-1D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
D6.12.7: Survey Report on KPIs for the
Control and Management of Smart Grids
Revision History
Edition
Date
Status
Editor
v0.1
14.11.2011
Created
Heli Kokkoniemi-Tarkkanen
V0.8
19.01.2012
Main texts
Pekka Savolainen
V0.9
31.01.2012
Comments
Seppo Horsmanheimo, Pekka Savolainen
V0.95
2.2.2012
Comments and additions
Heli Kokkoniemi-Tarkkanen
V1.0
6.2.2012
Final version
Pekka Savolainen
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
-2D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
Abstract
The focus of this report is on defining Key-Performance Indicators (KPI) and developing intelligent
decision-making algorithms for microgrid Energy Management Systems (EMS). A microgrid is a
low voltage (LV) electric network, which consists of a combination of generation sources, loads and
energy storages interfaced through fast acting power electronics. The purpose of the EMS is to
make decisions regarding the best use of the loads, storage units and generators for producing
electric power and heat in the microgrid. This report explains how the EMS makes those decisions
and what data it needs to make them. Useful decision-making techniques for this purpose are
presented, such as Self-Organizing Maps (SOM), Support Vector Machines (SVM) and Bayesian
Networks. In addition, useful complimentary techniques for data analysis are presented, such as
Principal Component Analysis (PCA) and Independent Component Analysis (ICA). Equally
important is the collection of data for these algorithms. This report explains what data is important,
how this data is collected and from where. In SGEM 2 Funding Period 3, the plan is to build a
microgrid EMS simulator, which optimizes the microgrid system according to a selected market
policy.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
-3D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
Table of Contents
Revision History............................................................................................................................ 1
Abstract ......................................................................................................................................... 2
Table of Contents.......................................................................................................................... 3
1 Preface.................................................................................................................................... 4
2 Scope ...................................................................................................................................... 4
3 Introduction ............................................................................................................................ 5
4 Scenario ................................................................................................................................. 7
4.1 Microgrid ............................................................................................................................. 9
4.2 Microgrid Control Strategies .............................................................................................. 10
4.2.1 Fully Decentralized Control (MAS) .............................................................................. 11
4.2.2 Hierarchical Control .................................................................................................... 13
4.2.3 Centralized Control ..................................................................................................... 14
4.2.4 Comparison of Control Strategies ............................................................................... 14
4.3 Microgrid Energy Management System (EMS) ................................................................. 16
4.3.1 Mathematical Formulation of the Optimization Problem .............................................. 18
5 KPIs of a Microgrid Energy Management System ............................................................. 22
5.1 KPIs of a Smart Grid ......................................................................................................... 26
6 Intelligent Decision-Making Techniques for a Microgrid Energy Management System . 27
6.1 Self-Organizing Map (SOM) .............................................................................................. 27
6.2 Bayesian Networks (BN) ................................................................................................... 28
6.3 Fuzzy Logic....................................................................................................................... 31
6.4 Support Vector Machine (SVM) ......................................................................................... 32
6.5 Complementary Techniques for Decision Algorithms ........................................................ 33
6.5.1 Principal Component Analysis (PCA) ......................................................................... 33
6.5.2 Independent Component Analysis (ICA) ..................................................................... 34
7 Discussion and Conclusions .............................................................................................. 36
Abbreviations .............................................................................................................................. 37
8 References ........................................................................................................................... 39
Appendix 1 .................................................................................................................................. 41
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
-4D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
1 Preface
This report was done as a part of the Finnish national research project "Smart Grid and Energy
Market" SGEM Phase 2 and it was funded by Tekes – the Finnish Funding Agency for Technology
and Innovation and the project partners.
2 Scope
This survey report contains information on Key-Performance Indicators (KPI) affecting the
operation, control and management of smart grids. The scope on smart grids was limited to
microgrids. More specifically, this report concentrates on intelligent decisions made by microgrid
Energy Management Systems (EMS). Equally important is the collection of data for the intelligent
decision making algorithms. This report explains what data is important, how this data is collected
and from where. All regulatory issues such as standby charges, net metering and public utility
status of the microgrid are outside the scope of this report. In addition, security issues in microgrids
are excluded from this report.
The material for this document has been collected from literature surveys, books, journals, papers
and expert interviews.
This report is organized as follows:





Chapter 3 presents the introduction for this survey report
Chapter 4 presents the microgrid EMS scenario. In addition, the definition of microgrids and
their possible control strategies are discussed. The chapter ends with the presentation of
the Energy Management System and the possible policies that can be used.
Chapter 5 is dedicated to the Key-Performance Indicators (KPI) of microgrid EMS systems.
Also, KPIs for the whole smart grid are considered briefly.
Chapter 6 introduces some suitable algorithms that can be used by the EMS for making
intelligent decisions. In addition, PCA and ICA are presented as possible tools supporting
data analysis.
Chapter 7 contains the conclusions of the survey report
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
-5D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
3 Introduction
In order to be called a smart grid, the electric grid should be able to perform a number of functions.
The functions of a smart grid system fall into both real-time and non-real-time categories. Some
important functions in the real-time category are distributed sensing, power grid device status and
health monitoring, failure detection and localization, power quality and reliability monitoring, and
safety and security monitoring. Important functions in the non-real-time category are the integration
of existing and new utility databases to support operational optimization, asset utilization
maximization and life cycle management, asset replacement optimization, strategic planning,
optimization of system performance metrics and regulatory reporting [1]. Transforming an electric
power grid into a smart grid that implements these functions, involves more than just hardware and
software updates. It also requires a utility-wide business transformation and fundamental changes
in grid operations, field service, inventory management, backoffice operations, and strategic
planning to obtain the full benefit of this added technology. This transformation does not happen
overnight. Therefore, there is a clear need for developing new algorithms as well as control and
management solutions.
The objective of this work was to write a survey report on Key-Performance Indicators (KPI) for the
control and management of smart grids. The focus of the report is on KPIs and intelligent decisionmaking algorithms for microgrid Energy Management Systems (EMS). A microgrid is a low voltage
(LV) electric network, which consists of a combination of generation sources, loads and energy
storages interfaced through fast acting power electronics. The purpose of the EMS is to make
decisions regarding the best use of the loads, storage units and generators for producing electric
power and heat in the microgrid. Up until the time of writing (end of 2011), microgrid projects
around the world have been research projects where the goal has been to gather information and
best practices for commercial investments in the future. Commercial microgrid projects are still at
least 5-10 years in the future. The technology will be ready earlier but many significant regulatory
and judicial issues still need solving. This report deliberately skirts all regulatory issues, such as
standby charges, net metering, public utility status of the microgrid etc., because the regulatory
and business structure of the Smart Grid (SG) industry could change significantly over the coming
years. The point of this survey report is to provide a condense blueprint for implementing the
technical control functions of a microgrid EMS simulator.
The microgrid can be owned by one customer or a group of customers. It can be operated and
managed by the owners themselves or by a 3rd party operator who can represent the owners as
one to the local Distribution System Operator (DSO). The microgrid can also be owned by the
DSO, excluding power generation, and be classified as part of the public distribution grid [2]. The
most important benefits of the microgrid are:

Improved energy efficiency

Lower overall energy consumption

Improved environmental impact

Improvement of energy system reliability and robustness

Lower total energy bill for the owners
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
-6D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
Every microgrid needs some kind of control strategy, whether it is manual or automatic. We believe
that the best and the most efficient way to implement the control functions is by using intelligent
decision-making algorithms. Decisions made by machines can only be as good as the input data,
so extra care should be given to checking the authenticity of the data.
In SGEM 2 Funding Period 3, our plan is to build a microgrid EMS simulation platform, which
optimizes the microgrid system according to some policies. These policies include the “good
citizen”, “ideal citizen” and “green” policies, presented later in this report. Other policies may be
added to the simulator as well. Methods and algorithms used previously in Tekes funded
telecommunications projects WISECITI [3] and MERCONE [4] will be exploited to the smart grid
domain, whenever found suitable.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
-7D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
4 Scenario
Figure 1 depicts the parts and entities of the smart grid that are included to our scenario.
Figure 1. Participants in the scenario
As depicted in the figure above, the microgrid is a low voltage community network that can be
separated from the distribution grid when needed. It is connected to the rest of the network through
a single substation. This interface is also called the Point of Common-Coupling (PCC). Because
the microgrid is connected to the distribution grid through a single PCC, it can be regarded as a
single customer by the Distribution System Operator (DSO) [5]. An aggregator is a medium voltage
(MV) level entity that connects different customers to the DSO or in some cases, straight to the
Transmission System Operator (TSO). Maybe the biggest difference between the traditional
electric grid and the smart grid is that private customers can own generation units, such as wind
farms, and sell their excess energy in open energy markets to other private customers. In smart
grid nomenclature, private customers who own generation units are called prosumers.
In the heart of the microgrid is the Microgrid Control Center (MGCC), which is responsible for
managing the microgrid resources (loads, sources and storage units). The MGCC needs to have a
fast, reliable and robust telecommunication infrastructure connecting the microgrid entities so it can
do its job effectively. Information and communication technology (ICT) has a major role not only in
microgrids but also in the whole smart grid concept. Not surprisingly, ICT in microgrids has been
studied actively for some time now and the International Electrotechnical Commission (IEC) has
published a standard (IEC 61850) for communications in European microgrids.
Figure 2 depicts possible technologies for building a telecommunication network based on IEC
61850 in a microgrid. Advanced Metering Infrastructure (AMI) consists of meters with smart
metering capabilities. These capabilities include functions such as automatic processing and
transfer of metering data, automatic management of meters etc. It was estimated in the ‘More
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
-8D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
Microgrids’ project that, at least a 100 Mbit/s capacity is needed to handle the entire traffic load of
the microgrid [6].
Figure 2. ICT in microgrids [6]
Players in the energy markets today as well as in the smart grid energy markets in the future are
divided into regulated and deregulated players. Microgrids, as envisioned in this report, are still
mostly research-oriented projects and their evolution into commercially exploitable solutions in the
energy market is still at least a decade away. This means, that the regulatory status of the
microgrid and its owner(s) is (are) yet to be defined. Some work has been done in this respect but
that is outside the scope of this report. Following table gives the definition of energy grid
participants depicted in Figure 1.
Table 1. Smart grid players / participants in our scenario
Distribution System Operator DSO is the operator of the MV (in IEEE/IEC standards, MV is 1(DSO)
35kV) grid, which carries power from the transmission network to
the consumer via a transformer. It is a regulated player in the
energy market.
Transmission
Operator (TSO)
System TSO is a regulated player in the energy market that transmits
power from energy plants to regional distribution network
operators.
Distribution
Management Current distribution networks are not controlled in real time. In
System (DMS)
future distribution networks there is a need for real time control,
because of distributed generation (DG), increasing demand levels
and the possibility of actively managing demand. DMS is a system
designed for this kind of control and it is located at the DSO’s
control centre.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
-9D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
Aggregator
Microgrid
(MGCC)
Aggregators are deregulated participants whose main role is to be
the mediators between the consumers who provide (sell) their
demand flexibilities (= modifications in consumption) and the
markets where the aggregators offer (sell) these flexibilities for the
use of the other electricity system players.
Control
Center Manages and controls the components of the microgrid in an
efficient way.
Substation
4.1
Transforms the electricity from MV to lower voltage, which is more
suitable for consumers.
Microgrid
A Microgrid is a low voltage (LV) electricity network with energy production, consumption and
storage. Low voltage network currents are typically 50 – 1000V of AC or 75 – 1500V of DC.
Microgrid can be connected to the public distribution grid or it can operate in an island mode, which
means that the microgrid is disconnected from the main grid and has to function autonomously. In
order to satisfy energy demand in island mode, microgrid can generate the required energy or it
can use stored energy in its local storage components. Going into island mode may be warranted
when the power quality of the main grid is not satisfactory or when it fails. When connected to the
public grid, microgrid can purchase energy from the DSO. Owners of microgrid generation
components can also sell excess energy at open energy market. The microgrid is connected to the
distribution network through a single Point of Common Coupling (PCC) and appears to the power
network as a single unit [1]. Components in a microgrid are divided into three categories:

Generation (source) components:
e.g. PV-cells, wind/hydro power plants, etc.

Load (demand) components:
e.g. heat pumps, power tools, electric vehicle recharging, etc.

Storage components:
e.g. batteries, super capacitors, flywheels, etc.
Figure 3 depicts an example of a microgrid consisting of aforementioned components.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 10 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
Figure 3. Microgrid participants
In the future, microgrids may take the form of a small to mid-sized suburb, a shopping center, an
industrial park, a farm or a college campus. Up to the end of 2010, the majority of microgrids have
been pilot projects and/or research related experiments. The year 2010 signaled a transition as
some of the first commercial-scale microgrid projects reached significant milestones. The shift from
pilot validation projects to fully commercial projects is expected to accelerate in the near future.
PikeResearch has a worldwide microgrid deployment tracker on their website [7]. At the time of
writing, there were 160 active microgrid projects around the world.
4.2
Microgrid Control Strategies
In order to provide energy of the required quality in an efficient and economical way, the resources
within the microgrid must be operated in a coordinated and coherent fashion. Therefore, a control
system for the microgrid is needed. Some factors that the control system must consider include
forecasted demand, potential for power generation, fuel and energy prices and technical
constraints on devices. These factors will be refined later in this document.
Figure 4 depicts a typical architecture of a microgrid for implementing the control functions of an
Energy Management System (EMS) [8]. Microgrid Central Controller (MGCC) implements the
centralized functions pertaining to the whole microgrid. In addition, there is a local controller for
each specific resource component. The local controllers are local Generator Controller (GC), local
Load Controller (LC) and local Storage Controller (SC).
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 11 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
LC
SC
PCC
LC
MVGrid
MGCC
GC
LC
GC
SC
Figure 4. Microgrid control architecture
The most popular control strategies for microgrids in smart grid publications and literature are fully
decentralized control, hierarchical control and centralized control. These control strategies are
presented in more details in next sub-sections. Consequently, their performances are compared
and suitability for our case is assessed.
4.2.1 Fully Decentralized Control (MAS)
Fully decentralized control is based on distributed multi-agent technology and therefore the
resulting microgrid system is often called a Multi Agent System (MAS). Microgrid control using
MAS decision making has been studied for example in [8]-[12].
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 12 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
Figure 5. Agent based microgrid control architecture [12]
In a MAS system, control is divided among intelligent agents. Each energy generator, load or
storage controller is represented as multiple agents that share a common communication interface.
