• No results found

Energy and Water Usage in the Manufacturing Industry : A study case to analyse, compare and decide where to reduce energy and water utilization

N/A
N/A
Protected

Academic year: 2021

Share "Energy and Water Usage in the Manufacturing Industry : A study case to analyse, compare and decide where to reduce energy and water utilization"

Copied!
87
0
0

Loading.... (view fulltext now)

Full text

(1)

ENERGY AND WATER USAGE IN THE

MANUFACTURING INDUSTRY

A study case to analyse, compare, and decide where to reduce energy and water

utilization.

JORGE LÓPEZ

YULLY RINCÓN

School of Business, Society and Engineering Course: Thesis work in sustainable energy systems Course code: ERA401

Credits: 30hp

Program: Civilingenjörsprogrammet I energisystem/Master of Science in Sustainable Energy Systems

Supervisor: Valentina Zaccaria, MDH. Examinor: Erik Dahlqvist, MDH. Customer: Johan Åhlund, Alfa Laval. Date: 2020-06-14th

Email:

jlo14001@student.mdh.se Yro18001@student.mdh.se

(2)

ABSTRACT

Increasing concern about global climate change has led to a growing interest in energy usage and water consumption. It is well known that changes in consumption habits lead to more efficient use of energy and water sources. Nowadays, globalization, environmental concerns, and the shortage of resources have led to an increase of stakeholder pressure on companies to expand their focus to sustainability. Also, the high impact that the savings can have in the financial status of the company. It is encouraging the headboards to study and improve the ways water and energy are being used within the processes. Significant economic savings and benefits for the environment could be achieved with slight changes in the company.

As an overview, this project starts with the extraction of data from a platform for energy management in an industrial company. Then, it goes through the understanding of the energy and water usage data set. Later, a methodology to handle and process the data will be set. It is intending to extract relevant information using clustering. The idea is to compare the usage profiles between different factories, using key performance indicators and reducing the initial data set. Once the benchmarking is performed, some critical parameters will be selected to support the decision-making process related to investments to reduce the energy usage and water consumption in a specific location. Finally, the case of study will be implemented with the measurements from Alfa Laval.

We will study how, from daily measurements with a very low investment and using the proper algorithms and methodologies, the main behaviours and features in an industrial location can be extracted from the utilization data. These characteristics can be used to develop strategies or productions schemes based on the interests of the energy manager and the company.

Keywords: Clustering, DB-index, Energy benchmarking, Energy management, Energy signature, Energy usage, k-means, KPI’s, load profile, Silhouette index, TOPSIS.

(3)

PREFACE

This is a Master of Science project degree in Energy Engineering at Mälardalen University in Västerås. We would like to extend our sincere thanks to colleagues at Alfa Laval in Eskilstuna, especially to Johan Åhlund, for this thesis proposal and all support in this work. Also, to our supervisor Valentina Zaccaria, Postdoctoral Fellow at School of Business Society and Engineering, Division of Automation in Energy and Environmental Engineering at Mälardalen University, for guiding us throughout this project. Finally, we would also like to thank Maher Azaza, Associate Senior Lecturer in School of Business Society and Engineering, Division of Civil Engineering and Energy Systems at MDH, for proposing the main methodology applied in this project degree.

Thank you so much for all the support and encouragement to my parents Maria del Pilar Franco Nino and Alvaro Rincon Cardenas, my brothers Alvaro and Juan Felipe Rincon Franco. My best friends Mario Capacho and Pablo Acevedo. And to Patrik Mäkinen. Finally, to the Swedish Institute (SI).

(4)

SUMMARY

Energy converted to electricity or heat, and water are extensively used by different chemical and physical processes in the manufacturing industry. Also, to create a conformable environment in the administrative areas or offices. But its utilization is not always formulated in a rational way. There are many improvement opportunities that are ignored, and it represents an enormous quantity of money, wasted. With very small changes, the resources protection can be drastically improved. Moreover, the cost of the manufactured products can be reduced, and the image of the company can be enhanced.

Due to the growing demand of water and energy by population and industries. And the fast and deep impact in the environment caused by the lack of concern in the topic. It is necessary to re-organize and implement changes within the company with the aim to protect the planet and reduce or eliminate the costs of the waste of energy and water.

This project starts with the definition of a methodology to handle an extend data set of (electricity and heat) energy usage, and water consumption in the industrial sector. Second, an extensive research in different scientific platforms was done to understand theoretical aspects, and previous applications in different countries, and industries. The research is related to a vast amount of measurements matrix organization, and handling, load profiles construction, data set analysis. All this to extract relevant characteristics, and look for possibilities to reduce the energy usage, and water consumption matrices without losing relevant information. Third, possible calculations to extract the most relevant data, and numbers that allow us to describe a location from the usage trends. Fourth, a methodology to compare, rank, and evaluate the consumptions profiles to find the best possibilities. In this case with the aim to reduce the utilization of resources in different locations.

From this project we conclude that it is possible to reduce an extensive data set without losing the relevant information. This is very useful to the analysis of energy and water usage. Because these measurements are usually a vast amount of data. Also, we find the advantages of using clustering to handle this kind of data. In addition, we compare Davies Boulding (DB) index and silhouette algorithms. Those are a significant input when grouping data is desired. It is possible to extract relevant features from energy and water utilization is the data is handled in a correct way. It was concluded that studying this type of measurement is a way to understand the processes in general and it carries a valuable amount of information to make decisions.

(5)

SAMMANFATTNING

Energi konverterad till el, värme och vatten, används i stor utsträckning av olika kemiska och fysikaliska processer inom tillverkningsindustrin men även för att skapa en bekväm arbetsmiljö. Dess användning är dock inte alltid formulerad på ett rationellt sätt. Det finns många förbättringsmöjligheter som ignoreras, vilket leder till ett enormt slöseri. Små förändringar kan leda till stora resursbesparingar. Dessutom kan kostnaden för de tillverkade produkterna minskas och företagets image förbättras. På grund av den ökande efterfrågan på vatten och energi samt dess påverkan på miljön, är det nödvändigt att omorganisera och genomföra förändringar inom företaget i syfte att skydda planeten och minska energi- och vattenanvändning.

Projektet är uppdelat i fyra delar och inleds med att definiera en metod för att hantera ett utvidgat dataset av (el och värme) energianvändning och vattenförbrukning inom industrisektorn. Detta följs av en omfattande studie på olika vetenskapliga plattformar för att förstå teoretiska aspekter och tidigare tillämpningar i olika länder och sektorer, vilket är relaterat till hanteringen av stora datamätningar, konstruktion av lastprofiler och dataanalyser. I den tredje delen tillämpas metoder för att få fram den mest relevanta informationen utan att förlora viktiga egenskaper (konsumtionsmönster). Till sist en metod för att jämföra, rangordna och utvärdera konsumtionsprofilerna tillämpades för att hitta de bästa möjligheterna. I detta fall med målet att minska resursutnyttjandet på olika platser. Från detta arbete drar vi slutsatsen att det är möjligt att minska en omfattande datauppsättning utan att förlora relevant information. Detta är användbart för analysen av energi- och vattenanvändning eftersom dessa mätningar vanligtvis uppstår av en enorm datamängd. Vi finner också fördelar med att använda grupper för att hantera den här typen av data. Dessutom jämför vi olika algoritmer som är en betydande input när gruppering av data önskas. Det är möjligt att extrahera relevanta funktioner från energianvändning om data hanteras på ett korrekt sätt. Att studera denna typ av mätning är ett sätt att förstå processerna i allmänhet och det innehåller en värdefull mängd information för att fatta beslut.

