Post on 13-May-2020
Industrial Energy Management in the Cloud
Hugo André Gomes Sequeira
Thesis to obtain the Master of Science Degree in
Information Systems and Computer Engineering
Supervisor: Prof. Dr. Paulo Jorge Fernandes Carreira
Examination Committee
Chairperson: Prof. Dr. Ernesto José Marques MorgadoSupervisor: Prof. Dr. Paulo Jorge Fernandes Carreira
Member of the Committee: Prof. Dr. Mário Serafim dos Santos Nunes
November 2014
ii
Acknowledgments
Gostaria de comecar por agradecer ao meu orientador, professor Dr. Paulo Carreira, pelo excelente
trabalho de orientacao, que inspirou tudo e todos durante a elaboracao desta tese e de todas as outras
que supervisionou ao longo deste ano. Obrigado por todo o suporte, todas as criticas, todos os elogios
e por acreditar no nosso potencial.
Gostaria tambem de agradecer ao meu supervisor, Dr. Thomas Goldschmidt, pela sabedoria, su-
pervisao, e pelo acompanhamento que prestou durante a minha estadia na ABB. Estarei para sempre
grato por ter acreditado em mim e aceite nesta grande empresa. Caso contrario teria perdido muita
coisa a nıvel pessoal e academico.
A minha famılia, em especial aos meus pais e avos, por todo o carinho que sempre me deram e por
todo o apoio na perseguicao dos meus sonhos. Ao Joao Loff, Alexandre Almeida, Viteche Ashvin, Hugo
Ramos, Sergio Isidoro, Tiago Aguiar, Edgar Santos, Nuno Teles e a todos os outros amigos e colegas
que foram companheiros de coracao durante a minha vida escolar. Sem eles, nunca teria chegado aqui
e por isso estarei eternamente grato.
A todos os amigos que fiz na Holanda e na Alemanha, durante o meu percurso no estrangeiro. Em
especial, ao meu amigo alemao Philipp Piroth, pelo o apoio e carinho desde o primeiro dia e que se
tornaram indispensaveis para conseguir viver na Alemanha com muita felicidade e realizar o meu tra-
balho com muito sucesso.
A todos aqueles que perdi, mas que sempre estiveram do meu lado e prontos para me apoiar nos
maus ou nos bons momentos. Nunca me esquecerei de vos.
A todos os colegas de trabalho que tive o prazer de conhecer na ABB e que tanto me ajudaram para
conseguir realizar os meus objectivos.
A todos vos e a todos os outros que fizeram parte da minha vida,
Um MUITO OBRIGADO e a vos vos dedico esta tese.
Hugo Sequeira
iii
iv
Resumo
Organizacoes industriais usam sistemas de gestao energetica (EMS) para monitorizar, controlar e op-
timizar o seu consumo energetico. Sistemas industriais como estes sao complexos e dispendiosos,
devido aos seus requisitos avancados de desempenho, confiabilidade e interoperabilidade. A industria
sente tambem algumas dificuldades na operacao dos actuais sistemas EMS quando pretende ter uma
monitorizacao centralizada do consumo energetico e emissao de CO2 nos varios locais de producao, na
integracao de dados energeticos e de automacao, e quanto pretende efectuar uma analise comparativa
da eficiencia energetica entre as diferentes producoes. Para alem disso, a industria sente tambem prob-
lemas de big data devido a evolucao tecnologica dos equipamentos de medicao. Estes produzem cada
vez mais medicoes com mais detalhe e com mais frequencia, resultando na geracao de grandes quanti-
dades de dados, que dificulta a gestao de toda esta informacao em tempo real. Esta tese propoe entao
uma solucao EMS na cloud para resolver estas dificuldades e derivar novas e mais informacoes em
tempo real. De facto, o impacto desta tese e deveras extenso, com possibilidades inovadores para as
organizacoes industriais detectarem padroes de ineficiencia no seu consumo energetico e conseguirem
reagir a mudancas de ambiente com mais rapidez. A relevancia da solucao proposta nesta tese foi con-
firmada atraves de uma avaliacao a forma como resolveu casos de uso que estao em falta nestes
sistemas industriais. A sua viabilidade de implementacao e o seu desempenho foram tambem avalia-
dos, atraves da implementacao de um prototipo e da avaliacao do seu comportamento em diferentes
testes de stress.
Palavras-chave: Eficiencia Energetica, Gestao Industrial Energetica, Sistemas Industriais
de Gestao Energetica, Demand Response (DR), Computacao em Nuvem, Computacao em Tempo
Real
v
vi
Abstract
Industrial organizations use Energy Management Systems (EMS) to monitor, control, and optimize their
energy consumption. Industrial EMS are complex and expensive systems due to the unique require-
ments of performance, reliability, and interoperability. Moreover, industry is facing challenges with cur-
rent EMS implementations such as cross-site monitoring of energy consumption and CO2 emissions,
integration between energy and production data, and meaningful energy efficiency benchmarking. Ad-
ditionally, big data has emerged because of recent advances in field instrumentation that led to the
generation of large quantities of machine data, with much more detail and higher sampling rates. This
created a challenge for real-time analytics. To address these needs and challenges, this thesis proposes
a cloud-native industrial EMS solution with cloud computing capabilities to enable the extraction of ac-
tionable knowledge from large amounts of real-time data. Indeed, the impact of this work is far reaching
as it enables organizations to detect hidden patterns of inefficient energy use and to react to changes
of events in real-time. The feasibility of our proposal was verified with the implementation of a proof of
concept and its usability and performance validated by respectively evaluating its approaches to solve
important use cases that the industry is lacking of and how it handles different amounts of workloads.
Keywords: Energy Efficiency (EE), Industrial Energy Management, Energy Management Sys-
tems (EMS), Demand Response (DR), Cloud Computing, Real-time Computing
vii
viii
Contents
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Resumo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi
List of Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Problem statement and objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Research methodology and contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Document organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Research Background 7
2.1 Energy Demand Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Smart Grids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.2 Liberalised Electricity Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.3 Demand-Side Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.4 Energy Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.5 Energy Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.6 Energy Management Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.7 Industrial Energy Management Systems . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Demand Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.1 Demand Response Programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.2 Demand Response Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.3 Survey on Energy Management Systems . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 Industrial Automation Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.1 Industrial Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.2 Industrial Control Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.3 Industrial Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.4 Cyber-Physical Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
ix
2.3.5 Internet of Things . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3.6 Industry 4.0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4 Cloud Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4.1 Cloud Computing Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4.2 Cloud Computing Service Models . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4.3 Cloud Computing Deployment Methods . . . . . . . . . . . . . . . . . . . . . . . . 32
2.4.4 Benefits of Cloud Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.4.5 Risks and Concerns of Cloud Computing . . . . . . . . . . . . . . . . . . . . . . . 34
2.4.6 Service-Level Agreements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4.7 Big Data and Real-Time Analytics . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3 Solution 39
3.1 Scope Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 Requirement Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2.1 Big Data Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2.2 Real-Time Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2.3 Functionalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2.4 Quality Attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2.5 Use Case Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.3 Conceptualization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.3.1 Energy Monitoring (Real-time computing) . . . . . . . . . . . . . . . . . . . . . . . 45
3.3.2 Energy Analytics (Batch processing) . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.4 Energy Cloud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.4.1 Dashboards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.4.2 Analytics API . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.4.3 Message-oriented Middleware (MOM) . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.5 Energy Monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.5.1 Storm Topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.5.2 Storm Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.5.3 Storm Clusters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.5.4 Energy Storm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.5.5 Real-Time Messaging Servers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.6 Energy Analytics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.6.1 Hadoop Cluster . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.6.2 Timeseries DB and Distributed Historical Storage . . . . . . . . . . . . . . . . . . . 53
3.7 Deployment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4 Evaluation 55
4.1 Conceptual Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.1.1 Use Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
x
4.1.2 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2 Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2.1 Data Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.2.2 Virtual Energy Cloud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.2.3 Test Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.2.4 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5 Conclusions 69
5.1 Achievements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Bibliography 76
A Virtual Energy Cloud Models 77
B Energy Cloud 82
xi
xii
List of Tables
1.1 Challenges and opportunities for industrial energy management using the cloud . . . . . 2
2.1 Demand Response programs used by energy providers and consumers . . . . . . . . . . 20
2.2 Survey of Energy Management Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Automation protocols and standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.1 Energy key performance indicators for the industry . . . . . . . . . . . . . . . . . . . . . . 56
4.2 Results of the solution proposal conceptual evaluation . . . . . . . . . . . . . . . . . . . . 68
xiii
xiv
List of Figures
1.1 Current industrial energy management model . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Research methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1 Smart grid environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Impacts of Demand-Site management programs in production . . . . . . . . . . . . . . . 10
2.3 The real-time requirement in the industrial sector . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Architecture of Energy Management Systems . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.5 The PDCA cicle model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.6 Load shifting technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.7 Industrial energy power load curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.8 Demand Response events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.9 Architecture of Industrial Automation Systems . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.10 Automation pyramid model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.11 Architecture of the Cloud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.12 Responsibilities of cloud providers and users . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.13 Costs of on-premises systems versus cloud systems . . . . . . . . . . . . . . . . . . . . . 34
2.14 The Lambda Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.1 Energy Cloud use case diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2 Energy Monitoring use case diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3 Energy Analytics use case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4 Data Collection use case diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.5 Architecture of a cloud-native industrial EMS . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.6 Energy Monitoring dashboard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.7 Energy Analytics dashboard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.8 Storm Topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.9 Storm Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.10 Energy Storm topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.1 Virtual Energy Cloud multi-site view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.2 Virtual Energy Cloud metrics view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3 Storm sampling validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
xv
4.4 Storm test cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.5 Storm benchmarking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
A.1 Virtual Energy Cloud domain and service layer . . . . . . . . . . . . . . . . . . . . . . . . 78
A.2 Virtual Energy Cloud gateway layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
A.3 Detailed Virtual Energy Cloud Dashboard - Sites . . . . . . . . . . . . . . . . . . . . . . . 80
A.4 Detailed Virtual Energy Cloud Dashboard - Metrics . . . . . . . . . . . . . . . . . . . . . . 81
B.1 Energy Cloud Performance Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
B.2 Detailed Energy Monitoring Dashboard . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
B.3 Detailed Energy Analytics Dashboard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
xvi
List of Acronyms
CPS Cyber-physical system
CS Control Server
CSA Cloud Security Alliance
CRUD Create, Read, Update,
and Delete
DCS Decision Control System
DER Distributed Energy
Resources
DLC Direct Load Control
DR Demand Response
DSM Demand Side
Management
EAF Electric Arc Furnace
EC Energy Conservation
EE Energy Efficiency
EMS Energy Management
Systems
FERC U.S. Federal Energy
Regulatory Commission
HMI Human-Machine
Interfaces
HRIS Human Resources
Information System
HVAC Heating, Ventilation, and
Air conditioning System
IAS Industrial Automation
Systems
IaaS Infrastructure as a
Service
ICS Industrial Control
Systems
ICT Information and
Communications
Technology
ISO Independent System
Operators
IoT Internet of Things
IT Information Technology
KDD Knowledge Discovery in
Databases
KPI Key Performance
Indicator
LM Load Management
MES Manufacturing Execution
Systems
MTU Master Terminal Unit
NIST National Institute of
Standards and
Technology
PaaS Platform as a Service
PLC Programmable Logic
Controllers
RTP Real-time Pricing
RTU Remote Terminal Unit
SaaS Software as a Service
SCADA Supervisory Control
and Data Acquisition
SEP Smart Energy Profile
SG Smart grid
SLA Service Level
Agreement
SMB Small and Medium-sized
Businesses
SOA Service Oriented
Architecture
SR Spinning Reserves
TOU Time-of-Use
WSDL Web Service Description
Language
xvii
xviii
Chapter 1
Introduction
The industrial sector produces the most CO2 and is one of the largest consumers of electricity worldwide,
at a rate that continues to grow annually (International Energy Agency, 2008). However, due to limited
resources and high costs, energy production is not growing at the same ratio, resulting in a demand-
supply mismatch (Inamdar and Hasabe, 2009). In an effort to close this ever-widening gap, energy
suppliers and consumers are working together to keep demand under acceptable and secure levels.
Energy suppliers run a set of Demand Response (DR) programs that influence consumers to amend their
energy consumption, through changes in the price of electricity or by financial incentives (Mohagheghi
and Raji, 2012). On the other hand, energy consumers can use their available energy more efficiently
by their own initiative. In order to accomplish this, industries must find inefficiencies and reduce energy
consumption without affecting their business and production processes. In residential and commercial
sectors, this essentially involves using energy efficient equipment or dimming out lights and heaters.
In contrast, energy efficiency initiatives encounter unique difficulties in the industrial sector, including
production and quality constraints, multiple energy tariffs, and consumption and emission restrictions
that make the task of saving energy more complex. EMS are tools that monitor, control, and optimize
energy consumption (Fiedler and Mircea, 2012). Nevertheless, literature, research projects, studies and
industry expertise, make clear that there is a need for a novel and robust platform capable of providing
more and better energy information monitoring, integration, repository, and analytics towards a future
energy efficient manufacturing (see Table 1.1). In addition, recent advances in hardware, networking
and software of sensor and control equipment, resulted in a massive increase of machine data in terms
of volume, variety and velocity (the three V’s of big data), that hinders the capability to collect, monitor,
and analyze all these data (Saha and Srivastava, 2014).
In the meantime, cloud computing is coming to the forefront and being applied to various fields. Its
main application is solving large scale computation problems by optimizing and combining distributed
resources (Jadeja and Modi, 2012). Cloud computing is a model for enabling ubiquitous, convenient, on-
demand network access to a shared pool of configurable computing resources (e.g., networks, servers,
storage, applications, and services) that can be rapidly provisioned and released with minimal manage-
ment effort or service provider interaction (Mell and Grance, 2011). It provides several benefits such as
1
Industry challenges Cloud solutions
Challenges Implications Cloud benefits Added value
Industries have complexinfrastructures and opera-tions
Industrial EMS are ex-pensive and harder to im-plement
Cloud solutions reduceinvestment and mainte-nance costs and withless time-to-benefit
Organizations can nowafford to have an EMSwith less investment andmaintenance costs
Sites are often geograph-ically distributed and self-managed
Hinders the ability tohave a multi-site energymanagement
Cloud solutions are cen-tralized with easy accessto data
Achieve energy monitor-ing and analytics acrosssites
Weak integration betweenenergy and productiondata
Manufacturing is notenergy-aware and de-cision making is notinformed nor based onreal-time data
Cloud can provide inter-operability and integratedata from external sys-tems
React faster to changeswith data driven and in-formed decisions
Weak and decentralizedenergy efficiency and costanalysis tools
Harder to benchmark en-ergy usage and strate-gies across sites
Centralized correlation ofdata from all productionlevels
Derive knowledge fromenergy use and identifyinefficiencies
Huge amounts of ma-chine and energy data toanalyze and process
Guesswork to find con-sumption inefficienciesand optimal schedules
Cloud can optimize theperformance of schedul-ing algorithms
Find hidden patterns andproduce new, faster andricher knowledge
Table 1.1: Summary of the challenges and needs that the industry is facing with current energy manage-ment solutions and how cloud could address them (adapted from (Walawalkar et al., 2010; Cannata andTaisch, 2010; Bunse et al., 2011; Thollander and Ottosson, 2010; Inamdar and Hasabe, 2009; Givehchiet al., 2013)).
saving of IT costs and maintenance, strong integration capabilities, short time-to-benefit, and scalable
computation on demand that keep up with customer needs (Voorsluys et al., 2011). These benefits have
pushed many residential and commercial EMS solutions to the cloud (Motegi et al., 2003; Byun et al.,
2012; Hong et al., 2012). Hence, based on our research across publications, research projects, and
industry trends (e.g., IMC-AESOP, IMS2020, Industry 4.0, Internet of Things, Cyber-Physical Systems),
we believe that the industrial sector will also incorporate cloud technologies in their energy management
and production processes in the near future. Therefore, the central motivation for this thesis is to study
the migration of industrial EMS to the cloud, evaluate the possibilities to achieve more energy and cost
savings, taking advantage of the latest cloud computing and big data technologies, and finally propose
a solution. The conclusions of this thesis will be validated by developing, deploying and evaluating the
feasibility and performance of a cloud-native industrial EMS proof of concept with big data capabilities.
1.1 Motivation
Companies with multi-site production usually follow the traditional energy management model with on-
site energy management (Ates and Durakbasa, 2012). In practice, for each facility, an EMS has to be
deployed and maintained. Moreover, these EMS may not even be from the same vendor. Each of them
uses different proprietary protocols, resulting in heterogeneity in the organization and integration issues
later on. This deployment method also implies some downtime and propagation time whenever there is
a need for a system update, which often results in complications.
Achieving a global energy and CO2 emissions management and an energy efficiency performance
2
evaluation, across equipment, production processes, departments, and facilities may involve many con-
tacts, integrations, and constant assessments on-site over time. Along with the energy information
systems, someone has to be responsible to manage the energy consumption on premises.
In conclusion, this management model hinders global energy management with the following inef-
ficiencies (see Figure 1.1): the same resources and costs spend in multiple sites to provide the same
functionality, inefficient capabilities to integrate data across locations, and harder and slower knowledge
sharing, affecting the business processes and competitiveness of organizations.
Conventional industrial EMS deployment model
IEMS
IEMS
IEMS
IEMS
IEMS
Proposed cloud-based industrial EMS model
9
Energy Managers Corporate Managers
Cloud-based IEMS
Figure 1.1: Conceptualization that compares the current industrial energy management model and theone proposed in this thesis.
1.2 Problem statement and objectives
EMS have evolved a great deal in the previous years, but there is still a gap between current solutions
and the industry needs (see Table 1.1). Recent studies show that the solution lies in better energy
monitoring and control systems, integration of energy efficiency into production information systems,
and useful energy usage and cost benchmarking across equipment and productions sites to evaluate
EE (Fiedler and Mircea, 2012; Kyusakov and Eliasson, 2012; Bunse et al., 2011; Arinez and Biller, 2010;
O’Driscoll and O’Donnell, 2013).
Therefore, the main objective of this thesis is to propose a solution that could solve these needs. This
solution intends to be as optimal as possible, to facilitate problem identification and decision-making, and
energy savings opportunities exploration. Following current industry standards and future trends, it aims
to be a sustainable solution that can prevail for many years. In addition, this proposal intends to be an
affordable solution oriented for industries of any size.
1.3 Research methodology and contributions
The challenging goal of this thesis requires an efficient research methodology to achieve the proposed
objectives on time. It is necessary to review wide ranging topics and analyze how they could be com-
3
bined to obtain an unified solution (see Figure 1.2). The research of this thesis was also influenced by
the following external contributions:
IMS2020 1 was an EU-funded research project that studied road maps towards Intelligent Manufactur-
ing Systems (IMS) until 2020. It concluded that research is needed in five key areas: sustainable
manufacturing, energy efficient manufacturing, key technologies, standards, and education. Fol-
lowing studies identified the main needs of the industry sector like: energy-aware manufacturing
processes (better measurement and control systems to improvement of energy efficiency) and
more integrating of energy efficiency into production information systems (Bunse et al., 2011).
IMC-AESOP 2 was an EU-funded research and development project that studied the concept of a cloud
of services in Industrial Automation Systems (IAS). Service Oriented Architecture (SOA) was the
platform chosen to provide system interoperability. Service Oriented Architecture (SOA) provides
an excellent platform for developing systems with various services offered by real time controllers,
data acquisition systems, and legacy systems (Mora et al., 2012).
ABB Corporate Research 3 in Germany, contributed with industry expertise and feedback in the areas
of Energy Management, Cloud computing and Industrial Automation Systems (IAS). ABB is one
of the largest engineering companies in the world. ABB is a leader in power and automation
technologies that enable utility and industry customers to improve their performance while lowering
environmental impact.
