DEGREE PROJECT IN ELECTRICAL ENGINEERING,

SECOND CYCLE, 30 CREDITS

STOCKHOLM, SWEDEN 2016

Reliability-Constrained
Microgrid Design
JESÚS PANIAGUA SÁNCHEZ-MATEOS

KTH ROYAL INSTITUTE OF TECHNOLOGY

SCHOOL OF ELECTRICAL ENGINEERING
Reliability-Constrained
Microgrid Design

Jesús Paniagua Sánchez-Mateos

Master of Electric Power Engineering Thesis 2016

KTH School of Electrical Engineering

Osquldas väg 10

SE-100 44 Stockholm
ii
 Master of Science Thesis 2016

Reliability-Constrained
Microgrid Design

Jesús Paniagua Sánchez-Mateos

Approved Examiner Supervisor

2016- Patrik Hilber Ebrahim Shayesteh

Commissioner Contact Person

Abstract
Microgrids are new challenging power systems under development. This report presents
a feasibility study of microgrid development. This is an essential task before
implementing microgrid systems. It is extremely important to know the number and size
of distributed energy resources (DERs) needed and it is necessary to compare
investment costs with benefits in order to evaluate the profitability of microgrids. Under
the assumption that a large number of DERs improves the reliability of microgrids an
optimization problem is formulated to get the accurate mix of distributed energy
resources. Uncertainty in physical and financial parameters is taken into account to
model the problem considering different scenarios. Uncertainty takes place in load
demanded, renewable energy generation and electricity market price forecasts,
availability of distributed energy resources and the microgrid islanding. It is modeled in
a stochastic way.
The optimization problem is formulated firstly as a mixed-integer programming solved
via branch and bound and then it is improved formulating a two stage problem using
Benders’ Decomposition which shortens the problem resolution. This optimization
problem is divided in a long-term investment master problem and a short-term operation
subproblem and it is solved iteratively until it reaches convergence.
Bender’s Decomposition optimization problem is applied to real data from the Illinois
Institute of Technology (IIT) and it gives the ideal mix of distributed energy resources
for different uncertainty scenarios. These distributed energy resources are selected from

iii
an initial set. It proves the usefulness of this optimization technique which can be also
applied to different microgrids and data. The different solutions obtained for different
scenarios are explained and analyzed. They show the possibility of microgrid
implementation and determine the most favorable scenarios to reach the microgrid
implementation successfully.
Reliability is a term highly linked to the microgrid concept and one of the most
important reasons of microgrid development. Thus an analysis of reliability importance
is implemented using the importance index of interruption cost ( ) in order to measure
the reliability improvement of developing microgrids. It shows and quantifies the
reliability improvement in the system.

Key words:
Microgrid planning, Microgrid investment, Uncertainty, Optimization problem,
Bender’s decomposition, Distributed energy resources, Islanded operation, Reliability
importance index.

iv
Acknowledgements
Firstly I would like to express all my love and thanks to my parents, Jesús and Rosario.
For their limitless daily support and love, without them all these years of study and
effort would not have been possible.
I want to say huge thanks to my friends Manuel Ferrán, José Lazuen, Raquel Mostaza
and Jaime Serrano. For all the advice given to me both in Spain and Sweden and all the
time spent together. You cheer me up in my bad moments and make me laugh and
smile. Thank you for your true and unconditional friendship.
I would also like to express my greatest gratitude to my supervisor Ebrahim Shayesteh
for his guidance and support during the whole time of this master thesis project and for
all the time he has spent with me. I also want to thank my examiner Patrik Hilber for his
suggestions to improve the results of this work.
Finally I would like to thank my mates and also friends Fernando García and Axel
Scheutz and all the people I have met during the 18 months I have spent in Stockholm:
Angel, Alessandro, Leo, Maïlys, Addis, Ander, Sebastian, Karin, Jasmin, Rahul,
Pritesh… You have made feel like home.

v
vi
List of tables
Table 1 Example of feed in tariff policy in Spain [19] ............................................................... 13
Table 2 Example of VOLL by country and customer [42] ......................................................... 23
Table 3 Data set for dispatchable generators [45] ....................................................................... 32
Table 4 Data set for non dispatchable units [87] ......................................................................... 32
Table 5 Useful life fron renewable energy systems [87]............................................................. 33
Table 6 Photovoltaic unit costs [87] ............................................................................................ 33
Table 7 Wind power onshore unit costs [87] .............................................................................. 33
Table 8 Energy Storage System's characteristic data .................................................................. 33
Table 9 Benders' Decomposition time improvement .................................................................. 45
Table 10 Comparison of non uncertainty models with and without microgrid ........................... 47
Table 11 Model with 9 hours of power cut at the peak load day ................................................ 51
Table 12 Model with 9 hours of power cut at the valley load day .............................................. 52
Table 13 Model with 9 hours of power cut at different days ...................................................... 52
Table 14 Investment and Operation cost comparison in power cuts placement ......................... 52
Table 15 Reliability Constraints with different locations (v1 and v2) at day 2 .......................... 55
Table 16 Reliability Constraints with different locations (v1 and v2) at day 206 ...................... 55
Table 17 Microgrid and no Microgrid Implementation with Power Cuts ................................... 56
Table 18 Load uncertainty implementation with 10% bound (VOLL in $/MWh) ..................... 58
Table 19 Investment and operation costs in model 21 and 22 (VOLL in $/MWh) ..................... 59
Table 20 Load uncetainty in microgrids with different bounds and VOLL=2000 $/MWh ........ 60
Table 21 Load uncetainty in microgrids with different bounds and VOLL=10000 $/MWh ...... 60
Table 22 Load uncetainty in microgrids with different bounds and VOLL=1000000 $/MWh .. 61
Table 23 Renewable energy uncertainty implementation at 20% bound .................................... 61
Table 24 Renewable energy uncertainty implementation at 10% bound .................................... 63
Table 25 Renewable energy uncetainty in microgrids with different bounds ............................. 63
Table 26 Investment and operation costs in models 41, 45, 46 and 48 ....................................... 63
Table 27 Load and renewable energy implementation together ................................................. 65
Table 28 Comparison of load and renewable energy uncertainty separately and altogether ..... 66
Table 29 Electricity market price uncertainty implementation (VOLL in $/MWh) (LS=Load
Shedding) .................................................................................................................................... 66
Table 30 Investment and operation cost study with electricity market price uncertainty (VOLL
in $/MWh) ................................................................................................................................... 67
Table 31 Electricity market price uncertainty implementation with a higher range (LS=Load
Shedding) .................................................................................................................................... 68
Table 32 Electricity market price uncertainty implementation without ESSs (VOLL in $/MWh)
(LS=Load Shedding) ................................................................................................................... 71
Table 33 Comparison between the electricity market uncertainty with and without ESSs ......... 71
Table 34 Comparison of the three kinds of uncertainties with the same hours and bounds
(LS=Load Shedding) ................................................................................................................... 72
Table 35 Load, renewable energy generation and electricity market price uncertainty
implementation ............................................................................................................................ 73
Table 36 Comparison of load, renewable energy and electricity market price uncertainty
separately and altogether with ESSs ........................................................................................... 74
Table 37 Comparison of load, renewable energy and electricity market price uncertainty
separately and altogether without ESSs ...................................................................................... 75
Table 38 Final models with the three types of uncertainty implementation without microgrid . 75
Table 39 Implementation of longer power cuts with the three uncertainties altogether ............. 77
Table 40 Importance Index comparisson among microgrids with different DERs implemented 79

vii
viii
List of figures
Figure 1 Distribution of developed microgrids’ capacity by sector in U.S. Source: GTM [28] . 15
Figure 2 Distribution of planned microgrids’ capacity by sector in U.S. Source: GTM [28] ..... 16
Figure 3 Capacity share in microgrids. Source: GMT [28] ......................................................... 16
Figure 4 Types of Electrical Energy Storage Systems. Source: [89] .......................................... 34
Figure 5 VOLL time dependence ................................................................................................ 37
Figure 6 Daily Average Electricity Market Price in a year ......................................................... 48
Figure 7 Daily Average Load Demanded in a year ..................................................................... 48
Figure 8 Hourly Electricity Market Price and Load demanded at days 7, 105, 187 and 300 ...... 49
Figure 9 ESSs' Charging and Discharging Average Times during the 6 years microgrid’s
performance................................................................................................................................. 50
Figure 10 Investment and Operation Cost Relation for models 1-5 ............................................ 53
Figure 11 Microgrid - Non microgrid cost difference at peak load demanded day ................... 57
Figure 12 Microgrid - Non microgrid cost difference at valley load demanded day .................. 57
Figure 13 Load Uncertainty Annual Average Cost Percentage Difference ................................ 60
Figure 14 Renewable Energy Uncertainty Annual Average Cost Percentage Difference .......... 62
Figure 15 Load and renewable energy uncertainty comparison with the same percentage bound
and different number of uncertainty hours .................................................................................. 64
Figure 16 Load and renewable energy uncertainty comparison with the same numbers of hours
but different percentage bounds .................................................................................................. 65
Figure 17 Comparison of the final annual average cost depending on the uncertainty kind ...... 73
Figure 18 Microgrid and no microgrid cost comparison with different VOLL .......................... 76
Figure 19 Microgrid and no microgrid cost comparison with different VOLL .......................... 76

ix
x
Contents
Acknowledgements ....................................................................................................................... v
List of tables ................................................................................................................................ vii
List of figures ............................................................................................................................... ix
1. Introduction ............................................................................................................................... 1
1.1 Background and motivation................................................................................................. 1
1.2 Aim, objectives and delimitation ......................................................................................... 1
1.3 Ethical aspects ..................................................................................................................... 2
1.4 Overview of the report ......................................................................................................... 2
2. Background ............................................................................................................................... 5
2.1 Microgrids ........................................................................................................................... 5
2.1.1 Microgrids and main grids ............................................................................................ 5
2.1.2 Operation modes ........................................................................................................... 6
2.1.3 Elements........................................................................................................................ 7
2.1.4 Benefits and costs ......................................................................................................... 9
2.1.5 Business models in microgrids ................................................................................... 11
2.1.6 Policies ........................................................................................................................ 12
2.1.7 Challenges ................................................................................................................... 13
2.1.8 Microgrid projects ....................................................................................................... 14
2.2 Reliability .......................................................................................................................... 18
2.2.1 Definition .................................................................................................................... 18
2.2.2 How to measure .......................................................................................................... 19
2.2.3 Value of loss of load [42] [43] [44] ............................................................................ 22
2.3 Uncertainty ........................................................................................................................ 24
2.3.1 Definition .................................................................................................................... 24
2.3.2 Uncertainty in microgrids ........................................................................................... 24
2.3.3 Deterministic and stochastic models ........................................................................... 24
3. Literature review ..................................................................................................................... 27
4. Methodology ........................................................................................................................... 31
4.1 Model ................................................................................................................................. 31
4.2 Formulation ....................................................................................................................... 34
4.3 GAMS................................................................................................................................ 39
4.4 Benders’ Decomposition ................................................................................................... 40
4.4.1 Benders’ decomposition problem formulation............................................................ 43
4.4.2 Benders’ decomposition applied benefits ................................................................... 45
4.5 Inputs and implementation ................................................................................................ 45
5. Results and discussion............................................................................................................. 47
5.1 Power cut addition ............................................................................................................. 51

xi
 5.2 Availability factor addition (reliability constraints) .......................................................... 54
5.2.1 Microgrid and no microgrid differences with availability constraints ........................ 56
5.3 Uncertainty in load, renewable generation and electricity market price ........................... 57
5.3.1 Uncertainty in load demand ........................................................................................ 58
5.3.2 Uncertainty in renewable energy generation............................................................... 61
5.3.3 Load and renewable energy generation uncertainty comparison ................................ 62
5.3.4 Load and renewable energy generation uncertainties together ................................... 65
5.3.5 Uncertainty in electricity market price........................................................................ 66
5.3.6 Uncertainty in electricity market price without energy storage systems (ESSs) ........ 71
5.3.7 Comparison of results with the three different uncertainties ...................................... 72
5.3.8 Uncertainty in load, renewable energy and electricity market price altogether .......... 73
5.4 Microgrid and no microgrid comparison ........................................................................... 75
5.5 Variation in the power cut hours ....................................................................................... 77
5.6 Importance index .......................................................................................................... 78
6. Conclusions ............................................................................................................................. 81
7. Future work ............................................................................................................................. 85
8. References ............................................................................................................................... 87
Appendix 1 .................................................................................................................................. 93
Appendix 2 .................................................................................................................................. 99

xii
1. Introduction

1.1 Background and motivation

Microgrids have emerged as a new kind of grid in the power systems field, where
distributed energy resources (DERs) take a fundamental role; microgrids make possible
diversity in power sources and allow penetration of renewable energy. The complexity
of microgrids and the mix of elements which compose them make investment studies
especially important; errors in investing and sizing elements in microgrids may arise in
huge not needed costs because of the high price of microgrid components. Therefore
before starting a microgrid project it is crucial to make sure the profitability of it and
measure it. In that way the investment optimization lower the risk of implementing
microgrids.
One of the difficulties in this kind of studies within microgrids is the uncertainty.
Uncertainty affects the model in several aspects such as the load demand, the renewable
energy generated, the electricity market price, the availability factor of the distributed
energy resources and the islanded periods in the microgrid. This uncertainty is taken
into account and modeled in a stochastic way.

1.2 Aim, objectives and delimitation

The aim of this master thesis is to optimize the investment of a new microgrid under
uncertainty, where it is possible to choose among different types of DERs which are
divided in dispatchable generators, non dispatchable generators and energy storage
systems. Different devices are available in these different DERs categories too.
The investment optimization is solved in a long term timescale but the optimization
problem also checks and optimizes the short term microgrid operation. It makes
possible, as explained before, to evaluate the profitability of the microgrid before
starting to implement it. The long time and large quantity of computer resources that
the problem optimization takes when it is solved as a normal mixed-integer
programming (MIP) resolution solved by branch and bound makes necessary the
introduction of algebraic algorithms to solve the problem. In this case Benders’
Decomposition is the one selected to cope with this task. It is important to remark that
Benders’ Decomposition algorithm has been implemented successfully avoiding
duality; in that way it is easier to apply it in other microgrid models for further
application.

1
Once the optimization problem is completed it is studied how the uncertainty affects the
microgrid both in unit investments and in the financial profitability. The five different
types of uncertainty (load demanded, renewable energy generation, electricity market
price, availability of the distributed energy resources and islanded operation of the
microgrid) are studied individually and finally altogether to find which uncertainty
affects more the microgrid. This analysis of uncertainty is performed only to the data of
the Illinois Institute of Technology and it is not possible to generalize all the
conclusions obtained to other microgrids.
Finally and in order to visualize and measure the reliability improvement in the system
it is performed an analysis of the reliability importance via the interruption cost index
(hazard rate index).

1.3 Ethical aspects

There are two main factors which also have motivated this master thesis giving an
ethical background:
 The enhance of reliability that microgrids contribute with and the possibility of
implementing them in areas where the supply of electricity is difficult via main
grids make an improvement in the social welfare. It is especially important in
remote areas with difficulties to obtain a reliable energy supply definitely
needed to ensure a high level of quality life.
 The possibility to add renewable sustainable power, which reduces
emissions level and makes the system achieve a lower environmental cost.

1.4 Overview of the report

The rest of the report is divided in the following chapters:
Chapter 2 - Background: In this part and after doing a literature review on microgrids,
reliability and uncertainty from a theoretical point of view these three main topics are
explained.
Microgrids are defined, explaining the differences between them and regular grids, how
they are operated, the elements which compose them, the benefits of implementing
them, the different business models about how implement them and the different
policies which can affect them. Also the challenges related with microgrids are
mentioned in this chapter as well as some examples of microgrid projects which are
already accomplished, under construction and future ones which will be implemented.
Reliability is defined and the main ways of measuring it are explained.
Finally uncertainty is defined in general and more specifically in the role that it plays
within microgrids, as well as the differences between deterministic and stochastic
models handling uncertainty.
Chapter 3 - Literature Review: In this chapter different approaches to optimization
problems in microgrids are mentioned. They bring reasons about how to define the final
optimization problem.
Chapter 4 - Methodology: In this part the different elements of the microgrid are
described and their main characteristics are given. After that the formulation of the

2
problem is defined as a mixed integer problem and explained. It includes the objective
function of the optimization problem, the constraints which it is subject to and how the
uncertainty is modeled. Afterwards a brief description of GAMS (General Algebraic
Modeling Systems) is given, this is the software used to program the problem
formulated previously.
Benders’ Decomposition is defined in theory being applied to a general problem and
explaining its way of operation. Finally Benders’ Decomposition is applied to the mixed
integer initial problem formulation and their benefits are tested.
Chapter 5 - Results and Discussion: In this chapter the results of the different
simulations of the system are presented, compared and explained in the context of the
Illinois Institute of Technology data which is applied to the model. The five different
types of uncertainty (load demanded, renewable energy generation, electricity market
price, availability of the distributed energy resources and islanded operation of the
microgrid) are studied individually and finally altogether.
Reliability importance analysis is performed to measure the reliability improvement
before and after the microgrid implementation. This is done via the interruption cost
index .
Chapter 6 - Conclusion: The summary and the main characteristics based on the results
previously taken are gathered and presented in this chapter.
Chapter 7 - Future Work: Different suggestions about how to continue this work are
given here in order to make the model more realistic and closer to the industry for a
future use.

3
4
2. Background

2.1 Microgrids
Since the world’s first power system was built in 1881 in England, electric power
systems have been developed continuously, step by step [1].
Throughout these different steps, the technology used has been improved both in
complexity and features: from the rudimentary waterwheels used to generate alternating
current to the large variety of both DC and AC power systems that nowadays are
possible to find all over the world.
A new kind of grid has been developed as an emerging field: it is called microgrid. Due
to the novelty of the concept, there is not an accorded and agreed definition, especially
in terms of the power capacity, where boundaries are not clearly defined. Some
definitions are given below by well-known and reputed organizations:
The US Department of Energy defines a microgrid as “a group of interconnected
loads and distributed energy resources within clearly defined electrical boundaries that
act as a single controllable entity with respect to the grid. A microgrid can connect and
disconnect from the grid to enable it to operate in both grid-connected or island-mode”
[2].
CIGRÉ defines microgrids as “electricity distribution systems containing loads and
distributed energy resources, (such as distributed generators, storage devices, or
controllable loads) that can be operated in a controlled, coordinated way either while
connected to the main power network or while islanded” [3].

2.1.1 Microgrids and main grids

One important question that should come up when reading about microgrids is: what are
the differences between main grids and microgrids?
Bulk grid, main grid or macrogrid is understood as the dominant power system for more
than one century, with large power plants. These systems generate energy and send it to
the final customer via transmission lines. It can be simply defined as: “A complex
assemblage of equipment and circuits for generating, transmitting, transforming, and
distributing electrical energy” [4].
On the one hand main grids’ power is generated by large generators, and the power
which flow in them is unidirectional. It implies that they are easier to scale due to the
big size of the generators but at the same time the system upgrading is costly and not
scalable; areas far away from the generation can suffer reliability problems. Also

5
macrogrid power systems have been built for decades and as a consequence they have
well established standards and codes.
On the other hand microgrids imply a faster installation due to the smaller size of the
distributed energy resources they are composed of compared to the large generators of
the power stations in main grids. Microgrids allow better integration of renewable
energy and have bidirectional flow. But since the technology is in a premature stage
microgrids are costly and they imply technical, financial and legal challenges which are
presented in chapter 2.1.7. [5] [6]

2.1.2 Operation modes

Microgrids have two different ways of operation: parallel and islanded, which are also
called interconnected and emergency operation. For both operation modes there is a key
element, the switch gear at the point of common coupling (PCC), which links or isolates
the microgrid to the main grid. In a microgrid there is at least one PCC, but it is possible
to have more than one. [6]
 Parallel operation: It is the common operation mode of a microgrid. Microgrids
work in this mode the vast majority of the time. During this time microgrids are
coupled with macrogrids and it is possible to exchange energy between them via
a transmission line. As it will be seen later, it makes possible to the microgrid to
get income and be profitable in a financial way.

It works as follows: on the one hand during low electricity market price hours,
when the price of the energy in the main grid is cheaper than the cost of
generating electricity in the different distributed energy resources placed in the
microgrid, there is an import of energy from the main grid to the microgrid. It
implies a cost in the microgrid due to it has to pay for the energy supply. On the
other hand during high electricity market price hours, when the energy price in
the main grid is more expensive than the cost of generating electricity in the
distributed energy resources placed in the microgrid, there is an export of energy
from the microgrid to the main grid, therefore the microgrid gets revenues which
make it profitable.

 Islanded operation: In this configuration microgrids work alone, it is a critical

factor for safety and reliability. It only happens in special situations when main
grids are down due to maintenance, accidents or they are avoiding possible
problems when severe weather is forecasted. The isolation can be done
automatically or manually. In this case a suitable capacity of distributed energy
resources must ensure the electricity supply to the entire load, or at least to the
critical load.

Although benefits exporting and importing energy explained above are

important, the islanding mode is the main reason to develop microgrids. In that
way the energy supply is ensured under circumstances when the main grid
cannot supply energy. It implies two main advantages interconnected:

Firstly a financial one because when microgrids are supplying energy when they
work in islanded operation the cost of loss load is avoided; this cost must be paid
when there are loads non supplied and it is dependent on the Value of Loss Load
(VOLL). Secondly the benefit caused to customers when there is not load

6
 shedding; it is especially important when the shedding can cause special
damages for customers. Both reasons are related since the more important is the
load for the customer the higher the VOLL should be.

2.1.3 Elements

Microgrids are complex systems composed by many components interconnected.

In this section the most important parts of microgrids are named and the most related
elements with this master thesis are explained and commented in more detail in order to
get an overall sight of what a microgrid is:
The most important elements are the distributed energy resources, the energy storage
systems, the controllable loads, the switchgear previously explained in section 2.1.2 and
the microgrid controls.
 Distributed energy resources (DERs): They are small energy resources which
provide local power supply. Adequate and varied generation in microgrids is
essential in order to help production in microgrids. Both renewable and non-
renewable DERs are desirable because microgrids need to be operative in cases
of non-favorable conditions. These conditions are varied and diverse: from bad
meteorology which could affect solar and wind power to high prices of oil and
gas which could affect combined heat and power units, engines, turbines and
other types of non-renewable units. The variety in generation also makes
possible to supply energy when there are outages in DERs, outages could
happen due to production errors; therefore having only one type of generators
could cause dramatic consequences for microgrids. It is possible to differ DERs
between dispatchable and non dispatchable:

1. Dispatchable: DERs which are available and controllable. It is possible

to modify their production when it is needed; i.e.: Diesel/gas/coal
generators, combined heat and power (CHP) and biomass generation
units. They also have restrictions as maximum/minimum capacity,
ramping rates, start up and shut down costs, fuel or emission limits.