(6)

CONTENT

1 INTRODUCTION ... 1 1.1 Background ...1 1.1.1 Alfa Laval ...2 1.2 Purpose ...2 1.3 Research questions ...2 1.4 Delimitation ...3 2 METHOD ... 4 2.1 Data gathering ...4 2.2 Data structure ...4 2.3 Computation ...4 3 THEORETICAL FRAMEWORK ... 6

3.1 Energy and water usage ...6

3.2 Energy and water usage profiles ...7

3.3 Clustering...8

3.4 K-means algorithm ...8

3.5 Cluster evaluation indexes ...9

3.6 Key performance indicators ...9

3.7 Benchmarking in the industry ...10

3.8 Multi-criteria decision-making process (MCDM) ...10

3.9 Weight definition methodologies ...11

4 CURRENT STUDY ... 12

4.1 Monitored sites ...12

4.1.1 Alfa Laval Eskilstuna, SEES...12

4.1.1.1. DATA STRUCTURE AT SEES ...12

4.1.2 Alfa Laval Kolding, DKKO ...14

4.2 Clustering load profiles for feature extraction ...14

4.2.1 Evaluating the number of clusters in a data set ...14

4.2.1.1. SILHOUETTE EVALUATION ...14

4.2.1.2. DAVIES BOULDIN INDEX ...15

4.2.2 K-means clustering algorithm to determine the cluster centroid ...15

(7)

4.3 Indicators ...17

4.4 Energy price forecasting ...18

4.5 Multi Criteria Decision Making ...18

4.5.1 Indicators criteria determination ...18

4.5.2 Indicators weight determination ...19

4.5.2.1. ENTROPY WEIGHTS ...19

4.5.2.2. CRITIC WEIGHTS ...20

4.5.3 Technique for Order of Preference by Similarity to Ideal Solution...21

4.6 Creation of different scenarios for comparison ...22

4.6.1 Electricity ...22

4.6.2 Heat ...22

4.6.3 Water ...22

5 RESULTS ... 23

5.1 Electricity ...23

5.1.1 Industrial electricity price ...23

5.1.2 Load profile ...24

5.1.3 Number of clusters evaluation ...26

5.1.4 Energy signature ...28

5.1.5 Electricity cluster analysis at SEES ...31

5.1.6 Possible clustering variations ...31

5.1.6.1. PARTITIONING ANALYSIS WITH 3 CLUSTERS ...32

5.1.6.2. PARTITIONING ANALYSIS WITH 4 CLUSTERS ...33

5.1.6.3. MANUFACTURING INTENSITY EXCLUDING NULL LOAD PROFILES .34 5.1.7 Multicriteria decision-making ...37

5.1.7.1. ELECTRICITY BENCHMARKING WITH KPIS ...37

5.1.7.2. ELECTRICITY BENCHMARKING WITH ANNUAL CONSUMPTION AND PRICE ...38

5.2 Heat ...38

5.2.1 Heat usage profile ...39

5.2.2 Heat demand cluster evaluation ...41

5.2.3 Energy signature ...43

5.2.4 Heating demand cluster analysis at SEES ...46

5.2.5 Multicriteria decision-making ...47

5.3 Water ...48

6 DISCUSSION ... 51

6.1 Data sets ...51

(8)

6.3 DB and Silhouette index ...52

6.4 Clustering...52

6.5 Performance indicators ...54

6.6 Benchmarking...55

7 CONCLUSIONS ... 57

8 SUGGESTIONS FOR FURTHER WORK ... 59

REFERENCES ... 60

APPENDIX 1: DATA STRUCTURE IN EXCEL ... 64

APPENDIX 2 MAIN CODE ... 65

APPENDIX 3 CLUSTER PARTITION SCRIPT ... 68

APPENDIX 4 WEIGHT DETERMINATION SCRIPT ... 70

(9)

LIST OF FIGURES

Figure 1 Clustering methods used for electrical load analysis. Source: (Zhou et al., 2013). ... 8

Figure 2 Energy and water structure for SEES. Source: The author. ... 13

Figure 3 Monthly seasonal distribution. ... 17

Figure 4 Electricity price forecasting. ... 23

Figure 5 SEES electricity load profile. ... 25

Figure 6 DKKO electricity load profile. ... 26

Figure 7. SEES electricity usage data set distribution using Davies-Bouldin evaluation. And load profiles distribution based on the calculations. ... 27

Figure 8. SEES electricity usage data set distribution using Silhouette evaluation. And load profiles distribution based on the calculations. ... 27

Figure 9. DKKO electricity usage data set distribution using Davies-Bouldin evaluation. And load profiles distribution based on the calculations. ... 28

Figure 10. DKKO electricity usage data set distribution using Silhouette evaluation. And load profiles distribution based on the calculations. ... 28

Figure 11. SEES clustering distribution following Silhouette index methodology. ... 29

Figure 12. SEES clustering distribution following DB index methodology. ... 29

Figure 13. DKKO clustering distribution following Silhouette index methodology. ... 30

Figure 14. DKKO clustering distribution following DB index methodology. ... 30

Figure 15 Daily distribution between 2 clusters based on electricity load profiles at SEES. .... 31

Figure 16 Seasonal distribution between 2 clusters based on electricity load profiles at SEES. ... 31

Figure 17 Partitioning results with 3 clusters for electricity usage in Eskilstuna. ... 32

Figure 18 Daily distribution with 3 clusters for electricity usage in Eskilstuna. ... 33

Figure 19 Seasonal partitioning results with 3 clusters for electricity usage in Eskilstuna. .... 33

Figure 20. Partitioning results with 4 clusters for electricity usage in Eskilstuna ... 34

Figure 21 Manufacturing intensity partition excluding non-production days, cluster 1. ... 35

Figure 22 Manufacturing intensity partition excluding non-production days, cluster2. ... 36

Figure 23. Heat demand profile in SEES. ... 39

Figure 24 Ambient temperature at SEES. ... 40

Figure 25. Estimated heat demand at DKKO. ... 41

Figure 26 Ambient temperature at DKKO. ... 41

Figure 27 Optimal number of clusters based on heating demand and Davies-Bouldin evaluation method for SEES site. ... 42

Figure 28 Optimal number of clusters based on heating demand and Silhouette evaluation method SEES site. ... 42

Figure 29 Optimal number of clusters based on heating demand and Davies-Bouldin evaluation method for DKKO site. ... 43

Figure 30 Optimal number of clusters based on heating demand and Silhouette evaluation method for DKKO site. ... 43

Figure 31 Heating demand partition based on Davies-Bouldin evaluation at SEES... 44

Figure 32 Heating demand partition based on Silhouette evaluation at SEES. ... 44

Figure 33 Heat demand partition based on Davies-Bouldin evaluation at DKKO. ... 45

Figure 34 Heat demand partition based on Silhouette evaluation at DKKO... 45

Figure 35 Heat demand daily distribution. ... 46

Figure 36 Heat demand season distribution between clusters. ... 46

Figure 37. Water consumption final rank based on the entropy methodology. ... 50

(10)

LIST OF TABLES

Table 1 Energy intensity indicators where LF refers to Load factor, EAC estimated annual

consumption, SHD space heating demand, MU manufactured units per year. ... 17

Table 2 Water intensity indicators. Where EAC refers to estimated annual consumptions, BWS baseline water stress and DRR drought risk. ...18