Defining the research road map of this thesis involved an initial scope of work analysis. This anal-
ysis revealed that an intense review of literature, research projects, and industry surveys and external
expertise was needed as follows:
1. Perform a comprehensive literature review on the main concepts that could provide more energy
efficiency: Energy Management, EMS, and DR.
2. Study research projects and industry trends to understand where the industry is moving towards
and provide a sustainable solution that could work on future smart manufacturing architectures.
3. Analyse current developments in the area of cloud computing that can support the solution pro-
posed in this thesis.
4. Review of current literature, surveys, and trends together with industry experts from ABB, in order
to identify the fundamental needs of industrial organizations and what should be provided by this
thesis.
5. The research finishes with topic consolidations and refinements from all information sources, com-
bining the gathered research and using it to support the proposed objectives.1http://www.ims2020.net/2http://www.imc-aesop.eu/3ABB is a multinational corporation headquartered in Zurich, Switzerland, operating in robotics and mainly in the power and
automation technology areas. ABB has operations in around 100 countries, with approximately 150,000 employees (November2013)—http://www.abb.com/
4
ResearchWorkload
ResearchTimeline(weeks)0 1684 12
Less
Less
ResearchCenter
More
DemandResponse
CloudComputing
IndustrialAutomation
Systems
The main focus of this work
EnergyManagement
(Energy Efficiency)
Figure 1.2: Illustration of the research workload of this thesis and the research centers involved duringits development.
6. Demonstrate the claims and feasibility of this thesis through the development of a proof of concept
and evaluate it using a fleet simulator that simulates virtual energy metering devices operating in
multiple sites.
This thesis brings the following contributions to the academic field:
1. Performs an extensive review of the existing literature regarding the state-of-art, benefits and risks
of the main discussed topics: Energy Efficiency (EE), Demand Response (DR), Industrial Automa-
tion Systems (IAS) and cloud computing.
2. Analysis of the different topics regarding automation and cloud computing towards a more energy-
aware and efficient industry.
3. Proposes a solution for the problems described earlier, following the current industry trends and
standards.
4. Conceptual and performance evaluation of the solution proposed.
1.4 Document organization
This thesis is divided in 5 Chapters. Chapter 2 describes the research background that supports this
thesis. It starts by describing the status of the energy management domain and current approaches
to tackle Energy Efficiency (EE) and Demand Response (DR) in the industry sector. Furthermore, it
5
presents a survey of current EMS in use today. Moreover, it introduces Industrial Automation Systems
(IAS) and modern trends towards a more smart manufacturing. This Chapter concludes with a review
of cloud computing and its modern developments that influenced this thesis. The proposed solution can
be found in Chapter 3 and its evaluation in the following Chapter 4. Finally, Chapter 5 presents the final
conclusions obtained from this thesis and summarizes the most important aspects.
6
Chapter 2
Research Background
2.1 Energy Demand Management
This section describes the concepts that define the outline of this work scope regarding the energy
domain. It follows a top-down approach, by starting from the outer concept, the Smart grid (SG), to
Demand Side Management (DSM). The Section DSM breaks into two sub approaches: Energy Effi-
ciency (EE) and Demand Response (DR). These latters represent the main research topics of this work
in the energy field.
2.1.1 Smart Grids
The rigidity of the traditional grid was a major hindrance to overcome the problem of demand-supply
mismatch. Nowadays, utilities and customers work as partners, they need to find ways to communicate
and help each other. The concept of a SG, a computerized power grid that provides many advanced
services using a two-way communication and information infrastructure linking utilities and customers,
has been called to the rescue (Report et al., 2011; Kyusakov and Eliasson, 2012).
The end goal of the SG is to enhance reliability of electricity distribution, reduce peak demand,
shift usage to off-peak hours, lower total energy consumption and carbon dioxide footprint (Kyusakov
and Eliasson, 2012). However, due to the enormous costs involved in this upgrade, grid operators are
looking for ways to leverage these new services provided by the SG to offset the costs. Although, the
economic benefits realized from the wide adoption of DR are expected to pay the largest share of the
investment on the SG (Faruqui et al., 2010).
2.1.2 Liberalised Electricity Markets
In traditional vertically-integrated electricity systems, supplies are maintained by a monopoly provider
who has the responsibility to ensure that adequate generating capacity is available. Prices are gener-
ally regulated wherever electricity is scarce or not. Electricity market liberalization was introduced with
the intention of creating a reliable, economically efficient electricity sector and increase price setting
7
Markets
OperationsServiceProvider
Consumer & Producer
DistributionTransmission
BulkGeneration
Secure data communication flowsElectrical flowsDomain
Figure 2.1: Conceptual model that describes the SG environment and the interactions between thedifferent actors (adapted from (Locke and Gallagher, 2010)).
transparency. Several countries have liberalized their energy supply system through energy markets
where different power suppliers can coexist and offer energy to customers through bidding. This way,
prices are formed through complex interactions between the demand and supply side of the market.
Notwithstanding, liberalization of electricity tends to substantially benefit large consumers like industrial
customers because these are more energy dependent and therefore more easily willing to adapt their
load on request. However, most buyers do not participate actively in the price-setting process and thus
the process is far from complete.
Several countries have liberalized their energy supply system through energy markets where different
power suppliers can coexist and offer energy to customers through bidding. They are usually regulated
by national and international authorities to protect consumer rights and avoid oligopolies1. With energy
markets liberalization , prices are formed through complex interactions between buyers and sellers, i.e.,
between the demand and supply side of the market. However, most buyers do not participate actively
in the price-setting process and thus the process is far from complete. As a result, prices fail to play
their normal role of balancing natural swings in supply and demand, leading to excessive instability.
Electricity market liberalization was introduced with the intention of creating a reliable, economically
efficient electricity sector and increase price setting transparency. Notwithstanding, liberalization of
electricity tends to substantially benefit large consumers, such as industrial customers, since these are
more energy dependent and are more easily willing to adapt their load on request.
In traditional vertically-integrated electricity systems, supplies are maintained by a monopoly provider
who has the responsibility to ensure that adequate generating capacity is available. Prices are gener-
ally regulated wherever electricity is scarce or not. In liberalized systems, by contrast, the function of
balancing supply and demand is performed in almost real-time, normally through a wholesale electric-1An oligopoly is a structured market where only a few dominant producers operate and where any action of one producer has
an influence on the overall market, prices, and payoffs (David and Wen, 2001).
8
ity market, where information about the current and future supply and demand balance is signaled by
electricity prices. Generally, efficient market prices are formed by interactions between suppliers and
customers. This interaction determines the value of supply at any point in time. However, in some
liberalized electricity markets, nearly all retail customers are exposed to prices that are fixed for rela-
tively long periods, regardless of the supply-demand balance in the market. Under such conditions, the
customers have no incentive to vary their consumption in response to actual market conditions.
2.1.3 Demand-Side Management
DSM is one important function in a SG that helps the energy providers reduce the peak load demand
and reshape the energy load demand profile. DSM includes everything that is done on the demand side
of an energy system, ranging from improving energy efficiency by using energy efficient materials, over
smart energy tariffs with incentives for certain consumption patterns changes, up to real-time control of
on-site power generation systems. There are different conceptions for DSM programs but generically
they can be categorized as follows (Palensky and Dietrich, 2011; Walawalkar et al., 2008; Albadi and
El-Saadany, 2007; Mohagheghi and Raji, 2012):
Energy Conservation (EC) focuses on user energy consumption behavioral changes to use less en-
ergy usually driven by education (e.g., use of natural lighting over electrical lighting).
Energy Efficiency (EE) means using building materials, equipment or techniques that are more energy
efficient, i.e., use less power to perform the same tasks (e.g., replacing an incandescent lamp with
a compact fluorescent lamp which uses much less energy to produce the same amount of light).
Demand Response (DR) refers to the changes in customers energy end-use from their nominal con-
sumption patterns. These changes in consumption are often in response to changes in the price
of electricity over time, or due to incentive payments (e.g., dimming down lights to comply with the
utility request to reduce energy consumption under a certain level, in exchange of a compensation).
Kyusakov et al. (Kyusakov and Eliasson, 2012) affirms that one of the main challenges for DSM and
SG is on the Information and Communications Technology (ICT) side. It is ICT interoperability, scala-
bility, information security, and network management that show more challenges. Fortunately, several
international research projects are working on smart grid, communication, security, interoperability, and
smart-metering standards for energy management (Palensky and Dietrich, 2011; Kyusakov and Elias-
son, 2012).
Manufacturing processes as paper pulping, steel smelting using Electric Arc Furnace (EAF) 2 con-
sume lots of energy. The industrial sector is often the bigger portion of the total load served by utilities.
At many utilities, industrial customers (2-10% of total customers) account for at least 80% of the elec-
tricity usage (Mohagheghi and Raji, 2012). This further emphasizes the importance of the role of the
industrial sector in these previous measures.2An EAF is a furnace that heats charged material by means of an electric arc. The use ofEAFs allows steel to be made from a
100% scrap metal feedstock. As EAFs require large quantities of electrical power, many companies schedule their operations totake advantage of off peak electricity pricing.
9
Figure 2.2: Impacts of DSM programs in the quality of industrial production processes (adaptedfrom (Palensky and Dietrich, 2011)).
2.1.4 Energy Efficiency
Efficient energy use, sometimes simply called Energy Efficiency (EE), is the goal to reduce the amount of
energy required to provide products and services without compromissing them (Palensky and Dietrich,
2011). Achieving this goal may involve using more energy efficient equipment, methods, processes,
energy management techniques, or others. For example, insulating a home allows a building to use less
heating and cooling energy, to achieve and maintain a comfortable temperature inside. Another typical
example would be the use of fluorescent lights or natural skylights, to reduce the amount of energy using
traditional incandescent light bulbs. Improvements in energy efficiency are usually achieved by adopting
a more efficient technology or production processes or by application of commonly accepted methods to
reduce energy losses. These measures imply immediate and permanent energy and emissions savings,
and therefore are the most accepted methods.
Surely, EE potential and operations of physical parts (motors, lights, Heating, Ventilation, and Air
conditioning System (HVAC), equipment, etc.) are important, but they are relatively well researched and
are out of scope of this work (Palensky and Dietrich, 2011). Hence, this work is only concern with EE
measures driven by Information Technology (IT) systems and processes in the industrial sector. There
are important drivers to introduce EE in this sector (Bunse et al., 2011):
• Rising energy prices makes energy consumption reduction more and more important to manu-
facturing companies. Especially in energy-intensive industries (e.g., steel, cement, pulp and paper,
chemicals), where energy can account for up to 60% of operating costs, turning energy costs in a
strong factor for competitiveness.
• New environmental regulations with environmental taxes, subsidies, emission permits, and green
certificates for CO2 emissions. For example, in Germany, the implementation of an certificated En-
10
ergy Management Systems (EMS), allows companies to save ten or hundred thousands of Euros
per year on environmental taxes reductions (Fiedler and Mircea, 2012).
• Changing purchasing behaviors based on products or services that have been manufactured in
a more environmentally friendly way, known as “green products”.
2.1.5 Energy Management
Energy management stands for all the measures and activities which are planned or executed in order
to minimize the energy consumption of a company or institution (Fiedler and Mircea, 2012). It influences
the organizational and technical processes as well as patterns of behavior and labor in order to reduce,
within economical constraints, the consumption of energy and increase energy efficiency. More specifi-
cally energy management includes control, monitoring, and improvement activities for energy efficiency.
In the end, energy management is beneficial for industrial companies for economic, environmental and
societal reasons (Bunse et al., 2011).
Nevertheless, several studies have identified a low status of energy management as a barrier to
energy efficiency (Bunse et al., 2011). One of the reasons lies on the energy managers that perform
energy management. Usually these managers are not qualified for the job or aren’t fully committed to
the role. A survey performed by Siemens in the UK to the top top 600 leading companies, showed that
only 1 in 10 energy managers spend up to 50% of their time on energy related issues and 94% of them
don’t have any qualification in energy management (Uk, 2011).
Small and Medium-sized Businesses (SMB) may not have the necessary capital or needs to hire an
energy manager. But sometimes even big companies with the necessary resources to recruit, lack of
energy managers and tools. Ates et al. (Ates and Durakbasa, 2012) surveyed 120 large companies with
a total annual energy consumption of 1000 toe or more and with 80% of them having more than 500
employees, from the top 2000 industrial companies in Turkey. The study revealed that 18% of surveyed
organizations don’t have energy managers. More importantly, it discovered that only 24% of those large
companies actually practice energy management. As for the SMB, it estimated a rate significantly below
20%.
2.1.6 Energy Management Systems
EMS is an energy management tool used in a wide variety of applications to effectively monitor, opti-
mize and control power generation, distribution and consumption. The main goal of these system is to
increase energy efficiency and thus achieve energy savings, through continuous monitoring and mainte-
nance of the facilities, improving the operation of equipment and decreasing energy consumption without
compromising the customer needs(Arinez and Biller, 2010).
The cost saving argument is probably the major driver for the majority of organizations implementing
an EMS (Fiedler and Mircea, 2012). Since lowering the energy costs increases the profit, the search for
energy saving potentials always merits. On the other side, one of the barriers to the adoption of EMS it’s
the capital investment necessary to deploy these systems.
11
EnergyQualityProduction
Timeline
Sudden changein operating conditions
Steady-stateoperation
Figure 2.3: Real-time is a requirement in the industrial sector because sudden disturbances affect sev-eral domains such as energy, production, and quality simultaneously (adapted from (Ma et al., 2010)).
Today, these systems offer broad scope of capabilities and features. These can be found in energy
suppliers (e.g., in electrical generation plants and power transmission supervision centers) and in energy
customers (e.e., in the industrial, commercial and residential sector). Nowadays both produce and
consume energy and data, they both need these systems to increase energy consumption awareness
and to provide informed decisions. Independent of the magnitude and application, each kind of EMS
has its own unique requirements depending on the user’s needs.
2.1.7 Industrial Energy Management Systems
The focus of this work is on those EMS found on the industrial customers, i.e., small to big manufacturing
facilities. Research shows that this sector is lacking complex and intelligent energy monitoring and
control systems when compared to EMS found in other sectors (Arinez and Biller, 2010; Kyusakov and
Eliasson, 2012; Bunse et al., 2011).
While on other sectors, as the residential and commercial, energy management may only involve
using more energy efficient equipment, dimming lights or switching off air condition equipment, the
industrial sector have some unique challenges that difficult the task.
To find out which activities can be dimmed or suspended in respond to a DR request or even, how
power demand can be reduced by rescheduling power dependent activities in response to a DSM ini-
tiative, we first need to understand how the facility is spending its energy and which constraints there
are.
Energy Management Systems Architectures
EMSs are intended to help managing and reducing energy consumption in any facility infrastructure. To
achieve this purpose, a standard EMS is usually architected in a multi-layer application as follows (Ma
et al., 2010):
12
Energy-use optimization
Performance evaluation
Energy-use interpretation
Data transmission
Energy data acquisition
EnergyMeters
Sensors
ApplicationLayers
Data Acquisition
Layers
IntegrationLayers
Harware
Network
Figure 2.4: Conceptualization of the generic architecture for EMS composed by a data acquisition layerto gather data from the field and transmit it to the upper layers, an integration layer to transform data intointernal representations, and an application layer to analyze these data (adapted from (Ma et al., 2010)).
Energy data acquisition layer includes the modules responsible to communicate with sensors and
metering devices and retrieve its data like: status, temperature, humidity, illumination intensity and
the current amount of energy consumed. This can be performed by pooling the devices periodically
or event-driven, where whenever some value changes the device has the responsibility to send the
data.
Data transmission is a middle-layer operating the data from the field level (aggregating it) and estab-
lishing connections with the central system. Industrial networks usually use communication proto-
cols standards like the M-Bus, Modbus, CAN, AS-I bus, Interbus or the Profibus to exchange data
between meters and servers (O’Driscoll and O’Donnell, 2013; Feuerhahn et al., 2011; Bayindir
et al., 2011; Kyusakov and Eliasson, 2012).
Energy-use interpretation containing the module responsible for evaluating, integrating, transforma-
tion and mapping retrieved data into the EMS energy data models repository.
Performance evaluation layer is responsible to benchmark energy performance like energy efficiency,
energy consumption, energy costs and others energy Key Performance Indicator (KPI) specific to
the customer bussines.
Energy-use optimization layer contains the modules responsible to reduce energy consumption and
to optimize the operation of equipment to increase energy efficiency.
13
The ISO 50001 standard
It is only through the use of standards that the credibility can be verified and requirements of inter-
connectivity and interoperability can be assured. With an issued EMS certificate, a company proves a
sustainable company strategy together with a reasonable usage of energy that strengthens its company
image.
ISO 50001 establishes an international framework for industrial plants and companies manage en-
ergy, including all aspects of processes and the energy management system model (Fiedler and Mircea,
2012). Based on the Plan, Do, Check, Act (PDCA) model, this standard provides an energy manage-
ment implementation strategy which involves (i) establishing an energy management policy (ii) forming
an energy management team to effectively implement an energy management system (iii) conduct an
energy review (iv) identifying and analyzing opportunities for improving energy performance (v) estab-
lishing a baseline and energy performance indicators for tracking the progress (vi) helping and guide
to set energy performance improvement targets (vii) implementing action plans to achieve customer
targets.
Like all International Organization (ISO)3 standards, such as the quality management ISO 9001 or
the environmental management ISO 140001, the ISO 50001 was designed to be implemented by any
type of organization, independently of its size, business, or geographical location.
It does not impose any energy performance improvement targets. The strategic and operative energy
targets are rather up to the organization itself. In other words, any organization, regardless of its cur-
rent level of energy management, can implement the ISO 50001 standard and achieve a improvement
baseline (Fiedler and Mircea, 2012).
PDCA (Plan, Do, Check, Act) is the model for energy management when employing the ISO50001.
The PDCA cycle provides a framework for continuous improvements of processes or systems. It is a
dynamic and repeating model, the results of one cycle are the input for the following (Fiedler and Mircea,
2012). This structure enables a continuous reassessment of the energy consumption and a sustainable
optimization and reduction:
1. Plan: conduct the energy review and establish the energy-use baseline, energy performance indi-
cators, objectives, targets and action plans necessary to deliver results in accordance with oppor-
tunities to improve energy performance and the organization’s energy policy.
2. Do: implement the energy management action plans.
3. Check: monitor and measure processes and the key characteristics of its operations that deter-
mine energy performance against the energy policy and objectives and report the results.
4. Act: take actions to continually improve energy performance and the EMS.
3http://www.iso.org/
14
Plan
Energy policy and energy-use review
Objectives and action planning
Do
Implementation and operational control
Awareness and training of staff
Check
Monitoring and analysis
Corrective and preventive actions
Act
Management Review
Optimisations
Internal audit of the EMS
Figure 2.5: Illustration of the different stages of the PDCA cicle model in ISO 50001 to perform energymanagement efficiently (adapted from (Chiu et al., 2012; Fiedler and Mircea, 2012)).
Load management
Load Management (LM) is the process of balancing the energy consumption over an electric network in
order to avoid consumption during high price periods and optimize utilization of valuable resources like
fuels, power generators, power transmission networks, and network distribution capacity (International
Union for Electricity applications, 2009).
Power demand during system on-peak demand is therefore more expensive since it requires expen-
sive generation power stations. If a customer can reduce his demand during a on-peak demand, hence
reducing the supplier requirement for network capacity, then the customer reduces the total electricity
charges, since saving costs to the supplier, distributor and to the producer. This also allows the post-
ponement of the need for additional capacity, while at the same time increasing the operating efficiency
of the energy system.