2. Non dispatchable: DERs which are neither always available nor

controllable. They depend on external factors like weather conditions
and produce intermittent output power; i.e.: wind turbines and
photovoltaic panels.
Some of the most extended DERs in microgrids are combined heat and power
and Renewable Energy Resources (RES):
Combined heat and power DERs are interesting because they are capable to
meet both heating and electrical power in the same region, which can be
extremely useful in areas which are not still connected to utility lines. These
areas are also a potential future market for microgrids because they are usually
far away from the large generation areas and can suffer reliability problems.
Also efficiency in combined heat and power DERs is improved using the waste
heat produced in electricity generation [7].
Wind turbines and photovoltaic panels are examples of renewable energy
resources. On the one hand they have the advantages of that operational costs are
very low and can be assumed to be 0 and that they do not produce

7
 emissions. But on the other hand they are not controllable so there is always
some degree of uncertainty with them. This is not desirable especially when
microgrids are working in islanded mode and they must supply all the energy
that is demanded by the loads.
During the last years solar and wind power have become more interesting due to
their cost effective improvement, which also has made more profitable to
include them in microgrids. But it has made more difficult to manage the control
and performance in microgrids due to their dependence on the weather
conditions and the uncertainty that this fact introduces. Fluctuations in
renewable energy DERs may produce load shedding.
It is important not to confuse renewable with non dispatchable, as an example
biomass energy is a renewable and dispatchable kind of energy at the same time.
Also it is important to not mismatch DERs and microgrids: To be considered a
microgrid, group of DERs must have clear boundaries, a master controller and
the generation must be higher than the critical peak load [8].
 Storage: Energy storage systems (ESSs) are essential components in microgrids.
The use of energy storage systems improves microgrid benefits and helps to
solve many problems in the system such [9], [10]:

1. Reducing peak generation capacity: Adding ESS capacity to existent

DERs can reduce the investment of adding more units. Without ESSs
these additional units would be needed to supply the amount of energy
required at peak times. ESSs also help the system to have a better
performance when there is load-shifting.

2. Maintaining voltage and frequency stability: ESSs can eliminate these

power irregularities. This makes the power system safer, improves the
electricity quality and helps customers whose loads may be damaged
with variations in voltage and frequency.

3. Improving the control of the microgrid.

4. Mitigating and smoothing the intermittency at non-dispatchable

generators: This is done by ESSs storing energy at peak production times
and releasing it at valley times.

5. Back-up power use: ESSs can act during short periods of time as back-up
generation when utility power is not available.

6. Supporting transmission and avoiding congestion: ESSs can store energy

at congestion times improving lines’ transmission and behavior.

7. Lowering emissions: This can be important in areas where there is a

restriction of greenhouse effect gases.

8. ESSs also intervene in storing energy when the market price is low and
leaving it when the price is high taking a fundamental role in the
microgrid performance and being essential to achieve revenues.

8
 There are different types of storage in a microgrid. The most important ones are
[11]:
1. Power storage: As power is understood the amount of energy change and
it is measured in megawatts (MW). It is particularly important for short term
applications like angular and voltage stability, where power must be injected
or absorbed.

2. Energy storage: As energy is understood quantity or volume and it is

measured in megawatt-hours (MWh). It becomes important in applications
which require a long duration load shift, where energy cost is more relevant
than power cost.

The use of ESSs has increased during the last years because although the
technology is still under development it is reaching an important degree of
maturity and prices are decreasing and being affordable for investments. But it is
important to size systems accurately to avoid extra investment in elements with
high capital costs.

 Controllable loads: Load management is essential in microgrids, especially in

islanded mode, when under some circumstances it could not be possible to
guarantee the supply of all of them. Therefore load control is needed. Loads are
divided in:

1. Vital loads: cannot be interrupted.

2. Essential loads: must be connected in seconds
3. Priority loads: must be connected in minutes.
4. Non-priority loads: can be disconnected for a long period.
Number 1 is also called fixed loads and numbers 2, 3 and 4 are called flexible
loads, they depend on economic interests and islanded requirements to be
disconnected [8]. These loads are associated with the value of lost load (VOLL)
which is higher when the load is more important.
 Microgrid controls and protections: There are vital elements enclosed here like
the microgrid central controller (one per microgrid) which is in charge of
connecting and disconnecting the microgrid based on economic, reliability and
security criteria; the microgenerator controller (one per microgenerator), the load
controller (one per consumer), the switchgear main connection, LV protection
devices at each LV branch and a communication system to link all of them.

2.1.4 Benefits and costs

Before starting deploying a microgrid it is important to assess the project both in

technical and financial ways. There must be always more advantages than drawbacks
before starting a project. The most important benefits and costs are now named and
briefly explained to get a global sight about why microgrids are important and the future
role that they will play in the society [12] [6] [13] [14]:
Benefits:
 Reliability: As explained before, microgrids are capable to work in parallel or
islanded from the main grid. This characteristic improves reliability making

9
 possible to disconnect the microgrid when there is a fault or any problem in the
main grid. In that way, generators in microgrids are capable to match the entire
load or at least a part of it, trying to ensure electricity supply to the most
important loads. It benefits both consumers and distributors because the first
ones do not suffer outages and the second ones do not have to pay for the
amount of load that would not be served if the microgrid was not implemented.
It is also important not to confuse microgrids with backup generation, which is
also used to supply energy when there are outages in the main grid.

Developing microgrids ensures critical load supply due to on-site generation. It

can be a strategic tool and it is one of the reasons why different countries and
organizations support the development of microgrids. Some examples of areas
where microgrids are important in terms of security are government areas,
hospitals, strategic industries or military bases. It is also important to say that
although microgrids can be developed with any kind of load, due to the high
investment cost they are mostly deployed in places like the ones mentioned
before. These places usually have vital loads which cannot be interrupted. On
the other hand residential customer loads are usually less important and
microgrid investments may not be profitable.

 Efficiency: On-site generation reduces losses in transmission lines. This

connection between microgrids and efficiency makes (as it was seen in section
2.1.3) combined heat and power generators one of the most implemented
distributed energy resources. Combined heat and power units’ main
characteristic is their good performance and efficiency improvement in areas
where both thermal and electric power are needed.

 Economic savings: Taking into account that most of microgrids have as

distributed generators (DGs) both combined heat and power and renewable
generators as wind turbines and photovoltaic panels. It is possible to sell to the
main grid the excess of power produced by them especially when there are
favorable weather conditions and the electricity market price is high. Moreover,
the existence of energy storage systems make possible to store energy when the
market prices is low and sell it when the price is high. This is one of the most
important ways to get profits by deploying this kind of systems. Although the
investment in these energy sources is high, due to their operation cost is lower
than conventional generators it is possible to make them economically profitable
exchanging energy between macro and micro grids.

 Sustainability: As commented previously in section 2.1.1, microgrids allow

better integration of renewable energy, this fact helps to reduce fuel
consumption and the carbon dioxide footprint. In [15] it is reflected how a
microgrid installed at Illinois Institute of Technology reduces up to 50% the
demand of energy and 20% the peak load.

Although a possible benefit of microgrids implementation could be the reduction

of carbon dioxide due to the addition of renewable generation, it cannot be
assured. The intermittency of the renewable wind and solar energy production
affects the main grid. If that generation is small compared with all the main grid
power (10-30% of the total power [6]) there is no problem because the main grid

10
 absorbs the intermittencies. But if the amount of renewable power is big
compared with the main grid power (>30%), it may cause instabilities in the
system. These instabilities cause that in the end it is not possible to use fully the
renewable energy generators’ capacity and that balancing resources as
conventional fuel or coal generators must be used to reduce the instabilities,
increasing in this way the emissions. [6]
Costs:
It is possible to divide the cost of microgrids in five different groups:
 Energy Resources (30-40%): Energy storage systems, distributed generators
and controllable loads.
 Switchgear, protections and transformers (20%): point of common coupling
switch, protection studies and devices.
 Communications and controls (10-20%): Systems like SCADA, real time
signal measuring systems and controllers, power electronics…
 Site engineering and construction (30%)
 Maintenance and operation cost.

2.1.5 Business models in microgrids

Ownership and operation in microgrids are controversial due to the developing state of
the technology and the non-extended deployment. There is not a common ownership
standard and depending on the ownership and operation microgrids are divided in [6]:
 Distribution company microgrids

The main advantage in this case is that both the macro and microgrid have the
same owner, which favors the coordination, control and integration. Also it is
the distribution owner company the one who is in charge of keeping the stability
in the microgrid. The reduced number of actors in the operation and control of
the grids reduces the challenge difficulty. Examples of distribution company
microgrids are the Consortium for Electric Reliability Technology Solutions
(CERTS) demonstration project in Ohio, owned by American Electric Power11
and the Borrego Springs microgrid owned by San Diego Gas & Electric.
 Single-user microgrids

In this case the single user is in control of its own expectations and performance
with respect to microgrid stability and critical load support capability. The more
complex the microgrid is the more difficult the stability maintenance is and a
higher level of expertise is required. It leads to simple microgrid configurations
which are usually composed by combined heat and power units. Even the
control of these simple microgrids may result in a difficulty for the user and a
lack of expertise could present an operational risk. For this reason this
ownership model is less developed than the other two models presented in this
section. An example of simple-user microgrid is Santa Rita Jail Microgrid in
California.
 Hybrid microgrids

This business model is a mix between the two ones explained before. Distributed
energy resources are owned by single users and the distribution company keeps

11
 the operation and maintains the stability in the microgrid. Therefore it combines
the best aspects of distribution company and single-user microgrids.
This method is profitable for both parts because the private owner is able to
make investments without the need of knowledge in controlling and operating
microgrids and the distribution operator is interested in a high number and
variation of DERs to consolidate the microgrid. On the other hand the more
complex the microgrid is the more difficult its control is.

2.1.6 Policies

Since the first electrical grid regulation was published electrical grids have been
developed and regulated reaching a wide and specific legal framework. Due to the
novelty of microgrid concept, it is necessary to create a new regulation framework as it
was done with common grids.
Many governments have proposed to promote microgrids to make them more
interesting for investors because of the benefits mentioned in section 2.1.4. Below there
are explained different policies adopted by regional or national governments to promote
microgrids [16] [17] [18]:
 Microgrids are granted with public funds which try to help renewable energy
installation and which come from taxes in electricity consumption. It is also
common that subsidies are applied to low emitting power sources and taxes to
high emitting power sources.

 The law applies emission limits also called emission rights which favor efficient
and low pollution systems like microgrids; in this case pollution emitted by
power producers is constrained under a certain amount.

 Certificates: Renewable energy sources receive certificates according with their

production level. Then these certificates are traded in a parallel market where
consumers are obliged to buy a number of them corresponding to a certain
consumption amount.

 Net energy metering: It produces benefits because prosumers (consumers that

also generate electricity) are able to sell the excess of electricity generated. It
requires accurate metering of the electricity consumption. An accurate metering
is extremely important in microgrids in order to know what amount of energy is
generated by distributed energy resources and what amount comes from the
main grid or big generators.

 Electricity disclosure: In this model there are different markets so with this
system consumers know the origin of the energy. Consumers are able to choose
the supplier that they want depending on their preferences, being able to choose
depending on the price, if the energy is renewable or not, etc.

 Feed in tariffs is another economic model which requires long term payment-
contracts for the energy produced by renewable generators in order to accelerate
and make the investment in them easier. This policy is more beneficial for
renewable energy with lower prices like wind power rather than for new
renewable sources like wave energy because their cost is higher therefore their

12
 profit is lower. An example of feed in tariffs payment in Spain is shown in table
1:

Energy Source Feed-in Tariff

Biomass and biogas up to 13.06 c€/kWh for the first 15 years

Cogeneration systems maximum FiT of 13.29c€/kWh during lifetime of system.

Geothermal, wave, tidal and sea-thermal 6.89 c€/kWh for the first 20 years

Hydroelectric 7.8 c€/kWh for the first 25 years

Solar thermoelectric 26.94 c€/kWh for the first 25 years

Waste combustion up to 12.57 c€/kWh for the first 15 years

Wind systems up to 7.32 c€/kWh for the first 20 years

Table 1 Example of feed in tariff policy in Spain [19]

 Feebate system: It is a portmanteau of fee and rebate. In this system high-

pollution generators compensate economically low-pollution generators favoring
the acquisition and use of no emission generators.

 “Based on property” energy policies make that grants and economic advantages
given by public institutions are given to the owners of green energy properties
instead of the final tenants/users of these properties. It is important in properties
whose final destination is the rent to other users. In that way the owners try to
make an effort because they are the ones who are benefited by the installation of
green energy generators.

2.1.7 Challenges

As a new concept, microgrids are surrounded by multiple challenges, both in technical,

economical and legal aspects [5]. The reduced number of elements that a microgrid
possesses makes difficult to keep the stability in the grid, it is especially difficult when
microgrids work in islanded mode. In parallel mode the macrogrid supports the stability
of the microgrid making the challenge easier due to the large number of elements that
compose it. The main challenges in a microgrid are enumerated below:
 Voltage Regulation: The fact that microgrids possess high impedance, a non-
radial source-demand configuration and multiple diverse sources and loads
with a wide range of dynamic characteristics implies a difficulty in the
voltage system regulation. Voltage regulation is essential because it can
cause important damages in the system; i.e. reduced voltages may produce
large currents that cause larger losses which may imply the overheating of
parts of the system and reduce its life time or cause its destruction. [20]

 Frequency Regulation: Opposite to large generators which are able to

manage the frequency stability due to their large inertia, microgrids with
small distributed generators have small inertia and changes in load might
cause frequency instabilities. This problem, like the voltage regulation

13
 described before becomes a challenge when microgrids work in islanded
mode and they have to maintain without any help from the main grid both
voltage and frequency. [21]

 Increased harmonics: The two challenges explained before can derive for
example in magnetic saturation for motors and transformers which can also
cause prejudicial harmonics when there is a frequency lower than the
expected or a voltage higher than the rated one. Harmonics may also cause
overheating, current redistribution producing electromagnetic torques and
what is worse loss of synchronization that could lead into a cascading
blackout with dramatic consequences. [20]

 Increased short circuit levels: It is caused by the fact that in microgrids there
are multiple sources close to the loads and that there is bidirectional power
flow. Most of conventional feeder protections are based on short circuit
current sensing but micro generators cannot provide high enough levels of
fault current. [22]

 Coordination of protection devices and sensitivity: as commented before,

microgrid protections are a challenge, they have to avoid short circuits and
faults in the system. But the coordination of these protections in order to
make them functional both for the microgrid and macrogrid is an added
challenge. Protection systems have to isolate the microgrid from the main
grid as fast as possible and also have to isolate parts inside the microgrid
when there is a fault in it. [23]

 Impact from high penetration: Protections against faults impede heavy

penetration of distributed generation and vice versa, multiple and complex
sources make microgrids difficult to model and design. [23]

 Islanded control: When islanding the microgrid from the main grid the large
number of distributed generators makes difficult the system synchronization
when the islanded mode starts. Active and reactive power must be
transmitted while keeping the stability of the system. Micro generator
production should be optimized and load must be classified in a hierarchical
way to supply from the most important to the less in case of failure. [24]

2.1.8 Microgrid projects

Microgrids, as it has been commented before, are new systems with huge future
potential and market opportunity all over the world. Different studies has predicted and
confirmed the fast growing both in number and total capacity.
Past studies measured that in 2012 the total amount of power installed in microgrids
was 1.1 GW but this amount was supposed to increase until 4.7 GW in 2017; that in the

14
moment meant a market opportunity estimated in 17.3$ billions [25]. Future studies
have confirmed that and even have measured a larger power capacity in the total
developed microgrids. The Navigant research [26] identified 4393 MW of power
capacity in microgrids in the second quarter of 2014 and 12031 MW in the second
quarter of 2015, which makes the current capacity higher than the estimated years ago.
The increase has almost triplicated the capacity in one year. Asia is the continent which
leads the data collected from projects and portfolios with the 47% of the capacity,
followed by North America, whose capacity sums 44% of the total. However North
America is still the continent which is leading the market share with the 66%.
Due to nowadays North America is the area leading the microgrid market share it is
taken as reference to show the current microgrid market state. Only in The United States
the increase in microgrids capacity is expected from 1051 MW in 2014 to 1843 MW by
2017 and to 2855 MW by 2020. The market value will reach $671 million by 2017 from
$133 million in 2014 and the cumulative value of the microgrid market, including
investments from prior years, will exceed $3 billion by 2018. [27]
Usually military bases and isolated communities have implemented microgrids but
lastly also public institutions like hospitals or universities and cities are deploying new
microgrids [27]. Although large institutions will be the ones which invest more money
in new microgrids, the cheaper the technology becomes the more commercial customer
will invest in them. If we take a look to the planned microgrids in US in 2015
commercial ones are leading among the different groups.

Developed microgrids in 2014 (1169 MW)

University/Research
3% 3% 3% Facility
Military
8%
40% City/Community
8%
Public Institution

Commercial
35%
Remote community

Island

Figure 1 Distribution of developed microgrids’ capacity by sector in U.S. Source: GTM [28]

15
 Planned Microgrids (779 MW)

1% 2% 4% University/Research
Facility
Military
22%
40%
City/Community
12%
Public Institution
19%
Commercial

Remote community

Island

Figure 2 Distribution of planned microgrids’ capacity by sector in U.S. Source: GTM [28]

Regarding the size of microgrids, in US more than the half of the microgrids has a
capacity lower than 1MW. Another important fact is that by now renewable energy in
microgrids deployed in US does not take an important role, taking only the 6% of the
total capacity among generators and energy storage systems. In the following years an
increase of the capacity share of the renewable generators is predicted reaching a 26%
of microgrids capacity in 2020. [28]

Figure 3 Capacity share in microgrids. Source: GMT [28]

Below is going to be presented the state of the art of projects already finished, projects
under construction and future planned projects with some examples all over the world.
[29]
 Projects already accomplished [30] [31]:

-Illinois Institute of Technology (IIT): It is the microgrid where the data is taken
to perform the study of this thesis. Power outages in the grid were the cause of
developing a microgrid due to the large costs they implied to the campus
(around $500000 annually). The microgrid has been granted with $12 million

16
 for five years and it provides a full islanding to the campus load. The campus is
provided with two on site combined cycle gas units (4 MW each), a small wind
turbine of 8 KW capacity, 160 KW of PV cells which are going to be increased
to 1.3 MW and an energy storage system of 500 KWh. As a consequence the
grid electricity demand has decreased 50% and the peak demand has decreased
20% since 2007. The system produces $1,3 million savings annually.

-Santa Rita Jail: Developed in 2012, it is a correctional facility with 4000

inmates and one of the best demonstrations of successful microgrids. It has a
peak demand of 3 MW of reliable and secure electricity. It is composed by 1.2
MW of photovoltaic power, 1 MW molten carbonate fuel cell, five 2.3 KW wind
turbines and an energy storage system of 2 MW- 4 MWh lithium-ion battery.

-Sendai Microgrid: It is probably the most well-known microgrid all over the
world which is fully operational. It is composed by two 350 KW natural gas
fired generators, 50 KW of solar photovoltaic power and a fuel cell of 250 KW.
The microgrid became notorious after the great behavior performed after the
earthquake and tsunami which took place in the region. The microgrid also
possesses six varying levels of power quality.

-Borrego Springs: This microgrid is an example of implementation in residential

communities. Borrego Springs is a desert community of 2800 inhabitants. In this
microgrid the utility is the owner of the distribution assets but some of the
distributed energy resources are property of the customers. The total capacity of
the microgrid is 4 MW, which is supplied by two 1.8 MW diesel generator units,
one battery of 500 KW-1500 KWh , three small 50 KWh batteries, 700 KW
solar photovoltaic power and six 4KW-8KWh home energy storage systems.

-University of California San Diego (UCSD): It is another example of

microgrids at universities and one of the most advanced in the world. It supplies
electricity to a campus with 45000 students. The components of the microgrid
are two gas turbines of 13.5 MW each, one steam turbine of 3 MW and 1,2 MW
solar cell installation. The microgrid is able to supply the 85% of electricity
needs, the 95% of heating and the 95% of cooling. There is also a 2.8 MW
molten carbonate fuel cell and a 14385 capacity thermal storage bank.

-Bornholm Island: It is a Danish emplacement with 1% of national electricity

consumption. The island has 28000 inhabitants and a peak load of 63 MW with
a high penetration of wind power generation (30.2% of energy consumed in
2007). It is one of the most important microgrids being tested in Europe. The
different generators in the island are: 14 diesel generators (34 MW), 1 oil fired
turbine (25 MW), 1 steam turbine (37 MW) 35 wind turbines (29 MW) and 2
biogas turbines (2 MW).

 Projects under construction and future projects:

-Niobara Energy Park, Colorado: It is the largest planned microgrid, with

capacity to supply energy for 52 data centers. It is composed by a 200 MW gas
fired plant, 50 MW of solar farms, 50 MW of solar cells and a large range of

17
 energy storage technologies like compressed air, batteries, fly wheels, thermal
and hydrogen storage, supercapacitors and super conductors. [32]

-Ha Kei School, Hawaii: Ha Kei is a project under development started in 2008
which includes the installation of 100 MW of solar PV and 25 MW of wind
power on microgrids. The goal is to achieve 70% of clean energy by 2030. The
project is carried by the U.S. State of Hawaii and Chevron Energy solutions. It
has started with three schools and the objective is that all the 225 school of the
state will be equipped with renewable generation. [33]

-Babcock Ranch: This city in Florida wants to be the first one powered by solar
power. In order to do that a 75 MW solar power plant will be built by Florida
Power & Light Company (FPL) working as microgrid connected to the FPL’s
main grid. It will be capable to avoid 61000 tons of . [34]

-Fort Carson: It is an example of a military base microgrid. The plan is to build

over 100 MW of PV power, wind, heat pumps and solar and water heating. By
now the microgrid is deployed with 1 MW of PV power, three diesel generators
with a total amount of 3 MW and 5 electric vehicles with V2G capacity. The
microgrid was operating successfully during a three days attempt in 2013. [29]

2.2 Reliability

2.2.1 Definition

Reliability is a wide spread concept, which covers many different areas from science
materials to medicine passing by transportation or manufacturing. This term refers to
the system ability to perform an assigned function [35] [36] [37] [38].
Reliability in power systems could be defined as the ability to provide an adequate
supply of electricity. There are fundamentally two aspects regarding with reliability in
power systems: system security and adequacy.
 System security is the part of reliability related with the behavior of systems
when disturbances appear. Disturbances can be caused by generator failures,
transmission line outages or unexpected electricity consumption.

