the SEP program. While the basis for energy savings calculation was varied, all other components of the determination were held constant, including the energy accounting methodology and statistical tests for normalization regression as defined by the updated SEP M&V Protocol. For all bases, data were analysed on a monthly basis for both the baseline and the reporting periods. Energy consumption and relevant variable data were recorded by energy type in units of delivered and primary. The Primary Energy Consump- tion (ECP), the source energy consumption for all energy types summed over a specified time period, was calculated for the baseline period and reporting period for each of the five facili- ties. The relevant variable quantities were individually summed for the baseline and reporting period for the energy intensity calculation methodology.
ABSOLUTE BASIS The absolute basis provides a simple, straightforward basis for calculating energy savings. With the absolute basis, ECP (^) b is compared directly to ECP (^) r to establish an absolute energy sav- ings value which can be translated to an energy performance improvement percentage value. Energy savings and energy performance improvement percentage for the absolute basis are described as:
INTENSITY BASIS Intensity basis uses a very simple form of energy adjustment model to adjust energy consumption values for a singular rel- evant variable. The intensity basis provides a basic level of con-
text and meaning to energy savings and energy performance improvement percentage values as at least one factor affecting energy consumption is considered. While the intensity basis accounts for the effects of one relevant variable, in many cases multiple relevant variables typically affect energy consump- tion. Variations in these other relevant variables will not be reflected in the intensity basis calculation, potentially leading to misinterpretation regarding the impact energy performance improvement actions have had. Intensity based EPI % values were calculated for all relevant variables provided. Energy sav- ings and energy performance improvement percentage for the intensity basis are described as:
NORMALIZATION BASIS Normalization basis uses adjustment models to adjust energy consumption values for multiple relevant variables. Normali- zation can be conducted in a forecast, backcast (retrospec- tive), or standard (reference) conditions approach. Regression analysis is a statistical technique that estimates the dependence of energy consumption on one or more independent relevant variables. Regression analysis can reveal the affect energy per- formance improvement actions have on facility-wide energy performance by removing the influence relevant variables such as production variation and weather. The analysis in this paper focuses solely on the forecasting approach to normalization. The U.S. DOE EnPI Tool, a regression analysis based Excel Add-in that follows the SEP M&V Protocol was used to calcu- late normalized energy savings and EPI % used in this paper. Forecast normalization requires the formation of a baseline period energy consumption adjustment model. This model is applied to reporting period relevant variable values to calcu- late adjusted baseline period energy consumption (ECPb|r a) for comparison with observed (actual) reporting period energy consumption (ECPr). The adjusted baseline period energy con- sumption is an estimate of the energy consumption that would have been expected given reporting period relevant variable values, if the baseline operating systems and practices were still in place during the reporting period. The adjusted baseline pe- riod energy consumption is described as:
where x (^) i is the relevant variable quantity during the reporting period, b 0 is the modelled primary energy consumption level assuming all relevant variables have a zero value (sometimes referred to as baseload energy or y-axis intercept of the model), and bi > 0 is the incremental energy consumption per unit of that relevant variable (coefficient), determined using multiple linear regression analysis on baseline period consumption and relevant variable quantities. Modelled baseline period primary energy consumption can be compared to the actual reporting
ESa = ECP! − ECP!
EPIa % = 1 −
ECP!
ECP!^ ×^100
EIb =
ECPb RVb
EIr =
ECPr RVr
EPIi % = 1 −
EIr EIb^100
ECP!^ !|^ != b 0 + b 1 x 1 + b 2 x 2 + … + bkxk
- ENERGY MANAGEMENT: THE NUTS AND BOLTS
ECEEE INDUSTRIAL SUMMER STUDY PROCEEDINGS 395
3-079-16 THERKELSEN ET AL
sectors including chemical, power generation equipment man- ufacturing, automotive assembly, and food processing. Additionally, facilities supplied relevant variable data. The type and number of relevant variables differed for each facility. As such, multiple statistically valid regression models for each facility were created.