Table 3 Decision matrix for electricity and heat. Where LF refers to Load factor, EAC estimated annual consumption, SHD space heating demand, MU manufactured units per year. ... 19

Table 4 Water decision matrix. Where EAC refers to estimated annual consumptions, BWS baseline water stress and DRR drought risk. ... 19

Table 5 Curve fitting goodness of fit ... 24

Table 6 Electricity KPIs and weights calculated with Entropy and critic method, having MU as variable. ... 37

Table 7 Electricity TOPSIS, performance score and rank between sites and MU as variable. 38 Table 8. Weight and TOPSIS methodologies result with electricity cost and electricity yearly usage. ... 38

Table 9. Heat KPIs and weights calculated with Entropy and critic method. ... 47

Table 10. Heat TOPSIS, performance score and rank between sites. ... 47

Table 11. Weight distribution using Entropy and Critic method for the water usage. ... 48

Table 12. Water usage yearly consumption rate and key performance indicators for AL locations around the world. ... 48

Table 13. Rank calculation using TOPSIS method depending on the critic and entropy weight methodology. ... 49

(11)

LIST OF EQUATIONS

Equation 1. Total electricity demand calculation for SEES location. ... 13

Equation 2. Definition of performance coefficient. ... 13

Equation 3. Total heat demand calculation for SEES location. ... 13

Equation 4. Total heat consumption for SEES location. ... 13

Equation 5. Silhouette coefficient. ... 14

Equation 6. Average distance between points to calculate Silhouette index. ... 14

Equation 7. Minimum average distance between points to calculate Silhouette index. ... 14

Equation 8. Davies-Boulding index calculation... 15

Equation 9. Davies-Boulding index calculation. ... 15

Equation 10. Distance calculation in k-means ++ algorithm. ... 16

Equation 11. Distance from each calculation to the centroid in k-means ++ algorithm. ... 16

Equation 12. Load factor calculation for the indicators. ... 17

Equation 13. Electricity usage estimation for one-year period. ...18

Equation 14. Matrix normalization for entropy method application in weight calculation. .... 19

Equation 15. Entropy method calculation. ... 20

Equation 16. Weight vector for the matrix. ... 20

Equation 17. Matrix normalization for critic method application in weight calculation. ... 20

Equation 18. Calculation for critic method. ... 20

Equation 19. Calculation for critic method. ... 20

Equation 20. Objective weights determination using critic method. ... 20

Equation 21. Matrix normalization for TOPSIS application. ... 21

Equation 22. Matrix calculation using normalized weight. ... 21

Equation 23. Possible ideal solutions calculation. ... 21

Equation 24. Euclidean distance for the possible solutions. ... 21

Equation 25. Final classification of the data using TOPSIS. ... 21

Equation 26 Heat demand. ... 22

Equation 27 Total water consumption. ... 22

Equation 28 Water consumption as a percentage. ... 22

Equation 29. Polynomial equation for price forecasting in Sweden. ... 24

(12)

NOMENCLATURE

Symbol Description Unit

A Area m2

BWS Baseline water stress 0-5

COP Coefficient of performance %

DRR Drought risk 0-1

EAC Estimated annual demand m3/year, kWh/year

MU Manufacturing units Units/year

Pt Power demand testing kW, kWh

Php Power demand heat pumps kW, kWh

Q Heat demand kW, kWh

Qhp Heat pump output kW, kWh

LF Load factor %

P Power demand kW, kWh

Pa Power demand average kW, kWh

Pm Power demand manufacturing kW, kWh

Pno Power demand non-operation kW, kWh

Po Power demand operation kW, kWh

Pp Power demand peak kW, kWh

Price Price per kWh €/kWh

SHD Specific heat demand kWh/m2

T Temperature °C

U Thermal transmittance W/m2,°C

W Water usage m3

Wf Water usage for facility m3

(13)

ABBREVIATIONS

Abbreviation Description

AL Alfa Laval

ALEM Alfa Laval Energy Management Team BEU Building energy usage

BWS Baseline water stress COP Coefficient of Performance CSV Comma-separated value

CWCP Characteristic water consumption profile DH District heating

DKKO Alfa Laval Kolding site DRR Drought risk

EAC Estimated annual consumption ES Energy signature

HP Heat Pump

LP Load profile

MCDM Multi-criteria decision making RA Resource Advisor

SEES Alfa Laval Eskilstuna site WS Water signature

(14)

DEFINITIONS

Definition Description

Comma-separated values file

Delimited text file that uses a comma to separate values.

Load profile Graph of energy demand or water consumption as a function of time.

Smart meter Electronic device that measure consumption.

Standard consumption band

(15)

1

INTRODUCTION

The variation in the energy prices, the scarcity of resources, and the rising interest in the protection of the environment are challenging companies to manage energy, water and raw material responsibly. In order to address the problem, the industrial sector must understand its consumption patterns and identify opportunities to improve its processes.

In the following sections, the problem background will be stated and narrowed to set the research questions about how the data from energy and water usage in the industrial sector can be handled in order to manage and reduce these resources consumption.

1.1 Background

The information technology has experienced a sharp increase in the last years, allowing the collection of more significant amounts of data in a more accurate way. This development has had an essential impact on the academic and industrial field due to the collection of real data from production processes. This information has been used as input in scientific research resulting in improvements with high impact in environmental and economic aspects.

Energy and water are the foundation of industrial production. Without improvements in these areas, companies’ growth will be slow, considering the current situation in the world where companies perform a vital role in society. Unfortunately, these processes have an environmental impact. It is therefore important to monitor and improve the consumption features allowing industries to achieve resources conservation and emissions reduction. However, this field of study is under development due to high demand from the governments and interested parties.

Water is essential for life on Earth. In the energy sector, water is important for power production. However, there are many indications around the world that human water use exceeds sustainable levels (Postel, 2000). On the other hand, water needs the energy to be treated, transported and distributed. There is, therefore, a linkage between energy and water. In 2017, the industry sector accounted for 37% of total global final energy use, an increment in energy consumption compared to previous years (IEA International Energy Agency, 2019). In 2014, the industry sector accounted for 10% of global water withdrawals, while in advanced industrial nations, this can reach up to 12% (IEA International Energy Agency, 2016). Increasing energy and water efficiency in the industry sector is, therefore, crucial to mitigate temperature changes and alleviate water scarcity.

(16)

1.1.1 Alfa Laval

Alfa Laval is a Swedish company founded in 1883. With 40 manufacturing sites and six distribution centres around the world, Alfa Laval is a world leader in technology areas of heat transfer, separation, and fluid handling. Their core businesses are Food and Water, Energy and Marine industry. The objective is to be more productive and competitive. For that reason, the board of directors and shareholders are focusing on finding solutions to manage energy and water consumption in their processes. (Alfalaval,2020)

In 2015, Alfa Laval set a company goal related to reducing their usage of energy and water in all their factories. It intends to develop production lines, and processes focused on the protection of the environment. Also, to have an economic impact (reduce costs) in their upcoming years. To achieve it, Alfa Laval created a project called Alfa Laval Energy Management (ALEM). Their objective is to define the methodologies and strategies, to find the tools and support needed, furthermore, to execute the project and reach the targets.

The first step was to hire a third party (Schneider Electric) and initiate a project to install an energy management platform. It consists of different meters to measure energy, heat, and water usage in selected processes, offices and production areas of the company. All this information is available in a platform to be consulted.