 Adequacy is the part of the reliability related with the number of facilities
needed to ensure electricity supply within systems. It includes both generation,
transmission and distribution layers

Reliability is not only important for final customers; both the transmission operator and
the generation company have interests. These interests arouse because while customers
do not want their supply affected, generation and transmission operators do not want to
stop that supply because it implies a payment for affected customers and also implies
that there is something wrong in the system that needs reparation. This reparation in
some cases can be extremely expensive. For this reason companies do periodic
maintenance in their equipment, especially in the most critical one.

18
2.2.2 How to measure

There is a huge variety of indexes to measure reliability. It is possible to differentiate

them in indexes based on load and generation levels and in indexes based on customers
[39] [40] [41]:
Indexes based on load and generation levels:
 Loss of load probability (LOLP): It is one of the most typical indexes. It shows
the probability when the demand is higher than the capacity. This index does not
give any information about how much is the difference between capacity and
demand. It is usually measured in hours/year or days/year. The formula of LOLP
is presented below:

∑ (1)

Where:
t=time
NT= Total number of time units in the study period (minutes, hours or days).
C(t)=Total generation capacity.
L(t)= Total system load.
=0 If C(t)-L(t)≥0
=1 If C(t)-L(t)≤0
 Loss of load expectation (LOLE): It describes the average time when generation
is lower than the demand. It is also measured in hours or days per year. In many
cases it is used to simulate penalties for electricity interruptions. This index also
does not give any information about how much is the difference between
capacity and demand. It is more commonly used than LOLP.

( 2)

Where:
T = 365 days (if the load model is a daily load curve throughout a year; the
LOLE unit is defined in days per year)
T = 8760 hours (if the load model is an hourly load curve throughout a year; the
LOLE unit is defined in hours per year)
 Expected energy not supplied (EENS): It is also called loss of energy
expectation (LOEE) or expected unserved energy (EUE). This index, rather than
the two explained above, gives an amount of the energy not supplied to the
customers when the load is higher than the capacity. EENS calculation
corresponds to the following formula:

19
 ∫( ( ) ( )) ( 3)

Where:
t=time
NT= Total number of time units in the study period (minutes, hours or days).
C(t)=Total generation capacity.
L(t)= Total system load.
=0 If C(t)-L(t)≥0
=1 If C(t)-L(t)≤0
 Loss of load frequency (LOLF): It is measured in occasions per year.

∑( ) ( 4)

Where:
i=Departing system state.
=Frequency departing system state.
=Portion of which corresponds to not going through the boundary wall
between the loss-of-load state set and the no loss-of-load state set.
 Loss of load duration (LOLD): It is defined by the division between LOLE and
LOLF.

( 5)

 Energy Index of Reliability (EIR): It is defined as the normalized value of

EENS.

Indexes based on customers:

 System Average Interruption Duration Index (SAIDI): This index measures the
total interruption duration per average customer in a defined period of time. It is
calculated dividing all the interrupted time by the total number of customers
served.

( 6)

20
 System Average Interruption Frequency Index (SAIFI): This index shows the
average number of interruptions that customers suffer in a defined period of
time.

( 7)

 Customer Average Interruption Duration Index (CAIDI): This index is very

similar to SAIDI, but it is based only on the interrupted customers and not in the
total number of served customers.

( 8)

 Customer Average Interruption Frequency Index (CAIFI): This index is very

similar to SAIFI, but it is based only on the interrupted customers and not in the
total number of served customers.

( 9)

 Average Service Availability Index (ASAI): It is defined as the ratio between the
time when customers have electricity supply and the time when they demand
electricity supply.

( 10)

 Average Service Unavailability Index (ASUI): The opposite of ASAI, it is the

time ratio when customers do not have electricity supply.

( 11)

 Momentary Average Interruption Frequency Index (MAIFI): This index is not

commonly used, and it is similar to SAIFI, showing in this case the momentary
interruptions per average customer.

( 12)

21
2.2.3 Value of loss of load [42] [43] [44]

Value of loss load (VOLL) is an essential factor to take into account modeling
optimization problems in microgrids. It is defined as the representation of the value that
electricity consumers give to security of the power supply.
The application of VOLL varies from a planning side to an operational side. In the
planning side VOLL is used mostly to create business plans and to analyze benefits and
costs of investments in power systems. The operation side of VOLL can be used as
criteria of generating adequacy and as a pricing tool.
VOLL is measured in $/MWh and there are two variations:
 Value of loss of load based on willingness to accept (WTA): It is the amount of
money that electricity users would admit (paid for the company in charge) as
compensation if there was a failure in the electricity supply.

 Value of loss of load based on willingness to pay (WTP): It is the amount of

money that electricity users would pay in order to have electricity supply when
there is a failure in the electricity supply.

WTA tends to be higher than WTP due to consumers, especially residential and small
ones, are typically not willing to pay for improving the service. For this reason WTA is
more used in terms of security supply.

VOLL can be marginal or average. Marginal VOLL is used for specific time points and
it is usually calculated for peak or worst case possible periods. Average VOLL is not
differentiated over time and it is used for long periods of time. It usually tends to be
lower than marginal VOLL.
In numerous references VOLL is used to evaluate costs when there is load shedding in
the system, this is the case of [45] [46] [14] [47]. These evaluations usually imply
VOLL simplifications; however VOLL is a complex index which depends on many
factors:
 Area: VOLL depends on each country and the activities which take place in it. It
means that VOLL cannot be generalized among different countries although
they seem to have similar characteristics. However developed countries tend to
have a higher VOLL than countries under development because of the high
dependence on electricity.

 Customer: VOLL differs considerably among customer classes. These classes

can be classified primarily in residential, small and medium commercial and
industrial customers. Secondarily it is possible to narrow down in sectors like
agriculture, mining, construction, transportation, etc.

22
 Small Large
Average
Residential Commercial & Commercial &
Country VOLL
(US$/kWh) Industrial Industrial
(US$/kWh)
(US$/kWh) (US$/kWh)

New Zealand 41,269 11,341 30,874 77,687

Republic of
9,538 17.976 3,302 10,272
Ireland

Austria 6,006 1,544

Netherlands 5,13 23,24

Table 2 Example of VOLL by country and customer [42]

In the table 2 is possible to see what it is explained in the two first bullet points
and see how these developed countries have similar characteristics but the
VOLL changes depending on customer classes. New Zealand and Austria have a
higher VOLL for non-residential than residential customers and in Ireland and
the Netherlands the opposite happens.

 Time dependence: The season, the day of the week and the time of the day affect
the VOLL. VOLL tends to be higher at weekends and evenings for residential
customers and it has the opposite behavior for commercial and industrial
customers where VOLL reaches the maximum at weekdays and daytime.
Seasons modify the need of electricity which also modifies VOLL; this factor
has different results depending on the meteorology of the country or region.

 Advance warning interruptions: VOLL changes considerably whether there is a

warning of the outage or there is not. When outages are known in advance
VOLL is lower because customers are able to make adjustments.

 Frequency of the load shedding: VOLL is higher as higher the frequency is

considering that outage durations do not vary. The relation may differ depending
on customers and duration.

 Duration of the load shedding: VOLL is affected by outages’ duration. In

general one hour outage has different consequences than six different outages of
10 minutes each. How this affects VOLL is determined mainly by the customer
and the type of load that is in use. Also the longer the outage is the higher the
VOLL is if the frequency does not change.

 Type of load: Different loads have different VOLLs depending on their

importance as it was explained in the chapter 2.1.3.

There are four different ways to calculate VOLL:

 Revealed Preference Method: The VOLL is estimated based on the behavior of

customers in the past. This information is gathered via observations.

23
  Stated Preference Method: The VOLL is estimated based on surveys and
interviews to customers where they are asked about hypothetical situations in the
future.
 Macroeconomic Analysis: The VOLL is calculated here estimating the value of
loss of production for non-residential customers or the leisure time for
residential customers. It is possible to use the following formulas:
o VOLL= Annual Gross Value Added/Annual Consumption
o VOLL=Annual Electricity Bill/Annual Consumption
 Case study: The VOLL in this approach is calculated using data from previous
outages and it is usually used to check the accuracy of the previous methods.

2.3 Uncertainty

2.3.1 Definition

Uncertainty is a spread multidisciplinary concept, which implies imperfect and/or

unknown information for future events, data, measurements or just simply the unknown.
It can also be defined as the impossibility to describe the existing or future state. It is
possible to find this concept in countless areas of knowledge like statistics, physics,
engineering, economics or actuarial sciences among others. [48]
The most common way to measure uncertainty is giving a margin of error based on two
values, one higher and one lower than the measured value.

2.3.2 Uncertainty in microgrids

Microgrids combine aspects from electric power systems and electricity markets, which
are fields plenty of uncertainty. In decision making problems within microgrids
uncertainty affects to parameters like electricity market price, demand, production,
equipment availability and microgrid isolation. There are even not mathematically
tractable uncertainties as uncertainty in laws and regulations. [49]

2.3.3 Deterministic and stochastic models

Taking into account the uncertainty it is possible to separate models among

deterministic and stochastic ones [50].
As deterministic models are defined models whose output is completely determined by
the input parameters. Deterministic models can be seen as an approximation of reality,
these models can be static or dynamic. Specific inputs always lead to a concrete output
without depending on the number of times that the problem is solved. In a deterministic
model the parameters are known or assumed. Depending on the time variation a
deterministic process can be separated in:
 Steady state: These models do not have temporal variations.

24
  Time series of steady state: These models are steady state models that are run
along a time series with input parameters.

 Dynamic: These models describe space-time variations. They are usually used to
prognosis and forecasting studies.

Stochastic models are the ones which are able to handle uncertainty having inherent
randomness and building uncertainty inputs per se. In these systems same initial
conditions result in different solutions, sometimes stochastic problems are run several
times in order to get probability distributions. A deterministic model can be considered
as a specific case of a stochastic one.
There are different types of stochastic process; examples of them are Poisson, Markov
or Brownian processes.
As it can be extracted from the definition stated above, a deterministic model does not
mean that it does not handle uncertainty because the uncertainty can be added
externally.
One of the most extended ways to create random numbers in simulations is the Monte
Carlo method. Monte Carlo Simulation is a non-deterministic method used to
approximate complex mathematic expressions. The method is based on repeated
random samplings to obtain numerical results and it is used in a huge range of problems
especially related with optimization, numerical integration and probability distribution.
Using the law of the large numbers Monte Carlo method runs one model a lot of times
generating random inputs in a deterministic model, aggregating afterwards the obtained
results. After different samples, the results are gathered (depending on the input the
result they have a different output) and then it is possible to determine a range of
possibilities.

25
26
3. Literature review

As it was described in the introduction the main aim of this master thesis is to choose
and size a group of different distributed generators in order to optimize the investment
cost in a long-term planning under uncertainty. These distributed generators supply
enough energy to a certain amount of load, this supply is also optimized in a short term
planning. The microgrid is composed by wind turbines, photovoltaic panels,
dispatchable generators and storage systems.
A vast literature review has been read before starting to write and develop the final
optimization problem. In this chapter many different methods to solve uncertainty
problems are referenced and briefly described. It is remarkable that although there are a
lot of references available in terms of uncertainty and reliability, it is difficult to find
papers referring these topics to microgrids both in long and short terms optimization
timeframes due to the novelty of the concept.
Among the optimization scientific papers which have been read, in [51] L. Guo and W.
Liu solve a stochastic optimization problem minimizing cost and emissions via a
stochastic chance-constrained programming for a microgrid composed by batteries, PV
panels, wind power and diesel generators. Markov process transition probability matrix
is used to implement the uncertainty in the problem and NSGA-II algorithm is adopted
to solve the problem. NSGA-II has been also applied in the first stage of a two-stage
combined heat and power multi-objective optimization problem [52] written by the
same authors. The objective of this paper is to optimize the net present cost and the
carbon dioxide emissions, but using mixed-integer linear programming (MILP) in the
second stage. The first stage optimizes the equipment investment (long term
optimization) and the second one the dispatch (short term optimization).
In [45] a model is implemented to solve the optimal generation mix in a microgrid,
being this reference one of the most important ones. The sources of uncertainties here
are the forecasted load, the microgrid islanding, the renewable generation, the market
price and the islanding connection. A robust optimization approach is done based in two
different scaled-time problems using Bender’s Decomposition method, method which is
explained in section 4.4 of this report.
The uncertainty reasons are:
 Wind and solar power have a high degree of uncertainty due to they do not
follow repetitive patterns during days and seasons.
 Market price forecasting offers uncertainty affected by the hourly supply,
demand (depending on the weather, special events, etc.).

27
  Regarding the load, it is divided by fixed and flexible load. The fixed one is
easy to forecast and the flexible one suffers a higher uncertainty due to the
variation in electricity market prices, weather conditions and consumers’
decisions.
 The last uncertain source is the islanding process. It happens when there is a
failure in the main grid or when there is some kind of maintenance process
which does not allow the supply to the microgrid area. The time of these
disturbances is highly unpredictable.

The same idea with small differences has been performed by the same author in [16]
and [47]. In [16] provisional microgrids are studied, they do not possess the islanding
capability and need coupled microgrids to work without connection to the main grid. In
[47] the study is performed in order to know which microgrid is more beneficial among
AC and DC microgrids. Both of them are optimization sizing problems with
uncertainty.
Also the same author uses Benders’ Decomposition in [53] to solve day ahead
microgrid optimal scheduling. The objective is to minimize the cost of local generation
and the energy bought from the main grid. In this case the master problem and the
subproblem are defined to optimize the grid coupled operation and the islanded
operation respectively.

In papers [39], [54] and [55] it is possible to find three similar optimization problems
about how to design microgrids enhancing reliability and supply security with
uncertainties in distributed generators. Tabu optimization is used in these researches.
The aim in these references is to study how to cluster a large distribution system into a
number of microgrids to facilitate both control and operation. A systematic strategy for
constructing reliable and supply-secure microgrids is developed. SAIFI and SAIDI
indexes are taken into account for reliability aspects and annually probability density
functions are created based on historical data.
Reference [9] optimizes an energy storage system (ESS) to minimize the investment via
mixed-integer programming (MIP) in a long term scale. This optimization problem can
lead to large money savings due to the high cost of these devices and it gives a deep
insight about the use of ESSs in microgrids and how to model them. Monte Carlo
simulation and scenario reduction is used to take into account random uncertainties.
Common MIP has been implemented and solved in multiple references. It is the case of
[14] where microgrids, generation units and transmission lines are co-optimized in two
different time scales, annual (long term) for the investment and daily (short term) for the
load and generation coupling. To do that the problem is divided in a planning problem
which includes investment and operation costs and an annual reliability subproblem
which calculates the expected energy not supplied. Monte Carlo simulation and scenario
reduction is applied to generate random outages.
One stage mixed-integer linear programming (MILP) optimization problem is solved in
[56] to optimize a microgrid operation planning. This kind of optimization problem is
effective when there are not long and short term variables in the problem. More day
ahead short term optimization problems in microgrids are [57], [58] and [59] which
solve energy management, capacity expansion, uncertainty analysis, payback period and
cash flow analysis. Sometimes they include two subproblems like [60], which schedules
a microgrid under uncertainty with two MILP short term stochastic problems, one for

28
the unit commitment and one for the economical dispatch. Unit commitment stochastic
short term problems are solved in [61], [62], [63], [64] and [65]; (in three last ones wind
power integration and stochasticity have a major importance).
There are more ways to solve and simulate stochastic problems. It is the case of Particle
Swarm Optimization (PSO), used in [66] and [67] to size distributed energy resources
under reliability uncertainty. This technique has been applied in many different areas in
power systems, and it uses swarm intelligence to find the solution of optimization
problems. Also reference [68] uses PSO and dynamic programming to maximize
economic benefits of deploying microgrids and PSO is used too to optimize the
operation problem of a microgrid under renewable power generation and load
uncertainties in [69] . There are varieties of PSO like Binary PSO used in [70].
HOMER (Hybrid Optimization Model for Multiple Energy Resources) is a simulation
tool designed by NREL (National Renewable Energy Laboratory) which allows
simulating, optimizing and making sensitive analysis of microgrids composed by
different distributed energy resources (DERs). This technique is used in [71] to perform
four different microgrid optimizations with different groups of DERs.
Another method to solve microgrid optimization planning with uncertainty is Analytical
Hierarchical Process (AHP), which is a useful technique to solve multi attribute
decision making problems. In [72] the first configuration of distributed generators is
obtained via HOMER software based on total cost plans and then uncertainties are
added to the problem; finally AHP selects the most convenient solution from
economical, technical and environmental viewpoints. [73] Also uses AHP to solve
strategic planning of distributed generators.
Microgrids can also be optimized with chaos algorithms as in [74].
Risk investment strategic problems are solved in [75], [76], [77] and [78]. They study
long term frame time problems and take into account uncertainty of generation and fuel
prices. They also study the volatility and flexibility of different parts of the system to
know if it is profitable to invest money.
In [79] a probabilistic approach for the energy and operation management of a
microgrid with renewable energy and under uncertainty in load, market price and
availability of renewable energy is modeled via 2m point estimate method.
In [80] a microgrid is designed under energy price and technology performance
uncertainties. The paper departs from a deterministic model and stochastic variables are
added via Monte Carlo simulation to create a probability density function; then these
variables are added to a deterministic model and simulations are repeated to get an
accurate probability. Islanded periods are also studied in the microgrid.
Regarding the concept explained in the section 2.2 there are different ways where
reliability indexes have been used in problem formulation.
As if, reliability in [82] is taken into account using loss of capacity (LOC), in [9] using
loss of load expectation (LOLE) and a curtailment indicator. The cost of unserved
energy, which derives from EENS, is used in [14] to reflect the load shedding effect for
reliability considerations minimizing a curtailment variable. In these references and in
several more like [45], [46], [14] and [47] VOLL have a tremendous importance in the
final result.
Another issue that is well embodied in the literature review is the stochastic
methodology. As commented before, one of the most famous methods to create

29
scenarios in a stochastic approach is the Monte Carlo Simulation, which creates a large
number of random scenarios taking into account probabilistic data; after this creation
the number of scenarios is usually reduced and each scenario has a specific assigned
probability. Monte Carlo simulation and scenario reduction have been implemented in
several references as [9], [14] and [80].
But there are more ways to model uncertain forecasts: Beta and Weibul probability
distribution functions (PDF) have been used in [55], [54] and [62] where four days of
different seasons have been taken to make the model of PV and wind turbines more
accurate. Another way to model uncertain forecasts is via ARMA (Autoregressive
Moving Average) model as it is used in [81] or ARIMA (Autoregressive Integrated
Moving Average) in [82].
Artificial Neural Network (ANN) is applied in [83] to forecast wind and solar power,
this method has a high adaptability to new conditions. It uses back propagation, learning
from the past information to minimize the error. The network is trained with historical
data, then forecasted data is introduced in the ANN and the program generates a new
forecast output. ANN can be applied with load forecast too.
More examples of modeling wind and solar power are the calculation of their respective
power via equations which include factors as wind speed and solar perpendicularity and
then apply equations depending on their mean value and standard deviation to calculate
the forecasted error as in [86] and [56].
In [84] both a stochastic and a deterministic part are used to optimize the management
of a microgrid. Firstly Probability Distribution Function and roulette wheel mechanism
are used to generate scenarios and afterwards (as in Monte Carlo simulation) a scenario
reduction is required. Then the stochastic variables are added to a deterministic problem
which is solved based on Adaptive Modified Firefly Algorithm (AMFA).
The paper [45] has been chosen as the main reference after reading all the other
references explained in this chapter. In our case the availability factor (which will be
explained in section 5.2) is added as a new uncertainty besides load demanded,
renewable energy generation, electricity market price and the islanding process. These
uncertainties are generated stochastically with different scenarios per year and
introduced in a deterministic problem where the optimal number of distributed energy
resources invested is solved for a long term planning. Also the short term planning
microgrid performance is solved showing us the power exchange between the microgrid
and the main grid, the load shedding of the system, when the distributed generators
previously implemented are working and when the energy storage systems are being
charged or discharged.
The range of uncertainty is modified and increased in different steps and the different
types of uncertainty are compared and analyzed in order to see how they do affect the
microgrid. VOLL is modeled to increase its value with the load shedding duration, thus
the longer is the load shedding duration the higher is the VOLL. Finally an importance
index comparison (hazard rate index) is performed in order to know how the
microgrid affects the reliability of the system.
Benders’ Decomposition method is modified in a way that it is not necessary to re-
formulate the problem in its dual form, which makes possible an easier formulation and
application to other similar problems.

30
4. Methodology

In this chapter the microgrid model and the different units which it is composed by are
described. Then the formulation is written and explained to get a full understanding of
the problem and the different steps carried out. Then the software GAMS is described as
the main tool used in the thesis and finally Benders’ Decomposition technique is
explained as the last step to reach the final problem formulation.