Results
For each of the five facilities analysed, absolute, intensity, and normalization basis EPI % values are shown in Figure 3. For each basis and facility, a box-and-whisker plot is shown. The upper and lower whiskers bound the maximum and minimum EPI % value while the box provides the first and third quartiles of the dataset. The median value is indicated with a horizontal line across the box. For each facility there is only one absolute basis EPI % value while the number of intensity and normaliza- tion (regression) EPI % values varies depending upon the num- ber of relevant variables provided by the facility and statistically valid adjustment models available. Facility 1 relevant variables are called out in Figure 2 as they are analysed later in greater de- tail. For each facility, an x marker indicates the regression basis energy performance improvement level for which the facility was certified to the SEP program. For each facility, there is a high degree of variability of the calculated EPI % values between the three calculation meth- odologies. While vastly different, none of the resulting energy performance improvement percentage values should be con- sidered as incorrect. While the absolute basis energy savings value clearly conveys information regarding the change in energy consumption, the intensity and regression bases en- ergy savings values should be understood to be estimations, or rather, representations of energy savings that provide more de- tail related to energy performance improvement than absolute energy savings values alone. Results generated from the appli- cation of each basis relay different contextualized information regarding energy performance at the facility. Depending upon their need, different end users may find more value in results from one basis over another. The absolute basis calculates actual energy savings by com- paring unadjusted energy consumption data from two time periods, the baseline period and the reporting period. This basis can be beneficial to organizations interested in report- ing progress towards absolute goals to reduce greenhouse gas emissions or for those that are interested in energy cost sav- ings. However, the absolute basis does not provide any context for the savings numbers or any basis for comparing facilities of varying size and sector, nor does it account for the seasonal and operational variability that may have a large effect on a facility’s energy consumption. Facility 1 provides an example of extremely high variability of EPI % in the use of the absolute, intensity and normalization bases. On an absolute basis, Facility 1 has a calculated EPI % value of -19.8 %, reflective of the fact that the facility consumed more energy in the reporting period than during the baseline period (top most graph in Figure 4). This result is of value to users seeking information related to information related to the abso- lute cost of energy consumed or total CO 2 emissions from the facility. However, the result does not provide context for chang-
es in relevant variables, such as production which increased from the baseline period to the reporting period (middle graph in Figure 4). The intensity basis provides a specific contextual relation- ship between energy performance improvement percentage and the changes in one relevant variable. However, as shown in Figure 3, intensity basis calculated EPI % can be highly vari- able and is greatly dependent on the relevant variable that is selected for use. For Facility 1, using production output as the normalization metric leads to a calculated EPI % of 24.0 %. Us- ing heating degree days (HDD) as the normalization metric, the calculated EPI % is -48.0 %. As seen in the middle and bot- tom graphs of Figure 4, reporting period production is greater than during the baseline period while reporting period HDD is lower than during the reporting period. The divergent direc- tion of these relevant variables is the cause of the dramatically different intensity basis calculated EPI % values. It should be noted that when using the intensity basis, the selection of the relevant variable used in the calculation is typically chosen based upon engineering knowledge of the most energy inten- sive operation in the facility. If there is a clear understanding that one relevant variable represents the dominant energy use at the facility, then use of that intensity basis EPI % may prove to be of value. Typically for an industrial facility this is a metric of production output though in some assembly facilities HVAC or other non-pro- duction metric may be known to be the largest energy use. The intensity EPI % values shown in Figure 3 reflect all potential intensity values based upon the provided relevant variables. As shown in Figure 5, electricity consumption is highly cor- related (high R 2 value) with production output while natural gas consumption is highly correlated with heating degree days. Total energy consumption shows a stronger correlation with production output than HDD; however, the correlation is lower than when the energy types were disaggregated. As seen in Fig- ure 2, electricity makes up approximately 75 % of Facility 1 total energy consumption, production would intuitively provide a
Figure 3. Comparison of Energy Savings Calculation Methodologies.