The initial plan was to install smart meters in twenty out of forty factories around the world. Those places were already defined within the scope of ALEM based on the highest electricity and water consumptions in each site. Another selection factor was the identification of areas with water shortages or water access problems define by ALEM as water scarcity areas. Currently, the meters are already installed in three out of 20 factories, Eskilstuna (Sweden), Kolding (Denmark), Lund (Sweden), and in the meanwhile, the project is being executed in Brazil, China, and India.

From 2019, Alfa Laval has been collecting an amount of data related to energy and water usage in their factories and divisions (offices building, production areas) within each location. However, the expertise within the company lacks on how to handle, process, analyse, and compare the data that they already have to start a decision-making process and define which is the place where investment will have the highest impact in the energy usage profiles of the company.

1.2 Purpose

This project aims to develop a methodology to handle, process, compare and analyse manufacturing sites measured data to carry a decision-making process subsequently, to help Alfa Laval in their energy management project. More specifically, the methodology is based on the energy usage and water consumption data to calculate indicators, to benchmark between factories and to define which are the places where and investment will be cost-beneficial to reduce energy usage and water consumption.

1.3 Research questions

• How can consumption features be extracted from a data set that is built from periodical measurements of energy and water usage in a specific location?

• How data from industrial processes can be used for decision-making project related to energy management?

(17)

• What is the methodology that Alfa Laval should follow to define where they should invest money to reduce energy usage and water consumption with the highest environmental and financial impact?

1.4 Delimitation

One of the main challenges for this project, it is that all the factories produce different goods. For instance, they have different processes, and it means different characteristics and energy usage and water consumption profiles. Then, the benchmarking is not under similar conditions. For that reason, the analysis should be more careful, and the consideration of the noise factor in the data should be noted.

Another point to think about is data resolution, which is usually a vast amount of measurements over time. For this project, the meters will register the consumption every fifteen minutes. Which mean a vast data set to be handled. The methodology will focus on reducing the size of the data to make more accessible the manipulation without losing information to make reliable the implementation.

On the other hand, this is a project that has faced challenges for the implementation. Currently, the measurements are done in two sites, for that reason, implementing the methodology is restricted to the available data. This project will also depend on the electricity, heat and water usage data available in AL. Also, the methodology will be created with information and prices of 2019/2020 without considering the expansion plans that AL will have.

(18)

2

METHOD

This project presents a methodology that helps the interested parts to extract information from measured data and take cost-beneficial decisions. The literature study was conducted using search engines such as google scholar and the University library, Primo. From such a methodology, the ALEM team can select the location where investments to reduce energy and water usage are profitable. AL was used as the overall case for the study.

General information about Alfa Laval was found on the internet and Alfa Laval’s internal portal. Technical specifications and details related to ALEM were informed in meetings with Johan Åhlund, ALEM’s project manager.

The first step is to gather measurements in the energy management platform used at AL. Once data has been gathered, it is organised using Excel by date in the columns and time in the rows. Excel files containing the data are imported into MATLAB® where all calculations are

performed. Maher Azaza (Associate Senior Lecturer in School of Business Society and Engineering, Division of Civil Engineering and Energy Systems at MDH) proposed the benchmarking concept applying key performance indicators based on feature extraction techniques used in this work.

2.1 Data gathering

All measured data was gathered from Schneider Electric internet platform, Eco Structure™ Resource Advisor (RA). The advantage of RA is that multiple data types of every facility can be aggregated on this platform (Schneider Electric, 2020). Measured data for a selected period is exported as comma-separated values (CSV).

2.2 Data structure

Excel is mostly used to structure and analyse data. The data file from RA is opened in Excel where every measurement is converted into a number. The data set is then transposed to form an 𝑚 × 𝑛 matrix where 𝑚 are dates and 𝑛 the number of measurements per day. See

APPENDIX 1: DATA STRUCTURE IN

EXCEL

. This file is called the raw matrix or raw data set. Using the raw data set, a general consumption profile is built for each site. On the other hand, a building can have different energy and water usage patterns depending on the days, the activities, seasons, and many other factors that can be studied. These patterns are analysed in a Pivot table.

2.3 Computation

Organised data is used as input data and imported into MATLAB® where a programming code,

(19)

Toolboxes are used in this work: Statistics and Machine Learning Toolbox, Symbolic Math Toolbox and Curve Fitting Toolbox. Several functions in toolboxes are used e.g. k-means, Davies-Bouldin and Silhouette evaluation, standard deviation, and correlation coefficient. The primary energy and water usage behaviours are extracted using a clustering algorithm to find similar characteristics within the data. A significant input in this stage of the research is the number of clusters that will be necessary to classify the data in an accurate number of groups. For this, DB-index and silhouette criterion are used. The energy signature is calculated as the centre of each cluster. Relevant indicators are then extracted from each energy signature to understand the consumption behaviour at the operating site fully. See APPENDIX 3: CLUSTER PARTITIONING.

To suggest the best-suited strategic plan, and identify operational and investment improvement opportunities, external information is required to analyse with each consumption profile. Different scenarios are created for comparison and validation. Once all indicators have been determined, an investment decision making process is executed by comparing sites to each other, applying a Multi-Criteria Decision Making (MCDM). See APPENDIX 4: WEIGHT DETERMINATION SCRIPT and APPENDIX 5: TOPSIS SCRIPT.

(20)

3

THEORETICAL FRAMEWORK

In this section an extensive review of scientific articles related to the main topics in this project is done. First, a theoretical explanation of the topic. And then, some examples from previous similar applications with similar objectives are recapitulated.

3.1 Energy and water usage

Governments have understood the importance of environmental protection. Therefore, they have decided to focus their attention on the reduction of water and energy usage. We certainly depend on the use of both. Nonetheless, an invitation to use them effectively has been done. Scientists has been working in understanding the nexus or co-dependency between energy food and water (Leung Pah Hang, Martinez-Hernandez, Leach, & Yang, 2016)(Martin & Fischer, 2012). In this report, we will focus on water and energy as electricity and heat utilization and the way to find improvement opportunities.

On one hand, water is essential for energy transformation processes. On the other hand, electricity is required to distribute, clean and process water. The dependencies in both directions are set to intensify rapidly. According to (U.N., 2014), almost 90% of global power generation is water intensive. The availability of water affects the viability of energy projects and must be considered when deciding on energy measures. Moreover, water dependency services on energy availability will impact the ability to provide clean drinking water and sanitation services.

Water is used in industry for processing, but also for fabricating, testing, and washing. The industry is the second largest water-consuming sector (after agriculture). According to Internationa Energy Agency (IEA), the industry sector accounted for almost 10% of water withdrawals in 2014. There is a significant potential for the wastewater industry, and municipalities to utilise existing technologies to improve process efficiency and harness the embedded energy in wastewater. It could even produce excess energy for other applications. (International Energy Agency, 2016)

Improved energy efficiency can reduce the amount of energy required by 13% in the industry (International Energy Agency, 2016). There are several ways in which business and policymakers can support energy efficiency within the industry. First, tools such as auditing and benchmarking should be used to identify problem areas and track progress in energy and water efficiency (International Energy Agency, 2016). With the current technology, it is straightforward to measure the energy and water usage in any industry. With a merger investment, an enormous amount of data can be extracted. If the data is handled and analysed in a proper manner, improving action can be taken to reduce the usage in the industry in order to protect the environment, achieve the government’s policies and improve the economic situation within the company.