4.1 Model
The model installed is a microgrid linked to the main grid with only one point of
common coupling (PCC). The microgrid has a peak demand of 8.5 MW and the line
between the main grid and the microgrid is assumed to have a power limit which does
not affect the power exchange between the two grids.
The forecasted values for the electricity market price in the system and the demand in
the microgrid are taken from historical data of the microgrid developed in the Illinois
Institute of Technology (Chicago, USA).
The value of lost load (VOLL) is based on average VOLLs for developed and industrial
economies as is the case of Chicago where the microgrid is placed. In these areas the
average range for residential customers goes from $0/MWh and $17937/MWh and from
$3000/MWh to $53907/MWh for commercial and industrial customers [88]. In the
model and in order to be conservative (the lower the VOLL is the lower chances there
are to develop a microgrid) there have been chosen two values of VOLL to compare and
see what happens depending on the different cases. These values are $2000/MWh and
$10000/MWh. The higher VOLL would represent more critical loads as explained in
sections 2.1.3 and 2.2.3.
The set of candidates DERs in the microgrid are six dispatchable units, two non
dispatchable units and three energy storage systems:
The characteristic data of the dispatchable units is shown in the following table:
Unit Number Rated Power (MW) Operational cost ($/MWh) Annualized Investment Cost ($/MW)
1 5 90 50000
2 5 90 50000
3 3 70 70000
4 3 70 70000
5 2 60 100000
6 2 60 100000

31
Table 3 Data set for dispatchable generators [45]

The data presented in the table 3 agrees with other sources like the Danish transmission
system operator (TSO) Energinet in its report “Technology Data for Energy Plants” [86]
where detailed characteristics about small DERs are catalogued. This agreement is
important because [45] does not specify the type of dispatchable generator of each unit,
which will be further needed in order to set the availability constraints in the problem.
Indeed units 1 and 2 are linked to gas engines, whose capacity goes from 1 to 10 MW.
Their annualized investment cost goes from 50000€/MW to 75000€/MW, their
operational cost varies from 7.4 €/MWh to 11 €/MWh (this cost does not include the
gas cost) and their availability factor is 95%.
In the same way, units 3 and 4 are linked to gas turbines, whose capacity goes from 0.1
to 5 MW, their annualized investment cost goes from 70000€/MW to 100000€/MW,
their operational cost is 9 €/MWh (without the gas cost) and their availability factor is
80-85% .
Finally units 5 and 6 are linked to biomass units, whose capacity varies from 0.6 to 4.3
MW, their annualized investment cost is180000€/MW and their availability factor is 90-
92%.
After looking for different public information about wind power finally Wintering Hills
wind farm in Calgary (Alberta, Canada) has been the one selected to take the data. This
wind farm has a power ratio of 0.38. The capacity has been adapted and truncated to a
maximum power capacity of 2 MW. The reason of taking a small farm as reference is
because taking entire countries data is improbable to find hours with null production of
wind energy while in a small farm it might happen, as it should be in a microgrid whose
generation is zero when there is no wind blowing. The power ratio has been calculated
as follows:

( 13)

Solar photovoltaic national French data from 2014 has been chosen as the input for this
kind of energy. In this case it has not been possible to find a small solar power farm so
data from one entire country has been selected. The power ratio of this data is 0.22 and
also the capacity has been adapted and truncated to a maximum power capacity of 2
MW. The difference between power ratio in wind and solar power is essential to
understand the investment of the units in the results.
The characteristic data for non-dispatchable units is presented in table 4:

Unit Number Rated Power (MW) Operational cost ($/MWh) Annualized Investment Cost ($/MW)
7 2 0 160000
8 2 0 154000
Table 4 Data set for non dispatchable units [87]

The annualized investment cost for both units 7 and 8 (wind and solar PV respectively)
has been taken from the National Renewable Laboratory (NREL) [87].
The life time of the renewable systems is provided in table 5:

32
 System Useful Life Years
PV 20 to 40
Wind 20
Biomass Combined Heat and Power 20 to 30
Table 5 Useful life fron renewable energy systems [87]

And in tables 5-6 are represented the cost for solar photovoltaic panels and onshore
wind technology respectively:
Non-Tracking Utility PV with 1 MW (DC) Install Size
Year Capital Cost Variable O&M Fixed O&M($/kW- Construction Schedule
($/kW) ($/MWh) year) (Months)
2008 4610 - - -
2010 3480 0 5 8
2015 3180 0 48 7,6
2020 3010 0 45 7,2
Table 6 Photovoltaic unit costs [87]

Onshore Wind Technology

Year Capital Cost Variable O&M Fixed O&M($/kW- Construction Schedule
($/kW) ($/MWh) year) (Months)
2008 2060 - - -
2010 1980 0 60 12
2015 1980 0 60 12
2020 1980 0 60 12
Table 7 Wind power onshore unit costs [87]

Solar photovoltaic power is evaluated depending on the power capacity, a device

capable to generate from 1 to 10 MW has an investment cost of 3180 $/kW and no
operational cost. It is assumed to have a lifetime from 25 to 40 years, therefore in the
end the annualized investment cost is around 106000 $/MW. If the fixed operation and
maintenance is added ($48000/MW year) the final annualized cost results in
$154000/MW as it is shown in table 4.
Talking about the onshore wind power, NREL places an investment cost of 1980$/kW,
this cost in a lifetime of 20 years results in an annualized investment cost of
$99000/MW. If the fixed operation and maintenance is added ($60000/MW year) the
final annualized cost results in $160000/MW as it is shown in table 4.
The Energy Storage Systems which have been implemented in the microgrid appear in
table 8:
Unit Rated Rated Energy Annualized Annualized
Number Power (MWh) Investment Cost Investment Cost
(MW) Power ($/MW) Energy ($/MWh)
9 1 6 60000 30000
10 2 6 30000 30000
11 3 6 20000 30000
Table 8 Energy Storage System's characteristic data

33
Regarding the ESSs, the characteristic data has been taken from reference [45], this
reference can be double checked with references [11] and [88].
There are multiple varieties of energy storage systems with different technologies as it
can be seen in figure 4.

Electrical Energy Storage Systems

Chemical Electric Thermal

Mechanical Electromechanical Magnetic
Field

Flywheels Hydrogen Conventional High Flow Molten Salt

Double-Layer
Batteries Temperature Capacitor
Batteries
Batteries
Pumped Hydro
Lead Acid Vanadium
Sodium
Sulfur Redox Superconducting
Compressed Air Nickel Magnetic Coil
Cadmium Zinc
ZEBRA Bromine

Lithium Ion

Figure 4 Types of Electrical Energy Storage Systems. Source: [89]

Although it is possible to choose among the different types of ESSs which are shown
before, electrochemical batteries have been finally implemented in the microgrid. And
among them, Lithium-ion batteries have been chosen due to their depth in charge,
frequency and efficiency characteristics. The final efficiency of the batteries is 90%.
Lithium-ion batteries have been implemented successfully in projects like the Pacific
Northwest microgrid in Oregon, the SDE&E-Greensmith demonstration in San Diego
(California) or in largest installations like in Elkins (West-Virginia) where 8MW and
32MWh capacity batteries have been implemented successfully providing and
improving the time-shift energy delivery, capacity credit, grid operational support,
transmission and distribution support, power quality and reliability and integration of
renewable generation.
Finally it is important that operational cost of ESSs is assumed to be zero.

4.2 Formulation

The units explained in the microgrid model are now included in the formulation of the
optimization problem, which is described below.
The aim is to minimize the cost function of the microgrid; in order to do that the
investment is a crucial study point because as commented in section 2.1 microgrids
need a higher investment per MW than conventional grids. This can be compensated
with the lower production cost of the distributed generators implemented compared to
the main grid production cost; the help of the energy storage system being charged at
low price times and discharged at high price times; and the benefits related with the

34
 improvement in the system reliability, which is one of the main purposes of the
microgrid.
The next assumptions are made before writing the formulation:
 Transient behavior in the microgrid is omitted due to it is a long-term
optimization problem. Timeframe is important; thus in a microgrid appears
different ones: (mili)seconds to provide stability, minutes to electricity exchange
with the microgrid, hours regarding the optimization flow, days to manage the
storage system and years to make investment decisions. [62]
 Despite it is a long term formulation, the efficiency of the elements does not
change with the time. It affects the generators and the energy storage systems.

 “The capacity of lines in a microgrid distribution network is typically much

higher than the power transferred through the lines; therefore, the power flow is
not considered in the proposed planning problem as the congestion is less likely
and would not impact the planning results” [47]. Therefore power losses are also
ignored. This assumption have been adopted after reading the next references
too:

o Network congestion is neglected because line flows are relatively small.

[45]

o Due to the number of distributed energy resources located close to the

loads, a microgrid can locally meet load growth and manage congestion
on distribution feeders and medium-voltage sub transmission networks.
[90]

The objective function of the problem is defined below:

(∑ ∑ ∑ ∑∑ ∑ ( )
* +
(14)
∑∑ ∑∑∑ ∑∑ ∑∑

∑∑ ∑∑ ) (∑ ∑ )

This objective functions is subject to the following constraints:

∑ ∑( ) (15)
* +

(16)

35
 (17)

(18)

(19)

(20)

(21)

∑( ) (22)

(23)

In the objective function (14) the first term represents the investment cost in both
dispatchable (D) and non dispatchable (ND) generation. The investment cost (InCost) is
measured in €/MW and is linked with the maximum capacity power ( ) of each
device (i). The second term makes reference to the investment cost in the energy storage
systems (S), which is split in two parts: the investment cost in power (CP) measured in
€/MW which is multiplied by the rated power of the ESS ( ) and the investment
cost in energy (CE) which is multiplied by the maximum energy capacity of the ESS
( ). Both terms are multiplied by which is a binary variable that indicates the
investment state (1 if the unit is deployed and 0 if not). The summation in the two first
terms is implemented per year (y), scenario (U) and device (i).
The third term of the equation refers to the operation cost of the dispatchable units
(OpCost), that is multiplied by the amount of power generated by generators
implemented by each hour (h), day (d), scenario (U) and year (y): ( ). Non
dispatchable generation cost is assumed to be zero.
The fourth term indicates the cost of the energy that comes from the main grid. In this
term the power exchanged between the micro and macro grid( ) is multiplied by
the electricity market price ( ). It is important that the power exchange can be
positive (from the main grid to the microgrid) or negative (from the microgrid to the
macrogrid), therefore this term can be an expense or a profit in the equation.
The last term of the equation indicates the cost of the non-served energy which is
calculated multiplying the value of loss load (VOLL €/MWh) by the load shedding (LS
MWh). This equation shows that the VOLL depends on the hour, day, scenario and
year. In this case the VOLL is modified in a way that it increases proportionally to the
continuous hours when there is load shedding. Thus the more hours in a row there is
load shedding the more expensive the VOLL is as it can be seen in figure 5:

36
Figure 5 VOLL time dependence

All the terms are divided by the summation of years (y) and scenarios (U) in order to
obtain the final annual average cost of the system.
Finally in all the terms of the equation (14) appear the factor . This factor is the
coefficient of present worth value, which is used to calculate the current value of the
money that is received or paid in the future.

(24)
( )

Where d=discount rate and y=year.

If the problem is simulated for more than one year the concept of present-worth value
makes sense in order to have the same money reference. It is possible to say that “A
dollar today is worth more than a dollar tomorrow” which implies that present value of
a certain amount of money is higher than the future value of the same amount of money
[91].
Also it is important to remark the difference between:
 Capitalization: “How much will $1000 today be worth in five years?”
 Discounting: “How much will $1000 received in five years be worth today?”
Obviously in this problem it must be used a discount rate to assess the future costs in the
current monetary status.
Regarding the constraints which the objective function is subject to:
The equation (15) is the load balance; it ensures that all the energy generated is
consumed in the system and it is equal to the demand ( ) each hour during the
simulation. It is remarkable that the power discharged by the ESSs ( ) is
multiplied by the efficiency of the storage system ( ) because it is not possible to
recover all the energy that charges the batteries; there is a penalty efficiency parameter.
The power exchanged between the micro and microgrid can be positive, negative or
equal to zero.
The next equation (16) sets limits to the exchange power between micro and macro grid
constraining the flow in the line which connects both grids ( ). The maximum
power exchange is multiplied by a binary parameter which indicates if there is an outage
in the main grid when the value is 0 ( ).

37
Equation (17) constraints the power generated by the dispatchable generators to a level
lower than their maximum capacity ( ); it also takes into account the availability
constraints which represent the possibility of failure in the units as it will be explained
in chapter 5 ( ). In the same way, equation (18) limits the non
dispatchable units to their forecast obtained previously and subject to uncertainty.
Equation (19) constraints also the non dispatchable units to their maximum capacity
which makes not possible that a forecast that exceeds the maximum capacity of the unit
could enter in the system resulting in a wrong performance of it.
Equations (20), (21) and (22) model the energy storage systems. Equations (20) and
(21) limit the discharging and charging power to a maximum ( ).
Equation (22) sets a limit in the amount of energy stored in the system. In this case
power is compared to energy because the lower timeframe in the problem is 1 hour;
therefore to convert power to energy it is needed to multiply by the time and in this case
multiplying by one (hour) does not change anything.
Finally equation (23) forces the load shedding ( ) to be, as it cannot be otherwise,
lower than the demand ( ) in every hour of the simulation.

UNCERTAINTY FORMULATION

As it was explained previously, in the microgrid optimization problem there are

uncertain variables like the demand, the electricity market price, the non dispatchable
generation and the main grid outages. That uncertainty must be somehow modeled in
the problem and it is done via the following equations:

̃ (25)

̃ (26)

̃ (27)

In these similar equations ̃ , ̃ and ̃ are respectively the forecasted demand, power
generation in non dispatchable units and electricity market price. The variables with
bars represent an upper or lower bound of the variables previously mentioned, thus the
forecasted variables change in a X% the forecasted value in a random way to simulate
the uncertainty. The upper and lower bounds are constrained by a binary variable (u)
which does not allow setting an upper and a lower bound at the same time as it is
defined in equations (28), (29) and (30):

(28)

(29)

38
 (30)

This upper/lower uncertainty is also constrained to a certain number of hours during the
simulations because not at all the hours there is a strong variation between the real value
and the forecasted one. This constraint is performed limiting the number of hours when
the binary variables are 1. When binary variables are zero the demand, electricity
market price or renewable generation values are the forecasted ones.
Changing the limitation number of binary variables per year ( and ) it is
possible to modify the conservatism of the solution. Therefore setting a high number
implies that a large number of the values have a determinate degree of uncertainty, and
it derives in a conservative result. In this case a conservative result implies a higher cost
and a bigger investment. On the other hand when the limitation number is low it means
that the majority of the values are the forecasted ones, with a low degree of uncertainty,
which ends in a less conservative and robust solution, more risky. As it will be seen in
the results chapter (5) it also depends on the type of uncertainty and its variation limits.
The larger number of uncertainty hours implies a more conservative solution in the load
and renewable energy cases but not in the electricity market price case; with the
electricity market price it happens the opposite, the larger the uncertainty is the less
conservative and therefore risky the system is.
The previously mentioned constraints which limit the number of uncertainty hours per
year are implemented with the following equations:

∑∑ (31)

∑∑ (32)

∑∑ (33)

∑∑ (34)

The uncertainty is implemented in different scenarios (U) per year in order to not rely
only in one scenario which could create improbable results due to the randomness of the
uncertainty. Generating the problem for different scenarios makes sure that random
uncertainty is better distributed which makes the final results more trustable.

4.3 GAMS

In this project GAMS (General Algebraic Modeling Systems), a high-level modeling

system for mathematical programming and optimization, is the software used to
program the problem formulated previously.

39
This software is designed to model linear, nonlinear and mixed integer optimization
problems (MIPs) as it is in this case. And it is especially suitable to solve large
problems with heavy data [92].
GAMS also makes easy the following aspects of the problem:
 The writing: GAMS does not require complexes codes to write the mathematical
formulation. And unlike other softwares it possesses few specific programming
key words and a clear interface.
 The compilation: GAMS checks the compilation errors pointing them and
describing their cause.
 The selection of the solver: Depending on the structure of the problem GAMS
selects the most suitable solver although it is also possible to change it and select
the one that the user prefers.
GAMS basic structure is divided in [93]:
 Sets: They can be defined as “building blocks” which make the problem be
easily stated and read.
 Data entry: Parameters, scalars and tables. It is data whose value is known
before starting to solve the problem.
 Variables: They are the entities whose value is unknown until the program is
solved.
 Equations: They are the mathematical expressions which make an algebraic
relation between the data entry and the variables to solve the problem. The
objective functions and constraints are defined here.
 Model and Solve Statements: In the model statement equations are gathered in
groups in order to be solved and in the solve statement the final optimization
order is given to the software indicating what is going to be maximized or
minimized and the model type that is going to be used (linear, nonlinear or
mixed integer).
GAMS also allows the use of Microsoft Excel and Matlab to import and export data,
which implies huge benefits when the problem that is going to be solved is large as in
this case.
Finally the code of the MIP problem implemented in GAMS is showed in the Appendix
1 in order to make the reading of this report easier and more convenient.

4.4 Benders’ Decomposition

Linear problems can be extremely large, which at the end causes computational
problems due to the thousands of equations and variables that must be solved. Special
techniques have been developed in order to solve these kinds of problems, among these
techniques decompositions ones have an important role. There are different techniques
like the Dantzig-Wolfe Decomposition Algorithm, Benders’ Decomposition, Lagrange
Relaxation Decomposition or the Augmented Lagrangian Relaxation. Among them,
Benders’ Decomposition has been selected due to the special suitability for the problem
characteristics and structure.

40
Benders’ Decomposition is a technique created by the Dutch mathematician Jacques F.
Benders whose objective is to simplify complicated large-scale optimization problems.
The technique divides the problem instead of solving it at one time. Thus the problem is
separated in smaller parts called master problem and subproblem.
Solving large optimization problems can take a vast amount of time and memory
resources in the computer because of the variable and equation number and sometimes
it is impossible to get the correct solution of the problem using a simultaneous
resolution method. By dividing the problem solving smaller parts the memory resources
decrease substantially. But when is Benders’ Decomposition especially needed?
Benders’ Decomposition is especially indicated for large problems which contain
difficult variables. These variables complicate the problem; therefore the problem is
easier solved when they are fixed. There are many reasons why a variable could be
marked as difficult; i.e. binary variables and variables involved in nonlinear functions.
In this problem binary variables in investment are the difficult variables. [94] [95] [96]
A general example of Benders’ Decomposition is showed below:

∑ ∑ (35)

Subject to:

∑ ∑ (36)

(37)

(38)

In this model x are the difficult variables and y the regular ones, so the problem is
formulated in a different way separating difficult and non-difficult variables:

∑ ( ) (39)

Subject to

(40)

Where:

( ) ∑ (41)

Subject to

41
 ∑ ∑ (42)

(43)

With this formulation, equations (39) and (40) depend only on the difficult variables (x),
which makes possible to perform a less resource demanding solution. It is possible now
to formulate a master problem and a subproblem:
Subproblem:

∑ (44)

Subject to

∑ ∑ (45)

(46)

(47)

Equation (47) shows that the values of the difficult variables x ( ) are given
previously by the master problem solution and each value has a dual variable associated
( ). When the subproblem is solved for the first time the values of the difficult
variables given ( ) are the solution of the first master problem, which is a trivial
solution. As a result of this optimization subproblem the solution of the non-difficult
variables is obtained: ( ).
The master problem is defined as:

∑ (48)

Subject to

∑ ∑ ( ) (49)

(50)

(51)

42
Equation (49) is the Benders’ cut; this cut is applied to the master problem when there is
an unfeasible solution or when the solution obtained previously in the subproblem is not
the optimal one. Benders’ cut does not apply the first time when the master problem is
solved and it reaches the trivial solution ( ). In equation (51) is a scalar and it is
constrained by a physical or economical value ( ) which is adopted depending on
the problem which is under evaluation. The solution of the master problem is
and .
Finally Benders’ Decomposition has two bounds:
 The lower bound is defined by the solution of the objective function in the
master problem at the last iteration (g+1).

∑ (52)

 The upper bound is defined by the solution of the objective function of the
whole problem in the last iteration of the subproblem.

∑ ∑ (53)

The order of the operations in Benders’ Decomposition is explained below:

 In the first step the master problem is minimized obtaining a trivial solution of
the difficult variables.
 In the second step the solution of the first step is included in the subproblem and
a solution of the rest of variables is obtained by solving the optimization
problem.
 In the next step lower and upper bounds are calculated and compared, if the
difference between them is lower than a stipulated threshold the solution
obtained is accepted, otherwise another iteration is performed in the problem.
 When the iteration is performed the Benders’ cut (49) is added to the master
problem using the dual variable values and the solution of the variables in the
previous iteration in order to obtain a new solution. After that the subproblem is
solved again and the lower and upper bounds are compared iteratively until the
difference between them is lower than the stipulated threshold.

4.4.1 Benders’ decomposition problem formulation

In this section the theoretical description of Benders’ Decomposition explained before is

applied to the mixed integer problem (MIP) formulation described in 4.2.

The master problem after applying Benders’ Decomposition is:

(∑ ∑ ∑ ∑∑ ∑ (
* + (54)
) ) (∑ ∑ )

43
Subject to

(∑ ∑ ∑ ∑ ∑

∑∑ ∑∑
(55)
∑∑ ∑∑ ) (∑ ∑ )

∑ ( )

(56)

The subproblem after applying the Benders’ Decomposition is:

(∑ ∑ ∑ ∑ ∑

∑∑ ∑∑ (57)

∑∑ ∑∑ ) (∑ ∑ )

Subject to

( ) ( )

(58)

The minimization objective function of the master problem (54) only includes the
investment terms of the normal MIP optimization problem because the binary difficult
variables ( ) are included in them. It is constrained by the Benders’ cut (equation 55)
and the scalar α is also constrained by 0 in order to make the master problem (which is
also the lower bound) higher and higher each iteration until there is a convergence with
the upper bound of Benders’ Decomposition. Constraints previously shown in equation
50 do not apply in this problem because x are binary variables.

In the subproblem the rest of variables are solved in equation (57) which is constrained
by the equations (15-23) previously explained. Also the dual variable (λ) is created in
equation 58 for each difficult variable (x) previously solved in the master problem. The
dual variables created at equation 58 are vital because they are added to the Benders’
cut (55) and they make the Benders’ Decomposition convergence possible.

44
Finally the GAMS code of the Benders’ Decomposition technique added to the MIP
problem described in section 4.2 is presented in Appendix 2.

4.4.2 Benders’ decomposition applied benefits

After the application of Benders’ Decomposition the resolution time of the model
implemented improves considerably. In simple examples the difference is not so
important; i.e. in a model with one year and one scenario the normal MIP problem lasts
2 minutes and 30 seconds to get the final solution and the Benders’ Decomposition
problem takes 39 seconds. When a problem with two years and two scenarios is solved
the time difference increases, the normal MIP problem takes 30 minutes to be solved
and the Benders’ Decomposition one only takes less than two minutes, which means a
reduction of more than 15 times in the resolution procedure. Even in larger examples
with 10 years and one scenario per year the Benders’ Decomposition program lasts 4
minutes to solve the problem while the normal MIP does not even reach a good
approximation to the final solution after the program is running more than 1 hour and 10
minutes.

Resolution time with Resolution time with MIP and

Problem length
normal MIP (m’ s’’) Benders’ Decomposition (m’ s’’)

1 year and 1 scenario per year 2’30’’ 39’’

2 years and 2 scenarios per year 30’ 1’57’’

10 years and 1 scenario per year >4000’’ 4’

6 years and 3 scenarios per year >>4000’’ 15’45’’

Table 9 Benders' Decomposition time improvement

Therefore Benders’ Decomposition reduces hugely the resolution time in large

problems. It makes possible create and solve more complex and useful models to
understand the behavior of the main grid and the microgrid; which helps to determine if
it is profitable or not to develop microgrids and which units should be invested.