In addition, the control of MGCC is done by multiple agents, each of them having specific
responsibilities. These agents form a communication network in a microgrid. Figure 5 depicts
agents inside control entities. Although there is not a unique definition of what intelligent agents
are, there is a common view that describes them as pieces of software with the following
characteristics [9]:
1. Agents are capable of acting in the environment, i.e., agents can change their environment
with their actions.
2. Agents may communicate with each other.
3. Agents have a certain level of autonomy. They can make decisions based on some data
without a central controller.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 13 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
4. Agents are partially or fully aware of the environment. In addition to knowing the status of
their own unit, agents also know the status of selected neighboring units or the status of all
agents in the network.
5. The behavior of each agent is determined by its goals. The behavior of MAS is determined
by the system goal.
4.2.2 Hierarchical Control
Hierarchical control of microgrids has been studied for example in [5] and [13]-[14]. In [13], to
optimize the general operation of the system, three critical control levels were identified. The three
levels are:
1. Local controller (GC, LC and SC) level decisions
2. MGCC level decisions
3. DSO level decisions
On the first level, generation controllers (GC) use local information to control the voltage and
frequency of the microgrid in transient conditions. The MGCC gives demands and requirements on
how local controllers should behave. Local generation controllers also have the autonomy to
perform local optimization of the generation components active and reactive power production.
Local load controllers (LC) are used by the MGCC to provide load control capabilities. This means
that LCs can lower the power demand by controlling the amount of power given to the load. MGCC
also controls storage controllers (SC) by giving them forecasted storage needs.
Level 3
DSO
Level 2
MGCC
Level 1
LC
GC
SC
LC
Figure 6. Hierarchical control architecture
Second level decisions are made by the microgrid control center. MGCC is responsible for the
maximization of the microgrid’s cost-efficiency and the optimization of its operation. It also has the
authority to disconnect the microgrid from the main grid. When going into island mode, it must
decide which loads are still served and which are shed, if demand is higher than local power
sources can generate. MGCC uses Key-Performance Indicators (KPI) collected from different
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 14 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
sources to construct a ‘state’ of what is currently happening in the system. By using the current
‘state’ and ‘states’ from the past, it is possible to construct a desired ‘state’ for the future: a
forecast. The job of the MGCC is then to make decisions that steer the system toward the
forecasted ‘state’. Data (KPIs) for constructing the ‘states’ is collected from anywhere it is available
(local controllers, energy boxes, weather forecast stations, aggregator, DSO, etc.).
Third level decisions in the hierarchy are done by the DSO or another entity on the distribution grid
level, but they still have an effect on the operation of the microgrid. DSO is responsible for the
operation of medium and low voltage grids in areas where several microgrids or other customers
(prosumers) may exist.
4.2.3 Centralized Control
Centralized control strategy is the classic “one entity takes care of everything” solution. In the
microgrid, this entity is the MGCC. Not only is the MGCC responsible for the optimization of
microgrids resources but it is also responsible for deciding the set-points of each unit [15]. In this
control strategy there is no need for local controllers (GC, SC and LC) because the MGCC takes
care of all their functions. In this control strategy, the MGCC must be computationally very powerful
because every single decision in the microgrid is done at the MGCC.
4.2.4 Comparison of Control Strategies
The main difference between the decentralized and centralized control lies in the amount of data
that is processed in each case. If the MGCC had available and could process all the information of
the local controllers, then its solution would be “at least as good” as the one provided by the
decentralized approach. This is because each local controller does not have direct access to the
information of its neighbor controller, although MAS technology allows asking for it [11].
Hierarchical control can be thought of as a hybrid of these two strategies. Following is a list of
advantages and disadvantages of each control strategy.
Decentralized (MAS) control
Advantages:
-
Minimizing data communication [11]: Because of local decision-making, data is only sent to
other entities if they need it. This decreases the capacity needs of the communication
infrastructure vastly.
-
Plug-n-Play [11]: In the MAS approach, each DG component or load manufacturer can
embed a programmable agent in the controller of their equipment. The agent would then be
programmed according to rules peculiar to that microgrid.
-
Scalability: Because of decentralized processing, adding components to the microgrid does
not increase the amount of processing in the MGCC notably.
Disadvantages:
-
Control responsibility ambiguity: If the responsibilities of each agent have not been clearly
defined, the agents might make decisions that are not theirs to make. On the other hand,
some functions may not be any agent’s responsibility.
-
Independence of microgrid components: It is difficult for the MGCC to steer the microgrid
toward a common goal, if agents act too independently.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 15 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
Hierarchical control
Advantages:
-
Minimizing data communication: Because of local decision-making, data is only sent to
other entities if they need it. This decreases the capacity needs of the communication
infrastructure vastly.
-
Optimization of the whole microgrid: In order to achieve the full benefits from the microgrid
operations, it is important that all local controllers in the system work towards a common
goal. This will provide a better optimization of the whole microgrid.
-
Common interface to the MV grid: With hierarchical control, the DSO is able to get a better
view of the status of the microgrid. The MGCC combines the local controller data into a
status report, which can be given to the DSO. In the decentralized approach, the DSO
would need to generate the status report of the microgrid by combining the local controller
data.
-
Scalability: As in MAS strategy, adding components does not affect the performance of the
system.
Disadvantages:
-
Plug-n-Play: When new equipment is installed to the grid, if there is no common standard,
manual configuration in both the MGCC and at the premises of the new equipment must be
done.
-
Partial Single Point of Failure problem: Because of the tree structure, the whole branch is
disabled if the parent of the branch nodes is disabled.
Centralized control
Advantages:
-
Total control: MGCC has total control to everything that happens in the microgrid
Disadvantages:
-
Data communication difficulties [11]: Centralized approach requires that all the information
is available at the MGCC almost instantly. This means, that the communication
infrastructure must have plenty of capacity. This is more expensive than the simpler
infrastructure that is sufficient for MAS technology.
-
Data acquisition difficulties [11]: In practice, it is very difficult for the MGCC to have access
to all available information in every component of the microgrid, especially when the grid
expands over time. For example, it is very difficult for the MGCC to know and handle the
temperature of the battery of a specific storage unit.
-
Implementation difficulties [11]: Implementing a system, that can bid in the market every
hour and, at the same time, has the ability to shut down a specific load or change a specific
generator set-point within the next 300 ms, is complicated.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 16 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
-
Scalability concerns: Requirements for the performance of the central controller increase
rapidly as new equipment is added. At some point, the limit is reached when the MGCC
cannot manage its tasks anymore.
-
Lack of Plug-n-Play: Every time new equipment is installed to the microgrid, the MGCC
requires extra programming.
-
Single Point of Failure (SPOF) problem: If the central controller is disabled due to e.g.
natural disaster or malicious software, the whole microgrid is crippled.
From the customer’s viewpoint, the microgrid should appear as an autonomous power system
functioning optimally to meet the requirements of the customer. Such issues as local voltage,
reliability, losses and quality of power should be those that support the customer’s objectives. From
the wider power system’s viewpoint, however, the microgrid should appear as a “good citizen” or a
“model citizen” to the power grid. The different “citizen” models are explained in the next chapter.
In SGEM2 FP3 our plan is to build a microgrid EMS simulation platform with a control strategy that
can deliver on both these aspects. It will be a choice between the decentralized (MAS) and the
hierarchical approaches. Both seem to have their supporters, based on the number of papers
published.
4.3
Microgrid Energy Management System (EMS)
An Energy Management System (EMS) is a system where the MGCC controls resources of a
microgrid according to some (market) policies. Control of the microgrid resources should be done
automatically and in an optimum way. The MGCC promotes technical and economical operation
and provides the set-points to local controllers (GC, LC and SC). It can be said that the MGCC acts
as an EMS. Market policies for the EMS include:

Market policy A: “Good Citizen” policy [13]:
o The MGCC aims to satisfy the total energy demand of the microgrid by using its local
production, as much as possible, without exporting power to the distribution grid. That
is, energy self-sufficiency of the microgrid is maximized. This is called “good citizen”
policy because this kind of behaviour is beneficial for the overall distribution grid. At
the time of a peak demand, when energy prices at the open market are high, the
microgrid relieves network congestion by taking care of its own energy needs. The
“good citizen” term was first used in [16]. From the consumer’s point of view, the
MGCC minimizes operational costs of the microgrid, by taking into account open
market prices, demand and DG bids. The consumers of the microgrid share the
benefits of the reduced operational costs.

Market policy B: “Ideal Citizen” policy [13]:
o In this policy, the microgrid participates in the open market, buying and selling active
and reactive power to the grid via an aggregator or a similar energy service provider.
The MGCC strives to maximize the value (profits) of the microgrid, i.e. maximize the
corresponding revenues of the aggregator, by exchanging power with the grid. The
consumers are charged for their power consumption at open market prices. From the
distribution grid point of view, the microgrid looks like a single generator, capable of
relieving possible network congestions by transferring energy to nearby feeders of the
distribution network.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 17 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids

Market policy C: “Green” policy:
o The MGCC should allow only the most vital loads in the microgrid to use energy.
Other loads should be switched off. Energy is provided by microgrids own Renewable
Energy Source (RES) components.
Some key functions of the EMS are [5]:

To provide the individual power and voltage set-point for each local generation controller
(GC)

To ensure that heat and electrical load demands are met

To ensure that the microgrid satisfies operational contracts with the bulk system

To minimize emissions and system losses

To maximize the operational efficiency of the generators

To provide logic and control for islanding and reconnecting the microgrid during events
All of these functions will not be implemented in the microgrid simulator. Figure 7 depicts the
interaction of the EMS with energy demand, generation and storage components of the microgrid.
Demand is divided into high and low priority loads. Providing for low priority loads may be delayed
or stopped (curtailed) if necessary.
Demand
High priority
loads
Low priority
loads
Load
plan
Load
forecast
Generation
forecast
Generation
RES
EMS
Conventional
generation
Generation
plan
Energy
Storage
Energy
storage
schedule
Figure 7. Energy Management System
The functionality of an EMS is incorporated in the MGCC. It collects data from all available sources
and makes decisions based on the acquired data. Data sources for the MGCC include local
electrical and heat needs, the weather, the price of electric power, the cost of fuel, the power
quality requirements, the wholesale/retail service needs, the special grid needs, the demand-side
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 18 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
management requests, the congestion levels, etc. The information exchange of the simulation
model between the local controllers, MGCC and the distribution grid is detailed in chapter 5. It is
assumed that local controllers have a real-time connection with the component they are controlling.
Information exchange in a microgrid goes as follows: Every m minutes (m is e.g. 15), each DG
source bids for production for the next hour in m minute intervals. These bids are prepared
according to the energy prices in the open market, the operating costs of the DG units plus the
profit for the DG owner. That is, each DG owners bid is tailored specifically for his equipment and
circumstances. The MGCC combines all the bids from DG owners to make a generation plan for
the next hour. Owners of energy storages can make similar bids to the MGCC, if they have stored
energy. For the customer, it does not make a difference if the power is coming directly from DG
sources or from storage units. However, power losses in storing and transmitting must be taken
into account when making bids. Similarly, consumers within the microgrid might bid for their loads
supply for the next hour in the same m minute intervals or they might bid to curtail their loads. If the
consumers don’t make bids for their loads, then the MGCC has to predict the demand for the next
hour. Either way, the MGCC combines the load bids or predicts them to make a load plan. In
addition, the MGCC predicts the usage of energy storages by making an energy storage schedule
for every hour.
It is the job of the MGCC to optimize the operation of the microgrid according to the open market
prices, generation and load plans and energy storage schedule, by dispatching signals to GCs,
LCs and SCs. The optimization procedure depends on the selected market policy.
Making bids concerning the customers demand for power is called Demand Side Bidding (DSB). It
is assumed that customers have low and high priority loads allowing them to send two types if bids
to the MGCC. The MGCC then has two options for dealing with low priority loads. The first option is
to delay the serving of the load until the price of energy is lower. The second option is to curtail the
load for which the customer will be compensated.
4.3.1 Mathematical Formulation of the Optimization Problem
The optimization problem depends on the selected market policy. In the following, no distinction is
made between active and reactive power bids even though reactive power markets at distribution
level are less developed. This might change in the near future as power markets grow due to the
increasing number of consumers with DG sources. Another thing that has not been mentioned yet
is Combined Heat and Power (CHP) generation units. With a CHP unit, a prosumer in the
microgrid can provide heat for his own demand periodically. During these periods, the prosumer
needs less imported power, because some of the imported power would be used in generating
heat anyway. If the microgrid is equipped with a suitable heat transfer system, the CHP unit
owners could provide heat for other consumers in the microgrid. In the following market policies
(and in the microgrid EMS simulator), CHP units are taken into account in that they decrease the
power demand of their owners but there is no heat transfer among microgrid participants nor with
the distribution grid. In the simulator, thermal loads (heat) of consumers are represented by
equivalent electric loads.
The basis for the market policies A and B have been taken from [17], but the formulas have been
modified to include energy storage bids.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 19 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
4.3.1.1 Market Policy A (Good Citizen)
In market policy A (good citizen), the MGCC aims to minimize the operational cost of the microgrid.
It is assumed that the consumers share the benefits (smaller energy bill). The function to minimize
for each m-minute interval is:
Minimize{cost}, where
𝑁
𝑀
𝐾
𝑐𝑜𝑠𝑡 = ∑ 𝐷𝐺_𝑏𝑖𝑑(𝑥𝑖 ) + ∑ 𝑠𝑡𝑜𝑟𝑎𝑔𝑒_𝑏𝑖𝑑(𝑧𝑙 ) + 𝐴𝑋 + ∑ 𝑙𝑜𝑎𝑑_𝑏𝑖𝑑(𝑦𝑗 )
𝑖=1
𝑙=1
(1)
𝑗=1
In (1),