Typically, data is organised to represent the customer’s consumption through daily load pattern. The duration of the data monitoring period must be long enough to guarantee the availability of enough data. For instance, the duration of the monitoring period should be at least two-three weeks in the same loading condition (Chicco, 2012). The building energy and water usage data analysis is a fascination field for researchers either for usage prediction or

(21)

classification. Every day the applications for this type of studies such as load forecasting and prediction, energy pattern profiling, regional energy-consumption mapping and benchmarking for building increases (Wei et al., 2018).

The three main thematic groups for researchers are: First, the general aspects of sustainability theory, integrated management of resources, and decision-making. The second group focuses upon improved efficiency of water use and wastewater minimisation. The third group includes developing cleaner production illustrations from within industries such as sugar, pulp and paper, oil refining and petrochemical (Klemeš & Huisingh, 2005).

3.2 Energy and water usage profiles

Knowledge on the shape of the energy and water consumption can be decidedly useful to deal with effective management for industry planning and operation. A Load Profile (LP), describes a systems energy pattern. Load profiles can be created by applying statistical methods to measurements. It is possible to determine a set of shape factors modelling specific aspects of the consumption pattern (Chicco, 2012). It is what in the work is defined as the Energy Signature (ES) of a building. In this study, electricity, thermal energy, and water profiles are investigated. In the process of data analytics, it may also be worth to determine shape features of the LP to reduce the amount of data.

The electrical energy, heat usage, and water consumption profiles describe the general characteristics in the utilization of water and energy. LP is used to find the underlying patterns of each customer or group of customers. It is usually implemented for load forecasting and demand response programs (Y. Wang, Chen, Hong, & Kang, 2019). In this study, the amount of energy (kW) and water (m3) are handled as a function of time (hour). Knowing that a load

curve is a time series representative of the Building Energy Usage (BEU) each building will have a typical shape of a daily load that will allow the energy manager to understand the energy use at the location (Jota, Silva, & Jota, 2011).

Identification of electricity usage is one of the key elements to motivate users to promote activities leading to more efficient use of energy. Governments and companies commonly use this methodology. In Finland, nearly 8000 customers in the city of Salo were studied to describe and characterised their annual electricity usage in order to create more personalised information about customers energy use (Räsänen, Ruuskanen, & Kolehmainen, 2008). In Portugal, three months of electricity usage measurement campaigns were carried out to describe each customer behaviour using their load profile curve and to find out how and when the consumer uses electricity (Figueiredo, Rodrigues, Vale, & Gouveia, 2005). In Taiwan, the load characteristic curve for different the customer was determined to create a load management strategy (C.S Chen, J.C.Hwang, Y.M. Tzeng, 1996). This information will allow the interested parts to have a demand/production scheme. Consequently, it will lead to a more effective and efficient industry with less impact in the environment and higher reward.

Monthly consumption water data over a three-year period is used as an input to create the consumption profile and calculate the exact value per volume of any water service. It refers to some other costs related to water usage that usually are not considered like maintenance, equipment depreciation and energy consumption for water usage in the industry (Walsh, Bruton, & O’Sullivan, 2017). Beside other studies show that from the analysis of a water energy profile, several calculations can be implemented that we can redesign, retrofit existing systems and technology development in different industries (Byers, Lindgren, Noling, & Peters, 2010). A building Energy Signature (ES) is defined as a set of parameters that describes its energy performance (Hammarsten, 1987). The parameters can be defined depending on the aim of the study. ES has been used to compare the energy performance of the heating system studying the building, and its residence impact (Sjögren, Andersson, & Olofsson, 2009). Also, the

(22)

correlation between energy conversion from solar source during different seasons is an application of the ES in the industry (Vesterberg, Andersson, & Olofsson, 2016).

3.3 Clustering

Clustering is the unsupervised classification of patterns (observations, data items, or feature vectors) into groups (clusters) (Jain, Murty, & Flynn, 1999). In addition to this, cluster analysis is the name given to a group of multivariate techniques whose primary purpose is to identify similar entities from the characteristics they possess. The essence of clustering approaches is the classification according to natural relationships. Moreover, clustering analysis has been used to classify time series data in groups (Jota et al., 2011).

The goal of clustering is to identify structure in an unlabelled data set by objectively organising data into homogeneous groups where the within-group-object similarity is minimized, and the between-group-object dissimilarity is maximized (Warren Liao, 2005).

Clustering methods can be grouped into five categories based on the clustering criterion, and each category contains many specific clustering methods. Such partitioning methods include

k-means, Fuzzy C-means (FCM), Partition around medoids (PAM). All the clustering methods

can be used for load classification, which is summarized in Figure 1 (Zhou, Yang, & Shen, 2013).

Figure 1 Clustering methods used for electrical load analysis. Source: (Zhou et al., 2013).

In a work from China, the general energy efficiency estimation was calculated and analysed using industrial production clustering methodologies (Yu, You, Zhang, & Ma, 2018). In another work, the performance of different clustering methodologies has been evaluated for 127 load curves of non-residential customers finding the more suitable algorithms depending on the central purpose of the research (Bidoki, Mahmoudi-Kohan, Sadreddini, Jahromi, & Moghaddam, 2010). Furthermore, based on the information collected from the water meter, a clustering algorithm was implemented to represent customer with similar consumption behaviour in North America (Malinowski, 2018).

3.4 K-means algorithm

K-means is the simplest and most used algorithm to cluster data sets employing a squared

error criterion. It starts with a random initial partition and keeps reassigning the patterns to clusters based on the similarity between the pattern and the cluster centres until a converge criterion is met (Jain et al., 1999)

This clustering algorithm is attractive in practice because it is simple, and it is generally very fast to implement. It partitions the input dataset into k clusters. Each cluster is represented by an adaptively changing centroid (also called cluster centre), starting from some initial values

(23)

named seed-points. K-means computes the squared distances between the inputs (also called input data points) and centroids and assigns inputs to the nearest centroid (Žalik & Žalik, 2008).

K-means algorithm was implemented to Algerian electricity load to create an electricity

forecasting strategy (Benabbas, Khadir, Fay, & Boughrira, 2008). In industrial aluminium production, k-means has been a feasible and effective tool in to order create a strategy for energy saving (Lou & Zou, 2010). It has been demonstrated that water analysis can be utilised as input to set policies and practices for water protection. Water use studies can provide essential water consumption information when, where how and why water is consumed. K-means algorithm has been also implemented to do market segmentation and identifies water distribution systems in Australia (Yang, Zhang, Stewart, & Nguyen, 2018) .

3.5 Cluster evaluation indexes

Validity indices have been developed for evaluating the quality of partitions to find optimal partitioning that consists of compact and well-separated clusters (Halkidi & Vazirgiannis, 2008). In other words, a key question when using clustering is how many groups or clusters should be created. This topic has been deeply studied, and different methodologies are available: DB index, C index, and partition coefficient (Žalik & Žalik, 2011).

Davies Boulding index is a popular measure to evaluate clustering performance through the separation between the 𝑖𝑡ℎ and the 𝑗𝑡ℎ cluster. It can indicate the clustering quality by the ratio

of intracluster similarity and inter-cluster similarity (Xiao, Lu, & Li, 2017). The Silhouette index bases on the comparison of a measure of closeness of each observation to the cluster where it has been allocated and a measure of separation from the closest alternative cluster. (Menardi, 2011)

3.6 Key performance indicators

A Key Performance Indicator (KPI) evaluates the performance of a process. KPIs development and implementation are described by Parmenter (2010). Indicators are the input when benchmarking is implemented as the factor parameter that will be compared and assessed (Sharp 1996). KPIs can help to identify, define, and communicate sustainability issues, and they can be used to forecast and monitor the results of choices. Companies can set their operation and process management strategies in the performance indicators.