This may include using on-site power generation, energy storage equipment, shifting demand to a
less expensive period of the day, such as lighting and heating, or through temporary shut-down of one
or more processes
Some of the most common demand changes techniques to achieve energy costs reduction without
sacrificing the performance and quality of the manufacturing processes are the following (International
Union for Electricity applications, 2009; Guntermann, 1982; Raghavendra Nagesh et al., 2010; Piette
et al., 2004):
15
Figure 2.6: Load shifting example where the cooling activity is shifted from a on-peak period (left side)to off-peak periods (right side) to keep consumption under the billing energy demand limit.
Load shifting is a technique where the energy consumption period is shifted to periods of the day with
lower energy consumption, or in particular, with lower prices. Although the same amount of energy
is used, the overall costs associated with the energy consumption will be reduced, because the
consumption will be shifted from on-peak to off-peak time slots.
Load shedding or Demand Limiting, is a technique that simply curtails energy loads, i.e., it reduces
current energy consumption by forcing equipments or processes shutdowns.
Load priority systems to avoid large loads interacting simultaneously, e.g., motors and ovens starting
at the same.
Energy storage units are charged during off-peak periods and used during peak hours, i.e., power
batteries.
On-site generation also called Distributed Generation, are systems with small-scale power generation
technologies used to provide an alternative to or an enhancement of the traditional electric power
system, e.g. solar panels.
Load curve profiling
Complex manufacturing facilities consume a significant amount of the industrial sectors electrical energy,
to power motors, compressors, machine tools and it is also required to maintain adequate heating,
ventilation and air conditioning (O’Driscoll and O’Donnell, 2013). Industrial energy use can be classified
as (Rahimifard et al., 2010):
Indirect energy is used to maintain the environment that surrounds the production processes, e.g.
energy used to power lights, sensors or HVAC.
16
Oven
Lighting
Heating
Hot water
Hot water
LightingLighting
Ventilation
Permanent processes in production
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24Time (h)
0102030405060708090
100110120130140150160170
Power (kW)
Figure 2.7: Example of an industrial power load curve, describing all the energy dependent activitiesand its consumptions over the day (adapted from (International Union for Electricity applications, 2009)).
Direct energy is defined as the energy used by various processes (e.g. casting, machining, spray
painting, inspection, etc.) required to manufacture products. Hence this is the essential energy
that production facilities depends on, making it the hardest one to be dimmed, because it can
jeopardized the whole production process, turning the power demand enhancement not costly
worth it.
Figure 2.7 provides an example of load curve of a production factory and in it, we can see all the
factory power dependent activities and their consumptions over time. As we can see, the energy used
in the activities ‘Permanent processes in the workshop‘ and ‘Oven‘ are classified as direct energy since
they power the production process. As for the other activities, they use indirect energy since they support
the production.
Load curve optimisation
After obtaining the site load curve, it’s necessary to analyse the curve and how it can be enhanced. This
process includes the following tasks (International Union for Electricity applications, 2009):
Definition of the load curve objectives according to the pricing mechanisms in effect. From the cus-
tomer point of view, a flatted curve may not be the the best solution. Industrial customers often
17
want to reduce the load during on-peak time and to increase off-peak consumption.
Incentives and opportunities checking the electricity bills to determine if any DSM promotion exist at
the moment and in which time periods.
Archiving load curves for a large period of time will provide a proper historic of the actions taken and
how well they reflected in the total energy efficiency. It may also be useful for load forecast in
similar situations.
Analysis of the processes and supporting operations energy consumptions, characterises the energy
load. This will allow to manage the amount of load reduction that can be achieved by means of
interruption or deferment of every single operation measured.
Constraints analysis of LM appliance. That is, to make sure that constraints do not exist in interrupting
or rescheduling loads, with respect to: safety of operations, impacts on quality and quantity of pro-
duction, preservation of the integrity of the equipment, and mutual interactions with other facilities
in the factory.
2.2 Demand Response
Electricity supply and demand must remain in balance in real time to ensure stability on the electricity
grid. Since seasons and weather influence electricity demand and since electricity cannot be stored in
large quantities, it is necessary to plan energy supply availability in advance. Without this task, supply
interruptions in the form of brownouts and blackouts would be common, causing considerable economic
damages.
Utilities rely on peaking power plants to meet these demand periods peaks, known as on-peak peri-
ods. However, these power plants are very expensive to run, thus suppliers try to keep energy demand
under control, to avoid running these power plants or to install extra capacity, and thus increasing the
electricity prices (Mohagheghi and Raji, 2012; International Union for Electricity applications, 2009). For
instance, the capital cost needed to produce 1MW of DR (see Section 2.2) capacity in collaboration with
the customers is about $240,000 vs. $400,000 for using a gas-fired peaking power plant. DR capacity
it’s even faster too, since it can potentially be dispatched in less than 5 minutes, whereas a peaking
power plant can take up to 30 minutes to ramp up to full capacity (Mohagheghi and Raji, 2012).
The solution relies on conserving more electricity, i.e., demand-side needs participate and reduce
their energy-use, by what is commonly called as virtual generation (Carreira et al., 2011). Either by their
own initiative, hence performing energy management on-premise to minimize their energy consumption,
or by the energy utility initiative. The fundamental idea is that on critical on-peak periods, the grid would
request the consumers to reduce their loads, hence, acting as if they had power generating capacity of
their own (though some may actually have) and economically compensate them for their participation.
It’s a win-win situation for both because utilities defer from investments and continue to have an available
and efficient power supply system, and customers benefit themselves from reduced costs and extra
compensations.
18
CustomerUtility Control CenterData Sources
DR Decision Engine Server
HMI
DR Validity Check
Energy MarketRates
Grid Status
DR Client Module
Load ManagementPolicy
EnergyMetering
Utility DR
Signal
ResponseSignal
EnergyLoadData
Signal Received
DR Initiation
Response Deadline
DR Release
Normal Operation
Load Forecast
Ram
pPeriod
SustainedPeriod
RecoveryPeriod
Dem
and Response
Event
Figure 2.8: Overview of the interactions between energy providers and consumers during DemandResponse events (adapted from (Dam et al., 2008)).
This is where DR comes into play. While the goal of EE is to reduce energy use (kW/h), the goal
of Demand Response (DR) is dynamic reduction of peak electricity demand (kW) (Kiliccote and Piette,
2005). DR is a DSM solution targeted to residential, commercial and industrial customers, with the
purpose of reducing or shifting power demand to a specific time for a specific duration, when energy
market prices are high wholesale or when system reliability is jeopardised. DR induce energy demand
alterations with changes in the price of electricity or with incentive payments (Manuel and Cardoso, 2013;
Mohagheghi and Raji, 2012; Cappers et al., 2010; Granderson and Piette, 2011; Albadi and El-Saadany,
2007; Cardoso, 2012; Dam et al., 2008). In other words, DR includes all intentional modifications to
consumption patterns of electricity of end-use customers that are intended to alter the timing, level of
instantaneous demand, or the total electricity consumption (Albadi and El-Saadany, 2007).
2.2.1 Demand Response Programs
There are many classifications for DR programs, but they can be roughly grouped in the ones based on
incentives and the ones based on time tarrifs (see Table 2.1). Other way to look at the various programs
of demand response is to distinguish in (i) Market DR for the plans that involves wholesale energy market
price signals and incentives (ii) Physical DR for the plans with utility grid load management signals and
utility emergency signals (Palensky and Dietrich, 2011).
Regarding industrial customers, they are usually billed according to Time-of-Use (TOU) rates (In-
ternational Union for Electricity applications, 2009). This means that the cost of the energy consumed
depends on the hour of the day and the season. Prices are higher during on-peak time and lower at
off-peak, hence customers are provided with these “price signals” that stimulate customers to change
their consumption.
19
Type Program Description
Incentive Based Direct Load Control (DLC) Utility or grid operator gets direct access tocustomer utilities.
Curtailable rates Customers get special contract with limitedloads.
Emergency signals Voluntary response to direct emergency sig-nals from the utilities.
Capacity markets Customers guarantee to pitch in when thegrid is in need.
Time-Based Rates Time-of-Use (TOU) Fix price tariffs based on periods of the daywith high and low power demand.
Critical peak pricing Price tariff based on seasonal demand peaks(e.g., 3—6 pm. on a hot summer weekday).
Real-time Pricing (RTP) Wholesale energy market prices are for-warded to end customers.
Table 2.1: Summary of Demand Response (DR) used by energy providers and consumers to keepenergy demand under control (adapted from (Palensky and Dietrich, 2011; Dam et al., 2008)).
2.2.2 Demand Response Standards
Customers can stabilize the power grid by increasing or decreasing their electricity consumption based
on the amount of energy available in the utility grid. The problem is that this requires a large amount
of customers. Hence, standards are required to make this work. A networking standard for demand
response, such as OpenADR 2.0B, will help grow demand response and enable large numbers of very
different customers to stable the SG (Kyusakov and Eliasson, 2012).
OpenADR 2.0 (Open Automated Demand Response) it’s a network protocol standard for energy infor-
mation and communication exchange, target to SG to standardize, automate and simplify DR. It con-
tains a set of data models and interfaces (exchange patterns) that define standard DR signals and the
interfaces between utilities, energy markets (dynamic and transaction pricing information), Independent
System Operators (ISO), Distributed Energy Resources (DER), and energy consumers (industrial or res-
idential buildings). The communication interfaces are based on SOA and are defined using Web Service
Description Language (WSDL) using SOAP Web Services. Unfortunately, the use of these technologies
arises technical challenges to resource-constrained devices due to its computing resource requirement,
therefore the OpenADR scope does not cover Internet of Things (IoT) devices.
Smart Energy Profile (SEP) 2.0 is an application layer specification target to IoT devices for on-
premise DR and LM management. Created by the Consortium for SEP Interoperability (CSEP), SEP
has been identified by the National Institute of Standards and Technology (NIST) as a primary candidate
specification for energy information and control on the consumer side. The specification includes smart
metering, pricing, DR, and LM applications for devices in residential and light commercial buildings
operating on a Home Area Network (HAN), sometimes called Premises Area Network. SEP 2.0 runs
20
on top of the IP protocol and therefore, it supports Ethernet, WiFi, powerline, and low-power radio
communications. Unlike OpenADR who uses SOAP as web services, SEP 2.0 relies on RESTful web
services and Create, Read, Update, and Delete (CRUD) operations.
2.2.3 Survey on Energy Management Systems
Energy Management Systems (EMS) exist for a few decades now, but due to lack of standards and
demand in previous years, this led to many proprietary implementations and systems outdated without
updates in years (Fan et al., 2005).
With the proliferation of smart energies and demand for EMS have pushed these systems further.
Even though, without standards on functionalities and architectures, many different implementation so-
lutions with different purposes, emerged in the market.
This work presents here a survey current EMS demonstrating exactly this phenomenon. There
are thousands EMS in the market now. For this work the focus rested in those who either had cloud
capabilities or focus to the industrial sector. The final set of EMS chosen were the following: McKinstry
- EEMSuite 4, KGS - ClockWorks 5, Powertech - EMS 6, Enernoc - DemandSMART 7, GE - XA/21 8,
and ABB - cpmPlus EM 9. This set results from the top search findings for specific EMS with Cloud
technologies and from the collaboration with ABB.
This has proved to be a hard task because it’s harder to make an apple-to-apple evaluation with
solutions target for different industries, even though, providing the same features. As intended this
surveys shows that current cloud EMS are still more focus for the residential and building sector. In
addiction, it shows that there are some gaps between cloud-based systems and industrial systems that
this work proposes to solve. As we can see, industrial EMS are stronger in monitoring and controlling
capabilities, but not so strong in integration and benchmarking capabilities.
4http://www.mckinstryeem.com/5http://www.kgsbuildings.com/clockworks.aspx6http://goo.gl/X4ElbA7http://www.enernoc.com/for-businesses/demandsmart8http://goo.gl/16hqj59http://goo.gl/CjcsLw, http://goo.gl/Ky6wv9
21
DemandSMART it’s the best example in this survey for a cloud-based EMS with DR capabilities using
OpenADR 2.0 (Deliso, 2013). Created by EnerNOC’s10 a demand response leader company, it ensures
that participating commercial, industrial, and institutional entities receive maximum payments for their
participation in demand response. They manage more than 8,500 MW of demand response capacity
worldwide on behalf of utility and grid operator clients.
According to EnerNOC moving to the cloud, allowed DemandSMART to be more capable of larger
amounts of demand response and energy efficiency than before. As this work solution proposal, they
believe that increasing the energy management scope from peripheral management to facility manage-
ment and responding to grid events and energy prices, it’s the natural energy management and energy
intelligence evolution.
Features surveyed
Several different set of features where analysed while performing this survey: (i) Cloud Technologies, (ii)
Systems Integration, (iii) Data Mining and Analysis, (iv) Consumption Measurement and Benchmarking,
(v) Monitor and Control, and (vi) Load Management capabilities. Each of the following set of features is
summarized in Table 2.2:
• Cloud Technologies refers to cloud-based functionalities provided by the solution;
• Data Integration refers to the ability of the solution to integrate data from other systems;
• Data Mining and Analysis refers to features to extrapolate information out of raw data and present-
ing them to the end-user;
• Energy-use refers to ability to measure the energy consumption over time and facing them against
key performance indicators;
• Monitor and Control refers to the capability to monitor data on different equipments and plants;
• Load Management capabilities refers to the solution load management capability.
10EnerNOC, Inc. engages in the business of providing energy management applications, services and products for the smartgrid, which include comprehensive demand response, data-driven energy efficiency, energy price and risk management andenterprise carbon management applications and services.—http://www.enernoc.com/
22
McK
inst
ry-
EE
MSuit
e
KG
S-
Clo
ckW
orks
Pow
erte
ch-
EM
S
Ener
noc
-D
eman
dSM
ART
GE
-X
A/2
1
AB
B-
cpm
Plu
sE
M
Target Market
Building Sector (HVAC) 4 4 4 4 - -
Industrial Sector 4 - 4 - 4 4
Cloud Technologies
Cloud storage - 4 4 4 7 7
Cloud capabilities (Scalablity, etc.) - 4 - - 7 4
Cloud deployment (SaaS, Paas, IaaS) 4 4 4 4 7 7
Data Integration
Energy meter data gathering 4 4 4 4 4 4
Equipment status gathering 7 4 4 4 4 4
Environmental data gathering 7 4 4 - 4 4
Production data integration 7 7 7 7 - -
Data Mining and Analysis
Unsual pattern detection 4 4 7 - - -
E�ciency advisory - 4 7 4 - 4
Cost allocation - 7 4 4 4 -
Demand forecast 4 7 7 4 4 4
Energy-use Evaluation
Equipment e�ciency benchmarking - 4 - 4 - -
Process e�ciency benchmarking 7 7 - 7 - -
Multi-site benchmarking 4 4 - - - 4
KPI for energy e�ciency 4 - - 4 - 4
Monitor and Control
Real-time monitor 4 4 4 4 4 4
Remote control - - 7 4 4 -
Load Management capabilities - 7 4 - 4 4
Demand Response capabilities - 7 7 4 - -
Alarms 4 7 4 4 4 4
Table 4: Summary of general features provided by current EMS solutions:4 feature supported, 7 : feature not supported, - : unknown information
22
Table 2.2: Summary of general features provided by current EMS solutions:4 : feature supported, 7 : feature not supported, - : unknown information
23
2.3 Industrial Automation Systems
Industrial Automation Systems (IAS) are very complex industrial automated systems and technologies,
working together for a common automated production. The common automation system pyramid model
of Industrial Automation Systems (see Figure 2.10), hierarchically layers the composition of systems
with different rights and purposes (Givehchi, Givehchi). The shape of a pyramid was chosen because of
the characteristics of information on the different levels i.e. size of data packages (highers layers have
less amount of data), frequency of transmission, real-time requirements, availability requirements, etc.
Each level can be briefly described as follows (Department of Electrical Engineering IIT Kharagpur and
Iit, Department of Electrical Engineering IIT Kharagpur and Iit):
Enterprise level deals with Enterprise resource planning (ERP) systems which are the less technical
and the more focus in commercial activities, such as supply and demand management, account-
ing, product marketing etc.
Management level include the Manufacturing Execution Systems (MES) systems which are in charge
of managing production and solving problems like production targets, resource allocation, task
allocation to machines, maintenance management etc.
Supervision level comprise the supervision and control systems such as Supervisory Control and
Data Acquisition (SCADA), Decision Control System (DCS), and Human-Machine Interfaces (HMI)
which are responsible for supervising and controlling the overall production process.
Control level comprehends the automatic control systems such as Programmable Logic Controllers
(PLC) that monitor and drive the field devices as sensors and actuators to fulfill the localized tasks.
Field level includes the devices in the field of operation that translate electrical signals into actions in
the physical world (actuators) and the devices responsible to measure the environment (sensors).
The automation pyramid describes the total automation functions and includes all devices from the
field to the enterprise level. From the bottom of the pyramid, where the information is framed by the
technical process, up until the top level, where the enterprise resource planning systems for business
management can be found. It is strictly and the different layers represent functions of similar type.
The model also highlights the decreasing amount of data: the higher the level on the automation
pyramid, the fewer amounts of data there are. From level to level, data is condensed and transformed
into knowledge. At the bottom many short data from field devices (sensors and actuators) need to be
gathered and transferred to the control level, where they are used to control the production process.
From the control level to the supervision level only summarized control data which are relevant for
the operator need to be transferred. Typically those data are less frequently transferred and in bigger
packets. At top of the pyramid only orders from the business IT are transferred to the control level and
shift protocols as well as production KPIs are transferred to the business IT (Givehchi, Givehchi; Bowers,
2013).
24
2.3.1 Industrial Manufacturing
Manufacturing is the process of transforming raw materials into finished goods. This involves several
industrial processes who trough a set of tasks transform the input material into desirable outputs. The
production of goods is performed by the following different manufacturing processes (Fraser, 2000):
• Process-based manufacturing is the production of goods in bulk quantities which cannot be
distilled back into the original basic components due to a production recipe. Process manufacturing
industries typically utilize two main processes:
– Continuous Manufacturing Processes are processes running continuously, often with tran-
sitions to make different grades of a product. E.g., fuel or stems flows in petroleum refinery,
chemical distillation, etc.
– Batch Manufacturing Processes have stage by stage production, conducted on a quantity
of material. There is a distinct start and end step to a batch process with pauses in between.
E.g., pharmaceutical ingredients, water purification, and inks.
• Discrete-based manufacturing industries typically conduct a series of steps on products that can
be individually counted and labeled. E.g., production of automobiles, smart-phones, and airplanes.
Both process-based and discrete-based industries need control systems, sensors, and networks to
orchestrate theirs processes in mass production.
2.3.2 Industrial Control Systems
Industrial Control Systems (ICS) is a part of IAS that encompasses the control systems, denoted as
Controllers, used in industrial production to orchestrate several process-control activities through Sen-
sors and Actuators systems (Stouffer et al., 2006; Department of Electrical Engineering IIT Kharagpur
and Iit, Department of Electrical Engineering IIT Kharagpur and Iit). These controllers are essential ele-
ments that operate in the field level in a Control Loop, i.e., they are continuously measuring the physical
world through sensors, deciding what do next using control hardware has PLC, and acting based on the
gathered data by interacting with their environment using actuators.
Production processes can be monitored by operators and engineers using Human-Machine Inter-
faces (HMI) devices. These are used to display the processes status information, historical information,
and adjust parameters in the controllers.
Supervisory Control and Data Acquisition (SCADA)
SCADA systems supervise distributed control sub-systems, such as DCSs and PLCs, which are usually
geographically dispersed (Galloway and Hancke, 2013; Stouffer et al., 2006). There are many types
of SCADA systems offering different features, sometimes including remote control functionalities, but by
definition these systems are tailored towards the monitoring of remotes sites. To achieve this, SCADA
control centers utilize a complex HMI to visualize the status of the remote sites and a Master Terminal
25
External Network Enterprise Network Management Network
Control NetworkFielbbus Network
SCADA Control Center
Gateway
Remote Site #NRemote Site #1
GatewayLine or Radio
Communication
DCS #N
CS
PLC #1
PLC #1
GatewayFiel
dBus
HMI
ERP LegacySystem
EMSMESGateway
TCP/IP
Gateway
DCS #1
CS
PLC #1
PLC #N
GatewayFiel
dBusSensors,
Actuators
Weather, Energy
prices, etc.