DG_bid(xi) is the bid from DG source i.
storage_bid(zl) is the bid from energy storage l. The cost for using the energy storage.
Using this energy makes sense if the cost of ‘stored energy’ < ‘generated energy’ < ‘market
energy’.
load_bid(yj) is the bid from load j, if demand side bidding is considered. If the customer is
offering to shed his load (curtailment option), the customer is compensated, which
increases the total operational cost.
xi is the amount of power offered by DG source i.
zl is the amount of power (including storage losses) offered by energy storage l.
yj is the amount of power used by load component j.
A is the cost of power from the open market.
X is the amount of power purchased from the open market for this m-minute interval.
N and M are the number of sources and storage units respectively.
K refers to the loads that have low priority loads and have the option to curtail their load if
needed. In the curtailment option the customer is compensated in his energy bill because
the cost of compensation for that customer is smaller than the cost of buying extra energy
from the distribution grid. This strategy is beneficial for the overall microgrid if the
operational cost is minimized.
The constraints for this optimization problem are.
1) Technical limits of the DG sources and storages, such as minimum and maximum limits of
operation. In addition, by setting the minimum threshold of energy storages to a certain
level, the microgrid can prepare for higher demand in the future.
2) Power balance (2) of the microgrid. P_demand is the power demand for one bidding
interval. In the equation, the imported energy X, DG energy xi and stored energy zl, must
satisfy local demand yj. The sign in front of stored energy zl depends on whether stored
energy is used or stored.
𝑁
𝑀
𝐾
𝑃_𝑑𝑒𝑚𝑎𝑛𝑑 = 𝑋 + ∑ 𝑥𝑖 ± ∑ 𝑧𝑙 − ∑ 𝑦𝑗 = 0
𝑖=1
𝑙=1
CLEEN OY
𝑗=1
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
(2)
- 20 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
4.3.1.2 Market Policy B (Ideal Citizen)
In this policy, the MGCC maximizes the profit from the power exchange with the grid. It is assumed
that consumers are charged with open market prices. This time, the optimization function for each
m-minute interval is,
Maximize{revenue-expenses} = Maximize{profit}
In this policy, the consumers of the microgrid not only buy energy but also sell excess production, if
any, from the DG sources to the upstream network at the market price. If the DG sources do not
produce enough power to cover the local demand or it is too expensive, power is bought from the
upstream network and sold to the customers. Another source of revenue (at least in theory) is the
power sold from energy storages. This is taken into consideration in (3). The revenue is then
defined as
𝑁
𝑀
(3)
𝑟𝑒𝑣𝑒𝑛𝑢𝑒 = 𝐴𝑋 + 𝐴 ∑ 𝑥𝑖 ± 𝐴 ∑ 𝑧𝑙
𝑖=1
𝑙=1
In (3), all notations are defined as in previous chapter.
If local energy generation is bigger than local demand, there is extra energy, which can be sold
(AX) outside the grid or stored. Whether extra energy is stored or sold depends on the market
price. If the price is low, it might be beneficial to store the extra energy. If energy is stored, the
stored amount decreases the revenue (minus-sign). When stored energy is sold, it increases the
revenue (plus-sign).
The term “expenses” includes costs for purchased power from the grid as well as compensation to
DG sources. If demand side bidding is considered, it is the same function as cost (1) in market
policy A.
𝑁
𝑀
𝐾
𝑒𝑥𝑝𝑒𝑛𝑠𝑒𝑠 = ∑ 𝐷𝐺_𝑏𝑖𝑑(𝑥𝑖 ) + ∑ 𝑠𝑡𝑜𝑟𝑎𝑔𝑒_𝑏𝑖𝑑(𝑧𝑙 ) + 𝐴𝑋 + ∑ 𝑙𝑜𝑎𝑑_𝑏𝑖𝑑(𝑦𝑗 )
𝑖=1
𝑙=1
(4)
𝑗=1
The MGCC must maximize
𝑁
𝑀
𝑁
𝑀
𝐾
𝑝𝑟𝑜𝑓𝑖𝑡 = 𝐴 ∑ 𝑥𝑖 ± 𝐴 ∑ 𝑧𝑙 − ∑ 𝐷𝐺𝑏𝑖𝑑(𝑥𝑖) − ∑ 𝑠𝑡𝑜𝑟𝑎𝑔𝑒_𝑏𝑖𝑑(𝑧𝑙 ) − ∑ 𝑙𝑜𝑎𝑑_𝑏𝑖𝑑(𝑦𝑗 )
𝑖=1
𝑙=1
𝑖=1
𝑙=1
(5)
𝑗=1
Note that the open market purchase and sale energy price (A) is considered the same in these
functions. Constraints for the optimization problem are the technical limits of the units and
minimum demand of the microgrid that must be met, as expressed in (6).
𝑁
𝑀
𝐾
𝑃_𝑑𝑒𝑚𝑎𝑛𝑑 ≤ 𝑋 + ∑ 𝑥𝑖 ± ∑ 𝑧𝑙 − ∑ 𝑦𝑗 ≥ 0
𝑖=1
𝑙=1
CLEEN OY
𝑗=1
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
(6)
- 21 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
4.3.1.3 Market Policy C (Green Citizen)
In this policy, the optimization is done to maximize the use of RES generation components. This
means, that the optimization formula needs to minimize the carbon footprint of power generation.
To make this happen, all generation components need a carbon footprint coefficient attached to
them. The more carbon dioxide the energy generator releases per produced kW, the higher the
coefficient. This can be described mathematically as
Minimize{carbon_footprint_cost}
where
𝑁1
𝑁2
𝑐𝑎𝑟𝑏𝑜𝑛_𝑓𝑜𝑜𝑡𝑝𝑟𝑖𝑛𝑡_𝑐𝑜𝑠𝑡 = 𝐶1 ∗ 𝑋 + ∑ 𝐶2 ∗ 𝑅𝐸𝑆_𝑥𝑖 + ∑ 𝐶3 ∗ 𝑇𝑅𝐴_𝑥𝑗
𝑖=1
(7)
𝑗=1
In (7), energy used in the microgrid is broken down to three parts. X is the power bought from the
distribution grid. RES_xi are the power produced with RES components inside the microgrid and
TRA_xj are the power produced with more traditional means in the microgrid. C1, C2 and C3 are the
carbon footprint coefficients for each type of energy. There can be more than three different
coefficients. Adding additional power generation types to (7) is straightforward.
Constraints for this optimization problem are the technical constraints of generation units and the
power balance in the microgrid.
𝑁
𝑀
𝐾
𝑃_𝑑𝑒𝑚𝑎𝑛𝑑 = 𝑋 + ∑ 𝑥𝑖 ± ∑ 𝑧𝑙 − ∑ 𝑦𝑗 = 0
𝑖=1
𝑙=1
CLEEN OY
𝑗=1
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
(8)
- 22 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
5 KPIs of a Microgrid Energy Management System
The data flow in the simulation model is depicted in Figure 8.
Figure 8. Data flow in the simulation model
Power generation using wind power, solar panels, photovoltaic cells, fuel cells, diesel engines,
microturbines, small hydropower plants or combined heat and power (CHP) systems depend on
surrounding conditions. Limiting factors for power generation can be e.g. wind speed, solar
radiation and shading, or availability and price of fuel or gas. To approximate the power production
of each form of power source, at least the following variables are considered in the simulation
model.
Table 2. Variables for power generation components
Variable [unit]
Description
Availability
1 if power generator is available, otherwise 0
Controllability
The variable tells if the power source is controllable (values 1-5) or not
(0). Some power sources like microturbines can be used e.g. for voltage
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 23 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
control or their capacity can be changed. Other sources like wind
turbines and PV cells can be seen as uncontrollable.
Fixed price [€]
Hourly payback amount for the investment including startup costs when
the power generator is not in operation or is in a startup state.
Price [€/kWh]
Generation cost including fuel cost.
MinLimit [kWh]
Minimum limit for the power generation.
MaxLimit [kWh]
Maximum limit for the power generation.
Efficiency [%]
Efficiency of power generation.
System losses [%]
Estimated system losses in power generation.
Wind speed [m/s]
Wind speed is the most important parameter for wind power production
estimation. The power from wind turbine can be calculated using the
formula P = k ½*CpρAV³, where k is a constant to yield power in
kilowatts (0.000133), Cp maximum power coefficient (ranging from 0.25
to 0.45), ρ air density, A rotor swept area and V wind speed. If
measured data is not available, wind speed can be approximated by
using the Weibull distribution based on the measured data of average
wind speeds at a site.
Air density [kg/m²]
Air density depends on the temperature and the elevation of the site. An
air density correction is needed for higher elevations but can be ignored
in case of predicting the long-term power production of wind turbine. For
more information, see description of wind speed.
Rotor swept area [m²]
Rotor swept area or A=π*D*2/4, where D is the rotor diameter. For more
information, see description of wind speed.
Solar
[W/m²]
radiation The power production of a PV module depends on the number of cells
in the module, the type of cells, and the total surface area of the cells as
well as solar radiation and the temperature of the PV module. Solar
radiation and temperature are used to estimate solar panel production. If
measured data is not available, daily or monthly solar panel production
estimates based on peak power and peak sun hours can be used in
simulation.
Number of PV cells
See description of solar radiation.
Surface area [m²]
See description of solar radiation.
Temperature [deg]
See descriptions of wind speed and solar radiation. If measured data is
not available, daily forecasts can be used instead.
Peak power [W p]
See description of solar radiation. Peak power of a PV module under
standard test conditions: (1000W/m² of sunlight (‘peak sun’), 25 ºC, and
air mass of 1.5)
Peak sun hours [h]
See description of solar radiation.
Flow rate
Electric power production at a hydroelectric plant can be estimated in
the simulations using the formula P = hrgk, where h is height [m], r is
flow rate [m³], g is acceleration due to gravity and k is a coefficient of
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 24 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
efficiency ranging from 0 to 1.
Power output [kW]
Power
forecast
For example, for gas turbines, micro turbines, and diesel generators
nominal full power performance and electrical efficiency are given by the
manufacturer.
production This variable is calculated from the raw data depending on the power
generation source.
DG bid
This variable is calculated from the power production and prices.
For controlling the energy consumption, loads should be prioritized in the simulation model. High
priority loads must be served when requested, but low priority loads can either be delayed (e.g. for
periods of lower market prices) or curtailed. In addition, priority can be used to equalize the hourly
energy consumption. The simulation model should contain at least the following variables for loads.
Table 3. Variables for power demand components (loads)
Variable [unit]
Description
Availability
1 if the load is available, otherwise 0
Controllability
The variable tells if the load is controllable (values 1-5) or not (0).
Fixed price [€]
Hourly payback amount for the investment including startup costs when
the load is not in operation or is in a startup state.
Price [€/KWh]
Load cost.
MinLimit [kWh]
Minimum limit for the load.
MaxLimit [kWh]
Maximum limit for the load.
Priority
Priority indicates the importance of the load. Value is 0 or 1.
Power demand [kW]
Power demand of the load.
Consumption
forecast
Hourly/daily forecast for the energy consumption.
LG bid
This variable is calculated from the power demand and prices.
In the simulation model, energy storages are needed for long-term decisions. Even if the decisions
are made in m-minute intervals, the energy storage can be directed to a certain state to be ready
for future load peaks. This is important especially when the green policy is simulated. Generation
from renewable energy sources (RES) might not produce enough electricity for the loads in
microgrid during peak consumption hours. One way to respond to this is to ramp up production
using e.g. the diesel generators, but in the long term, storage becomes cheaper. Energy storages
can also be used to respond to the deviation in voltage level and frequency. Thus, the simulation
model should include at least two types of energy storages; one for long-term and one for shortterm use. Variables for energy storages in the simulator are listed to the following table.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 25 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
Table 4. Variables for energy storages
Variable [unit]
Description
State
1 if the energy storage is set on, otherwise 0
Controllability
The variable tells if the energy storage is controllable (values 1-5) or not
(0).
Fixed price [€]
Hourly payback amount for the investment including startup costs when
the energy storage is not in operation or is in a startup state.
Price [€/kWh]
Energy saving cost.
MinLimit [kWh]
Minimum limit for the load.
MaxLimit [kWh]
Maximum limit for the load.
Priority
Priority (ranging from 1 to 5) can be useful to define which energy
storages are prioritized e.g. in short-term usage. The value of the
variable can be calculated from the raw data.
Efficiency [%]
Cycle efficiency in percentage.
Capacity [kWh]
Overall capacity of the energy storage.
Storage loss [%]
Overall loss in energy saving.
Stored energy [kWh]
Amount of energy stored.
Charging time [h]
Time needed for charging.
Power status
State of the energy storage. For example, total battery power remaining.
Power output [kW]
Power to get from the energy storage.
Power forecast
This variable is calculated from the raw data depending on the energy
storage.
SG bid
This variable is calculated from the power production and prices.
In the following table, the rest of the variables used in the simulation model are listed.
Table 5. General variables
Variable [unit]
Description
Power output [kW]
Power exported from the microgrid.
Power input [kW]
Power imported to the microgrid.
Market
purchase Market price if power is bought from the grid.
price [€/kWh]
Market selling price Market price if power is sold to the DSO.
[€/kWh]
Power quality
Power quality ranging from 1 to 5. 1 is poor, 5 is the best possible
quality.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 26 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
Voltage [V]
Voltage in the microgrid.
Frequency [Hz]
Frequency in the microgrid.
5.1
KPIs of a Smart Grid
When considering the whole smart grid, KPIs can be acquired from multiple sources, including the
data collection systems utilities already have. These include substation instrumentation, intelligent
grid devices, meters, smart sensors, utility databases, human input and external utility sources.
Some distribution grid control devices, such as capacitor controllers, recloser controllers and
sectionalizers, already contain digital processors and can collect power line data that is useful for
various power grid analytics. Many of these sources require only a communication channel to
make the data accessible. In some cases, new sensors must be deployed. In all cases, a
communication infrastructure must be in place to transport data to where it will be processed [1].
In the coming years, the electricity grid is going through a profound evolution. Evolution does not
happen overnight; it happens over a long period of time. Therefore, there is a need for a
methodology to measure the progress of the smart grid development. In [30], such a methodology
is presented. It is based on the framework by the U.S. Department of Energy but there is no reason
why it can’t be applied in Europe as well. The following characteristics are applied to assess the
status of American smart grid deployment:

Enable informed participation by customers

Accommodate all generation and storage options

Sell more than KWhs

Provide power quality for the 21st century

Optimise assets and operate efficiently

Operate resiliently to disturbances, attacks and natural disasters.
Starting from these six characteristics, KPIs have been constructed [30]. These KPIs are presented
in a table in appendix 1.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 27 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
6 Intelligent Decision-Making Techniques for a Microgrid Energy
Management System
This section introduces techniques that can be used for decision making in the microgrid energy
management system (EMS) simulator, implemented in SGEM2 FP3.
6.1
Self-Organizing Map (SOM)
The Self-Organizing Map (SOM) is an effective software tool for the visualization of highdimensional data. It implements an orderly mapping of a high-dimensional distribution into a
regular low-dimensional grid [18]. Applications of SOM can be found in various fields of science,
such as robotics, telecommunications, process control, pattern recognition, etc. A self-organizing
map consists of neurons organized on a regular low-dimensional grid. Each neuron is represented
by a d-dimensional weight vector (a.k.a. prototype vector or codebook vector) m = [m1, …, md],
where d is equal to the dimension of the input vectors. The neurons are connected to adjacent
neurons by a neighborhood relation, which dictates the topology of the map. Map topology is
divided into two factors: lattice structure and the global map shape. The lattice structure can be
hexagonal or rectangular and the map shape can be a sheet, a cylinder or a toroid [19]. Cylinder or
toroid map shapes can be used for circular data.
The most widely used methods for visualizing the cluster structure of the SOM are distance matrix
techniques, especially the unified distance matrix (U-matrix). It is a representation of a selforganizing map where the Euclidean distance between the codebook vectors of adjacent neurons
are depicted in a gray scale or a color image [20]. Figure 9 depicts an example of a U-Matrix
representing a SOM.
U-matrix
7.67
5.21
2.74
Figure 9. Self-organizing map visualized as a U-matrix
The SOM is often used to classify samples i.e. to find similarities in input samples. Before the SOM
can be used for this, it must be trained. Training of the SOM is done with training data, which can
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 28 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
be labeled or unlabeled. Data set is labeled if each sample in the set has a predetermined class.
On the other hand, a sample set with no classes is called an unlabeled data set. In supervised
learning, the SOM is trained with labeled data and in unsupervised learning, the SOM is trained
with unlabeled data. After the learning phase, an input sample can be classified with the SOM by
finding a Best-Matching Unit (BMU) for it. The BMU is the neuron on the map which codebook
vector is the most similar (by a selected metric) to the input sample. Now, in the case of supervised
learning, the class of the BMU signifies the class of the input sample. Unsupervised learning is, on
the other hand, useful for finding similarities in multivariate data. Learning type
(unsupervised/supervised) can be chosen depending on the application of the SOM and the
availability of training data. The SOM training algorithm resembles Vector Quantization (VQ)
algorithms, such as k-means [21]. The difference is that in addition to the best-matching weight
vector, also its topological neighbor’s weight vectors are updated in the training of the SOM. The
result is that the neurons on the map become ordered; adjacent neurons have similar weight
vectors.
The SOM is closely related to neural networks and their application areas are similar. In Japan,
neural networks are used in power systems for load forecasting and diagnosis [31]. SOM could be
used by the aggregator or the DSO to classify customers or to build generation forecasts. A Matlab
toolbox for SOM [22] developed by Helsinki University of Technology will be used in the microgrid
EMS simulator.
6.2
Bayesian Networks (BN)
Bayesian networks (aka belief networks), belong to the family of probabilistic graphical models.
These graphical structures are used to represent knowledge about an uncertain domain. Each
node in the graph represents a random variable, while the edges between the nodes represent
probabilistic dependencies among the corresponding random variables. A random variable is often
also called a feature. Conditional dependencies in the graph are often estimated by using known
statistical and computational methods. Graphical models with undirected edges are generally
called Markov random fields or Markov networks. Bayesian networks correspond to another
graphical model structure known as Directed Acyclic Graph (DAG), which is popular for example in
statistics and machine learning. The structure of a DAG is defined by two sets: the set of nodes
and the set of directed edges. The nodes represent random variables (features) and are drawn as
circles labeled by the variable names. The edges represent direct dependence between variables
and they are drawn as directed arrows between nodes. For instance, an arrow from node Xi to
node Xj represents a statistical dependence between the variables. The arrow indicates that a
value taken by Xj depends on the value taken by the variable Xi. Node Xi is then referred to as the
parent of node Xj and similarly Xj is referred to as the child of Xi. One reason for the popularity of
Bayesian networks is that they enable an effective representation and computation of the Joint
Probability Distribution (JPD) over a set of random variables [23].
Naïve Bayesian Classifier (NBC) is a simple Bayesian network specialized in classifying samples.
NBCs assume the features (random variables) independent given the class. Even though this is
rarely the case, NBCs have been proven surprisingly precise [24]. A reason for this is that when
performing classification, we are interested only in the class of maximal probability and not in the
exact probability distribution over the classes [25].
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 29 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
F2
F1
F3
C
F4
F5
...
Fx
Figure 10. Naïve Bayesian Classifier. Fx is a type variable (feature) and C is a class
variable. Variable C is the parent of all feature variables.
In an NBC, you have a set of variables, {F1, … , Fx}, called features and a class variable C. The
states of C correspond to the possible classes. A classifier is a function from F1 × … × Fx to C.
Before classifying can be done, the NBC must be trained with training data. When training the
NBC, one labeled set of variables is called a sample. All training samples must be labeled, which
means that all samples must be assigned a class from C. In practice it may happen, that while
collecting samples, some feature values may be missing. In this case, feature (random) variables
are extended with a state “?” for unknown (or “missing value”). A sample is said to be complete if
there are no missing values. A sample set is said to be consistent if two complete samples with the
same feature values belong to the same class. After training, the job of the classifier is to assign a
class (one state from C) for a sample with feature values {f1, … , fx}.
Let’s look at how the NBC works, by starting from a well-known theorem in probability theory.
Bayesian networks and Naïve Bayesian Classifiers are based on Bayes’ theorem [25], which is,
P A B  
PB AP A
likelihood  prior
 posterior 
P B 
evidence
.
(9)
The main idea behind the theorem is that the probability of an event A given an event B depends
not only on the relationship between events A and B but also on the marginal probability of
occurrence of each event. In other words, Bayes’ theorem gives us a method of updating our belief
about an event A given that we receive information about another event B. This is why, P(A) is
called the prior probability of event A and P(A|B) is called the posterior probability of event A given
B. The probability P(B|A) is called the likelihood of B given A and the probability P(B) is called
evidence because this term represents the new information that will update the posterior
probability. Of course, the events A and B have to depend on each other to use the Bayes’
theorem. If information about event B does not change our belief in the occurrence of event A, the
events are said to be independent. In this case, P(A|B) = P(A).
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 30 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
Figure 10 portrays a Naïve Bayesian Classifier. It has a number of features (Fx) with a single
parent variable C. Random variables in an NBC can be modeled as a discrete or a continuous
variable. Typically, the class variable (C) is modeled as a discrete variable. For example, if the
NBC is used in classifying e-mails, the states of the class variable could be ‘spam’ and ‘clean’. The
prior probabilities of these states are stored in a conditional probability table (CPT) as in table 2.
Table 6. CPT of P(C)
C
P(C)
spam
0,5
clean
0,5
The prior probabilities can be given equal probabilities (as in table 2) or they can be calculated as
the ratio between the ‘spam’ and the ‘clean’ samples in the training data. If the feature variables
(Fx) are modeled as discrete variables, then all of the prior probabilities must be calculated or
defined in a CPT before the NBC can be used. If the feature variables are modeled as continuous
variables, the user of the NBC has to specify a density function for each combination of states for
the parent variables. In an NBC, there is only one parent variable, which simplifies the analysis
considerably. A typical density function for this purpose is the normal (Gaussian) distribution.
By expanding the Bayes’ Theorem to include multiple features, the Naïve Bayesian Classifier
model can be written
PC F1 ,, Fx  
PC PF1 ,, Fx C 
.
PF1 ,, Fx 
(10)
In practice, we are interested only in the numerator because the denominator is just a
normalization constant. By using the fundamental rule for probability calculus [25], the numerator
can be rewritten as
PC PF1,, Fx C   PC, F1,, Fx  .
(11)
By using the chain rule for Bayesian networks [25]
PU    PAi pa Ai  ,
n
(12)
i 1
where U is the set of all variables in the network and pa(Ai) are the parents of Ai the numerator can
be written as
PC, F1,Fx   PC PF1 C PF2 C ,, PFx C  .
(13)
The posterior probability can then be written as
PC F1 ,, Fx   PC  PFi C  ,
x
i 1
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
(14)
- 31 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
where θ = 1/P(F1,…,Fx) is a normalization constant. Coefficients P(Fi|C) can simply be calculated
by integrating the density function of the appropriate normal distribution.
Bayesian networks containing both discrete and continuous variables are called hybrid Bayesian
networks. There are two constraints associated with a hybrid Bayesian network to ensure exact
probability updating [25]:
1. Each continuous variable must be assigned a (linear) conditional Gaussian distribution.
2. Discrete variables must not have continuous parents.
Bayesian networks can be a powerful tool if used properly. They can be used for e.g. automatic
generation control [26] and fault diagnosis [31].
6.3
Fuzzy Logic
Fuzzy logic is used in systems for which mathematical models are not available, but qualitative
information is present [27]. Fuzzy logic is a way to deal with uncertainty and imprecision. Fuzzy
logic permits using vague information, knowledge and concepts in an exact mathematical manner.
Terms such as 'fast', 'slow', 'very fast', 'quite slow', 'not very fast' can be used to describe
continuous, overlapping states.
Figure 11. An Example of Fuzzy logic member functions (Source:
http://www.answermath.com/Panels/fuzzy/fuzzy2.htm)
In classic logic, a variable can be either true or false (bivalent logic). In fuzzy logic, any given
variable does not have to be true or false, but partially true varying from 0 to 1 (multi-valued logic).
In fuzzy logic, a membership function (MF) defines how each point in the input space is mapped to
a degree of membership ranging from 0 to 1. The membership function can be an arbitrary curve
that suits from the point of view of simplicity, convenience, speed or efficiency. Commonly used
membership function types are piecewise linear function, Gaussian distribution function, sigmoid
curves, quadratic and cubic polynomial curves.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 32 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
There is distinction between fuzzy logic and probability. Both operate over the same numeric
range, and have similar values: 0.0 representing False (or non-membership), and 1.0 representing
True (or membership). The probabilistic approach yields the natural-language statement, "There is
an 80% chance that x is y," whereas the fuzzy terminology corresponds to "x has 0.80 degree
membership within the set of y". The fuzzy logic statement indicates that we have only 80%