3-079-16 THERKELSEN ET AL
396 INDUSTRIAL EFFICIENCY 2016
- ENERGY MANAGEMENT: THE NUTS AND BOLTS
stronger correlation to total energy consumption than HDD. However, by using only production output, any contextual in- formation associated with HDD is lost. The U.S. DOE EnPI Tool was used to conduct normaliza- tion basis regression analysis of Facility 1 energy consumption and relevant variable data. For each energy type consumed by Facility 1, only one regression model was statistically valid per the requirements of the SEP M&V Protocol. Electricity con- sumption was modelled using production as the independent relevant variable, and natural gas consumption was modelled using HDD as the independent relevant variable. The mod- elled consumption for each energy type was then aggregated
to establish the total modelled energy consumption. Using this model combination, the calculated EPI % was 5.6 %. The benefit of using the forecasting basis is twofold. First, it is capable of modelling the effect of multiple variables, for each energy type. Second, the model selection process relies on both statistics and intuition together, rather than just one of the two, to choose normalization metrics. In the case of Facility 1, the application of statistical tests and engineering logic resulted in an EPI % value that meaningfully reflects the energy perfor- mance improvement actions as well as changes in production. It is important to note that forecasting adjustment models are only as good as the data used to generate them, as well as the stringency of the statistical criteria that are required of them. Engineering logic should be applied to determine if the model is representative of the operations within the facility bounda- ries and when deciding which statistically valid model combi- nation to use to calculate a final EPI % value. The critical nature of this logical evaluation is demonstrated by the normalization basis results obtained for Facility 4. Generically for the five facilities, there tends to be a larger degree of variability in the intensity basis results than the fore- casting basis results when multiple relevant variables were provided. The primary reason for the decreased range in vari- ability is that the normalization basis determines and applies a weighting coefficient for each relevant variable. This reduces the relative impact of relevant variables that do no affect energy consumption as much as other relevant variables. In the case of Facility 1 this is clearly seen by the vastly different results when either production or HDD are used. HDD probably would not have been used as an intensity basis EPI value for Facility 1 as applying engineering logic shows the largest energy using sys- tems at the facility are directly related to production, not cli- mate control. However, without the use of regression modelling analysis, this cannot be quantitatively proven. Additionally, this may be due to the fact that the forecast- ing basis per the SPE M&V Protocol checks for the statistical strength of the correlation between the relevant variables and energy consumption. Model combinations with weak correla- tion are rejected prior to being considered by the user, while regression models that pass the statistical tests give rise to simi- lar calculated EPI % values. For the intensity basis, there are no checks or tests in place to determine the strength or validity of the normalization metrics. However, this trend does not hold for Facility 4, which has a larger degree of variability in the EPI % values calculated using the forecasting basis than with the intensity basis. Facility 4 is unique in that there is extreme seasonal variability in its opera- tion. There are two relevant variables that are assumed to affect energy consumption: monthly production level and average monthly ambient temperature. Therefore, for a given energy type, there are three possible models. Energy consumption as a function of:
- Production output,
- Ambient temperature, or
- Production output and ambient temperature. Facility 4 consumes electricity, natural gas and steam, which means there are 9 possible independent models (3 for each en- ergy type). The need to aggregate the modelled consumption
Figure 4. Changes in Baseline and Reporting Period Relevant Variables and Energy Consumption for Facility 1.
3-079-16 THERKELSEN ET AL
398 INDUSTRIAL EFFICIENCY 2016
- ENERGY MANAGEMENT: THE NUTS AND BOLTS
Absolute energy savings values represent the actual change in energy consumption at a facility, intensity and normaliza - tion basis savings values are adjusted and are thus estimations of actual savings. Absolute savings values however do not provide insights to the actual impact energy performance im- provement action have had on energy performance as chang- es in other factors such as production and weather may have changed. Intensity basis savings values provide additional meaning relative to changes in a singular variable as compared to abso- lute savings. However, the limitations imposed by the use of a singular variable, and with no statistical method of determin- ing if that variable has a large effect on energy consumption, results from the intensity basis can be misleading. Intensity basis should be used only if all external influences can be held constant over the period being analysed. Regression analysis adjusts energy consumption values in order to remove the influences of relevant variables such as production and weather so the effect of energy performance improvement actions can be clearly quantified. Unlike the in- tensity basis, he regressions basis can account for a large num- ber of variables and statistical tests can be applied, providing confidence in the relationship between energy consumption and the variables being controlled for. Engineering logic, in addition to statistical tests, must be applied to ensure that the relationship between relevant variables and energy consump - tion being used is reflective of operations in the facility. EPI % values based on each of the three bases were cal- culated from energy consumption and relevant variable data from five SEP certified industrial facilities. Results show that even with the same input data from a given facility, EPI % values can vary greatly depending upon the basis utilized. Ad- ditionally, when more than one outcome is valid for a given basis, variation in the results can occur. A deep examination into