KPIs are the input values to analyse, describe and compare electricity in companies to asses sustainability, as implemented in a work from Brazil (Sartori, Witjes, & Campos, 2017). In the literature, it is suggested to consider the water reuse, recycling, and regeneration. Because through the partial treatment of wastewater, a water regeneration unit allows for spent water to be reused and recycled within a process plant, thereby reducing freshwater requirements and wastewater generation (Hsieh, Sheen, Chen 1995). In the sugar cane industry, which requires a high demand for water, the environmental impact, the evaluation of freshwater use efficiency, and the wastewater performance in the industry were studied through the use of different performance indicators (Ingaramo, Heluane, Colombo, & Cesca, 2009).

(24)

3.7 Benchmarking in the industry

Benchmarking compares processes and performance to determine best practices. The benchmarking process typically refers to the collection of a large amount of quantitative data to ensure the quality of products (Venkatraman, Abraham, & Paprzycki, 2004). The term ‘building energy benchmarking’ was first used in the 1990s to refer to the comparison of energy use in buildings with similar characteristics (Pérez-Lombard, Ortiz, González, & Maestre, 2009). According to various authors, energy sector benchmarking includes a comparison of energy performance with other buildings, whereas baselining involves a comparison of the past energy performance of a single building with current energy performance (Nikolaou, Kolokotsa, & Stavrakakis, 2011).

A benchmarking program in a company is divided into two different activities. First, a Strategic Energy Review (SER), which are all the activities required to develop the comparison, measurements, calculations, methodologies. The purpose is to fully understand the energy situation at the operating site. The second one is an Energy Management System (EMS), which includes all the best practice documentation, organizational requirements and monitoring tools (Eggleston, 2015). Both are important in order to achieve the goals set in the company and to keep track of the project.

Furthermore, electricity use comparisons with other similar customers gives a more interesting and concrete point of view to examine their own consumption habits (Räsänen et al., 2008). Nevertheless, these constraints can rarely be satisfied in most energy benchmarking cases where the available samples may not meet these requirements on similarity. For example, the benchmarked buildings are often with diverse characteristics, e.g. in different climate conditions, or accommodating a distinct number of occupants (E. Wang et al., 2017). For this case study, the benchmarking will be internal. It will focus on the company sites and its internal processes to support the selection of where the investments need to be done to become more efficient. It is highly important since sites differ in size, location, and energy and water intensity.

Benchmarking has been commonly used in the brewery industry to identify improvement opportunities of water and energy performances (Kirstein & Brent, 2017). Benchmarking was carried out by using Performance Indicators (EPIs). These are specific water/energy consumption and waste/emission generation data to conduct a model to increase water and energy efficiency in the industry (Alkaya & Demirer, 2015).

3.8 Multi-criteria decision-making process (MCDM)

MCDM is a procedure that consists in finding the best alternative among a set of feasible alternatives (García-Cascales & Lamata, 2012). Multicriteria decision making process helps interested parts understand the problems to be solved and conduct analyses, comparisons, and rankings of alternatives. The ranking will allow to find a suitable alternative(s) that will be performed (Zaidan & Zaidan, 2018).

In the literature it is possible to find different techniques based on the ease of use, data requirement, or software availability. In the MAXMAX technique, an alternative is selected by its best attribute value. The SAW (Simple Additive Weighting) method multiplies the normalized value of the criteria for the alternatives with the importance of the criteria and the alternative with the highest score is selected as the preferred one. The TOPSIS (Technique for Order Preference by Similarity to the Ideal Solution) selects the alternative closest to the ideal solution and farthest from the ideal negative alternative. The AHP (Analytical Hierarchy Process) uses a hierarchical structure and pairwise comparisons. Hierarchy has at least three levels: the main objective An AHP of the problem at the top, multiple criteria that define alternatives in the middle and competing alternatives at the bottom. The ELECTRE

(25)

(Elimination and Choice Expressing Reality) method chooses the best action(s) from a given set of actions, but it can also be applied to three main problems: choosing, ranking and sorting. (Ewa, 2011).

These methodologies are used in a wide range of applications, for example, multi-criteria analysis is implemented to evaluate water allocations in multiyear scenarios (Srdevic, Medeiros, Faria, & Schaer, 2003). The evaluation of environmental impacts from a pilot organization defined in standard ISO14001 was studied using AHP and TOPSIS method to manipulate the data. The reliability of the methodologies is compared (Jovanovic, Shah, Vujovic, & Krivokapic, 2014). More specifically in the energy sector, it has been used to recommend future energy sources to be added into the grid in UK, taking into consideration water consumption and the possibility of desalination of seawater (Hunt, Bañares-Alcántara, & Hanbury, 2013).

3.9 Weight definition methodologies

When a multicriteria comparison is made, the output will be a ranking of possibilities. It will depend on the attribute of importance assigned to each indicator. The challenge is assigning the right value of priority to each ratio. Several methods are available in the literature to calculate the weight of priority for the different criteria.

The critic method proposes to embody the information which is transmitted from all the criteria participating in the multicriteria problem. In addition objective weights offer an insight into the nature of the dilemmas created by the existence of conflicting criteria and enable the incorporation of interdependent criteria. (Diakoulaki, Mavrotas, & Papayannakis, 1995) Entropy weight method is based on amount of information to determine the index’s weight, which is one of objective fixed weight methods (Li et al., 2011). It is made up of the monitoring values of evaluation index in objective conditions, and it can determine the target and the degree of order and effectiveness by referring to evaluation of information entropy. It avoids the subjectivity of the weights of various criteria, and therefore the results of evaluation can be better able to reflect the actual situation. (Bing & Denghao, 2001)

Comparisons between two different methods to calculate the weights, namely Critic and Entropy methods, conclude that the decision is independent of the MCDM method used. And exist correlation depending on the implemented weight method (Vujicic, Papic, & Blagojevic, 2017). Setting this as a critical input in the benchmarking because of its influence on the result then special attention should be paid to calculate in an objective way this factor. A guide in how to make better decisions in water resources management and the effects of weighting methods on multi-criteria decision making is studied in the available literature (Zardari, Ahmed, Shirazi, & Yusop, 2015).

(26)

4

CURRENT STUDY

This study comprehends energy demand and water consumption for different sites within Alfa Laval. A detailed methodology description is presented in this section. The key addressed aspects in this chapter are data gathering, data set description and the initial processing of the information using load profiles patterns. Second, the clustering phase is described, where using two methodologies the optimal number of groups is calculated. Following, k-means is used to group the load profiles and extract similar characteristics or behaviours that will describe and consequently reduce the whole data set. Performance indicators will be the input to compare the different locations. Subsequently, the locations will be categorized using two methods to assess the impact of the indicators.

4.1 Monitored sites

Smart meters capable of logging data are installed in several sites. The numbers of metering units installed in each factory differ due to technical reasons such as the complexity of the power lines and the layout of each factory. Alfa Laval sites located in Eskilstuna and Kolding are currently being monitored with a 15-minute interval. Water, electricity, and heat data have been measured in Eskilstuna. In Kolding, only measurements for electricity demand have been conducted. The history of energy can be represented by a time series which can have different seasonal cycles depending on the purpose of the study, daily, weakly, seasonal or yearly. In this report the load profile will be set daily. Measured data is exported to an Excel file where its structured in a matrix form. The data is organized with rows as 15 minutes interval and columns as days.