RTU
MTU
Sensors, Actuators
Figure 2.9: Illustration of a Industrial Automation Systems (IAS) architecture where typically a SCADAsystem supervises multiple DCS systems which are controlling production using PLC systems. In top ofthose, other management systems control the other areas of the organization.
Unit (MTU) which communicates with Remote Terminal Unit (RTU) in the sites. This RTU can be an
individual device or incorporated within a Control Server or a PLC. Due to its purpose these systems
are usually used in distribution grid systems such as power, water, oil, and gas grids.
Decision Control System (DCS)
DCS systems are used to control production systems within the same geographic location. Each DCS
uses a centralized control loop to supervise a process or a discrete part of a production facility, which
operates using a localized group of controllers. By modularizing the manufacturing in many DCS sys-
tems it reduces the impact of a single fault on the overall system. These control loops are controlled by
a real-time Control Server (CS) which is responsible for directly gather all the data from the controllers in
the network and commanding to execute some action. Since field controllers communicate through field-
bus protocols, sometimes CS may need a Gameteways to translate and inter-connect the CS network
with the fieldbus network (see Section 2.3.3).
Programmable Logic Controllers (PLC)
PLC are small industrial computers designed to perform the process logic functions by controlling the
connected sensors and actuators. These systems form the core of industrial control systems and oper-
ate in hard real-time using a power supply, processor, input/output, and communication module and with
multiple inputs and output arrangements, extended temperature ranges, immunity to electrical noise,
and resistance to vibration and impact.
26
Domain Protocols and standards
Power system automation IEC 60870, DNP3, IEC 62351, Modbus, Profibus
Automatic meter reading ANSI C12.18, IEC 61107, M-Bus, Modbus, ZigBee
Process automation CIP, CAN bus, ControlNet, DeviceNet, DF-1, DirectNET, Ether-CAT, EtherNet/IP, GE SRTP, HART, Honeywell SDS, HostLink,Modbus, Optomux, PieP, Profibus, PROFINET IO, SERCOS in-terface, SERCOS III, Sinec H1
Industrial control system OPC DA, OPC HDA, OPC UA, MTConnect
Building automation BACnet, C-Bus, DALI, DSI, KNX, LonTalk, Modbus, oBIX, X10,ZigBee
Table 2.3: Summary of protocols and standards used in automated domains.
2.3.3 Industrial Networks
Industrial networks differs from conventional networks found in residential and building sectors. These
have unique requirements such as the need for strong determinism (bounded and low latency variance),
real-time data transfer (250 µs-10 ms) and fixed sampling periods. Hence, there are different network
characteristics for each layer within the IAS hierarchy (Galloway and Hancke, 2013):
Control Network connects the supervisory control level like SCADA, DCS, HMI to lower-level control
modules such as PLC.
Fieldbus Network links field devices to a PLC or other controller. Use of fieldbus technologies elimi-
nates the need for point-to-point wiring between the controller and each device. The field device
communicates with the fieldbus controller using an industrial control protocol. The messages sent
between the sensors and the controller uniquely identify each of the devices.
Gateway/Router is a communications device that transfers messages between two networks. Common
uses for routers include connecting a LAN to a WAN, and connecting MTUs and RTUs to a long-
distance network medium for SCADA communication.
2.3.4 Cyber-Physical Systems
The term Cyber-physical system (CPS) refers to a new generation of control systems with integrated
computational and communication capabilities to monitor and control in the physical world (Rajkumar
et al., 2010). These systems interact with the physical world, and must operate dependably, safely,
securely, and efficiently and in real-time.
CPS is the confluence of technologies as embedded systems, real-time systems, distributed sensor
systems and controls. CPS are often referred to as embedded systems. But unlike traditional embedded
systems, a CPS is typically designed as a network of interacting elements with physical input and output
instead of as standalone devices (Lee, 2008).
Regarding industrial control systems, that means the merging of controllers as PLC with physical
devices as sensors and actuators, with extra intelligent and real-time computational capacities capable
27
Figure 2.10: Illustration of the current industrial trend to move from the traditional automation pyramidarchitecture to Cyber-physical system (CPS) (adapted from (Givehchi, Givehchi)).
of predicting and adapting the behavior of the system upon changes in the environment. Examples of
CPS are aerospace systems, transportation vehicles and intelligent highways, defense systems, robotic
systems, process control, factory automation, building and environmental control, etc.
Although the individual components of CPSs such as sensor networks, controllers, actuators, dis-
tributed systems, etc. have reached a research maturity level, research in the integrated whole called
CPS, is still in its infancy. Thus CPSs are considered to be an emerging discipline of research.
2.3.5 Internet of Things
The Internet of Things (IoT) is another novel paradigm gaining ground in the academic and industry
domain. A prove of this movement is the manifold definitions for IoT. The basic idea of this vision is that
the technology is going to move towards a world whereas a pervasive variety of things of objects, with
unique identification capabilities, are going to be able to interact with each other and cooperate with their
neighbors to reach a common goal (Atzori et al., 2010).
In this scenario, Things like objects, machines, or people, are provided with unique identifiers and the
ability to exchange data over a network as the Internet, without requiring human-to-human or human-to-
computer interaction, and therefore considered to be Smart. IoT is the confluence of technologies such
as machine-to-machine (M2M) communication, wireless technologies, micro-electromechanical systems
(MEMS), and devices with connection to the Internet.
2.3.6 Industry 4.0
The term “Industry 4.0” refers the arising fourth industrial revolution promoted by the German govern-
ment, under the premise of “Smart Factories” (April, 2013), with the basic principle that by connecting
machines, work pieces and systems, we are creating intelligent networks along the entire value chain,
that can control each other autonomously. Industry 4.0 can be a reality in about 10 to 20 years and
it will address and solve some of the challenges facing the world today such as resource and energy
28
efficiency, urban production, and demographic change.
The first three industrial revolutions resulted from the evolution of technology by mechanizing the
production, using electricity, and by embody industrial IT. Now, the introduction of the IoT, CPS, Big
Data, and Cloud Computing into the manufacturing environment is ushering a fourth industrial revolution
to increase productivity, quality, and flexibility within the manufacturing industry (SmartFactoryKL, 2014).
Using these novel technologies in the manufacturing environment means comprising smart machines,
storage systems, and production facilities capable of autonomously exchange information, trigger ac-
tions, and controlling each other independently (April, 2013).
29
2.4 Cloud Computing
The Cloud it’s a concept that interest many companies with needs for less maintenance overhead, less
costs, unlimited resources, quick deployment, and easy scalability. Thus, many businesses sectors
are trying to incorporate the cloud in their processes. The manufacturing industry is one of them (Gilart-
Iglesias, 2007; Givehchi, Givehchi; Givehchi et al., 2013; Langmann et al., 2012; Luo et al., 2011; Macia-
perez et al., 2012; Staggs and Mclaughlin, 2010; Qu and Yingjun, 2014; Tao et al., 2011; Xu, 2012). But
there are some common questions surrounding this new technology: What kind of information should
be stored there? What are the benefits and risks involved? Is moving toward cloud computing right for
the industry? How could it help to perform energy management?
The cloud is no “silver bullet” solution. It has strengths and weaknesses and it should not be applied
without thinking. Understanding the counterparts is necessary to take the right decision.
2.4.1 Cloud Computing Concepts
The cloud itself is a pool of resources—networks, servers, applications, data storage and services—
which the end user can access and use on-demand (Mell and Grance, 2011; Meenakshi, 2012). The
cloud is a congregation term of diverse technologies into one. Technologies such as clusters, grids, and
now, cloud computing, have all aimed at allowing access to large amounts of computing power in a fully
virtualized manner, by aggregating resources and offering a single system view. The aim of the cloud is
to provide computing as an utility (Voorsluys et al., 2011).
Many authors and entities have attempted to define what exactly cloud computing is and its char-
acteristics. The most accepted definition of cloud computing is a model for enabling ubiquitous, conve-
nient, on-demand network access to a shared pool of configurable computing resources (e.g., networks,
servers, storage, applications, and services) that can be rapidly provisioned and released with minimal
management effort or service provider interaction (Mell and Grance, 2011).
The objective of cloud computing is to make a better use of distributed resources, combine them to
achieve higher throughput and be able to solve large scale computation problems (Jadeja and Modi,
2012). This is possible due to the cloud’s following characteristics (Mell and Grance, 2011):
On-demand self-service permits that a consumer can unilaterally acquire computing capabilities, such
as server time and network storage, as needed automatically without requiring human interaction
with each service provider.
Broad network access in the sense that capabilities are available over the network and accessed
through standard mechanisms that promotes heterogeneous use by thin and thick client platforms
(e.g., mobile phones, tablets, laptops, and workstations).
Rapid elasticity means that computing capabilities can be elastically provisioned and released, in
some cases automatically, to scale rapidly outward and inward commensurate with demand. To
the consumer, the capabilities available for provisioning often appear to be unlimited and can be
appropriated in any quantity at any time.
30
model commonly referred to as cloud computing in 2000s.
Cloud computing has different definitions and understandings from different perspectives and applications.The National Institute of Standards and Technology (NIST) defined cloud computing as “a model for enablingubiquitous, convenient, on-demand network access to a shared pool of configurable computing resources (e.g.networks, servers, storage, applications, and services) that can be rapidly provisioned and released with min-imal management effort or service provider interaction.” [25]. From the scientific view point, the main goalof cloud computing is to provide on-demand computing services with high scalability and availability in adistributed environment with minimum complexity for the service consumers.
Cloud computing architecture can be divided into several layers: the hardware layer, the infrastructure layer,the platform layer and the application layer [28] as shown in Figure 1.
Figure 1: Cloud Computing Architecture [39]
Hardware layer handles the physical resources of the cloud, including physical hardware, network devicesand power systems. Typical issues at hardware layer include hardware configuration, fault-tolerance, trafficmanagement and power management. Infrastructure layer is also known as virtualization layer. This layerpartitions the hardware and provides a pool of computing resources and disk storage. Platform layer is mainlycovers operating systems and application frameworks depending on each specific platform. This layer tries tominimize the development efforts by providing development platform to the developers as a service withoutinstalling any software or framework on their computers. Application layer offers the cloud applications tothe end users as a service. These applications can be automatically scaled with high performance inside thislayer with lower maintenance costs comparing with traditional applications [39].
Cloud computing is based on a service-driven model. In this model, hardware and software resources will bedelivered as services on-demand. These fundamental cloud services are categorized into three models:
IaaS provisions infrastructural resources such as virtual machines on-demand. It is the most essential cloudservice model. IaaS providers or cloud owners e.g. Amazon EC2, GoGrid and Flexiscale offer their resourcesto the users with least complexity using this service model.
PaaS provides platform layer resources e.g. software development frameworks and deployment components.Software developers employ these services to develop and deploy applications with minimum installation andpreparation of resources. Google App Engine, Microsoft Windows Azure and Force.com are examples of PaaSproviders.
SaaS offers on-demand cloud applications to the users through network. This service offers complete complex-ity abstraction for the users. They do not need to deal with preparing required hardware and software resourcesand application is accessible through a standard interface e.g. web browsers. Examples of SaaS providersinclude Microsoft Office 365, Google Calender and SAP Business ByDesign.
Besides, based on applications and architecture of clouds, they can be divided into four different types.
Public Cloud offer cloud-based applications and services to the general public via the Internet. Numerousorganizations and users can use the resources from an infrastructure at the same time. Benefits of this cloudtype include less investment for users to install and maintain infrastructures at their location and outsource theseoperations to the providers. However, public clouds limit the control over data privacy and security settings
Figure 2.11: Conceptualization of the cloud architecture with the different layers and services of softwareand hardware depicted (source (Givehchi, Givehchi)).
Computing resources pool to serve multiple consumers using a multi-tenant model, with different
physical and virtual resources dynamically assigned and reassigned according to consumer de-
mand. There is a sense of location independence in that the customer generally has no control or
knowledge over the exact location of the provided resources but may be able to specify location
at a higher level of abstraction (e.g., country, state, or datacenter). Examples of resources include
storage, processing, memory, and network bandwidth.
Measurement services to control and optimize resource use by leveraging a metering capability at
some level of abstraction appropriate to the type of service (e.g., storage, processing, bandwidth,
and active user accounts). Resource usage can be monitored, controlled, and reported, providing
transparency for both the provider and consumer of the utilized service.
2.4.2 Cloud Computing Service Models
The cloud computing model is service-oriented. In this model, hardware and software resources are
delivered on-demand as services. The fundamental cloud services are categorized as (Mell and Grance,
2011):
Software as a Service (SaaS) is the model when the consumer has the capability to use provider’s
applications on a cloud infrastructure. The applications are accessible from various client devices
through either a thin client interface, such as a web browser (e.g., web-based email), or a program
interface. The consumer does not manage or control the underlying cloud infrastructure including
network, servers, operating systems, storage, or even individual application capabilities, with the
possible exception of limited user-specific application configuration settings. Famous examples are
31
Private Cloud
Applications
Data
Runtime
Middleware
O/S
Virtualization
Servers
Storage
Networking
IaaS
Applications
Data
Runtime
Middleware
O/S
Virtualization
Servers
Storage
Networking
PaaS
Applications
Data
Runtime
Middleware
O/S
Virtualization
Servers
Storage
Networking
SaaS
Applications
Data
Runtime
Middleware
O/S
Virtualization
Servers
Storage
Networking
Managed by Customer Managed by Vendor
Figure 2.12: Illustration of the responsibilities of cloud vendors and customers in different cloud servicemodels (adapted from (Kalakota, 2013)).
SalesForce.com, BaseCamp.com, and Microsoft Office 365 (Givehchi, Givehchi). Google Apps is
the most widely used SaaS (Jadeja and Modi, 2012).
Platform as a Service (PaaS) designates the model when the consumer has the capability to deploy
onto the cloud infrastructure, consumer-created or acquired applications created using program-
ming languages, libraries, services, and tools supported by the provider. The consumer does
not manage or control the underlying cloud infrastructure including network, servers, operating
systems, or storage, but has control over the deployed applications and possibly configuration set-
tings for the application-hosting environment. Key examples are Google App Engine, Heroku, and
Microsoft’s Azure (Jadeja and Modi, 2012).
Infrastructure as a Service (IaaS) specify the model when the consumer has the capability to acquire
processing, storage, networks, and other fundamental computing resources where the consumer
is able to deploy and run arbitrary software, which can include operating systems and applications.
The consumer does not manage or control the underlying cloud infrastructure but has control
over operating systems, storage, and deployed applications, and possibly limited control of select
networking components (e.g., host firewalls). Examples are Amazon EC2, GoGrid, Flexiscale,
CloudFoundry, Joyent and Rackspace (Jadeja and Modi, 2012).
2.4.3 Cloud Computing Deployment Methods
The cloud provides the means through which the computation can be delivered as a service. These
services can be deployed in several different ways (Mell and Grance, 2011):
32
Private cloud system infrastructure is provisioned for exclusive use by a single organization compris-
ing multiple consumers (e.g., business units). It may be owned, managed, and operated by the
organization, a third party, or some combination of them, and it may exist on or off premises.
Community cloud system infrastructure is provisioned for exclusive use by a specific community of
consumers from organizations that have shared concerns (e.g., mission, security requirements,
policy, and compliance considerations). It may be owned, managed, and operated by one or more
of the organizations in the community, a third party, or some combination of them, and it may exist
on or off premises.
Public cloud system infrastructure is provisioned for open use by the general public. It may be owned,
managed, and operated by a business, academic, or government organization, or some combina-
tion of them. It exists on the premises of the cloud provider. Examples of public cloud providers
are: Salesforce.com, Amazon Web Services, Microsoft,
Hybrid cloud system infrastructure is a composition of two or more distinct cloud infrastructures (pri-
vate, community, or public) that remain unique entities, but are bound together by standardized or
proprietary technology that enables data and application portability (e.g., cloud bursting for load
balancing between clouds).
2.4.4 Benefits of Cloud Computing
The features of the cloud and cloud computing brought many benefits for vendors and customers like
the following (Voorsluys et al., 2011):
Saving on IT costs and maintenance because it allows to avoid overhead costs on acquiring and
mantaining hardware, software, and IT staff. Consumption is billed as a utility, usually by hour
slots, with minimal upfront costs. Thus, the customer will only pay for what they need at each
moment. Since it stretch and grows without the need to buy hardware, extra software licenses,
or programs, turns cloud computing solutions more affordable over time (see SAP11 case study in
Figure 2.13). In many cases, customers are even offered with the latest updates as long as they
continue to acquire the service.
Easy access and up-to-date data because applications can be easily accessed from anywhere in the
world with an Internet connection and a browser, i.e., without having to download or install anything.
The cloud keeps all files in one central location and everyone access the same repository.
Less time-to-benefit with quick deployment of IT infrastructures and applications. The software de-
ployment times and resource needs associated with rolling out end-user cloud solutions are signif-
icantly lower than with on-premises solutions (Berggren et al., 2013).
11SAP AG is a german multinational software corporation that makes enterprise software to manage business operations andcustomer relations. Headquartered in Walldorf, Baden-Wurttemberg, Germany, with regional offices around the world, SAP is theleader in the market of enterprise applications in terms of software and software-related services.—http://www.sap.com/
33
Hardware
License
Database
On-Staff
Implementation
Subscription
On-premise
Cloud
$ 1,000
$ 2,000
$ 3,000
$ 4,000
$ 5,000
$ 6,000
$ 7,000
$ 1,000
$ 2,000
$ 3,000
$ 4,000
Year 1 Year 2 Year 3 Year 4 Year 5 On-Premise Cloud
Ann
ual C
ost
5-ye
ar T
otal
Cos
t of O
wne
rshi
p
Figure 2.13: Study about the typical costs of an on premises SAP’s Human-Resources informationsystem for a 10.000 employee company versus the costs for an equivalent SAP cloud solution. The on-premises solution cost increases in year 5 because there are upgrades, re-implementation, and licensecosts involved that do not exist in cloud deployments (adapted from (Berggren et al., 2013)).
Improving business processes with better and faster integration of information between different
entities and processes.
Scalability on demand to overcame constant environments and usage changes. To scale vertically
(or scale up) means to add resources to a single node in a system, typically involving the addition
of CPUs or memory to a single computer. To scale horizontally (or scale out) means to add more
nodes to a system, such as adding a new computer to a distributed software application.
2.4.5 Risks and Concerns of Cloud Computing
The adoption of a cloud approach on a traditional and sensitive domain as the industrial domain, raises
some valid concerns. The major concerns about the cloud are:
Security and privacy concerns are the firsts to rise when handing over sensitive control and data
(containing data of customers, consumers and employees, business know-how and intellectual
properties), to a third party (Xu, 2012). Having to share an infrastructure with unknown outside
parties, requires a high level of assurance in the security mechanisms used for logical separation.
One way to handle this concern is by using a proper cloud deplyment architecture, such as hybrid
clouds—with sensitive data kept on-premise—or with a private cloud and with the use of proper
authentication techniques (Combs, Combs) such as secure and encrypted connections (TLS/SSL
with X.509 digital certificates), encrypted data storage techniques (using AES-25big datalitary
encryption), authentication and identity management (using Active Directory/LDAP services), end-
to-end data integrity (using SHA-1, Secure Hash Algorithm), and private retained keys (ensures
that all information requests must involve the owner) (Mohamed, 2012; Archer and Boehm, 2009).
A more in-depth content on security, resides outside the scope of this paper and thus should be
researched in future work.