the variations in intensity and normalization bases was performed. Variations in EPI % values from these bases are due to the relationship and the relative statistical strength, or lack thereof, between individual relevant variables and energy consumption data. Additionally, normalization basis also allows for the direct comparison of EPI % across a set of facilities as well as mean- ingful comparison across different regions and types of facili- ties (e.g., comparing EPI % from an iron and steel facility in China to an automotive assembly facility in Germany). The other measures by their very nature are specific to the location and operation of the facility. By providing an analysis that nor- malizes local impacts like weather and facility specific variables like production, the normalized regression analysis provides a meaningful comparison of energy performance improvement regardless of location and facility type. Lastly, the accelerated pace of change in the global climate and economy necessitate the need for energy managers and facility managers to effec- tively understand their energy performance improvement in light of abnormal weather conditions is a key argument for the case of normalized basis regression analysis of EPI %. Normal- ized regression is the best tool for evaluating energy perfor- mance given the changing nature of what is considered normal weather. A logical conclusion of this analysis is that along with reason- able statistical tests and engineering logic, the normalization
leads to an EPI % of 48.8 %. A closer examination into the ex- treme variability in EPI % focuses on the use of ambient tem- perature as a relevant variable. Plotting the two relevant variables against each other in Fig- ure 7 shows that ambient temperature is highly correlated with production output, suggesting that the correlation between energy consumption and ambient temperature may be coin- cidental. Production at the facility requires large quantities of natural gas to operate heated process equipment during sum- mer months when ambient temperatures are independently elevated. Therefore, as production output is the result of the primary operation at the facility and that it has a stronger sta- tistical correlation to energy consumption, it should be selected as the relevant variable for use. Additional considerations for Facility 4 could include alter- ing the average temperature relevant variable to one that is more traditional of year-to-year weather variability, such as CDD. Though still weaker than production output, the use of CDD as the relevant variable shows stronger statistical perfor- mance than average ambient temperature for Facility 4. CDD is also strongly correlated with production output, once again indicating that the strong correlation with energy consump- tion may be coincidental, due to the seasonal dependence of production. Despite the statistical validity and greater EPI % of the other models, the adjustment model with the strongest statistical per- formance and agreement with engineering logic was selected for use in seeking SEP certification for Facility 4. While regres- sion basis EPI % values inherently pose greater context than those calculated via the absolute or intensity bases, it is the statistical and logical tests that give confidence that the EPI % value is representative and meaningful.
Conclusion
Energy savings and EPI % values can be calculated on an ab- solute, intensity, and normalization basis. Each of these bases has merit in their own right though all three are contextually different and convey different meaning. An appropriate scope of each basis has been presented and discussion relative to their use presented.
Figure 7. Relationship between the two Facility 4 Relevant Variables.
- ENERGY MANAGEMENT: THE NUTS AND BOLTS
ECEEE INDUSTRIAL SUMMER STUDY PROCEEDINGS 399
3-079-16 THERKELSEN ET AL
ESa Energy savings calculated on an absolute basis ES (^) i Energy savings calculated on an intensity basis ES (^) n Energy savings calculated on a normalization basis
References
Goldstein, D. and Almaguer, J. (2013) Developing a Suite of Energy Performance Indicators (EnPIs) to Optimize Out- comes, 2013 ACEEE Summer Study on Energy Efficiency in Industry, Niagara Falls, NY. Goldstein, D., McKane, A., and Desai, D. (2011) ISO 50001: Energy Management Systems: A Driver for Continual Im- provement of Energy Performance, 2011 Summer Study on Energy Efficiency in Industry, Washington D.C. International Standards Organization (2016) ISO 17747 – Determination of energy savings in organizations. International Standards Organization (2014) ISO 50015 – Measurement and verification of organizational energy performance – General principles and guidance. International Standards Organization (ISO) (2011) ISO 50001
- Energy management system standard – Requirement with guidance for use. U.S. Department of Energy (U.S. DOE) (2016a) Superior Energy Performance, http://www.energy.gov/eere/amo/ superior-energy-performance, accessed May 1, 2016. U.S. Department of Energy (U.S. DOE) (2016b) Superior Energy Performance Measurement and Verification Pro- tocol, Washington DC.
basis of calculating energy savings and EPI % values though ad- justment model regression analysis provides the most contex- tually meaningful energy savings value results that can be used to demonstrate energy performance improvement. However, consideration of why an energy savings value is needed should be taken into account, realizing that the energy performance improvement of an organization may be best represented by a combination of EnPIs generated by calculating energy savings value from a number of the bases detailed.
Nomenclature
EPIa Energy performance improvement (%), baseline to reporting period on an absolute basis EPI (^) i Energy performance improvement (%), baseline to reporting period on an intensity basis EPI (^) n Energy performance improvement (%), baseline to reporting period on a normalization basis ECP (^) b Primary energy consumption of all energy types summed over baseline period ECP (^) r Primary energy consumption of all energy types summed over reporting period RV (^) b Relevant variable for the baseline period RV (^) r Relevant variable for the reporting period EI (^) b Energy intensity for baseline period EI (^) r Energy intensity for reporting period ECPb|r a Modelled baseline period primary energy consumption adjusted to reporting period conditions