4.1.1 Alfa Laval Eskilstuna, SEES

Alfa Laval Eskilstuna is a manufacturing unit with focus on high speed separation technology for energy, food, water, biotech, and pharma applications. The site manufactures and sells separators, key components, and spare parts. Their main competence is cutting, machining, welding and grinding in stainless steel (Alfa Laval, 2020). The site consists of a production plant and an office building with a total of 19,812 m2 and 245 co-workers. All electricity demand

in the facility is purchased and heat demand is covered by district heating and heat pumps (HP).

4.1.1.1.

Data structure at SEES

There are fifteen metering units installed in Eskilstuna, twelve for electricity, three for water and one for district heating. The available dataset period is between 2019-06-20 and 2020-04-27. Energy demand and water consumption in Eskilstuna are categorised in Figure 2.

(27)

Figure 2 Energy and water structure for SEES. Source: The author.

Electricity demand is divided into four categories. Manufacturing intensity, 𝑃𝑚, comprehends

the demand in the manufacturing process such as machining, CNC, pressing, welding but also energy transformed for compressed air and cooling processes. Electricity demand of loads, lightnings, space cooling and heating for the manufacturing building are categorized as operation services, 𝑃𝑜. Corresponding demand of offices and R&D buildings are listed as

non-operation services, 𝑃𝑛𝑜. Equipment testing is assigned as 𝑃𝑡. Total electricity demand is

therefore calculated with Equation 1.

𝑃 = 𝑃𝑚+ 𝑃𝑜+ 𝑃𝑛𝑜+ 𝑃𝑡

Equation 1. Total electricity demand calculation for SEES location.

Heat demand is supplied by district heating and heat pumps. HPs power demand is converted to heat demand with the coefficient of performance (COP), which is defined as the ratio between heat output and electricity input into the system

𝐶𝑂𝑃ℎ𝑝=

𝑄ℎ𝑝

𝑃ℎ𝑝

Equation 2. Definition of performance coefficient. Therefore, total heat demand of the site is calculated as:

𝑄 = 𝑄ℎ𝑝+ 𝐶𝑂𝑃ℎ𝑝∗ 𝑃ℎ𝑝

Equation 3. Total heat demand calculation for SEES location.

Water consumption is distinct between water for facility and testing. Total water consumption for the site is calculated as:

𝑊 = 𝑊𝑓+ 𝑊𝑡

(28)

4.1.2 Alfa Laval Kolding, DKKO

Alfa Laval is Kolding’s largest company with 664 coworkers. Fluid handling is their main core. At Alfa Laval Kolding, pumps, valves and tank equipment are manufactured and sold. The site is divided into a production hall (14,600 m2), a distribution center (7,600 m2) and an

administrative building (9,300 m2) (Alfa Laval, 2020). Electricity demand is purchased. The

data available is between 2019-12-03 and 2020-04-27. Electricity demand is currently divided into three unspecified main meters. The sum of these three meters is assumed to be the total power demand at DKKO.

4.2 Clustering load profiles for feature extraction

Feature extraction procedure from load profiles is explained in this section. The first step is to import the measured data into MATLAB®. Then, the optimal number of clusters and the

partition algorithm are applied to determine the characteristic load profiles.

4.2.1 Evaluating the number of clusters in a data set

The optimal number of clusters is evaluated with two different criteria, Davies-Bouldin and silhouette evaluation. These methods are applied on the measured data. Two evaluation methods result in two optimal numbers. The idea is to compare the different outputs. If the optimal number of clusters is the same for the both methods, then it may be a strong indication that the result is accurate. If the optimal number of clusters is not equal, the output from each evaluation method may be used as input in k-means algorithm. The partitioning result can then be compared to determine the most accurate number of clusters.

4.2.1.1.

Silhouette evaluation

Silhouette evaluation is described in the book Finding Groups in data (Kaufman & Rousseeuw, 1990). The silhouette coefficient 𝑆𝑖, is defined as the measure of a point to its own cluster

compared to other clusters.

𝑠(𝑖) = 𝑏(𝑖) − 𝑎(𝑖)

𝑚𝑎𝑥{𝑎(𝑖), 𝑏(𝑖)}, 𝑖𝑓 |𝐶𝑖| > 1 Equation 5. Silhouette coefficient.

The average distance between a point 𝑖 to the rest of the points in its own cluster is

𝑎(𝑖) = 1 |𝐶𝑖| − 1

∑ 𝑑(𝑖, 𝑗)

𝑗∈𝐶𝑖,𝑖≠𝑗

Equation 6. Average distance between points to calculate Silhouette index.

The minimum average distance between a point 𝑖 to all the other points in a different cluster is

𝑏(𝑖) = min 𝑘≠𝑖 1 |𝐶𝑘| ∑ 𝑑(𝑖, 𝑗) 𝑗∈𝐶𝑘

Equation 7. Minimum average distance between points to calculate Silhouette index. The silhouette coefficient ranges between −1 ≤ s(i) ≤ 1.

(29)

s(i) =

{

1 −a(i)

b(i), if a(i) < b(i) 0, if a(i) = b(i)

b(i)

a(i)− 1, if a(i) > b(i)

A high value indicates that the point i is well matched to its own cluster and is therefore suitable. If a point has a low or negative value, the solution might not be suitable. Any distance metric can be used as a clustering evaluation. The highest silhouette coefficient indicates the optimal number of clusters.

4.2.1.2.

Davies Bouldin index

Davies Bouldin index is the second method used to evaluate the optimal number of clusters based on the data set. The method was introduced by David L Davies and Donald W. Bouldin (1979) and is determined as:

DBI = 1 N∑ Ri

N

i=1

Equation 8. Davies-Boulding index calculation. Where N is the number of clusters, 𝑅𝑖 = 𝑚𝑎𝑥 𝑅𝑖𝑗, 𝑖 ≠ 𝑗.

𝑅𝑖𝑗 =

𝑆𝑖+ 𝑆𝑗

𝐷𝑖,𝑗

Equation 9. Davies-Boulding index calculation. Where 𝐷𝑖,𝑗 = ‖𝑣𝑖− 𝑣𝑗‖ is the distance between the centroids 𝑖 and 𝑗, 𝑆𝑖 =

1

|𝐶𝑖| ∑𝑥𝑗∈𝐶𝑖‖𝑥𝑗− 𝑣𝑖‖ is a measure of scatter within the cluster, ‖∗‖ is the Euclidean distance and 𝐶𝑖 is the number of

vectors in cluster 𝑖.

The lowest Davies Bouldin value indicates the optimal number of clusters.

4.2.2 K-means clustering algorithm to determine the cluster centroid

The optimal number of clusters given by Silhouette evaluation and Davies Bouldin index, is used as input in k-means. The primary purpose in this stage is to identify similar profiles from the characteristics it possesses. This is a data pre-classification of the data. Giving as a result the grouping of information. The k-means++ algorithm is used as default in MATLAB® since

this algorithm is more efficient than the original k-means algorithm (Arthur & Vassilvitskii, 2007). Squared Euclidean distance is used as default in k-means++ algorithm in MATLAB®.