Network performance are concerns rather important to the industrial sector (Inductive Automation,
2011). Real-time monitoring systems are hard to implement in the cloud due to latency issues.
34
The term latency refers to the time that takes between the interaction and final response. In a local
network, data is constantly flowing back and forth through servers, routers, switches and other
hardware. Moving or accessing data to or from a cloud data center, will involve passing through
the cloud provider network—which it’s up to the cloud provider to decide their speed and quality
of service—that can be overloaded and through extra security layers, i.e. firewalls. The increased
and unpredictable latency and can lead to a very unsatisfactory real-time experience, cause errors
or affect the productivity of production lines.
Reliability it’s also a concern for the industry (Inductive Automation, 2011). Servers can crash, con-
nections can go down, and the more connections there are the more possibilities there are for
disconnections. The more dependent customer critical production processes are from the cloud,
the more dependent the customer is from the cloud Quality of Service (QoS). If something goes
wrong with the cloud system, the customer must wait for the cloud provider IT staff to fix it, and in
meanwhile production processes can stop and cause lost in revenues.
As with any new technology, issues must be addressed. But if the correct service model (IaaS, PaaS,
or SaaS) and the right provider is selected, the payback can far outweigh the risks and challenges. The
cloud performance and ability to scale up or down with much ease, means that companies can react
faster to changes of requirements like never before.
2.4.6 Service-Level Agreements
To protect providers and customers, Service Level Agreement (SLA) contracts are signed, which cap-
tures the agreedment upon guarantees, between the service provider and the customer (Sakr and Liu,
2012). SLAs for cloud services focus on characteristics of the data center and characteristics of the
network to support end-to-end communication. SLA management encompasses the SLA contract def-
inition which includes basic schema with the QoS parameters, SLA negotiation, SLA monitoring, and
SLA enforcement according to defined policies.
2.4.7 Big Data and Real-Time Analytics
Over the last years we have witnessed to an astonishing increase of data produced by social networks,
monitoring and controlling systems, scientific projects, financial transactions, mobile devices, and oth-
ers. This evolution is mostly due to recent technological advances. This incredible growth has affected
businesses in several ways. On the bright side, it made possible to do many things that could not be
done before: identify business trends, explore medical data, understand the universe and so on. On the
other hand, it created a series of new problems such as, short storage and processing capacities and
data security and privacy uncertainties.
The term big data was then coined to define a collection of data so large and complex that it becomes
difficult to process using traditional database management tools and processing applications. This in-
cludes capture, storage, search, sharing, transfer, analysis, and visualization of the data (Mark Beyer,
35
2012). Beyer (Mark Beyer, 2012) characterizes big data as high volume, high velocity, and/or high vari-
ety information assets that require new forms of processing to enable enhanced decision making, insight
discovery and process optimization. Big data is characterized by three metrics (the V’s of big data):
• Volume stands for the size of the data under consideration.
• Variety deals with the number of different data types and sources.
• Velocity refers to the speed of data generation or how fast data is processed.
These are the most accepted metrics by the community, but several other authors extended the previ-
ous metrics with more dimensions (more V’s) like veracity (refers to the messiness or trustworthiness of
the data), variability (stands for the variance of lexical meaning), value (measures the monetary amount
that can be produced out of raw data) and others.
Traditional database systems are one of the systems affected by this evolution. These have been
pushed to the limit and in an increasing number of cases they have failed to cope with this growth (Marz,
Marz). Traditional database systems handle internal and structured data sources, but big data sys-
tems handle unstructured and semi-structured data as well as internal and external data sources. This
is particularly interesting in applications with needs to process data to provide features like real-time
monitoring or real-time analytics.
The Lambda Architecture
A recent development entitled Lambda Architecture (LA), proposes an innovative architecture to provide
real-time analytics in big data systems (Marz, Marz). This architecture style emerged from a need that
the authors felt with previous experiences working with big data systems. The authors felt the need for
robust systems that are fault-tolerant, both against hardware failures and human mistakes, being able
to serve a wide range of workloads and use cases, and in which low-latency reads and updates are
required.
The LA describes a set of principles to enable both batch and real-time or stream data processing in
the cloud. The architecture consists in three layers as follows:
Batch layer processes high volumes of data by collecting, processing, and outputting a group of trans-
actions over a period of time. The storage in this layer is managed by the Apache Hadoop 12 (an
open source platform for storing massive amounts of data). The batch layer stores the master data
set using HDFS and computes the arbitrary results views by performing MapReduce (program-
ming model for processing large data sets with a parallel, distributed algorithms, using Map and
Reduce procedures) operations.
Speed layer processes in real-time views in distributed and fault tolerance stream processing solutions,
such as Storm13. The processing is done automatically every time new data enters the system.
12http://hadoop.apache.org/13http://storm-project.net/ and S414 http://incubator.apache.org/s4/
36
Figure 2.14: Conceptualization of the lambda architecture to achieve real-time analytics composed bya batch and speed layer to respectively perform batch processing and real-time computing, and theserving layer to merge and serve the output of these latter layers.
Serving layer indexes and exposes pre-computed views to be queried in ad hoc with low latency by
Hadoop Query implementations like the Cloudera Impala15.
Service-Oriented Architectures
Service Oriented Architecture (SOA) is a loosely-coupled architectural style that supports Service-
Orientation, i.e. it provides functionality as services to other systems, therefore it’s often used to in-
tegrate legacy and external systems (Microsoft, 2014a; Mora et al., 2012). A service acts as a black box
to the consumer of the service. A service is a self-contained logical representation of an activity that as
a specified outcome.
SOA targets system autonomy (i.e. systems whose functionality is independent from others), inter-
operability (i.e., heterogeneous systems who are capable of sharing information with each other), and
extensibility (i.e., systems which are able to be changed or enhanced with minimal costs). SOA archi-
tecture consists in a service provider, a service requester, a service broker and a service description
or interface, describing architectural principles and patterns. Each service provider publish to a broker
service server a description, also called as contract or interface, which describes their services, capa-
bilities and invocation requirements. This broker server is responsible to manage the list of available
services and to reply to service consumers whenever a service consumer asks about a service, this is
also known as Discovery Service. The service consumers must follow the description of the services to
be able to establish communication and use the services.
This architectural is being widely used in other sectors to interconnect heterogeneous systems,
where each component belongs to different vendors, offering different features by using different system
15http://www.cloudera.com/
37
specifications. The Industrial Automation Systems (IAS) has the exact same environment. Thus, is not
surprising that current literature in industrial automation (Delsing et al., 2011; Karnouskos and Colombo,
2011; Mora et al., 2012) and research projects has the IMC-AESOP look at SOA has the solution for a
more interoperable and efficient manufacturing.
38
Chapter 3
Solution
This thesis intends to provide a sustainable and conceptual architectural solution, common to any cloud-
native industrial EMS. Knowing that there are endless ways of implementing the same idea and knowing
that technologies quickly change over time, the main focus in this thesis are the architectural aspects
of a cloud system as the latter. Thus, business or external technical aspects are lightly depicted and
left for further decision of the organization providing or buying the system. For instance, is up the
organizations to decide which cloud deployment setup (private, public or hybrid cloud) or which business
model (Software as a Service (SaaS), Platform as a Service (PaaS), Infrastructure as a Service (IaaS))
fits their needs the best. Nevertheless, the authors developed a proof of concept to prove the feasibility
of this conceptual solution, using available technology at the moment and implementing only a specific
set of use cases that clearly demonstrate the application of the cloud in this domain.
The architecture here proposed followed a standard software development methodology with the
following stages: scope analysis, requirement analysis, conceptualization, implementation, deployment,
and evaluation. At the same time, a higher level project management activity managed, planed and
controlled its execution.
3.1 Scope Analysis
The industry sector uses complex heterogeneous systems that includes networks, protocols, and au-
tomation systems. Industrial organizations operate on multiple domains at the same time like energy,
quality, and production. Therefore, industrial EMS are more complex than EMS in other sectors, due to
all the constraints and requirements needed to operate in these domains. Even so, as it was described
in this thesis, industrial EMS are not at their full potential because the industry still have needs that
current EMS implementations don’t fulfil.
This chapter proposes a conceptual solution of a cloud-based industrial EMS capable of solving the
needs and challenges that the industrial sector faces today (see Chapter 1).
39
3.2 Requirement Analysis
This section presents a high level analysis of the requirements that this thesis solution needs to address
to achieve the objectives proposed.
3.2.1 Big Data Requirements
Recent advances in control and sensor devices have resulted in the generation of large quantities of
data (the volume of big data). These equipment now produce data with much more detail and in a much
higher frequency (the velocity of big data). Therefore, time series data automatically originated from
thousands of different sensors and customers (the variety of big data), presents a technical challenge
in terms of collecting, storing, and processing these data in a real-time basis, and at the same time
produce meaningful knowledge. Our solution proposal applies big data techniques, like batch and real-
time computation, to provide real-time analytics on large quantities of incoming energy and machine
data streams.
3.2.2 Real-Time Requirements
Various domains require real-time data processing for faster decision making: credit card fraud analytics,
network fault prediction from sensor data, security threat prediction, and others.
As it was described before, monitoring in today’s energy management system is usually performed
per minute or every 15 minutes. We proposed the introduction of real-time in the monitoring aspect of
these systems to enable organizations to react faster to quick changes of events and improve decision-
making with actions based on real-time data.
The concept of real-time has sometimes different definitions for different people. Therefore, it is
important to define the real-time requirement that this thesis discusses before any further development.
Whenever the term of real-time computing is used it means that the system guarantees that an event can
be completed computed in a short amount of time. In the case of this thesis, the real-time requirement is
in fact near real-time. This refers to the time delay introduced by automated data processing and network
transmission. Therefore, the throughput real-time time requirement is in terms of a few seconds, i.e. an
interval of 1 to 10 seconds maximum.
3.2.3 Functionalities
The functionalities proposed for a cloud-native industrial EMS are based on all the research background
described before and on ABB’s expertise. These represent the fundamental functionalities necessary to
fulfil the needs and challenges that the industry is facing and that are depicted in this thesis. Thus, not
every possible feature is here described but these are the ones that the authors find essential:
40
Energy Monitoring
• Real-time monitoring of energy information such as consumption, efficiency, intensity, quality,
cost and others, integrated with production data, to enable a more energy-aware manufactur-
ing and facilitate quicker more informed decisions.
• Real-time monitoring of energy consumption of multiple sites and meters to dynamically as-
sess the viability of participating in DR initiatives.
• Alarm and events notifications management to keep managers up-to-date with the latest
events.
Energy Analytics
• Energy efficiency benchmarking across entities such as metering and production equipment
(sensors, actuators, motors, machines, etc.), areas (manufacturing floors, departments, pro-
duction sites, etc.), and production processes to enhance energy efficiency.
• Energy efficiency KPI evaluation to analyze the success or failure of on-going energy and
manufacturing strategies.
• Cost allocation to characterize energy costs by entity (equipment, production processes, ar-
eas, etc.) in order to provide transparency and identification of energy costs savings oppor-
tunities.
• Pattern analytics and forecasting to be able to identify and predict anomalies and demand
peaks.
Data integration
• Integration of automation data from external system IAS to derive more knowledge and
achieve a more energy-aware production.
• Integration of external data sources as weather, day-ahead and real-time energy price mar-
kets and others.
3.2.4 Quality Attributes
The solution proposed by this thesis allied with cloud computing will provide the following quality at-
tributes:
System qualities
• Availability and reliability - Any system component failure is recovered by a redundant
copy. The complete cloud system can also be redundant by deploying it in multiple avail-
ability zones.
41
• Modifiability and portability – The architecture proposed offers low costs of changes by
providing a separation of concerns and coarse-grained modules through multiple decouple
components. Based on SOA, this architecture also enables system modifiability by separating
services interfaces and services implementations.
• Performance – With virtually infinite and redundant computation resources, the architecture
is able to quickly respond to requests and to enhance some types of mathematical algorithms.
• Security – The software and system architecture proposed enables security measures from
end-to-end such as built-in firewalls and secure connections.
• Interoperability - The adoption of communication standards enables communication be-
tween heterogeneous systems.
Business qualities
• Time-to-benefit and Costs – A cloud solution running and available benefits the customer
with quick deployment and cost savings, because it avoids acquiring, deploying and maintain
the necessary IT infrastructure and staff and paying only for the service in use. From the
vendor point-of-view, with a multi-tenant architecture in place, the solution is ready to handle
more customers without any architectural or system change, resulting in recurrent revenues
over time.
• Targeted market – This adaptable, scalable, and modular solution is target to manufacturing
companies of any kind and size.
• Integration – Data is centralized in the cloud, thus the system is able to integrate all kinds of
data and to perform data mining 24/7. In addiction, with its web services online, it’s able to
integrate external data into the system at any time. In opposite of EMS on-premises, this cloud
solution enables the vendor to evaluate the usability and performance of the solution over
time, making changes in system if necessary and to correlate data from multiple customers if
needed.
3.2.5 Use Case Diagrams
A set of use case diagrams were developed to portray the different types of users and the various
ways they can interact with the system. These provide a simplified and graphical higher-level view of
the capabilities of the system. Due to their simplistic nature, these are an ideal communication tool to
explain the solution here proposed.
These use case diagrams convey the requirements depicted before, in form of use cases. A use
case is represented by a circular form and addresses a set of requirements. Each use case perform a
list of steps that typically require an action from an actor or a system (represented by a human icon) to
achieve a certain goal. For the more complex use case diagrams, like the Energy Monitoring, Energy
Analytics, and Data Collection use case diagram, they are described in sub-use case diagrams.
42
Figure 3.1: Use case diagram with a high-level view of the Energy Cloud system that describes its maincapabilities and actors.
Figure 3.2: Use case diagram that describes the Energy Monitoring use case and depicts its variousmonitoring capabilities and the actors that interact with this part of the system.
43
Figure 3.3: Use case diagram that describes the Energy Analytics use case and depicts its variousanalytic functionalities and the actors that interact with this part of the system.
Figure 3.4: Use case diagram that describes the Data Collection use case and depicts how the systemgathers, deals, and stores data originated from the different systems.
44
3.3 Conceptualization
The architecture proposed is based on Marz and Warren lambda architecture (Marz, Marz) and Gold-
schmidt et al. time-series cloud architecture for industrial processes (Goldschmidt et al., 2014). Hence,
it uses two parallel layers that process continuous streams of time series data from energy metering
devices at different speeds (see Figure 3.5).
The system also responds to requests from users through Dashboards. These requests are then
processed and treated accordingly. For example, the execution of on-demand jobs or historical data
visualization, respectively uses the Analytics API to trigger batch processing jobs in the Energy Analytics
layer, or to request historical data from the Timeseries DB API. The point of input is the Data Collector
(DC) component. Data is pushed from the multiple sites to this component at any frequency possible
(e.g. one metering per minute or per second). Then, the DC forwards these data to the system Message-
Oriented Middleware (MOM) to be treated by the following layers:
3.3.1 Energy Monitoring (Real-time computing)
This high-speed pipeline layer processes all incoming time-series data, applies simple data transforma-
tions, and outputs the processed data without storing it. The main concern of this layer is to transfer
data from one end of the system (sites) to the other (desktop and mobile clients) as fast as possible, so
that users can react faster to quick changes of events.
Its Data Consumers are constantly listening and consuming incoming data. They then push these
data to other stage where they are converted, normalized, and sampled in configurable very short time
windows (e.g. 1 or 5 seconds) that keeps only the minimum and maximum values measured inside the
window. This way, we prevent data floods from meters that are incorrectly configured and are measur-
ing or pushing data at a frequency higher then accepted or even necessary. Thus, only the essential
data packages navigate through the layer, keeping performance under control. In the end, it calculates
the top energy usage consumers per site and transfers these results and all the individual energy me-
tering sampled values to the Real-Time Messaging Servers, so that they can be propagated to all the
connected desktop and mobile clients through WebSockets.
3.3.2 Energy Analytics (Batch processing)
This slow-speed layer also processes all incoming data. However, it autonomously stores and analyzes
these data using a series of pre-defined programs (“jobs”). Its concern is to compute complex algorithms
that extrapolate knowledge from all available data sources, also known as Knowledge Discovery in
Databases (KDD). Its objective is to produce valuable knowledge that is hard to unveil. This is made
possible because data are fed in parallel to several cloud computing clusters that perform different data
mining jobs. Hence, this layer takes more time to complete due to the complexity of the algorithms in
use. The partial outputs of stored in a Distributed Historical Storage and re-used in further computations.
45
Cloud Cluster
Clients
Web Browser
Real-Time Messaging
Servers
...
Dashboards
Industrial Sites
Cloud Connector
Energy Meters
Energy Monitoring (Storm Cluster)
Filtering, Normalization
& Sampling
Data Aggregation &
KPI Engine
Energy Efficiency
Benchmarking
Abnormal Energy Usage
Detection JobsData
Consumers
Message-orientedMiddlew
are
Distributed Historical Storage
Energy Analytics (Hadoop Cluster)
Top Energy Consumers
Timeseries DB
Data Collector
Batch Processing
Cluster
Knowledge Discovery
Cluster
Pattern Recognition
Cluster
Analytics API
R
R
R
Industrial Sites
Energy Meters
Cloud Connector
R
R
CassandraHBase
WebSocketsHTTPS
OPC UA
Sites Simulator
KairosDB Spring Freeboard.IO Node.JS
Apache Storm
SparkMahout
RabbitMQ
Spring
R
LegendR
Req./Response comm. channel
Bi-directionalcomm. channel
System Component
Technology Used
Storage Component
Read/Write Access
Figure 3.5: Conceptual architecture proposed for cloud-native industrial EMS with the Energy Monitoringlayer computing streams of data in real-time and calculating the top consumers per site (right side). TheEnergy Analytics batch processing layer computes the same data but derives knowledge autonomouslythrough a set of pre-defined algorithms and archives the data in full detail (center). Various other com-ponents enable interaction in real-time or at demand with users or external systems (top area) or withthe multiple sites and meters of the organization (bottom area).
3.4 Energy Cloud
This thesis intends to provide a sustainable and conceptual architectural solution common to any cloud-
native industrial EMS. Nevertheless, to prove its feasibility and claims, the authors developed a proof of
concept, entitled Energy Cloud. Running entirely in the cloud, this implementation is a scalable and real-
time industrial energy management system capable of monitoring (Energy Monitoring) and analyzing
(Energy Analytics) time series energy metering data from external or simulated energy meters. The
following sections depict the technologies and implementations for each architectural component and
how they work together as a unified solution.
46
3.4.1 Dashboards
The interaction between the user and the system is made through a presentation tier called Dashboards.
Its concern is to display the fleet of sites’ energy status and to provide data exploratory functionalities
(see Figure 3.9). This essentially is a collection of web pages that use JavaScript libraries like the Free-
board.io (an open source library to create real-time IoT dashboards), to visualize data in real-time with
lines (to display the variation of energy consumption over time), pies (to depict the ratio of consumption
per site), and columns (to have a side-by-side comparison of energy usage per site). These pages are
served by a cluster of web servers.
The dashboard page comes with a fixed widget that shows an overview of the energy consumption
per site and which are the top energy consumers. The user can then dynamically customize the rest of
the screen with other widgets. Each widget subscribes to a set of independent data sources and react
to incoming data by updating its visualization in real-time.
For this project, additional widgets were developed apart from the ones that come with the library
itself. In particular, a line chart widget was developed to display time-series data across time using the
Highcharts charting library. This widget is especially useful to visualize and compare energy load profile
curves of individual equipment or sites or the evolution of certain KPI, over a period of time. An additional
pie chart widget was developed to display the ratios of energy consumption within one site.