4.5 Inputs and implementation

It is important to mention, and it can be checked in both Appendices 1 and 2, that the
forecasted electricity market price, load demand and renewable energy produced are
inputs of the GAMS model from Microsoft Excel files. They are modified by different
kinds of uncertainties which are also inputs of the model, these uncertainties have been
generated randomly via Matlab. Other random parameters like the power cut time
between the main grid and the microgrid, and the availability factor that affects the
dispatchable units are generated randomly via Matlab and added to the GAMS model as
Microsoft Excel inputs.

It finally leads to a deterministic model as it was explained in chapter 2.3.3, because

each time that the model runs with the same inputs it reaches the same results. But it

45
does not mean that the model does not handle uncertainty, because it has been added
externally.

46
5. Results and discussion

Once the GAMS codes needed to solve the problem are completed, including power
cuts, availability constraints, and uncertainty in load demand, electricity market price
and energy produced by the non dispatchable units (wind and solar) it is the moment to
start doing the final simulations in order to get the final results and compare them.
The final number of years and scenarios of the final simulations of the model are 6 years
and 3 scenarios per year due to computational limitations. It is also important to know
that the limitations of the computer are not related with Benders’ Decomposition or any
code problem, simply the computer stops running because of all the memory that the
program demands. If the simulation with 6 years and 3 scenarios per year takes
approximately 15 minutes, the simulation with one more scenario per year is impossible
to solve because the computer cannot stand the resources demanded. This problem
would have been definitely higher if the model had been solved as a normal MIP instead
of using Benders’ Decomposition.
The VOLL is 2000 $/MWh if there is no special indication; in some models it is
changed to 10000 $/MWh (always with an appropriate indication) to study the behavior
of the model with different VOLLs.
In the first simulation done there are no power cuts (Um) in the line which links the
macro grid and the microgrid, no availability constraints (AC) and no uncertainty in
load demand, electricity market price and energy generated by non-dispatchable units.
There are presented two possible models (model 0 and model 1). In model 0 there is no
possibility to invest distributed energy resources (DERs) and in model 1 all of them are
available:

Model Name Annual average cost ($) Invested units

No Um, No AC, No
0 2021542 0
uncertainty + No DERs
No Um, No AC, No
1 1966755 7,10,11
uncertainty
Table 10 Comparison of non uncertainty models with and without microgrid

We can see that even without power cuts the microgrid is profitable. Energy storage
systems (ESSs) and the wind power unit are implemented when there is no power cuts
between the microgrid and the main grid. It is important that although the investment in
wind power is the most expensive one among the generators (160000 $/MW), the null
operational cost makes it profitable instead of the other and cheaper generators.

47
The final cost difference between them is 2.78% so we can see the importance of the
both wind power (unit 7) and ESSs (units 10 and 11).
Looking at these results it is possible to wonder why this implementation happens if
ESSs are the most expensive units. The reason can be explained looking at the general
pattern of the load demand curves and especially the electricity market price curves at
Illinois Institute of Technology (IIT) which are shown at figures 6, 7 and 8.

Daily Average Electricity Market Price

70

60

50

40
$/MWh

30

20

10

0
274
105
118
131
144
157
170
183
196
209
222
235
248
261

287
300
313
326
339
352
365
1
14
27
40
53
66
79
92

Figure 6 Daily Average Electricity Market Price in a year

Daily Average Load

9
8
7
6
5
MWh

4
3
2
1
0
105
118
131
144
157
170
183
196
209
222
235
248
261
274
287
300
313
326
339
352
365
1
14
27
40
53
66
79
92

Figure 7 Daily Average Load Demanded in a year

Looking at the figures above it is possible to observe that the daily average electricity
market price does not vary highly throughout the year and it has an average value of
52.08 $/MWh, however the daily average load has a larger variation from day to day
and it is possible to see how during the summer the load demanded reaches its peak

48
 with daily average loads up to 7.82 MWh and during the winter the demanded load
drops until 3 MWh.

Electricity Market Price day 7 Electricity Market Price day 105

140 140
120 120
100 100
$/MWh

$/MWh
80 80
60 60
40 40
20 20
0 0
1 3 5 7 9 11 13 15 17 19 21 23 1 3 5 7 9 11 13 15 17 19 21 23

Load Demand day 7 Load demand day 105

5 6
4 5
4
3
MWh

MWh

3
2
2
1 1
0 0
1 3 5 7 9 11 13 15 17 19 21 23 1 3 5 7 9 11 13 15 17 19 21 23

Electricity Market Price day 187 Electricity Market Price day 300
140 140
120 120
100 100
$/MWh
$/MWh

80 80
60 60
40 40
20 20
0 0
1 3 5 7 9 11 13 15 17 19 21 23 1 3 5 7 9 11 13 15 17 19 21 23

Load Demand day 187 Load Demand day 300

10 5
8 4
Título del eje

6 3
MWh

4 2
2 1
0 0
1 3 5 7 9 11 13 15 17 19 21 23 1 3 5 7 9 11 13 15 17 19 21 23
Figure 8 Hourly Electricity Market Price and Load demanded at days 7, 105, 187 and 300

49
The figure above shows the hourly electricity market price and the load demand for
random days at Illinois Institute of Technology one in each season, looking at them it is
possible to see how the electricity market price follows a similar pattern, this pattern as
it has been explained before has the same average value along the whole year. But
looking day by day there is a high price difference throughout the 24 hours, it varies
from less than 20 $/MWh at valley points to more than 120 $/MWh at peaks points.
However the load has the opposite behavior, with a larger variation along the year but
with an almost plane shape within each day as we can see in the figure above, where
there is as maximum 1.5 MWh of difference between the peak and the valley load
points.
The addition of these characteristic curves which barely vary along the year gives to the
ESSs a special importance. Trying to discharge energy at the peak points with a high
electricity market price and charge energy at low electricity market price hours.

Charging and Discharging ESS Daily Average Times

3

2,5

2
MWh

1,5 Discharge
Charge
1

0,5

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
Figure 9 ESSs' Charging and Discharging Average Times during the 6 years microgrid’s performance

In the figure above are presented the charging and discharging daily average times
during the 6 years’ microgrid simulation. It is possible to observe how there are 5
charging and discharging time periods per day, where the main discharging time period
matches with the highest daily electricity market price, and the other discharging times
period match with relative maximum points of the electricity market price. These points
are marked with red circles for the three main discharging points and with yellow ones
for the other two in the figure 8. The charging time periods occur at valley electricity
market price times and the main one takes place at the dawn when the electricity is
offered at its cheapest price. At this time (2 am) almost all ESSs are charging every day
and every hour which means an average charge of 2.44 MW when the maximum
possible is 2.5 (there are implemented one ESS unit with 2 MW rated power and
another one with 3MW). Thus it is possible to see the influence of ESSs trading with
energy depending on electricity market price and load demand.

50
5.1 Power cut addition

A first power cut of 9 hours between the main grid and the microgrid is implemented,
afterwards in section 5.5 it is going to be modified and compared with other models
whose power cut duration is higher (24 and 48 hours distributed along the year).
Now a system with 9 hours of power cut (Um) at the time when more power is
demanded by the system (day 206) is tested, forecasted values are used so there is still
no uncertainty either availability constraints.

Model Name Annual average cost ($) Invested units Cost variation
2 Um Day 206 9 hours, 1994797 3,5,6,7,10,11 +1.42%
No AC, No uncertainty Compared to model 1
Table 11 Model with 9 hours of power cut at the peak load day

Of course the cost is higher than when there is no power cut in the microgrid (model 1),
but it is still lower than when there is neither power cut nor possibility to implement a
microgrid (model 0). This shows how beneficial a microgrid might be.
Talking about the units invested, they are number 3, 5, 6, 7, 10 and 11. Units 7, 10 and
11 were implemented in the model before, even when there is not any power cut, so
their presence in this model is also understandable due to the inexistent operational cost
in the wind power generation (unit 7) and the trading energy role of the energy storage
systems (unit 10 and 11).
Non dispatchable units implemented are number 3, 5 and 6, which correspond to one
gas turbine of 3 MW and two biomass units of 2 MW each. This shows that even
though the units are the dispatchable ones with the highest annualized investment cost,
the operational cost is lower than in the others and this is more important at the time of
choosing the final units. The lower operational cost is the cause of avoid buying
electricity from the main grid when the electricity market price is expensive and sell all
the power that it is not demanded by the microgrid’s loads in order to get revenues and
be profitable.
An interesting point is that here it is possible to realize another role of ESSs, which not
only trade with energy between high and low electricity market price times. At day 206
at hour 15 there is a load of 8.49 MW which cannot be supplied by units 3,5,6 and 7
because units 3,5 and 6 supply 7 MW in total and at that time unit 7 (wind power) is
supplying 0.77 MW. Therefore ESSs avoid both the investment in a new DER and the
load shedding.
Also it is important to know that although both renewable energy generators (solar
photovoltaic panels and wind power turbines) have a similar investment cost: 154000
$/MW the solar panel (unit 8) and 160000 $/MW the wind turbine (unit 7), and also
both of them have the same operational cost (0 $/MWh); wind power is selected
because of the difference of the power ratio, as explained before in section 4.1. Wind
power in this case has a power ratio of 0.38 while solar PV power has a power ratio of
0.22, it depends of course of the meteorological conditions but solar photovoltaic power
cannot produce energy during the night while wind power is able to produce energy
throughout the whole day.
Regarding the energy storage systems, unit 11 is more profitable due to its higher rated
power and it can supply more energy at needed hours.

51
But let us see more different scenarios with power cuts at different times:
Now the system is run when the power cut takes place when the lowest load is
demanded by the microgrid (day 2) with forecasted values, without uncertainty and
without availability constraints.

Model Name Annual average cost ($) Invested units Cost variation
3 Um Day 2 9 hours, No 1972652 5,7,10,11 0.3%
AC, No uncertainty Compared to model 1
Table 12 Model with 9 hours of power cut at the valley load day

Since the load demanded by the microgrid is lower, fewer units are invested and
although it affects the energy trading (because it is not possible to sell as much energy
as with more dispatchable units) the annual average cost is lower.
In this model only dispatchable unit 5 (biomass unit) is implemented, which shows
again that lower operational costs are more important than investment costs. For this
reason units 5 and 6 are implemented before units 3 and 4, and these before 1 and 2.
In the next table are presented the results when the system is simulated with the power
cut taking place during 9 hours in different random days:

Model Name Annual average cost ($) Invested units Cost variation
1.5%
Day 195 9 hours, No AC,
4 1996444 3,5,6,7,10,11 compared to
No uncertainty
model 1
Day 131 9 hours, No AC,
5 1978068 3,5,6,7,10,11 0.57%
No uncertainty
Day 73 9 hours, No AC,
6 1978872 5,6,7,10,11 0.62%
No uncertainty
Table 13 Model with 9 hours of power cut at different days

As it has been explained before, the load demanded by the microgrid at the power cut
moment determines mainly the quantity of DERs which are needed. But in order to do a
deeper analysis the annual cost of the models previously commented are divided in
investment and operation cost in the table 14:

Model Name Invested Investment Operation Total Annual

units cost ($) cost ($) cost ($)
Day 206 *
2 3,5,6,7,10,11 1241768 753029 1994797
Day 2 *
3 5,7,10,11 877587 1095065 1972652
4 Day 195 * 3,5,6,7,10,11 1996444
1241768 754676
5 Day 131 * 5,6,7,10,11 1978068
1055236 922832
6 Day 73 * 5,6,7,10,11 1978872
1055236 923636
Table 14 Investment and Operation cost comparison in power cuts placement

* All power cuts are 9 hours long; there is neither uncertainty nor reliability constraints

52
At day 206 and 195 there is more load demanded, thus the more units invested imply a
more expensive investment cost. However the fact of having more capacity
implemented in the microgrid is favorable for the operation and this part of the cost is
lower than in the rest of the cases. The more units invested give the chance of take
better advantage of the internal trading between the microgrid and the main grid: more
energy is sold at high electricity market price hours when this price is higher than the
operational cost of the generators implemented and less energy is bought from the main
grid because it is cheaper to produce it with the units implemented; therefore the
expenses are reduced.

The same argument is provided for models 3, 5 and 6. Where the fewer units
implemented cause a reduction of the investment costs but an increase of the operation
costs because the internal trade is reduced.

Looking at the total annual cost it is higher when more units are invested. Thus in this
case the investment cost is more important than the operation cost to determine the final
cost for the system. But at the end the operation and the investment cost tend to
compensate each other having all models similar final annual costs although their
operation and investment costs vary considerably from one model to another. It is
presented in figure 10.

Investment and Operation Cost Relation

2500000

2000000

$ 1500000 Investment Cost

Operation cost
1000000
Total Annual Cost
500000

0
model 1 model 2 model 3 model 4 model 5
Figure 10 Investment and Operation Cost Relation for models 1-5

At the end the best or worst hours for the power cut implementation are determined by
the relationship between the load demanded and the electricity market price at the
power cut times. This relationship defines the amount of money that the microgrid can
get as revenue selling energy as much as possible (total MW invested – load demanded
at the moment) when there are high electricity market prices; and buying energy as
much as possible (load demanded energy + energy storage systems capacity) when there
are low electricity market prices. The load at power cut times determines the number of
distributed energy resources that must be implemented which are also a fundamental
factor in the final annual cost of the microgrid.

53
5.2 Availability factor addition (reliability constraints)

The availability factor represents the amount of time where units are able to produce
energy. The availability factor data gathered is based on the Danish transmission system
operator (TSO) Energinet report “Technology Data for Energy Plants” data [86]; and
the reliability constraint for each generator is:
 Photovoltaic solar panels = 100%
 Wind power generators=98%
 Gas generators= 80-99% where 85% correspond to Gas turbines (55 days per
year of failure) and 95% to gas engines (18 days per year of failure)
 Biomass generators=90% (36 days per year of failure)

Units 1 and 2 are gas engines (GE), units 3 and 4 are gas turbines (GT) and units 5 and
6 are biomass generators (Bio). We assume that the days of failure are placed in a row
because when there is a failure it takes time to fix it so it would not be realistic to place
these days alternating one day of failure and one day of good unit performing.
Therefore and assuming:
a) That units implemented work every day. This is because once the efforts of
spending money and developing the units are made; units are used every day
when electricity market price is high to sell energy to the main grid and feed the
microgrid loads because their operational cost is lower than the electricity
market price.

b) That all the reliability constraints last more than one day, so both high and low
electricity market price hours are unavailable during the unavailability period.

c) The shape of both electricity market price curve and demanded load for the
whole year and day by day (figures 6, 7 and 8).

There will not be interesting to study different places for these unavailable days of the
units.
As it was seen before, when there is not any power cut (model 1) only units 7,10 and 11
are implemented, which are not affected by reliability constraints (availability factor).
Thus availability factor is only added and simulated with models which have power
cuts.
Now the assumption of that the different places for the reliability constraints do not
affect the model is going to be checked. To do this model 3 (day 2, the lowest demand)
and model 2 (day 206, the highest demand) are run with two different locations of the
reliability constraints.

54
Model Name Annual average cost ($) Invested units Cost variation
3 Um Day 2 9 hours, No AC, 1972652 5,7,10,11
No uncertainty
7 Model 3 with Availability 1972975 5,7,10,11 0.016% compared
constraints location 1(v1) to model 3
8 Model 3 with Availability 1973070 6,7,10,11 0.0048 % compared
constraints location 2 (v2) to model 7
Table 15 Reliability Constraints with different locations (v1 and v2) at day 2

Model Name Annual average cost ($) Invested units Cost variation
2 Day 206 9 hours, No AC, No 1994797 3,5,6,7,10,11
uncertainty
9 Model 2 with Availability 1996520 3,5,6,7,10,11 0.086% compared
constraints location 1(v1) to model 2
10 Model 2 with Availability 1996760 3,5,6,7,10,11 0.012% compared
constraints location 2(v2) to model 9
Table 16 Reliability Constraints with different locations (v1 and v2) at day 206

V1 and V2 correspond to different location times for the availability constraints. The
cost difference between the placement of these constraints is almost inexistent with
variations of 100-300 $/year which means less than 0.012% of variation. This confirms
that the different placement of availability constraints does not affect the model at all so
it will not be studied. The similar shapes of the daily load and electricity market price in
the microgrid and the fact that generators works every day to take advantage of the
internal trading make the different timing for reliability constraint a non-important
factor in these models.
Also as we can see in tables 15 and 16, the difference between the models with
reliability constraints and without them is small (around 2000 $ per year) which means
a variation of 0.086% in model 2 and 0.016% in model 3. This is because in model 2
there are two more units implemented, unit 3 which corresponds to a gas turbine and
unit 5/6 which corresponds to a biomass unit, with 55 and 36 days of unavailability per
year.
The reasons of the small repercussion of the availability constraints in this model are:
a. The lack of generation that the constraints imply in the generators does not cause
load shedding so a high VOLL is not applied as penalty.

b. The lack of generation that the constraints imply in the generators in some
moments can be compensated by other distributed energy resources which are
not working at full capacity. Thus although they are not the optimal units to be
used at this moment they are capable to operate at that time.

c. The lack of generation that the constraints imply in the generators in case of that
the rest of DERs are operating at their maximum capacity only causes a loss of
benefits in the internal trading and a small increase of the operation cost.
Although in these models, with the specific data used as input for the electricity market
price, load demand and renewable energy generated, the availability factor does not
affect the model in deep, it must be said that in other systems with different load and
electricity market price characteristics the availability factor can be an important cause
of cost variation with percentages around 3.65 %. The availability factor can be

55
extremely important in cases when all generators are working at full capacity and an
outage in one or many of them implies a load shedding with a high VOLL; it can causes
high costs to the system. Thus it is important to keep the effect of the availability factor
in the formulation and in code equations although in some cases might not affect the
results in fact.

5.2.1 Microgrid and no microgrid differences with availability constraints

It is also interesting to know what the values of the models are in case of there is no
possibility to implement a microgrid. Models with power cuts at day 2 and 206 are
studied for this purpose.

Since now the availability constraints (AC) are only placed in one of the two location
times studied before because the difference between them does not imply important
changes in the models, in this case V2 is chosen.

Model Name Annual average Invested Load Shedding Cost variation

cost ($) units (MWh)
11 Um Day 2 9 hours, with 2062046 0 23,01 per year 4,5% compared
AC, No uncertainty to model 8

12 Model 11 with VOLL 2225555 0 23,01 per year 12.8% compared

10000 $/MWh instead to model 8
of 2000 $/MWh
13 Um Day 206 9 hours, 2146903 0 73,31 per year 7.5% compared
with AC, No to model 10
uncertainty
14 Model 13 with VOLL 2667841 0 73,31 per year 33.62%
10000$/MWh instead compared to
of 2000 $/MWh model 10
Table 17 Microgrid and no Microgrid Implementation with Power Cuts

Obviously in the models above there is load shedding because there is a power cut and
it is not possible to implement a microgrid to supply energy to the load demanded. The
load shedding depends on the day of the power cut; therefore at day 2 there is a load
shedding of 23,01 MWh per year (valley load point) and at day 206 there is a load
shedding of 73.31 MWh per year (peak load point). Looking at the results it is possible
to tell how important can be to implement a microgrid and how dependent it is on the
VOLL.

56
 Microgrid - Non microgrid cost difference
3000000

2500000
Microgrid with 9 hours of
2000000 powercut at peak load day
$ +7,5% +33,62%
1500000 Sytem without Microgrid
(VOLL=2000 $/MWh)
1000000
Sytem without Microgrid
500000 (VOLL=10000 $/MWh)

0
Annual average cost

Figure 11 Microgrid - Non microgrid cost difference at peak load demanded day

In figure 11 it is possible to see that the system without microgrid is 7,5% more costly if
the value of lost load is 2000 $/MWh and 33,5% if the value is 10000 $/MWh.
Therefore in determinate systems a microgrid could be really positive and profitable to
implement, especially when the VOLL is high.

Microgrid - Non microgrid cost difference

2500000

2000000 Microgrid with 9 hours

of powercut at valley
1500000 load day
+4,5% +12,8%
$ Sytem without
Microgrid (VOLL=2000
1000000 $/MWh)
Sytem without
500000 Microgrid (VOLL=10000
$/MWh)
0
Annual average cost
Figure 12 Microgrid - Non microgrid cost difference at valley load demanded day

In figure 12, with a lower load shedding than figure 11, the impact of developing the
microgrid is considerably lesser but still important, reaching cost difference of 12.8%
when the VOLL is 10000 $/MWh and reaching 4.5% when the VOLL is 2000 $/MWh.

5.3 Uncertainty in load, renewable generation and

electricity market price
After the primary study of the power cuts and availability constraints the model is
modified adding uncertainty to the load, renewable generation and electricity market
price inputs. These simulations are done as follows:
 Study of the load uncertainty alone.

57
  Study of renewable energy uncertainty alone.
 Study of both load and renewable energy uncertainties together.
 Study of the electricity market price uncertainty alone.
 Study of the three uncertainties altogether.

The day chosen to implement and then compare the different uncertainty cases is the
day 206, when the peak load takes place in the microgrid. The reason why this day is
chosen is because it is the day with the highest load demand and when the peak load
takes place there are more possibilities to have a failure in the system.
The uncertainty is implemented as a matrix input which multiplies the forecasted
values. In that way it is possible to modify the uncertainty both in hours per year and
percentage variation. The uncertainty matrix generates random hours of uncertainty
throughout the year; these hours place the uncertainty in one of its bounds, either the
upper or the lower one.

5.3.1 Uncertainty in load demand

Load uncertainty is initially applied with a degree of ±10% and a limit of 1000
hours/year. In table 18 the models’ results are presented with different variations both in
number of uncertainty hours and the percentage of the uncertainty bounds.