chance of knowing in which set x belongs to and the truth is a matter of degree.
Fuzzy logic can be understood as a superset of standard Boolean logic. Fixing the fuzzy values to
1 (completely true) and 0 (completely false) is equivalent to Boolean logic. Since in fuzzy logic the
truth of any statement is a matter of degree, the input values are not just ones and zeros but real
numbers. As a result, Boolean operations A AND B, A OR B and NOT A are presented in Fuzzy
logic with MIN(A,B), MAX(A,B) and 1-A functions respectively. The graphical representation of
these basic operations is presented in Figure 12.
Figure 12. Fuzzy logic operations AND, OR and NOT. (Source:
http://www.mathworks.com/access/helpdesk/help/toolbox/fuzzy/index.html?/access/
helpdesk/help/toolbox/fuzzy/).
In Japan, fuzzy logic or fuzzy inference is in use in actual power systems in application areas such
as load forecasting, stabilization control, restoration after fault and in simulators for training [31]
6.4
Support Vector Machine (SVM)
Support Vector Machine (SVM) is a methodology to solve a classification problem with two
possible classes via a transformation of the original feature space to a suitable higher dimensional
linear space. SVM is actually a set of supervised learning methods used for classification and
regression analysis. The goal is to transform the original (possibly nonlinear) decision boundary in
the original feature space to an optimal separating hyperplane in the transformed space through
maximizing the distance of either class from the hyperplane. With a suitable nonlinear mapping
(kernel function) the problem of finding optimum hyperplane is reduced to a constrained
optimization problem and solved using constrained quadratic programming techniques in the
transformed linear space. SVM methodology can be shown to minimize the risk of
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 33 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
misclassification. Advantage of SVM is that the optimization is straightforward. Unlike in neural
networks, SVM does not have local optima in quadratic programming and it is relatively well
scalable for higher dimensional data. Basic SVM is restricted to binary classification but it can be
adapted to multi-class classification by combining intelligently multiple pairwise binary classifiers or
changing the QP-formulation (Quadratic Programming) in the binary case. One weakness of SVM
is the need for choosing a ‘good’ kernel function. It is the most challenging task of using SVM. The
kernel function is critical because it creates the kernel matrix of the method and summarizes all the
data. Many principles to find the most suitable kernel function have been proposed: diffusion
kernel, Fisher kernel, string kernel, etc. In practice, a low degree polynomial kernel or RBF (Radial
Basis Function) kernel with a reasonable width is a good starting point. SVM with RBF kernel is
closely related to RBF neural networks, with the centers of the radial basis functions automatically
chosen for SVM [27].
SVM is an efficient and scalable technique for classification. In power systems, it could be used for
example to discriminate the normal conditions of protection relays from the fault conditions as
proposed in [32].
6.5
Complementary Techniques for Decision Algorithms
The following algorithms are not decision-making algorithms but they can be used in data analysis
and in reducing the complexity of data.
6.5.1 Principal Component Analysis (PCA)
Principal Component Analysis (PCA) [28] is a statistical technique used for finding patterns in high
dimensional data and expressing it in such a way as to highlight similarities and differences. It is
mostly used as a tool in exploratory data analysis and for making predictive models. PCA reduces
multidimensional data sets to lower dimensions for the analysis. Essentially, PCA involves only
rotation and scaling. It has been widely used in image compression and recognition fields. PCA
transformation begins by subtracting the mean of the data set from data points. PCA involves the
calculation of the eigenvalue decomposition of a data covariance matrix or singular value
decomposition of a data matrix. The significance of data components can be obtained by ordering
the eigenvectors based on their eigenvalues. The eigenvector having the highest eigenvalue is the
principle component of the data set. Eigenvectors are perpendicular to each other. Compression of
the data, if wanted, is done by discarding data components associated with smaller eigenvalues,
that is, less significant data is discarded. At this point, the original data has been transformed so
that it is being expressed relative to its eigenvectors.
To get the original data back, the data must be transformed again so that it is expressed relative to
the original axis. The original mean is needed because it was subtracted when PCA analysis was
performed. If some data components were discarded, the final data set will have fewer dimensions
than the original. Thus, PCA is a lossy compression method but provided the eigenvalues are
small, the loss is not significant. To be precise, data set with n dimensions has n eigenvectors and
eigenvalues. By choosing only the first p eigenvectors (p most significant) and discarding the rest,
the processed data set will have p dimensions, where p<n.
The advantage of PCA is that it is the simplest true eigenvector-based multivariate analysis
method and it reveals efficiently the internal structure of the data. It enables visualizing multivariate
dataset as a set of coordinates in a high-dimensional data space (1 axis per variable), which can
easily be projected to 2D-plane of two variables. PCA is especially useful for collinear data. In a
case where regression-based techniques are unreliable and can give misleading results, PCA can
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 34 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
combine all collinear data into a small number of independent (orthogonal) axes, which can then
safely be used for further analyses.
The drawback of this method is that it is a non-parametric analysis and the answer is unique and
independent of any hypothesis about data probability distribution. Thus, no prior knowledge can be
incorporated and PCA compressions often cause some loss of information.
The applicability of PCA is limited by the following assumptions [27]:
1. Assumption on linearity. The observed data set is assumed linear combinations of certain
basis. Non-linear methods such as kernel PCA have been developed without assuming
linearity.
2. Assumption on the statistical importance of mean and covariance. PCA uses the
eigenvectors of the covariance matrix and it only finds the independent axes of the data
under the Gaussian assumption. For non-Gaussian or multi-modal Gaussian data, PCA
simply de-correlates the axes. When PCA is used for clustering, its main limitation is that it
does not take into account class specific features. There is no guarantee that the directions
of maximum variance will contain good features for discrimination.
3. Assumption that large variances have important dynamics. PCA simply performs a
coordinate rotation that aligns the transformed axes with the directions of maximum
variance. It is only when we believe that the observed data has a high signal-to-noise ratio
that the principal components with larger variance correspond to interesting dynamics and
lower ones correspond to noise.
PCA has been used with SOM in previous telecommunication projects. It was found to be well
suited for the compression of data before the SOM map is trained. This enhances the
computational performance of the EMS without significant decrease in the accuracy of the SOM.
Also, PCA has made finding dependencies between variables easier.
6.5.2 Independent Component Analysis (ICA)
Most measured quantities are actually mixtures of other quantities. Typical examples are sound
signals in a room with several people talking simultaneously or an electroencephalogram (EEG)
signal, which contains contributions from many different brain regions. Under certain conditions the
signals underlying measured quantities can be recovered by using ICA. ICA is a member of a class
of blind source separation (BSS) methods.
The success of ICA depends on one key assumption about the nature of the physical world. The
assumption is that independent signals are generated by different underlying physical processes. If
two signals are independent, then at a given time the value of one signal cannot be used to predict
anything about the corresponding value of the other signal. As it is not usually possible to measure
the output of a single physical process, it follows that most measured signals must be mixtures of
independent signals. ICA works by finding a transformation for a set of measured signals (i.e.
mixtures), which produces independent signal components. It is assumed that each of these
independent signal components is associated with a different physical process. In the ICA
nomenclature, the measured signals are known as signal mixtures and the sought after
independent signals are known as source signals. Applications of ICA include for example
separation of speech signals, EEG data analysis, functional magnetic resonance imaging (fMRI)
data analysis, and image processing.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 35 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
Figure 13 depicts an example application of the ICA. If two people speak at the same time in a
room with two microphones, then the output of both microphones is a mixture of two voice signals.
ICA can recover the original voices or source signals when given these two mixture signals. It is
important, that there are at least as many different mixture signals as there are source signals. For
the Figure 13 case this means that there must be at least as many microphones as there are
speakers.
One common interpretation of ICA is a maximum likelihood (ML) method for estimating the optimal
unmixing parameters of mixture signals. For details about how to do the math of ICA, a good
reference can be found from [29].
Source 1
Mixture 1
Source 1
ICA
Source 2
Mixture 2
Source 2
Figure 13. ICA in a nutshell.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 36 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
7 Discussion and Conclusions
The focus of this report was management of microgrid Energy Management Systems (EMS) and
the discovery of KPIs for the optimization of microgrids. Three different control strategies and three
market policies were presented for optimizing the resources in a microgrid.
In the SGEM-project FP3, a microgrid EMS simulator is built with Matlab software. The EMS
simulator should optimize the microgrid production, demand and storage components according to
the policies presented in chapter 4.3.1. Namely, the policies are “good citizen”, “ideal citizen” and
“green citizen”. The “good citizen” and “ideal citizen” policies are well known from many papers
published about microgrids (e.g. [5], [13], [16] and [17]). The “green citizen” is a new policy,
proposed by the authors of this survey report. From the customer’s viewpoint, the microgrid should
appear as an autonomous power system functioning optimally to meet the requirements of the
customer. From the wider power system’s viewpoint, the microgrid should appear as a well
behaving entity that does not create problems to the power grid. The control strategy implemented
in the EMS simulator will be a choice between the decentralized (MAS) and the hierarchical
approaches. Both seem to have their supporters, based on the number of papers published.
Hierarchical control of microgrids has been studied for example in [13]-[14] and decentralized
control for example in [8]-[12].
With the EMS simulator, it can be calculated how much money is saved by using the proposed
EMS system. The simulator will also answer questions about when it is beneficial to store energy,
when to ramp up generation or when to import energy from the distribution grid.
Data for the simulator will be collected from multiple sources. For example, weather data can be
collected from meteorological websites or it can be simulated. Weather data has to be accurate
because RES generation (e.g. PV and wind power) forecast will depend on it directly. In addition,
demand forecast will depend on weather data. Market price for electricity should be quite easy to
retrieve from energy companies websites.
There are a number of publications (e.g. [5] and [17]), where neural networks are used for
decision-making in a microgrid EMS. Based on our survey, it is revealed that SOM will be a
potential candidate for that as well. The SOM based approach will be compared to fuzzy logic and
Neural Network techniques, which are used in actual power systems in Japan. This makes the
evaluation of these techniques interesting especially in the microgrid case.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 37 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
Abbreviations
AC
AMI
BMU
Alternating Current
Advanced Metering Infrastructure
Best-Matching Unit
BN
Bayesian Network