In the Freeboard.io library, visualization widgets are decoupled from data sources, meaning that
the same widget can display data from different types of data sources, e.g. JSON files, WebSockets,
WebServices, third-party messaging systems, etc. In this prototype, out of the box JSON data sources
plugins were used to collect data from external systems such as energy exchange markets, to obtain
one-day-ahead energy prices, production systems, and weather temperature web services. These could
also be used to autonomously pull data (such as computed energy KPI or other variables) from the
Analytics API with a certain refresh rate. A custom WebSockets data source plugin was developed to
be able to receive data from the internal Real-Time Messaging Servers. These were then bound to the
visualization widgets that display real-time data.
The analytics page allows the user to interact with the Analytics API to pull and visualize historical
data. In this page the user specifies the time range to investigate and which metrics to query. In this im-
plementation, the metrics available are energy consumptions measurements and previously calculated
KPI. Once the user confirms its query, the page sends an asynchronous request to the Analytics API to
deal with this call (see Figure 3.7).
3.4.2 Analytics API
The Analytics API is the interface between the Dashboard or any external systems, and the systems
components regarding the Energy Analytics layer. Decoupling the Dashboard from internal components
brings many benefits.
The first one is security. This intermediate layer acts as a black box, i.e. the user doesn’t know how
the system works internally and doesn’t have any direct access to internal systems.
47
Figure 3.6: Energy monitoring dashboard where the user can visualize current energy usage consump-tions of multiple sites and equipment with line charts and gauges, and monitor the most significantenergy consumers with pie charts and columns.
Secondly, by abstracting data from presentation, we enable different HMI, apart from the Dashboard,
to be further developed such as mobile and desktop applications. These all obtain the same data but
present it in different ways.
It also allows the addition of extra logic per operation. For example, a simple request to query for
existing metrics in the system can perform additional actions like sorting and filtering. Another example
is the when requests require the use of transactions, i.e. complex operations that follow a series of steps
and that rolls-back in case some step fails.
Furthermore, it creates interoperability, by enabling the exchange of information with external sys-
tems that can use these data into their processes.
On the other side, this extra layer creates additional round trip time to get data from the persis-
tent data stores. Although, this time is so short due to the speed of today’s internet connections, that
becomes irrelevant. As for the additional time that takes to process each request, it depends on the
specifications of the machines running the API and on the implementation of each operation. Another
point to take in consideration is that this one entering point to the system might become an attractive
point of security attacks. Therefore, proper security and authentication measurements must be used.
In this proof of concept, the authors used the Spring Web Framework, to create the Analytics API
and it offers a set of operations to pull historical energy consumption measurements and KPI from
the Timeseries DB. It also offers options to execute on-demand Energy Analytics computation jobs.
Background threaded daemons hold on to these requests and reply to them as soon as the operations
are completed.
48
Figure 3.7: Energy analytics dashboard where the user can query, visualize, and correlate the evolutionof historical energy data or previously calculated KPI.
3.4.3 Message-oriented Middleware (MOM)
The message-oriented middleware (MOM) is the cross-layer component that enables communication of
messages between several distributed systems. In this case, there are many open source options to
choose from. But since the AMQP industry standard protocol was needed to support a wider variety
of developer platforms, the authors chose RabbitMQ because it is a robust, yet easy to use and deploy
queue messaging system. A collection of exchanges and queues were created to queue all input data
that comes from the Data Collector to be consumed by both computing layers, and all output data from
the Energy Monitoring layer, to be consumed by the Real-Time Messaging Servers.
3.5 Energy Monitoring
The concern of the Energy Monitoring layer is to compute streams of data containing timestamped en-
ergy data in real-time. To implement this requirement, the authors needed an open source, distributed
and scalable real-time processing system capable of computing large amounts of data. Storm, a dis-
tributed real-time computation system, designed to be scalable, fault tolerant (at-least-once or exactly-
once) and programming language agnostic, seemed like the perfected choice. Storm makes it easy to
reliably process unbounded streams of data, doing for real-time processing what Hadoop does for batch
processing. However, in Storm, topologies run forever or until explicitly killed or un-deployed.
49
Spout
Data Source
Spout
Bolt
Bolt
Bolt
Bolt
Data Source
Bolt
Output System
Figure 3.8: Conceptualization of a Storm topology with various Spouts generating streams of data fromsome data source and Bolts processing the data from these Spouts or other Bolts.
3.5.1 Storm Topologies
In Storm, the structure of a distributed computation is referred to as a Topology. These are graphs of
stream computation where each node is either a spout or a bolt. Spouts essentially produce/fed data
in form of tuples (ordered list of elements) into the topology. These can read data from HTTP streams,
databases, files, message queues, etc. For example, a spout may connect to the Twitter API and emit a
stream of tweets.
A bolt is a component that performs stream transformations or operations. Bolts may subscribe to
any number of streams emitted by other spouts and bolts and produce new streams. Therefore, it is
possible to create complex network of stream transformations. Typical operations that bolts perform
include: filtering, joining, and aggregating tuples, calculations, and external database reads and writes.
3.5.2 Storm Architecture
Storm clusters are composed by two kinds of nodes: master and worker nodes. The master node runs
a daemon called Nimbus (master node), responsible for distributing code around the cluster, assigning
tasks to machines, and monitoring for failures.
Inside the worker nodes, there are several processes running: a single Supervisor process (slave
node) and multiple Workers processes, i.e. Java Virtual Machines (JVM) processes. The Supervisor
starts and stops worker processes as necessary based on what Nimbus has assigned to it. Each
worker process executes a subset of a certain topology. There may exist various topologies running
using worker processes spread across many machines.
50
Worker (JVM)
Executor (Thread)
Tasks
(spout/bolt)
Executor (Thread)
Task(bolt/spout
instance)
Nimbus
Zookeeper(backup)
Zookeeper(elected leader)
Zookeeper (backup)
Supervisor
Supervisor
Supervisor
Supervisor
Supervisor
Worker (JVM)
Executor (Thread)
Tasks
(spout/bolt)
Executor (Thread)
Task(bolt/spout
instance)
Worker (JVM)
Executor (Thread)
Tasks
(spout/bolt)
Executor (Thread)
Task(bolt/spout
instance)
Worker (JVM)
Executor (Thread)
Tasks
(spout/bolt)
Executor (Thread)
Task(bolt/spout
instance)
Figure 3.9: Conceptualization of the software and hardware architecture of a Storm cluster deployment.The Zookeeper nodes do not belong to Storm but they are used to perform coordination and machinediscovery.
Executors are threads that are spawned by worker process and runs within the worker’s JVM. An
executor may run serially one or more tasks for the same component (spout or bolt).
Tasks are bolt or spout instances and they perform the actual data processing. The number of tasks
for a component (bolt or spout) is always the same throughout the lifetime of a topology, but the number
of executors (threads) for a component can change over time and thus scaling the system.
3.5.3 Storm Clusters
Storm relies on the ZooKeeper system to coordinate the Nimbus process and its Worker processes.
By using ZooKeeper, Nimbus will automatically take care of discovering and integrating new Supervisor
nodes into the cluster without any interaction from the user.
Additionally, both Nimbus and Supervisor daemons are fail-fast and stateless, because their state
is kept in Zookeeper or on local disk. The result is stable clusters with great performance. In a recent
benchmark, a Storm topology clocked one million 100 byte messages per second per node (2xIntel
E5645 2.4Ghz processors and 24Gb memory).
51
3.5.4 Energy Storm
The authors developed a custom topology (a graph of computation nodes) to perform the necessary real-
time computation on streams of data, entitled Energy Storm. The first nodes of this topology, consume
data in parallel from the MOM, using the AMQP protocol in a round-robin way due to the specifications
of the binding established with RabbitMQ. These are then converted from its raw data type (JSON, CSV,
text, etc.) to internal data representation (Java objects).
These individual messages are then shuffle by their unique ids and transfer to the following node
responsible to sample that individual meter. Hence, data originating from the one meter goes always to
the same sampling node so that it can then have all the necessary data to sample.
Nevertheless, each node can handle data from meters from different sites and thus load balancing
incoming data. For the sake of simplicity, these time-windows are essential classes that hold only the
minimum and maximum values read. Storm’s supervision mechanism then orders these classes to
dump these data to the next nodes after a configurable time. Finally, the following nodes transfer these
sampled energy measurement data to the MOM.
At the same time, a set of parallel nodes also receives these data and calculates the top energy
consumers per site. These nodes hold these computed data in-memory according to a configurable
Time-To-Live (TTL) variable. This enables further calculations until a new measurement arrives and it
also supports brief periods of missing data. Like the time-window sampling nodes, Storm orders these
nodes to continuously transfer these calculations to the MOM for further distribution.
3.5.5 Real-Time Messaging Servers
Real-time messaging servers are network servers that consume data from the MOM and distribute them
to any subscribed client in real-time through WebSockets. The authors used Node.js to manage all the
WebSockets connections and to push in real-time any incoming data from the MOM queues to every
client listening.
3.6 Energy Analytics
The batch processing cluster performs complex computations. There are endless opportunities here and
much research has been made in the last years in the fields of data mining, machine learning, pattern
recognition and others. These clusters might even be based on active on premises systems that perform
the same computations. In that case, they must be migrated to reliable cloud computation systems that
could scale and process incoming data in batches against historical data, such as the Apache Hadoop.
3.6.1 Hadoop Cluster
In this proof of concept, a simple use case was chosen to demonstrate the inherent possibilities. Based
on the work of Anna Koufakou et al. (Koufakou et al., 2008), the authors developed a simple outlier
52
Legend
Energy Monitoring (Storm Topology)
Raw Input Data Queue
Sampling(Max/Min. Windows)
Sampling Archiving(Java to JSON)
Top Consumers(Expiring Windows)
Output DataQueues
Spout: a source of data stream
Bolt: processing unit
Raw Data
Computed Data
Data CollectionConversion
(JSON to Java)
Figure 3.10: Conceptualization of the Storm topology developed in the Energy Cloud project withstreams of data coming from the RabbitMQ queue, being processed through several stages by a setof distributed bolts that output data to various other queues.
detection system (MR-AVF) to identify energy consumption peaks per site. The user is then aware of
significant energy usage that might be inefficient or unexpected, and later lead to higher energy costs,
or worst, result in penalties from the energy providers from crossing a contracted limit. Therefore, it is of
the utmost importance to keep these peaks under control.
The MR-AVF is based on the Map Reduce paradigm for parallel programming. It provides high-speed
and scalable outlier detection by analyzing each individual point and categorizing it by the average rate
of occurrence. The more infrequent or irregular a value is, the more likely it is to be an outlier. This
rather simple approach is easy to parallel and implement. Even though, it is faster and sometimes more
efficient than other more complex calculations.
3.6.2 Timeseries DB and Distributed Historical Storage
Data from the computation layer are persistent stored in a distributed storage. Following the lambda
architecture principals, every detailed raw data is time-base stored and immutable. The authors used
KairosDB, a fast distributed scalable time series database, that works on top of Cassandra, an industry
standard for distributed NoSQL databases. KairosDB provides us with an easy to use abstraction layer
(API) to push and pull time series data from the Cassandra cluster.
53
This setup was proved to be very robust, scalable, and suitable for industrial processes by Gold-
schmidt et al. (Goldschmidt et al., 2014). A cluster of 24 KairosDB nodes could handle the workload of
a large city (6 million smart meters).
3.7 Deployment
This proof of concept was deployed on ABB’s development cluster that runs Openstack, an open source
software for building private and public clouds. All the components described before, except the Storm
cluster, were deployed on top of Cloud Foundry, a PaaS system that made the deployment of the devel-
oped applications in Spring and Node.js, and services, such as the KairosDB, RabbitMQ and Cassandra,
much easier. Furthermore, it provided us with easy to use commands to vertically and horizontally scale
all the components independently.
54
Chapter 4
Evaluation
The major contribution of this thesis is a conceptual solution for modern cloud-native industrial EMS. It
intends to solve the needs and challenges that the industry is facing with current EMS implementations
and provide extra capabilities from the use of the cloud.
The evaluation of this proposal is based on three criteria: feasibility, relevancy, and performance.
Its feasibility is proved through the implementation of a proof of concept. It shows that the envisioned
solution can indeed be implemented using today’s technologies. Although it’s not a full product, with
every possible aspect implemented, this proof of concept developed by the authors implements the
necessary components to prove this thesis proposal claims.
The second criteria validates this solution proposal relevance. Here, an evaluation framework eval-
uates the importance of this work and how it contributes to the community, by testing it against a set of
important use cases and decisive questions.
Finally, the third part validates its performance. In this part the authors submit the proof of concept
that sustain this proposal, under a series of benchmarking tests to assess its capabilities under different
workloads. It is essential that this solution continues to work properly under any kind of circumstances,
to prove its reliability, robustness, and scalability.
4.1 Conceptual Evaluation
Evaluating this thesis solution proposal from a conceptual point of view, is important to asses if the
decisions taken and if the proposals here proclaimed, are in fact relevant to the community and valid to
solve the objectives discussed.
To accomplish this evaluation the authors gathered a set of use cases and analyzed how this proposal
approaches each of them.
4.1.1 Use Cases
To prove the application and benefits of the cloud in current EMS, the authors focused on five use cases,
extrapolated from recent literature and industry experts’ expertise, that represent innovative functional-
55
KPI Indicator Focus Description
Power Energy consumption Instantaneous or average power used by a process
Energy consumption Energy consumption Energy input into a process during a defined timeperiod
Production energyconsumption
Energy consumption Energy consumption per manufactured product(items or units)
Energy costs Energy costs Monetary cost of energy used including fixed andvariable components
Production energycosts
Energy costs Energy costs per manufactured product
Energy losses Energy efficiency Energy use associated with non-value adding pro-cess steps or operating states
Energy efficiency Energy efficiency Ratio between the total energy used and the oneused only in production
Table 4.1: Summary of the energy key performance indicators (KPI) that the industry is in need of andthat were implemented in the Energy Cloud project (adapted from (Vikhorev et al., 2013)).
ities that the industry lacks or that could not be easily obtained in a multi-site management without the
cloud.
UC1 - Monitor the most significant energy consumers
Monitoring is necessary to derive knowledge of current energy use. Especially the most significant en-
ergy consumers need to be autonomously identified, monitored and analyzed in real-time, to facilitate
the use of the system and increase industrial energy efficiency, e.g. supporting the judgment as to
whether anticipated energy savings such as DR could be achieved or not, the scheduling of tasks to
avoid peaks loads, and as to take advantage of on-site power generation (Bunse et al., 2011). This re-
quires standardization of data collection, on-line data processing and visualization techniques (Vikhorev
et al., 2013).
UC2 - Calculation of energy performance indicators (KPI)
The system must calculate energy related KPI to enable actors within an organization to react to negative
developments. Nowadays, there is a need for effective energy efficiency KPI to track the changes and
improvements on both process and on plant level (Bunse et al., 2011) (see Table 4.1).
UC3 - Historical data visualization and correlation
Besides monitoring, the system must also provide targeting. With targeting one compares current en-
ergy consumption behaviours of sites or individual equipments with a set of targets, in order to identify
management priorities for action, e.g. a certain percentage reduction over a given period. The en-
ergy consumption behaviours usually known as load curves or power profiles, should be analyzed using
different timescales to develop holistic energy efficiency strategies. For example, at plant level such
56
analysis may be useful to identify peak loads that would attract a surcharge from the energy utility com-
pany (Vikhorev et al., 2013).
UC4 - Pattern matching and data usage analysis
The system must derive real-time intelligence from data streams to enable low-latency decisions in
response to changing conditions. Using pattern matching techniques the system must be able to unveil
patterns in a data stream of events, such as abnormal peaks and troughs, deviation of energy use
from reference operating state, which can indicate a malfunction of the equipment, and identify different
stages of production to quantify idle time and others (Vikhorev et al., 2013).
UC5 - Benchmark energy usage and efficiency
Benchmarks for similar equipment should be facilitated. Benchmarks should be available, stating where
other sites or even other companies with the same challenges stand, in order to increase energy effi-
ciency with the same process quality (Bunse et al., 2011). Due to the complexity involved with this use
case and time restrictions, this task was left as further work.
4.1.2 Results and Discussion
After gathering the use cases necessary to conceptually evaluate this proposal, the authors together with
ABB’s industry experts, evaluated how this proposal address each of one by to answering the following
questions:
• Q1 - What is the value added by the use case?
• Q2 - How easy is it to obtain the same results without the use of the cloud?
• Q3 - Which aspects of the cloud does this use case highlight?
• Q4 - How is this use case implemented?
The answers to these questions can be found in Table 4.2. These results show that the solution
proposed positively supports and addresses every depicted use case and provide additional value that
could not be obtained before or that could not be obtained with such ease. Therefore, we conclude that
this proposal is in fact an innovative, beneficial, and relevant solution that could enhance the processes
of actual organizations in many ways.
4.2 Performance Evaluation
The industrial sector is a very sensitive domain regarding the performance of every system in use. In-
dustrial processes operate in a supply chain system with every system affecting the following one across
all levels of the organization, from production, to distribution and management. Systems are more and
57
more connected and dependent of each other and their performance directly affects the ongoing busi-
ness processes. Therefore, it is important that these systems keep running and successfully operating,
so that these processes are not troubled.
The energy management is an important part because it directly affects production and quality. Es-
pecially if the production processes are highly dependent on energy. If this is the case, in the worst
scenario a failure on the EMS system can dictate the interruption of a production process. This can
have serious impacts in the organization.
Thereby, the solution here proposed has to comply with these requirements and prove to be robust
and reliable to operate in such environment. In addition, this solution might be deployed in a multi-tenant
deployment, i.e. it may be the case that an organization implementing and deploying this proposal runs
the system to cope with multiple customers in the same cluster. It can also be the case that only one
customer operates in a single cluster. In any case, the system must be ready to scale to the number of
meters managed because the number of customers or the number of meters that one customer have
can increase spontaneously.
The architectural decisions that lead to the solution proposed in this thesis, intent to address these
performance requirements. To evaluate its performance means evaluating the proof of concept de-
veloped, i.e. if this proof of concept copes with these requirements then it shows that it is a correct
implementation of the conceptual solution and that the conceptual solution is in fact valid for the indus-
trial sector. Thus, the authors developed a series of test cases that put the prototype under different
workload situations and evaluated its performance according to a set of metrics.
4.2.1 Data Sets
Energy and automation data is the fundamental input to the proposal solution detailed in this thesis.
Hence, to be able to evaluate its proof of concept, energy and automation data originated from metering
equipment has to be provided.
These data can be simulated, but it has to follow the characteristics of real data to be accepted as
valid. Taking that in consideration, the authors conceived a data set based on typical industrial energy
load curves and use it to evaluate the proof of concept. This data set was also essential during the
development of the latter, to test the system during the different stages of development. Nevertheless,
real data provided by ABB was also used in this prototype.
Simulated Data
Simulating the operation of industrial organizations involved the creation of a dataset with fictional data
with sites, meters and energy curves. The data regarding the hierarchical logic between sites and meters
was provided by an ABB’s WebService. From all the information provided by this service, only the data
regarding the unique identification of each site and meter was used.
On the other hand, the energy curves were manually created by the authors and were based on
typical industrial energy consumers, e.g. industrial ovens with sudden peaks of energy consumption
58
and heating systems with a peak during the hottest part of the day. To create these curves, the authors
adapted the use of LIMBO, an eclipse-based tool for modeling load variations curves (von Kistowski
et al., 2014). The main purpose of this tool is to generate curves to be benchmark s distributed systems
by recreating the variation of number of requests of multiple clients. The authors then adapted these
curves to instead of representing the number of requests per minute, they would in fact represent energy
load curves. Each curve generated contained a list of 1440 Cartesian points, i.e. one point per minute
during a full day, with an incremental delta in the x-coordinate and a generated energy value in the
y-coordinate, e.g. [(0, 10),(1, 30)...(1440, 23)].
For this data set we created a collection of 100 sites with at least two meters per site: one meter
simulating the operation of an oven and another simulating a heating system. Each simulator instance
represented one customer.