Model Name Annual average Invested units Load Cost

cost ($) Shedding variation
(MWh)
10 Day 206 9 hours, + 1996520 3,5,6,7,10,11 0
Availability constraints,
No uncertainty
15 Model 10 +Addition of 1996682 3,5,6,7,10,11 0 +0.008%
+1000 hours +-10% compared
load uncertainty to model 10
16 1000 hours +10% 2019592 3,5,6,7,10,11 0 +1.15%
load uncertainty
17 2000 hours +10% 2043001 3,5,6,7,10,11 0 2.32%
18 3000 hours +10% 2065486 3,5,6,7,10,11 0 3.45%
19 5000 hours +10% 2112137 3,5,6,7,10,11 0 5.79 %
20 8000 hours +10% 2182445 3,5,6,7,10,11 0.76 per year 9.31%
VOLL 2000 $/MWh
21 8000 hours +10% 2186519 3,5,6,7,10,11 0.76 per year 9.51%
VOLL 10000 $/MWh
22 8000 h +10% 2193345 3,4,5,7,10,11 0 9.85%
VOLL 100000 $/MWh
Table 18 Load uncertainty implementation with 10% bound (VOLL in $/MWh)

Looking at the table above it is possible to see that when the uncertainty hours are
divided equally among its upper bound (+10%) and its lower bound (-10%) the cost
variation is almost non-existent and the result barely changes (the model 15 is only
0.008% more costly than the model 10). However when the uncertainty hours tend more
to be at its upper bound the final cost is higher than in the model 10 without uncertainty.
It is understandable because the higher the load demand is the more energy must be
generated. At importing energy hours (when the electricity market price is low and it is
profitable to import energy rather than generate it) there is more energy needed so the

58
cost increases because the microgrid imports more energy; and at exporting energy
hours (when electricity market price is high and it is profitable to generate energy in the
microgrid rather than import it) there is not possible to export as much energy as before
because the load demands more energy so there is a reduction in the microgrid benefit.
Therefore with positive load uncertainty the microgrid income is lower and the final
cost higher.
On the other hand when there are more load uncertainty hours at lower bounds than at
upper ones the final annual average cost is lower than the forecasted cost. Hence this
uncertainty is favorable for the microgrid.
In this master thesis it is intended to place the microgrid in the worst case possible to
consider whether developing it is profitable or not. Therefore models with load
uncertainty at its lower bound (-10%) and future models with favorable uncertainties to
microgrid implementation are not presented in the report.
In figure 13 it is possible to visualize the difference between models when the
uncertainty hours increase. If there are 1000 hours of load uncertainty per year at upper
bounds the system costs are 1.15% higher, with 2000 hours they increase until 2.32 %
,with 3000 hours of load uncertainty throughout the year the cost difference reaches
3.45% and so forth until the number of uncertainty hours reaches 8000. At this time
there is a load shedding in the microgrid of 0.76 MWh per year which is avoided with a
large VOLL (100000 $/MWh) and the replacement of the unit 6 by the unit 4 which
increases the capacity of the microgrid in one MW.
In all these models (15-21) the units invested do not change so the increase in the final
cost is only due to the augmentation in the operation cost. If models 21 and 22 are
studied in detail it is possible to see how in table 19 the change of the units implies
again a higher investment cost and a lower operation cost as happened in table 14.

Model Name Invested Investment Operation Total Annual

units cost ($) cost ($) cost ($)
8000 h +10%
21
VOLL 10000 3,5,6,7,10,11 1241768 944751 2186519
8000 h +10%
22
VOLL 1000000 3,4,5,7,10,11 1250650 942695 2193345
Table 19 Investment and operation costs in model 21 and 22 (VOLL in $/MWh)

59
 Load Uncertainty Annual Average Cost Percentage
Difference 1000 hours of load uncertainty
12 +- 10%
1000 hours of load uncertainty
10 + 10%
2000 hours of load uncertainty
8 + 10%
3000 hours of load uncertainty
% 6
+ 10%
5000 hours of load uncertainty
+10%
4 8000 hours of load uncertainty
+10%
2 8000 hours of load uncertainty
+10% VOLL 10000 $/MWh
8000 hours of load uncertainty
0
+10% VOLL 10000000 $/MWh
Percentage difference
Figure 13 Load Uncertainty Annual Average Cost Percentage Difference

Also the higher the upper bound is the more expensive the final cost of the system is as
we can see in the following tables 20, 21 and 22:

Model Name Annual average Invested units Load Shedding Cost

cost ($) (MWh) variation
10 Day 206 9 hours, + 1996520 3,5,6,7,10,11 0
Reliability
constraints, No
uncertainty
23 + 1000 h +20% 2042692 3,5,6,7,10,11 0 +2.31%
Load uncertainty compared to
model 10
24 + 1000 h +30% 2065767 3,5,6,7,10,11 0.74 +3.46%
25 + 1000 h +40% 2090304 3,5,6,7,10,11 12.62 4.70%
26 + 1000 h +50% 2114162 3,5,6,7,10,11 22.67 5.89%
27 + 1000 h +70% 2164477 3,5,6,7,10,11 55.14 8.41%
28 + 1000 h +100% 2235793 3,5,6,7,10,11 84.85 11.98%
Table 20 Load uncetainty in microgrids with different bounds and VOLL=2000 $/MWh

Model Name Annual average cost Invested units Load Shedding Cost
($) (MWh) variation
10 Day 206 9 hours, + 1996520 3,5,6,7,10,11 0
Availability
constraints, No
uncertainty
29 Model 10 2066081 3,5,6,7,10,11 0.74 +3.48%
+ 1000 h +30% compared to
Load uncertainty model 10
30 + 1000 h +40% 2095090 3,5,6,7,10,11 12.62 4.93%
31 + 1000 h +50% 2123864 3,5,6,7,10,11 22.67 6.37%
32 + 1000 h +70% 2176461 3,4,5,6,7,10,11 1.32 9.01%
33 + 1000 h +100% 2253927 3,4,5,6,7,10,11 17.84 12.89%
Table 21 Load uncetainty in microgrids with different bounds and VOLL=10000 $/MWh

60
Model Name Annual average cost ($) Invested units Cost variation
10 Day 206 9 hours, + 1996520 3,5,6,7,10,11
Availability constraints,
No uncertainty
34 Model 10 2077677 3,4,5,6,10,11 +4.06%
+ 1000 h +30% compared to
load uncertainty model 10
35 + 1000 h +40% 2101136 3,4,5,6,10,11 5.24%
36 + 1000 h +50% 2130103 3,4,5,6,7,10,11 6.69%
37 + 1000 h +70% 2227192 3,4,5,6,7,9,10,11 11.55%
38 + 1000 h +100% 2339785 1,3,5,6,7,10,11 17.19%
Table 22 Load uncetainty in microgrids with different bounds and VOLL=1000000 $/MWh

In the three tables above it is possible to see the differences in annual cost and units
implemented when the uncertainty percentage increases.
In the first table, with VOLL=2000 $/MWh , the increase in the upper bound does not
imply any modification in the units invested, although the higher the uncertainty bound
is, the higher the final annual cost and the total load shedding (LS) are.
When the VOLL increases until 10000 $/MWh, there is only a change of the units
implemented when the upper bound is greater than or equal than 70%, where unit 4 is
implemented and the load shedding drops drastically. Also the higher the VOLL and the
uncertainty bound are the higher the cost in the system is.
Finally when the VOLL=10000000 $/MWh, this extremely high price makes that more
units are invested due to load shedding leads to a cost that is not affordable for the
microgrid. In this cases load shedding is 0 MWh MWh and the costs are higher than in
tables 20 and 21.
These three tables present again the importance of VOLL at the time of implement or
not a microgrid. Therefore the higher the VOLL is the easier to implement a profitable
microgrid is and the more DERs the microgrid is composed by in order to avoid more
expensive load shedding costs.

5.3.2 Uncertainty in renewable energy generation

Uncertainty in renewable energy generation is initially applied with a variation degree

of ±20% and a limit of 1000 hours per year. In tables 23, 24 and 25 are presented the
results with different variations both in number of uncertainty hours and the placement
of the percentage bounds:

Model Name Annual average cost ($) Invested units Cost variation
10 Day 206 9 hours, + 1996520 3,5,6,7,10,11
Reliability constraints, No
uncertainty
39 Model 10 +1000 hours 1996997 3,5,6,7,10,11 0.024%
+-20% renewable energy compared to
uncertainty model 10
40 1000h -20% 2003146 3,5,6,7,10,11 0.33%
41 2000h -20% 2008691 3,5,6,10,11 0.61%
42 3000h -20% 2008691 3,5,6,10,11 0.61%
Table 23 Renewable energy uncertainty implementation at 20% bound

61
As it happened before with the load uncertainty, when the uncertainty takes place both
in the upper and in the lower bound there is not a significant change in the annual
average cost. Both bound uncertainties tend to compensate each other and the variation
with 1000 hours of uncertainty results in a 0.024% more expensive annual cost for the
model with uncertainty compared to the model 10 which does not have it.
In this case the annual average cost increases when there are more uncertainty hours in
the lower bound than in the upper one. It is understandable because the renewable
energy has a variable cost equal to zero. Therefore at the lower bound uncertainty times
there is a reduction of “free” energy so the annual average cost increases because it is
not possible to export as much as before at times of high electricity market price; and at
moments of importing energy it is needed more power from the main grid or from
dispatchable units to supply the load demanded by the microgrid.
In an opposite way the upper bound uncertainty hours imply a reduction of the annual
cost because with the extra energy it is possible to export more energy at high electricity
market price hours and import less energy at low electricity market price times.
Therefore in that way the microgrid gets a higher income being more beneficial in
financial terms.
As it has been done with the load uncertainty, only lower bounds have been considered
because they result in the worst case possible to consider developing the microgrid is
profitable or not with renewable energy uncertainty.

In figure 14 the table 23 is presented graphically to see the cost variation with
renewable energy uncertainty.

Renewable Energy Uncertainty Annual Average Cost

Percentage Difference
0,7
Model without uncertainty
0,6
0,5 1000 hours of renewable
energy uncertainty +- 20%
0,4
1000 hours of renewable
% 0,3 energy uncertainty - 20%
0,2 2000 hours of renewable
energy uncertainty - 20%
0,1 3000 hours ofrenewable
0 energy uncertainty - 20%
Percetage difference
Figure 14 Renewable Energy Uncertainty Annual Average Cost Percentage Difference

5.3.3 Load and renewable energy generation uncertainty comparison

Although the original referenced uncertainty in renewable energy generation is 1000

hours and -20%, this percentage bound has been modified and reduced to -10% to be
able to compare load and renewable uncertainty:

62
Model Name Annual average cost ($) Invested units Cost variation
43 Model 10 1999787 3,5,6,7,10,11 0.16%
+1000 hours -10% compared to
renewable uncertainty model 10
44 +2000 hours -10% 2003059 3,5,6,7,10,11 0.33%
45 +3000 hours -10% 2006383 3,5,6,7,10,11 0.49%
46 +4000 hours -10% 2008691 3,5,6,10,11 0.61%
Table 24 Renewable energy uncertainty implementation at 10% bound

Model Name Annual average cost ($) Invested units Cost variation
40 Model 10 + 2003146 3,5,6,7,10,11 0.33%
+1000 hours -20% compared to
renewable uncertainty model 10
47 +1000 hours -30% 2006426 3,5,6,7,10,11 0.49%
48 +1000 hours -40% 2008691 3,5,6,10,11 0.61%
Table 25 Renewable energy uncetainty in microgrids with different bounds

Looking at the final results in tables 24 and 25 above and comparing them to the load
uncertainty in tables 18 and 20 it is possible to see that the variation in the final cost
with renewable energy uncertainty is lower than with load uncertainty; so load
uncertainty affects more the final system cost. It is possible to explain because
renewable energy uncertainty only affects to the unit 7 (wind power generation), which
only possesses 2 MW out of the 14 MW which compound the microgrid among
dispatchable units, non dispatchable units and energy storage systems. Then the worst
case created by the renewable uncertainty makes the annual average cost 0.61% more
expensive than in the model without uncertainty.
On the other hand renewable energy uncertainty causes an easier variation of the DERs
implementation while load uncertainty does not do that. Hence with 2000 hours or more
with 20% of uncertainty or 4000 h with 10% there is not a wind power turbine
implementation because of the drop in the unit power ratio, it also implies that more
hours of uncertainty do not affect the model anymore due to the non-existence of non-
dispatchable units.
However in the load uncertainty case, even with 1000 hours of uncertainty and an upper
bound of +100% there is not a change of the invested units although the variation of the
annual average cost is 11.98%; and it is needed to change the VOLL to 10000 $/MWh
to see the first modification of the units implemented with 1000 hours and 70% of
uncertainty.

Model Name Invested Investment Operation Total Annual

units cost ($) cost ($) cost ($)
Model 10
+1000 h -20%
40 renewable energy 3,5,6,7,10,11 1241768 761378 2003146
uncertainty

45 3000 h-10% 3,5,6,7,10,11 1241768 764615 2006383

2000h -20%
41,46,48 4000 h -10% 3,5,6,10,11 2008691
968188 1040503
+1000 h -40%
Table 26 Investment and operation costs in models 41, 45, 46 and 48

In the table 26 it is possible to see that although the importance of non dispatchable
units in the final annual cost variation of the microgrid is not very important with a

63
maximum of 0.61%; the really cheap operation cost and the high investment cost of the
non dispatchable generation (which is assumed to be 0) make an important difference
between the operation and the investment costs of the microgrid. It remarks the
importance of a good forecast because an investment in wind power implies a high cost
(22% of difference between models with and without the non dispatchable unit 7). If the
investment cost is not reflected in the amount of non-dispatchable energy generated and
at the end it is lower than the forecasted, there would not be a decrease in the operation
cost and it could lead to a waste of invested money with a negative economic
repercussion.
In the figures 15 and 16 there is a comparison of the influence between load and
renewable energy uncertainties which shows the results explained in this section about
how the load uncertainty affects more the cost in the model than the renewable energy
uncertainty. In the figure 15 there is a comparison with the same percentage bounds
(10%) and different number of uncertainty hours and in the figure 16 there is a
comparison with the same numbers of uncertainty hours (1000h) but different
percentage bounds:

Load and Renewable Energy Uncertainty Influence (I)

7
Model without uncertainty
6
1000 hours with both upper and
5 lower 10% bound
4 1000 hours with the worst case
% 10% bound
3 2000 hours with the worst case
2 10% bound
3000 hours with the worst case
1 10% bound
0 5000 hours with the worst case
Percentage difference with Percentage difference with 10% bound
load uncertainty renewable energy
uncertainty

Figure 15 Load and renewable energy uncertainty comparison with the same percentage bound and different
number of uncertainty hours

64
 Load and Renewable Energy Uncertainty Influence (II)
5
Model without uncertainty
4,5
4
3,5 1000 hours with the worst case
10% bound
3
% 2,5 1000 hours with the worst case
2 20% bound
1,5 1000 hours with the worst case
1 30% bound
0,5 1000 hours with the worst case
0 40% bound
Percentage difference with Percentage difference with
load uncertainty renewable energy
uncertainty

Figure 16 Load and renewable energy uncertainty comparison with the same numbers of hours but different
percentage bounds

5.3.4 Load and renewable energy generation uncertainties together

The tables below show the results of combine both load and renewable energy
uncertainties at the same time. These combinations have been performed with few of the
cases studied before.

Model Name Annual average cost ($) Invested units Cost variation
10 Day 206 9 hours, + 1996520 3,5,6,7,10,11
Availability constraints,
No uncertainty
49 1000 h +- 10% load and 1997159 3,5,6,7,10,11 0.032%
1000h +-20% renewable compared to
energy model 10
50 1000 h +10% load 2026218 3,5,6,7,10,11 1.49%
1000h -20% renewable
51 2000 h +10% load 2056123 3,5,6,7,10,11 2.98%
2000h -20% renewable
52 3000 h +10% load 2081818 3,5,6,10,11 4.27%
3000h -20% renewable
Table 27 Load and renewable energy implementation together

In this case it is possible to see how when both uncertainties are combined the final
annual average cost is almost equal to the results if the uncertainties are implemented
separately. It can be seen in the next table (the variation is measured respect the model
10, which only has power cut and reliability constraints):

65
 Percentage Percentage variation Sign Percentage variation
variation with with renewable with load and
load uncertainty energy uncertainty renewable energy
uncertainty
1000 h +- 10% 0.008% 0.024% = 0.032%
load and 1000h
+-20% renewable
1000 h 1.15% 0.33% ≈ 1.49%
+10% load 1000h
-20% renewable
2000 h 2.32% 0.61% ≈ 2.98%
+10% load 2000h
-20% renewable
3000 h 3.45% 0.61% ≈ 4.27%
+10% load 3000h
-20% renewable
Table 28 Comparison of load and renewable energy uncertainty separately and altogether

There is equality in the first case and inequalities in the following rows of the table, but
these inequalities are paltry. Therefore the addition of both uncertainties at the same
time barely changes the addition of both uncertainties separately.
Also with the union of both uncertainties there is a distributed energy resources
implementation change, so if with only the load uncertainty implementation there was
not a unit implementation change and with only the renewable energy uncertainty
implementation there was a unit implementation change with 2000 hour of uncertainty
per year, now with both uncertainties at the same time and due to with more load more
power capacity is needed, the wind power unit remains when there are 2000 hours of
uncertainty per year but it is not implemented when the number of hours with
uncertainty is 3000.

5.3.5 Uncertainty in electricity market price

The initial hours of uncertainty and the percentage variation for the electricity market
price are 2000 hours per year and 20% respectively. In the next tables the results with
different bounds are presented.

Model Name Annual average cost ($) Invested units Cost variation
10 Day 206 9 hours, + 1996520 3,5,6,7,10,11
Reliability constraints,
No uncertainty
53 Model 10 1967550 3,5,6,7,10,11 -1.45%
+2000 h +-20% price compared to
uncertainty model 10
54 +2000 h -20% price 1983815 3,4,7,10,11 -0.64%
uncertainty LS=2.43 MWh
55 +2000 h -20% price 1992520 3,5,6,7,11 -0.2%
uncertainty
(VOLL=10000)
56 +2000 h +20% price 1930896 3,4,5,6,7,10,11 -3.28%
uncertainty
Table 29 Electricity market price uncertainty implementation (VOLL in $/MWh) (LS=Load Shedding)

66
As it is possible to see looking at table 29 this case is different from the load and
renewable energy uncertainty cases studied before. The table shows that the electricity
market price uncertainty does not imply an increase in the annual average cost, indeed it
causes a decrease in the in the final annual average cost. Only when there are 2000
hours with uncertainty of -20% the annual average cost remains similar to the one
without uncertainty (only being 0.2% lower).
Besides the reduction of the final annual average cost it is possible to see more variation
in the units implemented than in the previous cases. But why does it happen?
The reason of this effect with the electricity market price uncertainty is the interaction
between the shape of the electricity market price (figure 8), the generation units invested
and the energy storage systems. The variations in the prices makes possible more
profitable trading with the energy and be able to get more income selling energy at
favorable times. Therefore when the bound is in its upper position the result is more
favorable to develop the microgrid and the average annual cost decreases. It is proved in
the table 30, where the annual average cost is split in investment and operational cost
and in section 5.3.6 where the model without energy storage systems is studied to
confirm the high dependence of this microgrid with these devices.

Model Name Investment Operation Total Annual

Units Invested
Cost cost Cost

10 3,5,6,7,10,11 1241768 754752 1996520

53 +2000 h +-20% 3,5,6,7,10,11 1241768 725782 1967550

54 +2000 h -20% 3,4,7,10,11 1073001 910814 1983815

+2000 h -20%
55 (VOLL=10000) 3,5,6,7,11 1028589 963931 1992520

56 +2000 h +20% 3,4,5,6,7,10,11 1428292 502604 1930896

Table 30 Investment and operation cost study with electricity market price uncertainty (VOLL in $/MWh)

As it is possible to see the more the uncertainty tends to the upper bound the more units
are implemented and the lower the operation cost and also the final annual average cost
are. It also highlights the relation of electricity market price, units invested and
operation cost because now the operation cost is more important than the investment
one, and when there is not uncertainty it happens the opposite (section 5.1). So now
when there are more units implemented of course the investment cost is higher than
when there are few units but the operation cost is much lower so at the end the total cost
is lower with more units than with fewer units.
The electricity market price uncertainty at different bounds affects the unit
implementation in the microgrid more than in the previous uncertainty cases studied
because of the higher benefits they imply in the operation cost.
Thereby when the uncertainty bound is at its lowest point (model 54) units 5 and 6 are
changed by unit 4 so the total capacity of the microgrid is reduced in one MW compared
to mode 10 and 53. This happens because the low electricity market prices do not allow
a more favorable internal trading between micro and main grid and it also implies 2.43
MWh of annually load shedding. Using a higher VOLL (model 55) units 5 and 6 remain
and they are not substituted by unit 4, but the lack of advantage in energy trading

67
between the main grid and the microgrid is reflected in the no implementation of the
energy storage system number 10. Finally with the electricity market price uncertainty
at its highest bound (model 56) all units implemented without uncertainty remain in the
system and unit number 4 (3 MW) is added to take advantage of the more favorable
internal trading which is viable when the electricity market prices are high.
Uncertainty in electricity market price is different and more complex than load and
renewable energy uncertainties, so a table with variations in uncertainty hours and
percentages is presented to get a better sight of how electricity market price uncertainty
affects the system:

Model Name Annual average cost ($) Invested units Cost variation
57 Model 10 1953604 3,5,6,7,10,11 -2.15%
+3000 h +-20%
electricity market
price uncertainty
58 3000 h +20% 1890260 3,4,5,6,7,10,11 -5.32%
59 3000 h -20% 1966878 3,4,7,10,11 -1.48%
LS=2.43 MWh per year
60 2000 h +-30% 1930798 3,5,6,7,10,11 -3.29%
61 2000 h +30% 1873723 3,4,5,6,7,10,11 -6.15%
62 2000 h -30% 1948839 3,4,7,10,11 -2.38%
LS=2.43 MWh per year
63 3000 h +-30% 1910270 3,4,7,10,11 -4.32%
64 3000 h +30% 1809745 3,4,5,6,7,10,11 -9.35%
65 3000 h -30% 1941426 3,4,7,10,11 -2.76%
LS=2.43 MWh per year
66 2000 h +-10% 1988839 3,5,6,7,10,11 -0.38%
67 2000 h +10% 1978192 3,5,6,7,10,11 -0.92%
68 2000 h -10% 2002055 3,5,6,7,11 0.28%
69 3000 h +-10% 1985508 3,5,6,7,10,11 -0.55%
70 3000 h +10% 1961153 3,4,5,6,7,10,11 -1.77%
71 3000 h -10% 2004517 3,5,6,7,11 0.4%
Table 31 Electricity market price uncertainty implementation with a higher range (LS=Load Shedding)