The algorithm is as follows (MATLAB, 2020):

1. A measurement is randomly chosen from the data set and set as the first cluster centroid 𝐶1.

2. The distance for each measurement to 𝐶1 is computed. The distance between 𝐶𝑗 and

the measurement 𝑚 is denoted as 𝑑(𝑥𝑚, 𝑐𝑗).

A new measurement is randomly chosen, next centroid 𝐶2, with weighted probability

(30)

𝑑2(𝑥 𝑚, 𝑐1) ∑ 𝑑2(𝑥 𝑗, 𝑐1) 𝑛 𝑗=1

Equation 10. Distance calculation in k-means ++ algorithm.

3. Distances from each measurement to each centroid is computed and assigned to its closest centroid. For 𝑚 = 1, … , 𝑛 and 𝑝 = 1, … , 𝑗 − 1, select centroid 𝑗 at random from X with the probability.

𝑑2(𝑥 𝑚, 𝑐𝑝)

∑ 𝑑2(𝑥 ℎ, 𝑐𝑝) {ℎ;𝑥∈𝐶𝑝}

Equation 11. Distance from each calculation to the centroid in k-means ++ algorithm. 4. where 𝐶𝑝is the set of all measurement closest to centroid 𝑐𝑝 and 𝑥𝑚 belongs to 𝐶𝑝.

5. Step 4 is repeated until 𝑘 centroids are chosen.

Each centroid is the mean of the measurements in that cluster. The cluster centroids represent the characteristic consumption patterns of a group of load profiles. These characteristic patterns are used for feature extraction. The cluster centroid of water consumption is defined as the characteristic water consumption profile, (WS). While the characteristic LP of electricity and heat is defined as the ES.

4.2.3 Cluster analysis

Cluster partitioning is further investigated to determine the characteristics of each cluster. Cluster analysis is also conducted with a different value than the optimal number of clusters given by the evaluation methods. The k-means function output “idx” is used. This output is a n-by-1 vector containing cluster indices for each daily load profile. The vector is then, exported to Excel where it is analysed in a Pivot table by days and seasons.

According to (SMHI, 2014), it is difficult to draw a line between seasons since it depends on the context. The seasons in this work are assumed to contain three months and are set as shown in Figure 3.

(31)

Figure 3 Monthly seasonal distribution.

4.3 Indicators

Several indicators are selected to determine the intensity performance of a site. Some indicators are extracted from the characteristic load profiles while others are selected based on different criteria. Indicators chosen for electricity and heat demand are shown in Table 1. Load factor and manufactured units per year are indicators exclusively for electricity while space heating demand is a heating demand indicator.

As an overview, mean and peak indicators are statistical average and maximum calculations, respectively. The load factor refers to the ratio of energy consumed in a given period, and it is calculated with Equation 12

𝐿𝐹 = 𝑃𝑎𝑣𝑒 𝑃𝑝𝑒𝑎𝑘

Equation 12. Load factor calculation for the indicators.

In other words, how long the peak demand was sustained, this indicator will be relevant in locations where the energy cost is based on peaks. The estimated annual consumption refers to energy usage along one-year period. The space heat demand describes how much heat is consumed per squares meter in a year. The manufactured units per year, shows the production per electricity usage.

Table 1 Energy intensity indicators where LF refers to Load factor, EAC estimated annual consumption, SHD space heating demand, MU manufactured units per year.

Indicators Mean Peak LF EAC SHD Price MU

Electricity kWh kWh % kWh/y - €/kWh Unit/y

Heat kWh kWh - kWh/y kWh/m²y €/kWh -

Water indicators are presented in Table 2. Besides annual water consumption, baseline water stress (BWS), and drought risk (DRR) are added to the decision matrix. BWS and DRR are

Jan-Mar Apr-Jun Jul-Sep Oct-Dec

(32)

Aqueduct 3.0 water risk indicators described by (Hofste et al., 2019a). BWS is defined as the ratio between total water withdrawals and available renewable surface and groundwater supplies. DRR measures the likelihood of a drought based on a specific location and is scored between 0 and 1.

Table 2 Water intensity indicators. Where EAC refers to estimated annual consumptions, BWS baseline water stress and DRR drought risk.

Indicators EAC BWS DRR

Water m3/y 0-5 0-1

4.4 Energy price forecasting

Electricity price is an important factor for investments. Future or current electricity prices are estimated by applying regression techniques on previous prices. Energy prices in different regions or countries are converted to the same currency, Euro (€). The average prices on electricity, including network charges, taxes, and charge for green certificate, were obtained from statistics from the respective country. In Sweden from Statistics Sweden (2020) and in Denmark from Statistics Denmark (2020).

The electricity price paid by industrial consumers depends on the standard consumption band. Since electricity data from the sites has been measured for less than a year, the yearly electricity demand may be estimated with the following equation:

𝑃𝐸𝐶 =

∑𝑛𝑑=1𝑃𝑑

𝑛 ∗ 365

Equation 13. Electricity usage estimation for one-year period. Where 𝑃𝑑, is the daily electricity consumption and 𝑛 is the number of days.

A curve fitting application in MATLAB® is used to estimate the actual electricity price.

Electricity prices based on standard consumption band from the last 5 years are used as input.

4.5 Multi Criteria Decision Making

A multi criteria decision making is carried out to determine the performance score of different sites. The first step is to build a decision matrix with the selected indicators for each consumption profile with respective criteria. Then, indicators weight determination is performed. At last, a technique for order of preference by similarity to ideal solution is applied for benchmarking the sites.

4.5.1 Indicators criteria determination

An indicator criterion is defined as beneficial if its value is desired to be as high as possible while as non-beneficial if its value is desired to be as low as possible. Once criteria are determined a decision matrix can be constructed. Decision matrices for electricity, heat and water are presented in the tables below. Beneficial indicators are set to 1 since the maximum value is desired while non-beneficial indicators are set to zero.

Figure

Figure 2 Energy and water structure for SEES. Source: The author.
Table 1 Energy intensity indicators where LF refers to Load factor, EAC estimated annual consumption, SHD  space heating demand, MU manufactured units per year
Table 4 Water decision matrix. Where EAC refers to estimated annual consumptions, BWS baseline water stress  and DRR drought risk
Figure 7. SEES electricity usage data set distribution using Davies-Bouldin evaluation
+7

References

Related documents

46 Konkreta exempel skulle kunna vara främjandeinsatser för affärsänglar/affärsängelnätverk, skapa arenor där aktörer från utbuds- och efterfrågesidan kan mötas eller

Uppgifter för detta centrum bör vara att (i) sprida kunskap om hur utvinning av metaller och mineral påverkar hållbarhetsmål, (ii) att engagera sig i internationella initiativ som

Däremot är denna studie endast begränsat till direkta effekter av reformen, det vill säga vi tittar exempelvis inte närmare på andra indirekta effekter för de individer som

Based on a combination of the previous studies and a quantitative study presented in this paper a discussion about the optimal number of names and persons in a case

Our simulations for test particles in a reasonably realistic Galac- tic potential suggest that at least for the last 5 Gyr, the observed heating of the thin Galactic stellar disk can

Industrial Emissions Directive, supplemented by horizontal legislation (e.g., Framework Directives on Waste and Water, Emissions Trading System, etc) and guidance on operating

How will different inundation risk levels caused by a range of different increases in sea level, and a combination of sea level and highest projected high water, affect the

In the previous report, some propositions have been mentioned in order to improve the Mix method. This method seeks to combine the straights of the Direct