Real Data
Evaluating the performance of the proof of concept when dealing with real data was also took into
consideration. To perform this evaluation, ABB provided a data set with real energy data from one of
their sites, measured during a period in time. In fact, the data contained in these data did not differ much
from the simulated data set. The main difference was that the amount of energy measurements, i.e. the
amount of points per energy curve, was much higher.
4.2.2 Virtual Energy Cloud
To simulate the operation of energy meters in multiple industrial sites, the authors developed a software
tool to virtually generate time-series data streams to emulate physical energy meters, entitled Virtual
Energy Cloud. This multi-thread application enabled the concurrent operation of multiple virtual energy
meters in one machine. This was fundamental to dynamically change the number of active sites and
meters and evaluate the solution proposed easily.
In this case, the authors developed a Java application for the desktop, in a stacked three-layered
software architecture. This application proves the cloud can also be used as a remote service from
the desktop. The three-layer model is a software architecture pattern focused to enterprise business
applications with complex system and communication requirements (Microsoft, 2014b). It consists of
coarse-grained modules in a unidirectional relation with each other that allows that any layer to be
changed independently. Hence, this design pattern promotes modifiability, portability, and code reuse.
Three-layer architecture has the following layers:
• Presentation layer is the topmost level of the application. It communicates with other layers by
which it puts out the results to the front-end application layer.
• Business layer contains the business logic that controls the functionality of the application.
• Domain layer consists of database servers and services gateways that store and retrieve data.
This layer keeps data independent from application servers or business logic.
59
Figure 4.1: User interface of the Virtual Energy Cloud simulator to visualize the active sites and meters(left side) and the data being generated by the active threads (right side).
This architecture is a big advantaged in fields where requirements and features changes are con-
stant. In this case, it proved to be particularly beneficial during the development of this thesis’ proof of
concept, because it enabled the experimentation of different techniques in each layer without affecting
the others.
This simulator was created using the Spring Framework, an open source application development
framework and inversion of control container for the Java platform. The Spring Framework core features
provide popular functionalities that are common to most of the Java applications. Thus, it facilitated the
development of this project.
Domain Layer
The domain layer was implemented with Hibernate, an open source Java persistence framework project.
This Object-Relational Mapping (ORM) library was important to abstract the databases used by the
simulator from its internal object representation. The library takes care of the mapping Java classes to
database tables and provides data query and retrieval operations. Hence, in case the database system
used, changes there is no need to change the simulator code. Internally, data and tables are treated as
normal Java objects. The final set of classes used to represent the domain entities used by the simulator
tool consisted of (see Figure A.1):
• Metering class to represent each individual load curve points.
• LoadCurve class to represent kinds of energy curves, each with a collection of a Metering objects.
60
• MeterType class to represent different types of metering equipment.
• Meter class to represent energy meters, each with its LoadCurve association to characterize the
curve measured by the meter and an association with MeterType to depict its type.
• SiteMeters class to represent the different sites, each with a collection of Meter objects.
The conclusive data storage configuration used under Hibernate to persistently store the simulated
dataset involved a WebService, a MySQL server, a H2 database, and Ehcache. The WebService was a
common service used by ABB to provide data about sites and equipment data. Several different ABB’s
projects use this service to not re-implement and re-create these data in their systems. The MySQL
server, an open-source relational database management system, was running in a local machine and
stored all the points from all the energy curves. This allowed the authors to run multiple simulators simul-
taneously in different machines but sharing the same data from the same MySQL server. Each instance
running the simulator had its own H2 database, an open-source in-memory relational database man-
agement system, and Ehcache, an open-source memory cache system. The H2 database is created
and loaded on start-up with data from the MySQL server.
Without this local database in each simulator, the MySQL server would be a bottleneck for simulta-
neous simulators running, because theirs request to load the energy curves would take so much time
that would hang any other concurrent request. This way data is once pushed from the server and once
all the simulators are loaded their operation can resume without affecting others. The EhCache system
was used to boost the performance by caching on-memory common energy curves, i.e. collection of
Java objects, used by multiple meters. Thereby, all the data needed was loaded to memory. Thus, the
execution of these simulators was very fast, with data being generated and published by each meter up
to a rates of a couple of hundred milliseconds if needed.
Business Layer
The business layer of the Virtual Energy Cloud project performs the simulating operations based on
the data provided by the domain layer and also exchanges data with the upper layers through services.
By separating the implementation and logic from the other layers enabled us to change these inner
operations without affecting the surrounding layers.
Hence, the presentation layer is not dependent on the service layer and the only way they communi-
cate it is through the use of events that are exchanged among them. Therefore, the only thing that the
methods in the presentation layer need to know are the specifications of the events that the service layer
expects.
In this application, the business layer is divided in two parallels layers with distinct concerns. The
service layer is responsible to intermediate data in form of objects with the upper layers, e.g. get the
list of available sites and meters, get the energy curve of one meter, and others. The gateway layer
however, provides mechanism to publish data out of the simulation tool.
Internally, the gateway layer is composed by classes that manage the active sites. The ActiveSites-
Manager class keeps track of all the actives sites with internal Java concurrent HashMap collections.
61
Each entry is identified by the site id and contains an ActiveSite object. The latter represents an active
site and its concern is to keep track of all the active meters of the site. Like the previous class, the
ActiveSite class also uses Java concurrent HashMaps to manage the collection of active meters (see
Figure A.2):
The active meters are in fact threads. Once a thread is activated, it starts publishing data at the
defined rate to the RabbitMQ system. As soon as an event to disable a certain meter arrives to this
layer, the ActiveSitesManager catches the event and replicates the order to the corresponding ActiveSite
instance which kills the running thread.
Presentation Layer
During the development of the proof of concept there was a need to enhance this simulator and change
its output from log files to a user interface, to facilitate the debug and testing of the system. It was hard
to keep track of all the generated data of each meter with so many outputs at the same time. Therefore,
a user interface was created to provide a visual representation of the energy consumed in each active
site. This way it was easy to visualize the past, present and future energy measurements generated and
compare it with the data that the proof of concept system had in storage and in its real-time monitoring
dashboards (see Figure 4.1).
In a secondary area of the interface, one can configure the frequency of the messages per minute that
are being published. Additionally, two tables show all the active sites and their total energy consumptions
and a list of all the active meters ordered by consumption, to quickly identify the top consumers. In top
of that, these tables are updated as soon as the measurements are published. Thereby, this view
concentrates all the necessary data to quickly evaluate the real-time aspect of the proposed system in
evaluation (see Figure 4.2).
4.2.3 Test Cases
Evaluating the performance of the proof of concept involved submitting the proof of concept under dif-
ferent kinds of conditions and evaluates how it performs. To do so, the authors created test cases where
the ability of the system to deal with its workload is tested under different configurations.
These tests had multiple variables changing over time and different metrics to assess its perfor-
mance. The goal of these tests is to understand if the system sustains its claims.
Variables
In a distributed system like the one here depicted, the number of nodes is one of the variables that can
influence its performance. Most of the distributed systems have higher throughput and have far more
capabilities to deal with bigger workloads when the number of nodes increases. Hence, this is one of
the important variables to take in consideration.
The next variable is the number of active meters. Without knowing the architecture in use, one could
argue that the number of connections could be limited or that could in fact have some impact in the link
62
Figure 4.2: User interface of the Virtual Energy Cloud simulator to quickly visualize the current generatedtotals of energy consumption per site (left side) and per meter (right side).
and operation of a remote system. Thus, the number of active meters should vary over time to try to
bring the system to its limits.
Another important variable to take in consideration is the number of messages per second that are
inputted into the system. Although there might exist many active meters, that does not directly implies
that frequency of messages is higher. These meters can be configured to send messages very often.
Therefore, the solution here proposed must be ready to deal to an uncertain amount of messages
per second because in a real life deployment the configurations for metering sampling, or even the
equipment itself, can suddenly change. This is actually the real workload for the solution proposed.
Metrics
A series of metrics were defined to measure the performance of the proof of concept during the test
cases. The critical computation of this proof of concept is performed in the real-time computing layer
(Energy Monitoring). Therefore, the performance of each Storm node was evaluated by monitoring their
hardware status. Hence, the metrics monitored in each node were the CPU activity, the memory usage,
the network traffic, and the number of packages exchanged.
In addition, the Storm UI also provides more metrics for the running topology that were also taken
in consideration: number of messages emitted, number of messages transferred, and the average time
take to process a tuple along the entire topology. These metrics were important to tune the configuration
of the topology in use by the proof of concept and to evaluate it afterwards.
63
4.2.4 Results and Discussion
Since this thesis solution proposal intends to provide real-time analytics using real-time computation,
the Energy Monitoring layer is the main target of evaluation. As for the Energy Analytics layer, its
performance depends on the algorithms that run inside of it. Their performance is discussed in the
literature that describe the algorithms. Therefore, the work implemented in this thesis with relevancy
to be evaluated, is in fact the Energy Monitoring layer. This is going to be evaluated accordingly to its
efficiency, performance, and scalability.
Efficiency
The first aspect under evaluation was the efficiency of the real-time computing system. This means
having to perform a test case where a known data set by the authors is inputted to this system and its
output is saved over time to be compared with the initial data. To achieve this the authors developed a
plugin for KairosDB to autonomously pull data from the RabbitMQ that followed a specific pattern in its
id. This was then used to record input and output data from the Storm cluster. Therefore, in this first
test case we loaded the Virtual Energy Cloud simulator with the simulated dataset, deployed the Storm
topology, and activated a series of meters over 10 minutes. After this time everything was stopped and
the results of were analyzed.
The results were very good. During this time a collection of 1440 points from one meter (circa 2.5
points per second) entered the system. From those, Storm’s output sample had only 254 points. The
result is a much simpler energy curve with much less detail. Although, if we look at both curves (see
Figure 4.3) we see that no important point was lost, i.e. every maximum and minimum peak points were
outputted by Storm and all of those in between ignored because they don’t add any more information.
Like so, the number of packets that circulate through the system is much less, resulting in a much higher
degree of performance. If we compare the times to obtain both curves from the persistent storage, we
need 6,107ms to obtain the raw data and 290ms to obtain the sampled data (21x faster).
The conclusion is that Storm was successful used for sampling and aggregation operations. Hence,
only the necessary points are send to the desktop clients, avoiding extra overhead that could hinder the
real-time communications and performance of Storm, intermediate systems, and client’s browsers.
Scalability
Evaluating the Energy Monitoring layer scalability capacity is an important task to assess if the real-
time and big data requirements can be solved. To solution relies on the distributed system used in the
Energy Monitoring layer that must scale horizontally and increase its throughput linearly. In the Energy
Cloud project, the authors used Storm to implement this layer. Thus, two test cases were created
to test the scalability of the Storm topology. As described before, Storm runs on multiple nodes with
multiple threads. These nodes share multiple topologies on their available slots. Thereby, scaling in
Storm means scaling the number of slots (threads) used. To prepare this test the MOM queues were
64
Figure 4.3: Comparison between the raw data (1440 points) that is inputted to the system (top area) andthe one that is outputted by the Energy Monitoring layer (254 points) after being sampled and normalized(bottom area). As we can see, the sampled data have much less detail but it has all the important pointsthat define the energy curve. The result is a sampled data which is 21x faster to obtain than the rawdata.
pre-loaded with thousands of messages to make sure that the topology could consume the maximum
amounts of messages as possible.
In the first test case, the topology was deployed with one executor (thread) per component and the
topology runs for about 5 minutes. The metrics of each component are then read at regular intervals.
However, in the second test case, the RabbitMQ spout and the Converter bolt have now two executors
instead of one. The initial results of the first test case, indicated that these components were taking a
considerable amount of time in comparison with the others. Therefore, these were scaled by adding
more executors. As we can see in Figure 4.4, the speed of computation of each component, i.e. the
number messages processed, remained linear during the time for both test cases. For example, the
RabbitMQ spout has an average of about 35 messages emitted per second with a total of 10k messages
in the first case. However, in the second test case, the same spout but now with two executors, had an
average of 80 messages emitted per second with a total of 23k. This is an increase of almost 2.5 times.
The conclusion is that by horizontally scaling the topology we scale the throughput of the system
linearly. Meaning that in case this system reaches its limits, just by scaling the topology we will have
much more capacity. In Storm, it is possible to scale either by re-deploying the topology with a new
configuration or by using the command-line and issuing a re-balancing command.
65
Figure 4.4: Storm topology test cases to evaluate scalability. The first case every component have onlyone executor (thread) (left side). On the second test case, the RabbitMQ Spout and the Converter Bolthave two executors (right side). The difference is almost 2.5 times more throughput.
Performance
Another point to take in consideration is the amount of resources used by the system. It is important to
evaluate how the resources of such a system are used to understand the degree of resource consump-
tion versus the workload. In an optimal solution, doubling the resources should at least cover the double
of workload. Otherwise, extra resources have to be allocated whenever the system scales.
To assess this evaluation, the authors collected the status of the hardware used by a simple topol-
ogy of one Nimbus (Storm master) and one Supervisor (Storm slave). Both components had Collectd
installed, a popular software daemon to collect system performance statistics periodically. All statistics
were then pushed every 10 seconds to a Graphite server, a scalable real-time graphing system.
In this case, the number of active meters and hence the number of messages per second, was
increase over time by the power of 2 each minute. The master node in used in this test case included
1xVCPU and 1Gb of memory. However, the slave node was running 2xVCPU and 2Gb of memory.
The results are presented in Figure 4.5 and show the use evolution of CPU, memory, and network
for both components, over a period of almost 10 minutes, time where the topology was killed. As we can
see, the master resources remained unchanged during the entire time, until it received the order to kill
the topology. As for the slave node, its resources were used linearly over time.
In the slave node, the results show that the CPU use increases almost at the same ratio than the
number of packages received. Which means that the CPU followed the workload almost at the same
pace. As for the memory, there was an increase of about 45 megabytes at a certain point, but the results
do not show that is directly related with the workload.
Therefore, the conclusion is that this topology is CPU intensive, i.e. the CPU is the resource that
matters when deploying this topology because its use increases almost linearly with the workload in
comparison with the memory. Hence, scaling the CPU resources of this topology results in a significant
computational of the same ratio of the scale itself.
66
Figure 4.5: Storm resource (CPU, memory, network) benchmarking with the number of active metersincreasing constantly.
67
UC1 - Monitor the most significant energy consumers
Q1 Increase industrial energy efficiency by monitoring and identifying unaware high consumers ofenergy
Q2 Moderate, because with current on premise systems only results from individual sites can beobtained. Performance is also limited to system’s capabilities
Q3 Real-time analytics
Q4 The Energy Monitoring layer computes the top consumers per site in a real-time basis
UC2 - Calculation of energy performance indicators (KPI)
Q1 Increase industrial energy efficiency by identifying energy improvement opportunities, increaseenergy consumption awareness and assist the development of strategies
Q2 Hard for large and complex sites because performance is limited to the local system capabilities
Q3 Data integration and cloud computing
Q4 Both computing layers compute and archive different KPI
UC3 - Historical data visualization and correlation
Q1 Increase industrial energy efficiency by deriving knowledge from energy use, identifying energymanagement inefficiencies, and increase energy consumption awareness
Q2 Hard with multi-site correlation, because without the cloud, this involves having to transfer andimport data between systems
Q3 Data integration and easy access to data
Q4 With the Dashboard, the user can correlate and visualize external and historical data stored in theDistributed Persistent Storage
UC4 - Pattern matching and data usage analysis
Q1 Increase industrial energy efficiency by deriving knowledge from energy use, identifying energymanagement inefficiencies, and increase energy consumption awareness
Q2 Hard for large and complex sites because performance is limited to the local system’s capabilities
Q3 Data integration and cloud computing
Q4 The Energy Analytics layer continuously analyzes and computes incoming data and perform com-plex computations (e.g. MR-AVF) to autonomously derive more knowledge
UC5 - Benchmark energy usage and efficiency
Q1 Increase energy efficiency by providing benchmark metrics to improve production processes andenergy strategies
Q2 Very hard, because on premises, there may not exist the necessary statistic population numberto benchmark
Q3 Data integration and cloud computing
Q4 On-demand data mining jobs in the Energy Analytics layer can browse through the multipledatasets and compare energy usage profiles
Table 4.2: Summary of the answers to the questions of each use case used to evaluate the solutionproposal from a conceptual point of view.
68
Chapter 5
Conclusions
Following the current industry trends regarding smart and future production, this thesis focuses how
industrial EMS can be enhanced with the use of the cloud, to fulfill the needs that this sector faces
today. To achieve the defined goals, a research methodology was created around this thesis’ main
concepts: Energy Efficiency (EE), Demand Response (DR), Industrial Automation Systems (IAS), and
Cloud Computing. The research included academic literature, research projects, ABB’s expertise and
surveys on current EMS solutions. The conclusion of this research is that the current gap between the
industry’s needs and current EMS products, can be diminished by incorporating cloud technologies. In
addition, big data is a present challenge due technological advances in field equipment that led to the
generation of larger amounts of data, with much more detail and higher sampling rates. Therefore, this
thesis proposes a novel cloud-native architectural solution for future EMS solutions, to address these
concerns and to cope with the big data challenges involved. This solution proposal provides means to
unveil patterns of inefficiency and to enhance decision-making with real-time analytics. The feasibility,
relevancy, and performance of this solution were validated with the implementation and evaluation of a
proof of concept called Energy Cloud.
5.1 Achievements
The development of this thesis has been proved to be very successful, with many achievements accom-
plished during it development. The greatest of them all, was the publication of an article entitled “Energy
Cloud: real-time cloud-native Energy Management System to monitor and analyse energy consumption
in multiple industrial sites”. This paper was accepted for the 7th IEEE/ACM International Conference
on Utility and Cloud Computing (UCC) in London, England. UCC is the premier IEEE/ACM conference
covering all areas related to Cloud Computing as a Utility. Publication is coming soon.
In top of that, the solution proposed in this thesis and included in this latter paper, was selected as a
finalist for the UCC 2014 Cloud Challenge. The cloud challenge is a competition within the UCC confer-
ence where participants develop solutions for real-world problems by utilizing virtualization technologies
and cloud computing. The competitors have to submit their proposals outlining the nature of the problem
69
that might range from business, scientific to socio-lifestyle applications, the methodology used to solve
the problem and the means of validation of the solution. Furthermore, the competitors have to outline
how characteristics of the solution can lead to business value, or to improve existing processes, prac-
tices, tools and applications. Later this in the year in the conference, all the finalists will present their
solution and be judged according to the previous criteria.
Furthermore, this article was also featured in the conference Energia@IST2014 as a workshop pub-
lication. This conference it’s organized by the Instituto Superior Tecnico university, to share all the
ongoing research work regarding the energy domain.
Besides academic publications, part of the work developed in the Energy Cloud project was also
accepted by the community. Namely, the Freeboard.io widgets and plugins developed for the Energy
Monitoring dashboard were included in source of this library by the company running this open source
project1. In addition, a plugin developed to extend KairosDB was also accepted and imported into
KairosDB’s plugin directory2. This plugin is a scalable-ready plugin for KairosDB that autonomously
pulls time series data from RabbitMQ3.
5.2 Future Work
This thesis presents the fundamental solution for modern cloud-native EMS. Therefore, there are still
many spaces for improvement and development. Always with the goal to enable a more energy-aware
and smart production, much research can be done in the Energy Analytics part of the solution proposed.
New algorithms can be developed and techniques can be applied to further improve these systems. This
however, requires the action of experts of different areas such as mathematics, machine learning, data
mining, etc.
The solution here depicted can of course be fully implemented with features that weren’t discussed
due to their low importance for this thesis. For example, a back-end system to manage the information
about sites and meters, a dashboard system to save and load custom dashboards from application
database, and many more could be implemented.