In model 57, 58 and 59 there is an increment in the number of uncertainty hours. In

these models with 3000 hours of uncertainty the pattern marked by models 53, 54 and
56 is followed and the final annual average costs are even lower. When the uncertainty
is both at the upper and lower bound (±20%) the units invested do not change compared
to the base model 53 but with 3000 hours of uncertainty the system is able to reach
more advantages in the internal trading and as a result the cost decreases. When the
uncertainty is at its upper bound the annual average cost experiments an important
decrease and also the system keeps the same number of units invested as in the model
with 2000 hours of uncertainty (model 56). Finally when the uncertainty is at its lower
bound (model 59) the units invested remain the same as in the model 54 with 2000
hours of uncertainty, but the difference in the final annual average cost is lower than in
the other two cases, it is due to the load shedding reduces the margin of the system
performance.
In model 60, 61 and 62 there is an increment in the percentage variation, the upper and
lower bound are plus and minus 30% respectively. As it happened before, these three
models also follow the pattern of models 53, 54 and 56 and the final annual average
costs decrease compared with the model without uncertainty. Comparing them with
models 57, 58 and 59 the units invested are the same, which also are the same as in

68
model 53, 54 and 56 but the decrease in the annual average cost is even higher. So a
variation of 10% in the lower and upper bounds affects more the model than an increase
of 1000 hours per year in the uncertainty time.
Obviously when hours of uncertainty are incremented from 2000 to 3000 (models 63,
64 and 65) the variation in the final annual average cost is even deeper, and the
differences between them and the model without uncertainty (model 10) reach -4.32%, -
9.35% and -2.76% respectively when the bounds of the uncertainty are placed both at
the upper and lower extremes, only at the upper and only at the lower extreme.
Regarding the units invested there is only a difference when the uncertainty bounds are
placed both at the upper and the lower point at model 63; units number 5 and 6 are not
invested and unit 4 replaces them.
When the uncertainty in the models is 10% and the number of hours is 2000, the
variation in the final cost is considerably lower. When the uncertainty is at the upper
and lower bound (±10%, model 66) the variation is only -0.38% with the same units
invested as models with ±20% of uncertainty and ±30% with 2000 hours (models 53, 57
and 60). When the uncertainty is at its upper level there is an important change
compared with the rest of the models and unit 4 is not added to the units implemented
before; it is because with only +10% of electricity market price uncertainty, although
the internal trade improves there are not as many chances as before to trade between
high and low electricity market price hours and it is not worth to invest more units.
Furthermore in model 68 when the uncertainty is at its lowest there is not benefit
anymore and the annual average cost is higher (+0.28%) than in model 10 without
uncertainty. The lack of benefit in the internal trading is also reflexed by the no
investment of the energy storage system unit 10, which implies also that units 5 and 6
remain instead of being replaced by unit 4 as in the models with uncertainty at their
lower bound previously commented.
Finally when there is an uncertainty variation of 10% but there are 3000 hours of
uncertainty per year as it happened in the cases with 20% and 30 % of uncertainty, the
final costs are incremented compared to when there are only 2000 hours of uncertainty
per year. Thus it follows the pattern of the 10% and 2000 hours uncertainty models,
with a higher variation of the final annual cost. Due to there are more hours of
uncertainty, now it is worth to invest the unit 4 when the uncertainty is at its upper
bound. When the uncertainty is at its lower bound (model 71) the same units as in
model 68 are implemented and the final cost becomes higher following the pattern.
Models 68 and 71, both with a lower bound of -10%, clearly indicate that not all the
uncertainty in the electricity market price results in a decrease of the annual average
cost. But what is possible to conclude after looking at tables 26 and 28 is that:
 Upper bounds (+X%) in electricity market price lead the microgrid to:

 Sell more expensive: It happens when the randomness places the upper
bound in an hour in which previously (without uncertainty) the microgrid
was exporting energy.
 Sell more hours: The increase in the electricity market place at some
hours makes that the price changes in a way that previously it was lower
than the operation cost of generators and finally it becomes higher than
that operation cost. It results in a change from purchase hours to sale
hours.

69
  Buy more expensive: It occurs at the hours when the system was
previously (without uncertainty) buying energy from the main grid and
now with the upper bound the market price raises but the increment is
not high enough to surpasses the operation cost of the generation units.
 Buy less hours: It is a consequence of the before explained bullet point 2.
 The four points explained before make that the microgrid tends to invest
more units, and this results in:
-Increase of the investment cost.
-Decrease of the operation cost.

 Lower bounds (-X%) in electricity market price lead the microgrid to the
opposite points explained before:

 Sell cheaper: It happens when the randomness places the lower bound in
an hour in which previously (without uncertainty) the microgrid was
exporting energy and this drop keeps the electricity market price higher
than the operation cost of the generators.
 Sell less hours: Due to the drop of the electricity market place, at some
hours previous prices which were more expensive than the operation cost
of generators become cheaper than the generation operation cost.
Therefore there is an increment of purchase days and a decrement of sale
days.
 Buy cheaper: It happens when the randomness places the lower bound in
an hour in which previously (without uncertainty) the microgrid was
importing energy. Thus with the drop of the electricity market price this
importation is less expensive.
 Buy more hours: It is a consequence of the before explained bullet point
2.
 The four points explained before make that the microgrid tends to invest
less units, and this results in:
-Decrease of the investment cost.
-Increase of the operation cost.
In both cases energy storage systems have a tremendous importance maximizing the
benefits of the internal trading as it is proved in the next section.

70
5.3.6 Uncertainty in electricity market price without energy storage systems
(ESSs)

Let us see what happen when there is electricity market price uncertainty without
energy storage systems:

Model Name Annual average cost ($) Invested units Cost variation
72 +2000 h +-20% price 2035926 3,5,6,7 1.97%
uncertainty LS=4.7 MWh
VOLL 2000
73 +2000 h +-20% price 2037023 3,4,5,7 2.02%
uncertainty
VOLL 10000
74 +2000 h -20% price 2038207 3,5,6,7 2.08%
uncertainty LS=4.7 MWh
75 +2000 h -20% price 2044608 3,4,5,7 2.40%
uncertainty
VOLL 10000
76 +2000 h +20% price 2007293 3,4,5,6,7 0.54%
uncertainty
VOLL 2000
77 +2000 h +20% price 2007293 3,4,5,6,7 0.54%
uncertainty
VOLL 10000
Table 32 Electricity market price uncertainty implementation without ESSs (VOLL in $/MWh) (LS=Load Shedding)

Looking at the table 32 is possible to realize two important points:

 The first one is to check that energy storage systems, as commented before, are
essential in this system and one of the causes of that with electricity market price
uncertainty the final cost in the microgrid generally decreases. It is because the
internal trading they make between the high market price hours and the low
ones. Thereby the results in the models with electricity market price uncertainty
and without ESSs show an increase in the annual average microgrid cost. It
clearly shows that the investment cost is lower because the ESSs are not
invested and the operation cost is much higher because the lack of internal
trading.

Type of uncertainty Cost variation with ESS Cost variation without ESS
+2000 h +-20% price uncertainty -1,45% 2,02%
+2000 h -20% price uncertainty -0,20% 2,40%
+2000 h +20% price uncertainty -3,28% 0,54%
Table 33 Comparison between the electricity market uncertainty with and without ESSs

Thus in the table 33 is possible to see how the variation changes in 3.47% when
there is 2000 hours of uncertainty per year at both upper and lower bounds, how
it changes 2.60% when the bound is at its lower level, and 3.82 % when the
bound is at its upper level.
 The second point is how important is the value of loss load (VOLL) in specific
occasions. Thus, looking at model 72 and 73, and 74 and 75, it is possible to see
that a change in VOLL can avoid the load shedding almost without increasing

71
 the cost in the microgrid. So model 72 has an annual average cost of $ 2035926
and a load shedding of 4.7 MWh per year, while model 73 has no loss of load
and the annual average cost is $ 2037023, which results in 0.05% cost
difference. This paltry cost difference can make in a huge improvement to the
system when the load is specifically critic as in research or data centers. The
same conclusion results to compare model 74 and 75. Where it is possible to
avoid 4.7 MWh of annually load shedding increasing the cost in 0.31%.

5.3.7 Comparison of results with the three different uncertainties

Now the three uncertainties: load, renewable energy and electricity market price are
compared being applied to the system with the same number of hours per year (2000)
and the same percentage bound (20%).
The percentage bound is placed in the worst case possible for each type of uncertainty:
 +20% for load.
 -20% for renewable energy.
 -20% for electricity market price.
Model Name Annual average cost ($) Invested units Cost variation
78 Model 10 +2000 h +20% 2089802 3,5,6,7,10,11 4.67% compared
load uncertainty LS=1.66 MWh to model 10
40 Model 10 + 2000 h -20% 2008691 3,5,6,10,11 0.61%
renewable energy
55 Model 10 +2000 h -20% 1992520 3,5,6,7,11 -0.2%
market price
56 Model 10+2000 h +20% 1930896 3,4,5,6,7,10,11 -3.28%
price uncertainty
Table 34 Comparison of the three kinds of uncertainties with the same hours and bounds (LS=Load Shedding)

On the one hand, comparing the annual average cost variation, the load uncertainty is
the one which affects more the model, followed by the renewable energy and finally the
electricity market price uncertainty, which makes the model more favorable in
economic terms. Even if we include in the comparison the model 56, which applies
+20% of electricity market price uncertainty but does not lead to the worst case possible
and causes a larger variation, the variation in absolute terms caused by the electricity
market price uncertainty is lower than the caused by the load uncertainty.
On the other hand, even though the model with the load uncertainty suffers the largest
variation of the annual average cost the units implemented remain the same. The other
types of uncertainties affect less the cost variation but more the changes in unit
implementation, specially the electricity market price uncertainty.

72
 Percentage Cost Variation depending on the type
of uncertainty
6
2000 h 20% load
4 uncertainty
2000 h -20% renewable
2 energy
% 2000 h -20% market price
0

-2 +2000 h +20% price

uncertainty
-4
Figure 17 Comparison of the final annual average cost depending on the uncertainty kind

5.3.8 Uncertainty in load, renewable energy and electricity market price

altogether
Finally the three types of uncertainty are implemented altogether as it can be seen in the
table 35:

Model Name Annual average Invested units Load Shedding Cost

cost ($) (MWh) variation
10 Day 206 9 hours, + 1996520 3,5,6,7,10,11 0
Reliability constraints,
No uncertainty
79 Uncertainty add.: 1968171 3,5,6,7,10,11 0 -1.42%
+1000 h +-10%load compared
+1000 h +-20% to model
renewable +2000 h +- 10
20% electricity price
80 +1000 h +10%load 1997229 3,5,6,7,10,11 0 0.035%
+1000 h -20%
renewable +2000 h +-
20% price
81 Model 80+NO ESS 2067175 3,5,6,7 5.61 3.54%
82 Model 80+NO ESS+ 2069457 3,4,5,6,7 0.63 3.65%
VOLL 10000
83 +1000 h +10%load 2013943 3,4,7,10,11 3.42 0.87%
+1000 h -20%
renewable +2000 h -
20% price
84 +1000 h +10%load 2019465 3,5,6,10,11 0.9 1.15%
+1000 h -20%
renewable +2000 h -
20% price
+VOLL 10000
85 Model 83+NO ESS 2068072 3,5,6,7 5.61 3.58%
86 Model 83+NO ESS 2075658 3,4,5,7 0.63 3.96%
+VOLL 10000
*Load Shedding is considered per year, VOLL in $/MWh
Table 35 Load, renewable energy generation and electricity market price uncertainty implementation

73
In this case it is possible to see how when all three uncertainties are combined the final
results are almost equal to the results if the uncertainties are implemented separately. It
can be seen in the next table (the variations are measured compared to the model 10,
which only has power cut and reliability constraints). It happens both when there are or
not energy storage systems (ESSs) implemented:

Percentage Percentage Percentage Sign Percentage variation

variation variation variation with with load,
with load with electricity renewable energy
uncertainty renewable market price and electricity
energy uncertainty market price
uncertainty uncertainty

Model 10 +1000 h +- 0.008% 0.024% -1.45% = -1.42%

10%load +1000 h +-
20% renewable
+2000 h +-20% price
Model 10 +1000 h 1.15% 0.33% -1.45% = 0.035%
+10%load +1000 h -
20% renewable
+2000 h +-20% price

Model 10 +1000 h 1,15% 0,33% -0,59% = 0.87%

+10%load +1000 h -
20% renewable
+2000 h -20% price
Model 10 +1000 h 1,15% 0,33% -0,20% ≈ 1.15%
+10%load +1000 h -
20% renewable
+2000 h -20% price
VOLL 10000
Table 36 Comparison of load, renewable energy and electricity market price uncertainty separately and
altogether with ESSs

Percentage Percentage Percentage Sign Percentage variation with

Systems variation variation with variation with load, renewable energy
with no ESSs with load renewable electricity and electricity market
uncertainty energy market price price uncertainty
uncertainty uncertainty

Model 10 +1000 h 1.15% 0.33% 1.97% <= 3.54%

+10%load +1000 h -
20% renewable
+2000 h +-20% price

Model 10 +1000 h 1,15% 0,33% 2.02% <= 3.65%

+10%load +1000 h -
20% renewable
+2000 h +-20% price
VOLL 10000

Model 10 +1000 h 1,15% 0,33% 2.08% = 3.58%

+10%load +1000 h -
20% renewable

74
+2000 h -20% price
Model 10 +1000 h 1,15% 0,33% 2.40% <= 3.96%
+10%load +1000 h -
20% renewable
+2000 h -20% price
VOLL 10000
Table 37 Comparison of load, renewable energy and electricity market price uncertainty separately and
altogether without ESSs

As commented before in the case of electricity market price uncertainty, when the three
uncertainties occur there are some cases like models 81, 82, 85 and 86 when there are
not ESSs implemented: that if the VOLL is low (2000 $/MWh) there is load shedding,
but if the VOLL is higher this shedding is almost completely avoided and the final
annual average cost of the system barely changes, which can be a prominent
improvement in the customer’s service. This remarks again the importance of the value
of loss load in microgrids.

5.4 Microgrid and no microgrid comparison

In the table 38 the results of the final models with all the three uncertainties and without
the possibility of implementing a microgrid are shown:

Model Name Annual Invested units Load Shedding Cost variation

average cost (MWh)
87 Model 10 +1000 h 2171179 0 74.16 8.7%
+10%load +1000 h - Compared to
20% renewable model 80
+2000 h +-20% price
VOLL=2000
+NO MICROGRID
88 Model 87+VOLL 2698278 0 74.16 35.1%
10000
89 Model 10+1000 h 2077849 0 74.16 3.17%
+10%load +1000 h - Compared to
20% renewable model 83
+2000 h -20% price
VOLL=2000
+NO MICROGRID
90 Model 89+VOLL 2604947 0 74.16 29.34%
10000
*Load Shedding is considered per year, VOLL in $/MWh
Table 38 Final models with the three types of uncertainty implementation without microgrid

Figures 18 and 19 show graphics which represent the cost differences when the
microgrid is implemented in the system and when it is not; both with a VOLL= 2000
$/MWh and a VOLL=10000 $/MWh. It is important to realize how large is the
difference in the annual average cost, especially when the VOLL is 10000 $/MWh,
which reaches up to 35,1% of annual cost difference. Obviously the models without
microgrid have load shedding due to the existent power cut and the no possibility to
invest distributed energy resources.

75
 Microgrid with 9 hours of
3000000 powercut at peak load day,
2698278 reliability constraints, 1000 h
+10%load +1000 h -20%
2500000 renewable +2000 h +-20%
2171179 +35,1%
1997229 price
2000000 Sytem without Microgrid
+8,7% (VOLL=2000 $/MWh)
$
1500000

1000000
Sytem without Microgrid
(VOLL=10000 $/MWh)
500000

0
Annual average cost
Figure 18 Microgrid and no microgrid cost comparison with different VOLL

Microgrid with 9 hours of

3000000 powercut at peak load day,
2604947 reliability constraints, 1000
2500000 h +10%load +1000 h -20%
2077849 +29.34% renewable +2000 h -20%
2013943 price
2000000
+3,17% Sytem without Microgrid
$ (VOLL=2000 $/MWh)
1500000

1000000

500000 Sytem without Microgrid

(VOLL=10000 $/MWh)
0
Annual average cost

Figure 19 Microgrid and no microgrid cost comparison with different VOLL

The two systems studied in the graphics above have the same load and renewable
energy uncertainty but one has the electricity market price uncertainty placed both at its
upper and lower bound (±20%) and the other one has the electricity market price
uncertainty at its lower bound (-20%). It is important to point out that when the
microgrid is not implemented, in the case with +-20% uncertainty the cost is higher than
in the case with only -20% of uncertainty. Therefore and knowing that the model with -
20% uncertainty and microgrid implementation is more costly than the model with
±20% uncertainty and microgrid implementation: the final difference between the
model with and without microgrid is even higher in the model with ±20% of electricity

76
market price uncertainty. It is understandable because without microgrid it is necessary
to import as much energy form the main grid as needed and the lower the electricity
market price is the lower the annual average cost is. It also proves that microgrids are
more favorable when there is a variation to the upper bound in the electricity market
price as it was explained in section 5.3.5; because with high electricity market prices
and the microgrid it is possible to sell energy to the main grid and get revenues and
without the microgrid the high electricity market prices imply big expenses to buy the
electricity.

5.5 Variation in the power cut hours

The final models with the three different uncertainties implemented (load demand,
renewable energy generation and electricity market price) are simulated for longer
power cut time periods (Um) of 24 and 48 hours distributed along the year in groups of
12 hours. The results are shown in the table below:

Model Name Annual average Invested units Cost variation

cost ($)
91 Model 10 +Um 24 hours +1000 1998759 3,5,6,7,10,11 Compared to model
h +10%load +1000 h -20% 80
renewable +2000 h +-20% price 0.076%
+VOLL 2000
92 Um 24 hours +1000 h +10%load 2017469 3,5,6,10,11 Compared to model
+1000 h -20% renewable +2000 LS=0.95 83
h -20% price + VOLL 2000 0.175%
93 Um 24 hours +1000 h +10%load 2022440 3,5,6,10,11 Compared to model
+1000 h -20% renewable +2000 LS=0.95 83
h -20% price + VOLL 10000 0.42%
94 Um 24 hours +1000 h +10%load 2023385 3,5,6,7,10,11 Compared to model
+1000 h -20% renewable +2000 83 0.47%
h -20% price +VOLL 1000000 Compared to model
93 0.046%
95 Um 48 hours +1000 h +10%load 2001638 3,5,6,7,10,11 Compared to model
+1000 h -20% renewable +2000 80
h +-20% price +VOLL 2000 0.22%
96 Um 48 hours +1000 h +10%load 2020353 3,5,6,10,11 Compared to model
+1000 h -20% renewable +2000 LS=0.95 83
h -20% price +VOLL 2000 0.31%
97 Um 48 hours +1000 h +10%load 2025324 3,5,6,10,11 Compared to model
+1000 h -20% renewable +2000 LS=0.95 83
h -20% price +VOLL 10000 0.56%
98 Um 48 hours +1000 h +10%load 2026214 3,5,6,7,10,11 Compared to model
+1000 h -20% renewable +2000 83 0.61%
h -20% price +VOLL 1000000 Compared to model
97 0.044%
* Load Shedding (LS) in MWh per year and VOLL in $/MWh
Table 39 Implementation of longer power cuts with the three uncertainties altogether

Comparing models 91 and 95 with model 80 (they have the same uncertainty
percentages but different duration of the power cut) the differences in the results are
almost inexistent. The units implemented do not change and the annual average cost is

77
0.076% higher when there are 24 hours of power cut instead of 9 and 0.22% higher
when there are 48 hours. In all of them there is not load shedding.
In the models with an electricity market price uncertainty of -20% there are larger
differences between the models with different power cut times. Models 92, 93, 94, 96,
97 and 98 are compared to model 83 (they have the same uncertainty percentages but
different duration of the power cut); model 83 has units 3, 4,7,10 and 11 implemented
and a load shedding of 3.42 MWh per year. In this case models 92, 93 and 94 (which
have 24 hours of power cut) have a reduced load shedding and a different number of
units implemented in the microgrid. Specifically in models 92 and 93 units 4 and 7 are
substituted by units 5 and 6, so the capacity in the microgrid is reduced in 1 MW; but in
model 83 with units 4 and 7 there were 3 MW of dispatchable power (unit 4) and 2 MW
of non-dispatchable (unit 7) and in models 92 and 93 although the total capacity of the
microgrid is 1 MW lower the dispatchable capacity is 1 MW higher so it finally results
in that the load shedding is reduced from 3.42 MWh per year to 0.95 MWh. In these
models with VOLL=2000$/MWh (model 92) and VOLL=10000 $/MWh (model 93)
there is still load shedding, and the annual average cost increases 0.175% and 0.42%
respectively. In models 96 and 97, which have a power cut of 48 hours and a VOLL of
2000 $/MWh and 10000 $/MWh respectively, the results are the same as in models 92
and 93 previously explained, having a decrease in load shedding and the same change in
units invested, but the increase in the annual average cost is higher reaching 0.31% and
0.56% comparing it with model 83 because of their larger power cut periods.
Finally in models 94 and 98 which have a VOLL of 1000000 $/MWh and with 24 and
48 hours of power cut respectively there is a change again in the units invested and unit
7 is implemented again beside units 5 and 6, which results in a removal of the load
shedding due to the high VOLL. Comparing these models (94 and 98) with models 93
and 97 (which have same input data but a lower VOLL) it happens again that an
increase in the VOLL barely changes the final cost being only 0.046% and 0.044%
higher respectively but it implies that the load shedding disappear in the system. If we
compare these models (94 and 98) with model 83 the final increase of the cost is 0.47%
and 0.61% respectively.

5.6 Importance index

In this section it is going to be performed an analysis of the reliability importance for

the power line between the main grid and the microgrid. This analysis is measured via
the interruption cost index (hazard rate index). This index studies the interruption cost
with respect to component reliability [100].

(59)

In the equation 59 stands for total yearly interruption cost and for component i
failure rate. Hence measures the interruption cost of the system if the element i
(which in our case is the line between the micro and the main grid) fails [97].
The customer interruption cost is defined as follows:

78
 ∑ ( ) (60)

[kW] is the average power of the interruption, [f/y] and [h/f] are reliability
indices and [$/f, kW] and [$/kWh]are cost constants.
The importance index is compared in different cases when the microgrid is implemented
and when it is not. Hence in case of the microgrid is developed is reduced because
the average power affected in case of line’s failure is lower due to the distributed energy
resources (DERs) implemented in the microgrid satisfies the entire or part of the load
demanded.
Microgrid scenarios affect the importance index depending on the number of DERs
invested in the microgrid. Therefore if the microgrid is designed to cope with power
cuts at peak demand times the more units invested make a higher reduction of the
importance index of the power line.
Figure 19 shows a graphic comparing the importance index of the power line without
microgrid and with microgrid designed with DERs number 3, 5, 6, 7, 10 and 11
implemented.
The graphic shows the importance index calculated hourly throughout one year. At table
40 the average of the hourly importance index during a year is presented depending on
the number of units developed in the microgrid.