BSS
CAIDI
Blind Source Separation
Customer Average Interruption Duration Index
CHP
CPT
Combined Heat and Power (generator)
Conditional Probability Table
DAG
DC
Directed Acyclic Graph
Direct Current
DER
DG
DMS
DSM
DSO
Distributed Energy Resources
Distributed Generation
Distribution Management System
Demand Side Management
Distribution System Operator
EMS
ESCO
FP3
GC
HV
Energy Management System
Energy Service Company
Funding Period 3
Generation Controller
High Voltage
ICA
Independent Component Analysis
ICT
IEC
JPD
KPI
Information and Communication Technology
International Electrotechnical Commission
Joint Probability Distribution
Key Performance Indicator
KWh
Kilo Watt hour
LC
LV
Load Controller
Low Voltage
MAIFI
MAS
Momentary Average Interruption Frequency Index
Multi Agent System
MGCC
ML
Microgrid Control Center
Maximum Likelihood
MV
MW
PCA
PCC
Medium Voltage
Mega Watt
Principal Component Analysis
Point of Common-Coupling
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 38 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
PHEV
Plug-in Hybrid Electric Vehicle
PV
QP
RBF
RES
Photovoltaic
Quadratic Programming
Radial Basis Function
Renewable Energy Sources
RTP
Real-Time-Pricing
SAIDI
SAIFI
System Average Interruption Duration Index
System Average Interruption Frequency Index
SC
SCADA
Storage Controller
Supervisory Control & Data Acquisition
SG
Smart Grid
SOM
SPOF
SVM
T&D
TSO
Self-Organizing Map
Single Point of Failure
Support Vector Machine
Transmission & Distribution
Transmission System Operator
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 39 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
8 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
Jeffrey Taft, “The Intelligent Power Grid”. IBM, 2006.
R. Pasonen: “Community Microgrid – A Building Block of Finnish Smart Grid”. Master’s
Thesis, Tampere University of Technology. June, 2010.
Wiseciti project homepage: http://www.cs.helsinki.fi/group/wiseciti/ (checked 30.12.2011)
Mercone project homepage: http://www2.cs.hut.fi/~pmrg/index.cgi?id=256 (checked
30.12.2011)
G. Celli, F. Pilo, G. Pisano and G. G. Soma: “Optimal Participation of a Microgrid to the
Energy Market with an Intelligent EMS”. In: IPEC the seventh international power
engineering conference, 2, 2005. p. 663-8.
J. Oyarzabal, Project coordinator: “Advanced Architectures and Control Concepts for
Microgrids”. WP E DE4: Report on applied data structures and mapping to communication
means. More Microgrids project, 2009.
http://www.pikeresearch.com/research/microgrid-deployment-tracker-2q11
J. Jimeno, J. Anduaga, J. Oyarzabal and A. G. de Muro: “Architecture of a microgrid
energy management system”. European transactions on electrical power 2011. 21: 11421158.
T. Logenthiran, D. Srinivasan and D. Wong: “Multi-Agent Coordination for DER in
MicroGrid”. IEEE International Conference on Sustainable Energy Technologies, 2008.
ICSET 2008. Issue Date: 24-27 Nov. 2008. p. 77-82.
A. Dimeas and N. Hatziargyriou: “A MultiAgent System for Microgrids”. IEEE Power
Engineering Society General Meeting 2004, 6-10 June 2004; 1: 55-58. DOI:
10.1109/PES.2004.1372752.
A. Dimeas and N. Hatziargyriou: “Operation of a Multiagent System for Microgrid Control”.
IEEE Transactions on Power Systems, Vol. 20, No. 3, August 2005.
J. Oyarzabal, J. Jimeno, J. Ruela, A. Engler and C. Hardt: “Agent based Micro Grid
Management System”. International Conference on Future Power Systems, 16-18
November 2005.
N.D. Hatziargyriou, A. Dimeas, A.G. Tsikalakis, et al.: “Management of Microgrids in
Market Environment”. Int. Conf. Future Power Syst. 2005:1-7.
A. Afzalian and A. Nabavi Niaki: “Hierarchical DES Supervisory Control Approach to
Microgrid Energy Systems”. 21th International Power System Conference, 2006.
A. G. Tsikalakis and N.D. Hatziargyriou: “Economic scheduling functions of a microgrid
participating in energy markets”. In Proc. CIGRE Symp. Power Systems with Dispersed
Generation, Athens, Greece, Apr. 2005.
R. Lasseter A. Akhill, C. Marnay, J. Stephens, J. Dagle, R. Guttromson, A. Meliopoulos,
R. Yinger and J. Eto: “White Paper on Integration of Distributed Energy Resources. The
CERTS MicroGrid Concept”. Consortium for Electric Reliability Technology Solutions
(CERTS), CA, Tech. Rep. LBNL-50829, Apr. 2002.
A. G. Tsikalakis and N.D. Hatziargyriou: “Centralized Control for Optimizing Microgrids
Operation”. Power and Energy Society General Meeting. Issue Date: 24-29 July 2011
IEEE.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 40 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
[18] T. Kohonen: “The self-organizing map”. Neurocomputing, Vol. 21, Issues 1-3, Pages 1-6,
November 1998.
[19] J. Vesanto, J. Himberg, E. Alhoniemi and J. Parhankangas: “SOM Toolbox for Matlab 5”.
Technical Report A57, April 2000. http://www.cis.hut.fi/projects/somtoolbox/
[20] J. Vesanto and E. Alhoniemi: “Clustering of the Self-Organizing Map”. IEEE Transactions
on Neural Networks, Vol. 11, No. 3, May 2000.
[21] Robert M. Gray: “Vector Quantization”. IEEE ASSP Magazine, pages 4-29, April 1984.
[22] SOM Toolbox for Matlab: http://www.cis.hut.fi/projects/somtoolbox/
[23] Irad Ben-Gal: “Bayesian Networks”, in Encyclopedia of Statistics in Quality and Reliability.
Wiley & Sons 2007.
[24] Olivier Pourret, Patrick Naim and Bruce Marcot: “Bayesian Networks, A Practical Guide to
Applications”. John Wiley & Sons Ltd. 2008.
[25] Finn V. Jensen and Thomas D. Nielsen: “Bayesian Networks and Decision Graphs”.
Second Edition. Springer, 2007.
[26] H. Bevrani, F. Daneshfar, P. R. Daneshmand: “Intelligent automatic generation control:
Multi-agent Bayesian networks approach”. IEEE International Symposium on Intelligent
Control (ISIC), 2010. Issue: 8-10 Sept. 2010.
[27] S. Horsmanheimo, J. Eskelinen, et al.: Wiseciti project deliverable: “D4.1 Monitoring
information and requirements for an expert system”. 2009.
[28] Lindsay I. Smith: “A tutorial on Principal Component Analysis”. 2002.
http://www.cs.otago.ac.nz/cosc453/student_tutorials/principal_components.pdf
[29] James V. Stone: “Independent Component Analysis”. In Encyclopedia of Statistics in
Behavioral Science, Volume 2, pp. 907–912, Editors Brian S. Everitt & David C. Howell,
John Wiley & Sons, Ltd, Chichester, 2005.
[30] B. Dupont, L. Meeus, and R. Belmans: “Measuring the ‘smartness’ of the Electricity Grid”.
7th International Conference on the European Energy Market (EEM), 2010.
[31] H. Bevrani and T. Hiyama: Intelligent Automatic Generation Control”. CRC Press, Taylor &
Francis Group, LLC, 2011.
[32] Yi Zhang, M.D. Ilic, O. Tonguz: “Application of Support Vector Machine Classification to
Enhanced Protection Relay Logic in Electric Power Grids”. Large Engineering Systems
Conference on Power Engineering, 2007. Issue: 10-12 Oct. 2007.
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 41 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
Appendix 1
Characteristics, Categories and Key Performance Indicators of a Smart Electricity Grid [30].
Enable informed participation by customers
Advanced Meters
1A: Number of advanced meters installed
1B: Percentage of total demand served by advanced meters
Dynamic Pricing Signals
2A: The fraction of customers served by RTP tariffs
2B: The fraction of load served by RTP tariffs
Smart Appliances
3A: Total yearly retail sales volume for purchases of smart
appliances [€]
3B: Total load capacity in each consumer category that is
actually or potentially modified by behaviours of smart
appliances [MW]
Demand Side Management
4A: Fraction of consumers contributing in DSM [%]
4B: Percentage of consumer load capacity participating in
DSM [MW/MW]
4C: Potential for time shift (before start-up and during
operation) [h]
Prosumer
5A: Total electrical energy locally (decentralized) produced
versus total electrical energy consumed [MWh/MWh]
5B: Minimal demand from grid (maximal own production)
versus maximal demand from the grid (own production is zero)
[MW/MW]
5C: Fraction of time prosumer is net producer and consumer
[h/h]
Accommodate all generation and storage options
Distributed Generation
6A: Amount of production generated by local, distributed
generation (MW/MW)
Storage
6B: Potential for direct electrical energy storage relative to
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 42 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
daily demand for electrical energy [MWhc,/MWhd]
6C: Indirect electrical energy storage through the use ofheat
pumps: time shift allowed for heating/cooling [h]
PHEVs
7A: The total number and percentage shares of on-road lightduty vehicles, comprising PHEVs
7B: Percentage of the charging capacity of the vehicles that
can be controlled (versus the charging capacity of the vehicles
or the total
power capacity of the grid) [MW/MW]
7C: Percentage of the stored energy in vehicles that can be
controlled (versus the available energy in the vehicles or the
total energy
consumption in the grid) [MWh/MWh]
7D: Number of charging points that are provided to charge the
vehicles
DER interconnection
8A: The percentage of grid operators with standard distributed
resource interconnection policies
Sell more than kWhs
New Energy Services
9A: Number of customers served by ESCO's
9B: Number of additional energy services offered to the
consumer
9C: Number of kWh that the consumer saves in comparison to
the consumption before the energy service
Flexibility
10A: The number
aggregators
of
customers
offering
flexibility
to
10B: The flexibility that aggregators can offer to other market
players [MWh]
10C: The time that aggregators can offer a certain flexibility [h]
10D: To what extent are storage and DG able to provide
ancillary services as a percentage of the total offered ancillary
services
10E: Percentage of storage and DG that can be modified vs.
total storage and DG [MW IMW]
Customer Choice
11A: Number of tariff plans available to end consumers
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
- 43 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
Support Mechanisms
12A: The average percentage of smart grid investment that
can be recovered through rates or subsidies
12B: The percentage of smart grid investment covered by
external financing
Interoperability
Level
Maturity 13A: The weighted average maturity level of interoperability
realised among electricity system stakeholders
Provide power quality for the 21st Century
Power Quality
14A: Amount of voltage variations in the grid [RMS]
14B: Time of a certain voltage variation [hl
14C: The percentage of customer complaints related to power
quality problems (excluding outages)
Required Power Quality
15A: Range of frequencies [Hz] contracted and range of
voltages [V] contracted
Microgrids
16A: The number of microgrids in operation
16B: The capacity of microgrids [MW]
16C: The total grid capacity of micro grids to the capacity of
the entire grid [MW/MW]
Optimize assets and operate efficiently
T&D Automation
17A: Percentage
technologies
Dynamic Line Rating
18A: Number of lines operated under dynamic line ratings
of
substations
applying
automation
18B: Percentage of kilometers of transmission circuits
operated under dynamic line ratings [km]
18C: Yearly average transmission transfer capacity expansion
due to the use of dynamic (versus fixed) line ratings [MW-km]
Capacity Factors
19A: Yearly average and peak generation capacity factor (%)
19B: Yearly average and average peak capacity factor for a
typical kilometer of transmission line (%-km per km)
19C: Yearly average and
transformer capacity factor (%)
Efficiencies
average
peak
distribution
20A: Efficiency of generation facilities [energy output (MWh) I
energy input (MWh)]
20B:
CLEEN OY
Energy
losses
in
transmission
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi
and
distribution
- 44 D6.12.7 Survey Report on KPIs for the Control and Management of Smart Grids
[MWh/year]
Operate resiliently to disturbances, attacks and natural disasters
Advanced Sensors
21A: Number (or percentage) of grid elements (substations,
switches, ... ) that can be remotely monitored and controlled in
real-time
21B: The percentage of substations possessing advanced
measurement technology
21C: The number of applications supported by these various
measurement technologies
information Exchange
22A: Total SCADA points shared per substation (ratio)
22B:
Fraction
of
transmission-level
measurement points shared multilaterally (%)
synchrophasor
22C: Performance (bandwidth, response speed, availability,
adaptability, ... ) of the communication channels towards grid
elements
T&D Reliability
23A: SAIDI represents the average number of minutes
customers are interrupted each year [Minutes]
23B: SAIFI represents the total number of customer
interruptions per customer for a particular electric supply
system [Interruptions]
23C: CAIDI represents the average outage duration that a
customer experiences [Minutes]
23D: MAIFI represents the total number of customer
interruptions per customer lasting less than five minutes for a
particular electric supply system [Interruptions]
Standards
telecommunication
infrastructure
in 24A: The compliance of electric power industries with
European and international telecommunication standards and
protocols
CLEEN OY
Eteläranta 10, P.O. BOX 10, FI-00131 HELSINKI, FINLAND www.cleen.fi