Further more advanced visualization widgets could also be developed to provide more and better
visual and exploratory capabilities.
1http://goo.gl/qDSQme2http://goo.gl/xQU5ka3http://goo.gl/uwwSFq
70
Bibliography
Albadi, M. H. and E. F. El-Saadany (2007, June). Demand Response in Electricity Markets: An Overview.
In 2007 IEEE Power Engineering Society General Meeting, pp. 1–5. IEEE.
April, W. G. (2013). Recommendations for implementing the strategic initiative Industrie 4.0. (White
Paper) (April).
Archer, J. and A. Boehm (2009). Security guidance for critical areas of focus in cloud computing. Cloud
Security Alliance, 0–176.
Arinez, J. and S. Biller (2010). Integration requirements for manufacturing-based Energy Management
Systems. In Innovative Smart Grid Technologies (ISGT), pp. 1–6.
Ates, S. A. and N. M. Durakbasa (2012). Evaluation of corporate energy management practices of
energy intensive industries in Turkey. Energy 45(1), 81–91.
Atzori, L., A. Iera, and G. Morabito (2010, October). The Internet of Things: A survey. Computer
Networks 54(15), 2787–2805.
Bayindir, R., E. Irmak, Colak, and A. Bektas (2011, January). Development of a real time energy
monitoring platform. International Journal of Electrical Power & Energy Systems 33(1), 137–146.
Berggren, E., E. Bjorn, and F. Froschl (2013). The Cloud: When & Why? (SucessFactors an SAP
Company).
Bowers, N. (2013, November). The perfect pyramid of automation. Available at http://www.
electronicspecifier.com/design-automation/the-perfect-pyramid-of-automation (visited in
Jan. of 2014).
Bunse, K., M. Vodicka, P. Schonsleben, M. Brulhart, and F. O. Ernst (2011). Integrating energy effi-
ciency performance in production management - gap analysis between industrial needs and scientific
literature. Journal of Cleaner Production 19(6–7), 667–679.
Byun, J., I. Hong, and S. Park (2012). Intelligent Cloud Home Energy Management System Using
Household Appliance Priority Based Scheduling Based on Prediction of Renewable Energy Capability.
58(4), 1194–1201.
71
Cannata, A. and M. Taisch (2010). Introducing Energy Performances in Production Management: To-
wards Energy Efficient Manufacturing. Advances in Production Management Systems New Chal-
lenges New Approaches 338, 168–175.
Cappers, P., C. Goldman, and D. Kathan (2010, April). Demand response in U.S. electricity markets:
Empirical evidence. Energy Volume 35, Issue 4, April 2010, 35(4), 1526–1535.
Cardoso, T. (2012). Efficient Integration of Energy Data. Ph. D. thesis, Instituto Superior Tecnico.
Carreira, P., R. Nunes, and V. Amaral (2011, October). SmartLink: A hierarchical approach for con-
necting smart buildings to smart grids. 11th International Conference on Electrical Power Quality and
Utilisation, 1–6.
Chiu, T.-Y., S.-L. Lo, and Y.-Y. Tsai (2012, December). Establishing an Integration-Energy-Practice
Model for Improving Energy Performance Indicators in ISO 50001 Energy Management Systems.
Energies 5(12), 5324–5339.
Combs, L. Cloud Computing for SCADA.
Dam, Q. B., S. Mohagheghi, and J. Stoupis (2008, November). Intelligent Demand Response Scheme
for Customer Side Load Management. 2008 IEEE Energy 2030 Conference, 1–7.
David, A. K. and F. Wen (2001). Market power in electricity supply. Energy Conversion, IEEE . . . 16(4),
352–360.
Deliso, R. (2013, September). Openadr: Transforming the smart grid
through standards. Available at http://energysmart.enernoc.com/bid/334732/
OpenADR-Transforming-the-Smart-Grid-Through-Standards (visited in Jan. of 2014).
Delsing, J., J. Eliasson, R. Kyusakov, A. W. Colombo, F. Jammes, J. Nessaether, S. Karnouskos, and
C. Diedrich (2011, November). A migration approach towards a SOA-based next generation process
control and monitoring. IECON 2011 - 37th Annual Conference of the IEEE Industrial Electronics
Society , 4472–4477.
Department of Electrical Engineering IIT Kharagpur and E. E. Iit. Architecture of Industrial Automation
Systems. pp. 1–21.
Fan, R., L. Cheded, and O. Toker (2005). Internet-based SCADA: a new approach using Java and XML.
Computing and Control Engineering.
Faruqui, A., D. Harris, and R. Hledik (2010). Unlocking the EUR53 billion savings from smart meters
in the EU: How increasing the adoption of dynamic tariffs could make or break the EU’s smart grid
investment. Energy Policy 38(10), 6222–6231.
Feuerhahn, S., M. Zillgith, C. Wittwer, and C. Wietfeld (2011, October). Comparison of the communi-
cation protocols DLMS/COSEM, SML and IEC 61850 for smart metering applications. 2011 IEEE
International Conference on Smart Grid Communications (SmartGridComm) (c), 410–415.
72
Fiedler, T. and P.-M. Mircea (2012, October). Energy management systems according to the ISO 50001
standard — Challenges and benefits. 2012 ICATE .
Fraser, R. E. (2000). Process Measurement and Control: Introduction to Sensors, Communication,
Adjustment, and Control. Premtc HaII.
Galloway, B. and G. P. Hancke (2013). Introduction to Industrial Control Networks. IEEE Communica-
tions Surveys & Tutorials 15(2), 860–880.
Gilart-Iglesias, V. (2007). Industrial Machines as a Service: Modelling industrial machinery processes.
Industrial Informatics (INDIN), 2012 10th IEEE International Conference, 737–742.
Givehchi, O. Industrial Automation Services as part of the Cloud: First Experiences. Conference Pro-
ceeding.
Givehchi, O., H. Trsek, and J. Jasperneite (2013, September). Cloud computing for industrial automation
systems — A comprehensive overview. 2013 IEEE 18th Conference on Emerging Technologies &
Factory Automation (ETFA), 1–4.
Goldschmidt, T., A. Jansen, H. Koziolek, J. Doppelhamer, and H. Pei-Breivold (2014, July). Scalabil-
ity and robustness of time-series databases for cloud-native monitoring of industrial processes. In
Proceedings 7th IEEE Int. Conf. on Cloud Computing (IEEE CLOUD 2014) Industry Track. IEEE.
Granderson, J. and M. Piette (2011). Energy Information Handbook: Applications for Energy-Efficient
Building Operations. Lawrence Berkeley National Laboratory.
Guntermann, A. E. (1982, November). Are Energy Management Systems Cost Effective? IEEE Trans-
actions on Industry Applications IA-18(6), 616–625.
Hong, I., J. Byun, and S. Park (2012). Cloud Computing-Based Building Energy Management System
with ZigBee Sensor Network. In Innovative Mobile and Internet Services in Ubiquitous Computing
(IMIS), 2012 Sixth International Conference on, pp. 547–551.
Inamdar, H. P. and R. P. Hasabe (2009). It based energy management through demand side in the
industrial sector. In INCACEC 2009.
Inductive Automation (2011). Cloud-Based SCADA Systems: The Benefits & Risks. White Paper..
International Energy Agency (2008). World energy outlook 2008.
International Union for Electricity applications (2009). Electric Load Management in Industry.
Jadeja, Y. and K. Modi (2012, March). Cloud computing - concepts, architecture and challenges. In 2012
International Conference on Computing, Electronics and Electrical Technologies (ICCEET). IEEE.
Kalakota, R. (2013). Cloud taxonomy. Available at http://cloudblueprint.wordpress.com/
cloud-taxonomy/ (visited in Dec. of 2013).
73
Karnouskos, S. and A. Colombo (2011). Architecting the next generation of service-based SCADA/DCS
system of systems. IECON 2011-37th, 359–364.
Kiliccote, S. and M. A. Piette (2005). Advanced Control Technologies and Strategies Linking Demand
Response and Energy Efficiency.
Koufakou, A., J. Secretan, J. Reeder, K. Cardona, and M. Georgiopoulos (2008, June). Fast parallel
outlier detection for categorical datasets using mapreduce. In IJCNN 2008.
Kyusakov, R. and J. Eliasson (2012). Emerging energy management standards and technolo-
gies—Challenges and application prospects. Emerging Technologies & Factory Automation (ETFA),
2012 IEEE 17th Conference.
Langmann, R., O. Makarov, L. Meyer, and S. Nesterenko (2012, July). The WOAS project: Web-oriented
Automation System. 2012 9th International Conference on Remote Engineering and Virtual Instru-
mentation (REV), 1–3.
Lee, E. a. (2008, May). Cyber Physical Systems: Design Challenges. 2008 11th IEEE International
Symposium on Object and Component-Oriented Real-Time Distributed Computing (ISORC), 363–
369.
Locke, G. and P. Gallagher (2010). NIST framework and roadmap for smart grid interoperability stan-
dards, release 1.0. National Institute of Standards and Technology .
Luo, Y. L., L. Zhang, D. J. He, L. Ren, and F. Tao (2011, February). Study on Multi-View Model for Cloud
Manufacturing. Advanced Materials Research 201-203, 685–688.
Ma, X., R. Cui, Y. Sun, C. Peng, and Z. Wu (2010, August). Supervisory and Energy Management Sys-
tem of large public buildings. 2010 IEEE International Conference on Mechatronics and Automation,
928–933.
Macia-perez, F., J. V. Berna-martinez, D. Marcos-jorquera, I. Lorenzo-fonseca, and A. Ferrandiz-
colmeiro (2012). A new paradigm: cloud agile manufacturing. 45, 47–54.
Manuel, T. and F. Cardoso (2013). Efficient Integration of Energy Computer Science and Engineering
Examination Committee. (August).
Mark Beyer, D. (2012). The importance of ’big data’: A definition.
Marz, N. Big Data: Principles and best practices of scalable realtime data systems.
Meenakshi, M. (2012). An overview on cloud computing technology. Advances in Computing and
Information Technology (3), 244–246.
Mell, P. and T. Grance (2011). The NIST Definition of Cloud Computing, Recommendations of the
National Institute of Standards and Technolog. National Institute of Standards and Technology .
74
Microsoft (2014a, January). Service oriented architecture (soa). Available at http://msdn.microsoft.
com/en-us/library/bb833022.aspx (visited in Dec. of 2014).
Microsoft (2014b, January). Three-layered services application. Available at http://msdn.microsoft.
com/en-us/library/ff648105.aspx (visited in Dec. of 2014).
Mohagheghi, S. and N. Raji (2012). Intelligent demand response scheme for energy management of
industrial systems. In Industry Applications Society Annual Meeting (IAS), 2012 IEEE, pp. 1–9.
Mohamed, E. (2012). Enhanced data security model for cloud computing. Informatics and Systems
(INFOS), 2012 8th International Conference, 12–17.
Mora, D., M. Taisch, A. W. Colombo, and J. M. Mendes (2012). Service-oriented architecture approach
for industrial System of Systems: State-of-the-Art for Energy Management. In Industrial Informatics
(INDIN), 2012 10th IEEE International Conference on, pp. 1246–1251.
Motegi, N., M. A. Piette, S. Kinney, and K. Herter (2003). Web-based Energy Information Systems for
Energy Management and Demand Response in Commercial Buildings. Lawrence Berkeley National
Laboratory .
O’Driscoll, E. and G. E. O’Donnell (2013). Industrial power and energy metering – a state-of-the-art
review. Journal of Cleaner Production 41(0), 53–64.
Palensky, P. and D. Dietrich (2011). Demand Side Management: Demand Response, Intelligent Energy
Systems, and Smart Loads. IEEE Transactions on Industrial Informatics 7, 381–388.
Piette, M., O. Sezgen, and D. Watson (2004). Development and evaluation of fully automated demand
response in large facilities. (January).
Qu, X. and W. Yingjun (2014). Software Resource Re-sharing in Middle-sized Enterprise Cloud Manu-
facturing System. 12(1), 711–717.
Raghavendra Nagesh, D. Y., J. V. Vamshi Krishna, and S. S. Tulasiram (2010, January). A real-time
architecture for smart energy management. 2010 Innovative Smart Grid Technologies (ISGT), 1–4.
Rahimifard, S., Y. Seow, and T. Childs (2010, January). Minimising Embodied Product Energy to support
energy efficient manufacturing. CIRP Annals - Manufacturing Technology 59(1), 25–28.
Rajkumar, R., I. Lee, L. Sha, and J. Stankovic (2010). Cyber-physical systems: the next computing
revolution. Design Automation Conference (DAC), 2010 47th ACM/IEEE , 731–736.
Report, T., U. P. Mar, and Y. Simmhan (2011). An informatics approach to demand response optimization
in smart grids. Proceding Conference.
Saha, B. and D. Srivastava (2014, March). Data quality: The other face of big data. In ICDE Data
Engineeringa 2014, pp. 1294–1297.
75
Sakr, S. and A. Liu (2012, June). SLA-Based and Consumer-centric Dynamic Provisioning for Cloud
Databases. In 2012 IEEE Fifth International Conference on Cloud Computing, pp. 360–367. IEEE.
SmartFactoryKL (2014, January). Industry 4.0 - from hype to realization. Available at http:
//smartfactory.dfki.uni-kl.de/en/content/news/2013/innotag-nachlese (visited in Jan. of
2014).
Staggs, K. P. and P. F. Mclaughlin (2010). Cloud computing for an industrial automation and manufac-
turing system.
Stouffer, K., J. Falco, and K. Kent (2006). Guide to Supervisory Control And Data Aquisition (SCADA)
and industrial control systems security. Recommendations of the National Institute of Standards and
Technology Keith.
Tao, F., L. Zhang, V. C. Venkatesh, Y. Luo, and Y. Cheng (2011, August). Cloud manufacturing: a
computing and service-oriented manufacturing model. Proceedings of the Institution of Mechanical
Engineers, Part B: Journal of Engineering Manufacture 225(10), 1969–1976.
Thollander, P. and M. Ottosson (2010). Energy management practices in Swedish energy-intensive
industries. Journal of Cleaner Production 18(12), 1125–1133.
Uk, S. (2011, January). The green league: How businesses are reacting to the green agenda. Available
at http://w3.siemens.co.uk/home/uk/en/aboutus/Documents/greenleaguereport.pdf (visited in
Sept. of 2013).
Vikhorev, K., R. Greenough, and N. Brown (2013, March). An advanced energy management framework
to promote energy awareness. Journal of Cleaner Production 43, 103–112.
von Kistowski, J. G., N. R. Herbst, and S. Kounev (2014, March). LIMBO: A Tool For Modeling Variable
Load Intensities. In Proceedings of the 5th ACM/SPEC International Conference on Performance
Engineering (ICPE 2014), ICPE ’14, New York, NY, USA, pp. 225–226. ACM.
Voorsluys, W., J. Broberg, and R. Buyya (2011). Introduction to cloud computing. Cloud computing:
Principles and Paradigms.
Walawalkar, R., S. Blumsack, J. Apt, and S. Fernands (2008, October). An economic welfare analysis
of demand response in the PJM electricity market. Energy Policy 36(10), 3692–3702.
Walawalkar, R., S. Fernands, N. Thakur, and K. R. Chevva (2010, April). Evolution and current status of
demand response (DR) in electricity markets: Insights from PJM and NYISO. Energy 35(4), 1553–
1560.
Xu, X. (2012, February). From cloud computing to cloud manufacturing. Robotics and Computer-
Integrated Manufacturing 28(1), 75–86.
76
Appendix A
Virtual Energy Cloud Models
77
Figure A.1: UML diagram to describe the domain and service layer of the virtual energy cloud project.
78
Figure A.2: UML diagram to describe the gateway layer of the virtual energy cloud project.
79
Figure A.3: Actual size view of the Virtual Energy Cloud dashboard to manage active sites and meters.
80
Figure A.4: Actual size view of the Virtual Energy Cloud dashboard to visualize the metrics simulated.
81
Appendix B
Energy Cloud
82
Sto
rm
Me
ssag
es
Qu
eu
ed
Ge
t sp
ee
dU
pti
me
Emit
ted
Co
mp
lete
Lat
en
cyEx
ecu
te L
ate
ncy
(m
s)Ex
ecu
ted
Pro
cess
Lat
en
cy (
ms)
Exe
cute
Lat
en
cy (
ms)
Exe
cute
dP
roce
ss L
ate
ncy
(m
s)Ex
ecu
te L
ate
ncy
(m
s)Ex
ecu
ted
Pro
cess
Lat
en
cy (
ms)
Exe
cute
Lat
en
cy (
ms)
Exe
cute
dP
roce
ss L
ate
ncy
(m
s)
2069
415
00:3
912
2048
66,1
60,
778
180
0,1
1,90
362
02,
355
1112
01,
167
0,03
360
043
33,7
23
2267
218
01:4
739
0048
14,0
540,
5198
00,
224
1,32
1940
1,44
93,
103
580
0,51
70,
784
2040
4616
,755
2409
122
02:2
658
2047
16,3
30,
474
1560
0,24
71,
1829
001,
292,
255
940
0,55
30,
711
2080
4606
,591
2579
420
03:2
080
0046
02,2
650,
441
2220
0,24
31,
140
001,
141,
572
1340
0,58
20,
909
4160
4633
,939
2776
9220
04:2
810
800
4550
,085
0,21
516
200
0,22
81,
0754
001,
044
1,24
1920
0,46
90,
824
5580
4663
,652
Sto
rm
Me
ssag
es
Qu
eu
ed
Ge
t sp
ee
dU
pti
me
Emit
ted
Co
mp
lete
Lat
en
cyEx
ecu
te L
ate
ncy
(m
s)Ex
ecu
ted
Pro
cess
Lat
en
cy (
ms)
Exe
cute
Lat
en
cy (
ms)
Exe
cute
dP
roce
ss L
ate
ncy
(m
s)Ex
ecu
te L
ate
ncy
(m
s)Ex
ecu
ted
Pro
cess
Lat
en
cy (
ms)
Exe
cute
Lat
en
cy (
ms)
Exe
cute
dP
roce
ss L
ate
ncy
(m
s)
1537
243
00:4
837
0047
49,0
120,
960
00,
552
1,29
318
401,
554
1,06
332
02,
375
0,59
618
8347
29,1
57
1663
943
01:5
487
0047
00,3
170,
6316
200,
646
0,88
543
600,
977
0,64
490
00,
956
0,56
144
2047
63,7
85
1663
943
02:2
011
320
4639
,011
0,50
422
600,
487
0,72
5640
0,82
30,
492
1260
0,77
80,
618
5760
4695
,409
1741
240
03:0
314
840
4639
,05
0,41
731
200,
429
0,68
774
200,
763
0,40
217
400,
716
0,59
7560
4613
,017
1832
140
04:1
121
000
4623
,025
0,34
646
200,
390,
6410
500
0,64
10,
321
2620
0,50
40,
515
1068
045
43,3
85
1870
148
04:3
923
200
4623
,505
0,33
251
800,
355
0,63
411
600,
605
0,32
429
000,
473
0,46
611
800
4519
,516
Rab
bit
MQ
Spo
ut
RM
Q (
2)B
olt
-Arc
hiv
er
(1)
Bo
lt-C
on
vert
er
(2)
Bo
lt-T
op
Co
nsu
me
rs (
1)B
olt
-Win
do
w (
4)
Rab
bit
MQ
Spo
ut
RM
Q (
1)B
olt
-Arc
hiv
er
(1)
Bo
lt-C
on
vert
er
(1)
Bo
lt-T
op
Co
nsu
me
rs (
1)B
olt
-Win
do
w (
4)
Figure B.1: Detailed results of the test cases to evaluate the Energy Cloud project performance.
83
Figure B.2: Actual size view of the Energy Monitoring dashboard.
84
Figure B.3: Actual size view of the Energy Analytics dashboard.
85
86