Invested units Importance factor (k$/failure)

without microgrid 0 136,266822
with microgrid with power cut at day 206 3,5,6,7,10,11 7,39945774
with microgrid with power cut at day 2 5,7,10,11 51,7514786
with microgrid without power cut 7,10,11 102,436475
Table 40 Importance Index comparisson among microgrids with different DERs implemented

Importance index comparison

300
250
200
150
k$/ 100
failure
50
0
hour 267

hour 4789
hour 5055
hour 533
hour 799
hour 1065
hour 1331
hour 1597
hour 1863
hour 2129
hour 2395
hour 2661
hour 2927
hour 3193
hour 3459
hour 3725
hour 3991
hour 4257
hour 4523

hour 5321
hour 5587
hour 5853
hour 6119
hour 6385
hour 6651
hour 6917
hour 7183
hour 7449
hour 7715
hour 7981
hour 8247
hour 8513
hour 1

Ih without microgrid Ih with microgrid day 206

Figure 19 Importance Index Comparison between no microgrid and microgrid

79
In this case = 9 hours/failure, = 0.73 $/f, kW and =3.07 $/kWh.

Looking at table 40 and figure 19 it is possible to see how the higher the number of
invested units in the microgrid is the lower the importance index ( ) results.

80
6. Conclusions

First of all it is important to say that the system studied is tremendously case dependent
and it is especially highly influenced by the load demand and the electricity market
price. Therefore small differences modify considerably the results: both the units
invested and the annual average cost.
Based on the results commented in chapter 5 it is possible to conclude that for this
system the microgrid implementation is highly recommended due to its large
profitability, this implementation of the microgrid results in savings for the system even
without power cuts between the main grid and the microgrid. But it is not possible to
generalize for other microgrids with different components and with other conditions of
electricity market price, load demand, renewable energy generation and/or power cuts.
Therefore the system must be always simulated before implementing a possible
microgrid to know its behavior; for this reason Benders’ Decomposition technique helps
future simulations reducing both the resolution time and the memory resources
demanded by the computer.
Besides this main conclusion there are more conclusions obtained which are divided
among general conclusions applicable to all microgrids and specific conclusions only
applicable to the system studied.
General conclusions:
High electricity market prices help the implementation of microgrids. This is because
high prices imply high expenses to buy the electricity from the grid. Thus the higher the
electricity market prices are the more interesting to develop microgrids is because it is
easy to find distributed energy resources whose total costs are cheaper than the
electricity market price and it results in profitable microgrids.
High values of loss load also help the implementation of microgrids because microgrids
avoid load shedding and the payments to customers which derive from them. Therefore
the higher the value of loss load is the more chances a microgrid has to be developed;
on the other hand if values of loss load are low it is more profitable for the companies to
pay for load shedding than to implement microgrids. Moreover reliability is also
improved when microgrids are implemented as it was measured with the reliability
importance index ( ).
Load and renewable energy generation uncertainties have a straightforward behavior
increasing the cost of microgrid performances when the load demand grows or the non
dispatchable generation decreases. This is because with high load demanded the
microgrids need to buy more energy and consequently the quantity of energy that they

81
can sell is reduced. When the non dispatchable generation is reduced, and always
assuming that its price is zero or a very low one, the microgrid operation cost increases
because it is needed to substitute this energy for other generated in more expensive
sources. However the electricity market price uncertainty has a more variable behavior
due to the interaction between the operation cost of the distributed generators
implemented and the days when the uncertainty takes place; it results in different
variation of the microgrid final cost and the units implemented, which can be higher or
lower depending on the hours of uncertainty and their percentage bound.
Specific conclusions for the system studied are commented below:
The operation cost in dispatchable generators is more determinant that the investment
cost, thus units with a lower operation cost are the ones implemented in the microgrid as
far as the units are able to supply the load demanded. It happens due to the long time
that they are working once they are implemented, which prioritizes the operation cost.
The power ratio is the main factor in order to implement the non dispatchable generators
in the system. They have similar investment costs so a high power ratio implies a high
quantity of energy with null operation cost.
In this system the availability factor has a small repercussion in this microgrid and does
not produce a high increase of the annual cost because it does not imply load shedding
in the system and the value of loss load is not applied as a penalty.
Regarding the value of loss load in some cases which have load shedding, a high value
of loss load results in the load shedding termination with a low increase of the annual
cost of the microgrid. It may imply great benefits to the microgrid customers because it
may be better to pay a bit more to avoid the load shedding which could affect the
customers’ loads.
Talking about the influence of the different uncertainties:
 Load uncertainty produces the highest variation of the annual average cost while
implies a low variation of the units’ implementation.
 Renewable energy generation uncertainty results in a small variation of the total
cost due to non dispatchable units have a low power share in the system
capacity, but it affects considerably to the implementation or not of this kind of
units. Due to the high investment cost of the non dispatchable units and their
null operation cost, the forecast and the uncertainty in them have a special
importance because if the units are finally implemented but the forecast is
wrong: the low operation cost may does not compensate the high investment
cost and there is a waste of invested money.
 Electricity market price uncertainty affects to the annual average cost in the
microgrid more than the renewable energy generation uncertainty but less than
the load uncertainty. As commented previously it produces a different variation
in the annual cost depending on the number of hour and the percentage bound in
the uncertainty but it is the uncertainty that implies the highest variation of units’
investment in the system studied.
Other important elements in microgrids are the energy storage systems, which have an
exceptional importance in this one, taking a decisive role in the internal trading between
the main grid and the microgrid. In that way the microgrid gets revenues and the energy
storage systems also reduce the investment in generation capacity discharging energy at
peak demand periods. These units even have a more important effect in the presence of
electricity market price uncertainty.

82
Finally the difference on the results when the load demand, renewable energy
generation and electricity market price uncertainties are implement separately or
altogether is almost inexistent. This shows how these different uncertainties barely
affect each other in this system.
In order to extend the last specific conclusions to other microgrids it would be necessary
to simulate other microgrids with different components and with other conditions of
electricity market price, load demand, renewable energy generation and/or power cuts;
and after that study and compare them to see different characteristics which follow the
same pattern. To do that the optimization program implemented in GAMS helps this
task because it makes possible to add the different parameters which affect the
microgrid via external files in Excel or Matlab so it is not needed to implement
important changes in the code of the program.

83
84
7. Future work

In order to make this topic more comprehensive and complete there are some future
suggestions and recommendations:
With the actual formulation units tend to generate always the maximum capacity, so as
a future recommendation it could be interesting to add ramp rates and on/off times to
see how that affects the results once it is known that the implementation of the
microgrid is profitable.
Another suggestion is to make an approach and divide the total load demanded among
fixed and adjustable load, adjusting a different value of loss load for each type of load.
This may allow us to be more precise about when and where there could be load
shedding. Fixed load must be supplied while adjustable ones can be curtailed or
deferred in response to economic incentives or islanding requirements. One step further
could be to make the load price sensitive.
Also adjust the value of loss load for different scenarios regarding when the load
shedding takes place would help to create a better simulation trying to place possible
load shedding in the moments when affect less the system. In this way changing the
value of loss load depending on season, day of the week and hour of the day would
create a more realistic system but it would also add more uncertainty because the these
different future values of loss load are not known and are difficult to know in a long
future.
Finally the last recommendation for a future work is to use Bender’s acceleration
algorithms to try to make the model faster, or try Benders’ Decomposition dual
formulation. In this thesis Benders’ Decomposition has been successfully implemented
without changing the optimization problem formulation to a dual problem, which has as
main advantages that the optimization can be easily modified and adapted to other
similar problems without implementing important changes in the code program and it
has reduced hugely the resolution time. The use of Bender’s acceleration algorithms
and/or Benders’ Decomposition dual formulation could lead to reduce even more the
time resolution and could make possible to simulate longer periods of time until reach
to simulate the whole life time of the system. On the other hand the implementation of
these techniques makes more difficult to adapt the problem to similar ones for future
variations.

85
86
8. References
[1] (2016, January) Engineering Timelimes. [Online]. http://www.engineering-
timelines.com/scripts/engineeringItem.asp?id=744
[2] The US Department of Energy. (2015, September) Microgrids at Berkeley Lab. [Online].
https://building-microgrid.lbl.gov/microgrid-definitions
[3] CIGRÉ. (2015, September) Microgrids at Berkeley Lab. [Online]. https://building-
microgrid.lbl.gov/microgrid-definitions
[4] Library of Congress, Subject Headings Volume 2 C-F.
[5] S. Won, "Introduction to smart Microgrids in Remote Communities," Canmet Energy,
2013.
[6] KEMA DNV, "Microgrids – Benefits, Models,Barriers and Suggested Policy Initiatives
forthe Commonwealth of Massachusetts," 2014.
[7] S. M. Moghaddas-Tafreshi, S.Mahdi Moosavian, N.A.Rahim, J.Selvaraj, S. M. Hakimi,
"Optimization of a Reliable Combined Heat and Power Microgrid Considered
Uncertainty," IEEE, 2013.
[8] H. Lotfi, A. Khodaei, S. Bahramirad, S. Parhizi, "State of the Art in Research on
Microgrids:A Review," IEEE, 2015.
[9] W. Reder, A. Khodaei, S. Bahramirad, "Reliability-Constrained Optimal Sizing of Energy
Storage System in a Microgrid," IEEE, 2012.
[10] California Public Utility Commission, "Electric Energy Storage: An Assessment of
Potential Barriers and Opportunities," 2009.
[11] D. Gotham, D. Nderitu, R. Carnegie, "Utility Scale Energy Storage Systems," State Utility
Forecasting Group.
[12] I. Sanchez, "Microgrid Technology: Enabling Energy Reliability and Security –
Opportunities in Campus, Commercial & Industrial Communities," in MAYA Smart Energy
Consulting.
[13] M. A. Matos, P. M. Costa, "Economic Analysis of Microgrids Including Reliability
Aspects," IEEE, 2006.
[14] M. Shahidehpour, A. Khodaei, "Microgrid-Based Co-Optimization of Generation and
Transmission Planning in Power Systems," IEEE, 2012.
[15] J. F. Clair, M. Shahidehpour, "A Functional Microgrid for Enhancing Reliability,
Sustainability, and Energy Efficiency," The Electricity Journal, 2012.
[16] A. Khodaei, "Provisional Microgrid Planning," IEEE, 2015.
[17] C. Abbey, S. Wong, G. Joos, G. Y. Morris, "Evaluation of the Costs and Benefits of
Microgrids with Consideration of Services beyond Energy Supply," IEEE, 2012.
[18] L. Söder, Electricity Market Analysis.: KTH Royal Institute of Technology, 2012.
[19] BOE Spanish Government. (2015, November) Boletín Oficial del Estado Real Decreto
661/2007. [Online]. http://www.agenbur.com/docftp/fi1182RD%20661-2007.pdf
[20] L. J. Rashkin, S. D. Pekarek, R. R. Swanson, O. Wasynczuk, "Voltage and Frequency
Regulation Strategies in Isolated AC Micro-Grids," IEEE, 2012.
[21] J. W. Kimball, L. D. Watson, "Frequency Regulation of a Microgrid Using Solar Power,"
IEEE, 2011.
[22] Dr Z. Bo, "The Challenge of protecting microgrids," Alstom Grid.
[23] A. Fidigatti, A. Oudalov, "Adaptive Network Protections in Microgrids," ABB Switzerland,
2008.
[24] K. Kauhaniemi, O. Palizban, "Microgrid control principles in island mode operation,"
IEEE, 2013.

87
[25] (2016, February) Smartgrid IEEE. [Online]. http://smartgrid.ieee.org/march-2012/527-
microgrids-an-emerging-technology-to-enhance-powersystem-reliability
[26] (2016, February) Navigant. [Online].
http://www.navigantresearch.com/research/microgrid-deployment-tracker-2q15.
[27] GTM. (2016, February) GreenTechMedia. [Online].
http://www.greentechmedia.com/articles/read/US-Microgrid-Capacity-Will-Exceed-1.8-
GW-by-2018
[28] GTM. (2016, February) GreenTechMedia. [Online].
http://www.greentechmedia.com/articles/read/Microgrid-Capacity-Will-More-Than-
Double-by-2020
[29] Berkeley Lab. (2016, February) Microgrids at Berkeley lab. [Online]. https://building-
microgrid.lbl.gov/examples-microgrids
[30] Berkeley Lab. (2016, February) Microgrids at Berkely lab. [Online]. https://building-
microgrid.lbl.gov/bornholm-island
[31] NEMA National Electrical Manufacturers Association. (2015, December) NEMA.
[Online]. https://www.nema.org/Communications/Documents/Microgrids-Macro-
Benefits.pdf
[32] E. Wood. (2016, February) Microgrid Knowledge. [Online].
http://microgridknowledge.com/worlds-largest-planned-microgrid-seeks-buyer/
[33] U.S. Department of Energy State of Hawaii. (2016, February) Hawaai.gov. [Online].
http://www.hawaiicleanenergyinitiative.org/hawaii-schools-to-install-100-mw-of-solar-pv-
on-microgrids/
[34] (2016, February) Naples Daily News. [Online].
http://www.naplesnews.com/news/environment/babcock-ranch-to-be-united-states-first-
solar-powered-city-ep-399025656-344079152.html
[35] Z. Darabi, B. Falahati, "Reliability Assessment of Power Systems Considering the Large-
Scale PHEV Integration".
[36] J. F. Prada, "The Value of Reliability in Power Systems - Pricing Operating Reserves-,"
Energy Laboratory Massachusetts Institute of Technology, 1999.
[37] D. Zhu, "Power System Reliability Analysis with Distributed Generators," 2003.
[38] M. Shahidehpour, S. Wang, "New Metrics for Assessing the Reliability and Economics of
Microgrids in Distribution System," IEEE, 2013.
[39] Y. A.-R. I. Mohamed, T. H. M. EL-Fouly, S. A. Arefifa, "Optimum Microgrid Design for
Enhancing Reliability and Supply-Security," IEEE, 2013.
[40] J. Romera, "Demand Response Application for Reliability Enhancement in Electricity
Market," 2015.
[41] J. Yu, "Maintenance Optimization Scheduling of Electric Power System Considering
Renewable Energy Generators," 2015.
[42] Deloitte Consulting, "A STUDY OF THE VALUE OF LOST LOAD (VOLL) FOR
GEORGIA," United State Agency International Developement, 2014.
[43] London Economics, "The Value of Lost Load (VoLL) for Electricity in Great Britain,"
2013.
[44] London Economics International, "Estimating the Value of Lost Load".
[45] S. Bahramirad, M. Shahidehpour, A. Khodaei, "Microgrid Planning Under Uncertainty,"
IEEE, 2015.
[46] A. Khodaei, "Provisional Microgrid Planning," IEEE, 20015.
[47] A. Khodaei, H. Lotfi, "AC Versus DC Microgrid Planning," IEEE, 2015.
[48] (2015, November) Udacity. [Online]. https://www.udacity.com/wiki/cs271/unit1-notes
[49] M. Carrión, J. M. Morales, A. J. Conejo, Decision Making Under Uncertainty in Electricity

88
 Markets., 2010.
[50] (2015, November) Slideshare. [Online]. http://es.slideshare.net/sohail40/deterministic-vs-
stochastic
[51] W. Liu, B.i Jiao, B. Hong, C. Wang, L. Guo, "Multi-objective stochastic optimal planning
method for stand-alone microgrid system," IET, 2014.
[52] W. Liu, J. Cai, B. Hong, C. Wang, L.Guo, "A two-stage optimal planning and design
method for combined cooling, heat and power microgrid system," ELSEVIER, 2013.
[53] A. Khodaei, "Microgrid Optimal Scheduling With Multi-Period Islanding Constraints,"
IEEE, 2014.
[54] Y. A.-R. I. Mohamed, T. El-Fouly, S. A. Arefifar, "Supply-Adequacy-Based Optimal
Construction of Microgrids in Smart Distribution Systems," IEEE, 2012.
[55] Y. A.-R. I. Mohamed, T. El-Fouly, S. A. Arefifar, "Optimized Multiple Microgrid-Based
Clustering of Active Distribution Systems Considering Communication and Control
Requirements," IEEE, 2015.
[56] C. Bovo, A. Berizzi, G. G. Moshi, "Optimal Operational Planning for PV-Wind-Diesel-
Battery Microgrid".
[57] S. Roggia, M. Dicorato, G. Forte, M. Trovato, M. Laera, "A procedure for day-ahead
optimal operation planning of a Microgrid," IEEE, 2013.
[58] W. Gao, H. Ren, "A MILP model for integrated plan and evaluation of distributed energy
systems," Elsevier, 2009.
[59] S.-Y. Lu , Y.-R. Chang , T.-T. Lee , M.-C. Huc, Y.-H. Chen, "Economic analysis and
optimal energy management models for microgrid systems: A case study in Taiwan,"
ScienceDirect- Elsevier, 2012.
[60] E. F. El-Saadany, M. M. A. Salama, A. M. Zein Alabedin, "Generation Scheduling in
Microgrids under Uncertainties in Power Generation," IEEE, 2012.
[61] S. S.Oren, A. Papavasiliou, "A Comparative Study of Stochastic Unit Commitment and
Security-Constrained Unit Commitment Using High Performance Computing," IEEE, July
2013.
[62] M.A. Leach, A.D. Hawkes, "Modelling high level system design and unit commitment for
a microgrid," ELSEVIER, 2008.
[63] C. R. Philbrick, E. Zak, K. W. Cheung, P. W. Sauer, P. A. Ruiz, "Uncertainty Management
in the Unit Commitment Problem," IEEE, 2009.
[64] P. Meibom, E. Denny, M. O’Malley, A. Tuohy, "Unit Commitment for Systems With
Significant Wind Penetration," IEEE, 2009.
[65] M. Shahidehpour, Z. Li, J. Wang, "Security-Constrained Unit Commitment With Volatile
Wind Power Generation," IEEE, 2008.
[66] S.M.M.Tafreshi, M. Barzegari, A. J. Shahrood, A. Navaeefard, "Optimal Sizing of
Distributed Energy Resources in Microgrid Considering Wind Energy Uncertainty with
Respect to Reliability," IEEE, 2010.
[67] S. M. Moghaddas-Tafreshi, S.Mahdi Moosavian, N.A.Rahim, J.Selvaraj, S. M. Hakimi,
"Optimization of a Reliable Combined Heat and Power Microgrid Considered
Uncertainty," IEEE, 2013.
[68] S. B. Patra, S. J. Ranade, J. Mitra, "Reliability Stipulated Microgrid Architecture Using
Particle Swarm Optimization," International Conference on Probabilistic Methods Applied
to Power Systems, 2006.
[69] G. Wu, A. Sobu, "Optimal operation planning method for isolated micro grid considering
uncertainties of renewable power generations and load demand," IEEE, 2012.
[70] S. Ushakumari, H. Kumar, "Optimal Management of Islanded Microgrid using Binary
Particle Swarm Optimization," IEEE, 2014.

89
[71] K. Bhattacharya, O. Hafez, "Optimal planning and design of a renewable energy based
supply system for microgrids," Elsevier, 2012.
[72] S.S. Mousavi-Seyedi, A. Rahimikian, S. Rezayi, F. Aminifar, "AHP-based Prioritization of
Microgrid Generation Plans Considering Resource Uncertainties," IEEE, 2013.
[73] S. V. Kulkarni, S. A. Khaparde, A. P. Agalgaonkar, "Multi-attribute Decision Making
Approach for Strategic Planning of DGs," IEEE, 2005.
[74] S. Zhang, Y. Wang , G. Sun, L. Zhou, X. Wang, P. Li, "Research on Economic Optimal
Operation of Microgrid," IEEE, 2012.
[75] C. Marnay, A. Siddiqui, "Distributed Generation Investment by a Microgrid under
Uncertainty," ERNEST ORLANDO LAWRENCE BERKELEY NATIONAL LABORATORY,
2008.
[76] K. M. Maribu, I. Wangensteen, S.-E. Fleten, "Optimal investment strategies in
decentralized renewable power generation under uncertainty," Munich Personal RePEc
Archive, 2005.
[77] C. Marnay, A. Siddiqui, "Distributedgenerationinvestmentbyamicrogridunderuncertainty,"
ELSEVIER, 2008.
[78] C. Marnay, A. Siddiqui, "Distributed Generation Investment by a Microgrid Under
Uncertainty," Lawrence Berkeley National Laboratory, 2006.
[79] F. Golestaneh, A. Malekpour, T. Niknam, "Probabilistic energy and operation management
of a microgrid containing wind/photovoltaic/fuel cell generation and energy storage
devices based on point estimate method and self-adaptive gravitational search algorithm,"
Elsevier, 2012.
[80] A. D. Hawkes, "Optimal Selection of Generators for a Microgrid Under Uncertainty,"
IEEE, 2010.
[81] R. Villafafila, N.M. Bellaaj, I. Sansa, "Optimal Sizing Design Of An Isolated Microgrid
Using Loss Of Power Supply Probability," IEEE, 2015.
[82] J. Zhao, Z. Guo, "A corrected hybrid approach for wind speed prediction in Hexi Corridor
of China," Elsevier, 2011.
[83] B. Francois, D. Abbes, X. Yan, "Operating Power Reserve Quantification Through PV
Generation Uncertainty Analysis of A Microgrid".
[84] S. Soleymani, B. Mozafari, S. Mohammadi, "Scenario-based stochastic operation
management of MicroGrid including Wind, Photovoltaic, Micro-Turbine, Fuel Cell and
Energy Storage Devices," Elsevier, 2013.
[85] S. Keane, J. Ng, J. Frayer, "Estimating the value of lost load," London Economics
International LLC, Boston, 2013.
[86] Energinet, "Generation of Electricity and District Heating, Energy Storage and Energy
Carrier Generation and Conversion. TECHNOLOGY DATA FOR ENERGY PLANTS,"
Energinet, May 2012.
[87] National Renewable Energy Laboratory (NREL), "COST AND PERFORMANCE DATA
FOR POWER GENERATION TECHNOLOGIES," National Renewable Energy
Laboratory, 2012.
[88] S. Schoenung, "Energy Storage Systems Cost Update, A Study for the DOE Energy
Storage Systems Program," Sandia National Laboratories, 2011.
[89] D. Gotham, D. Nderitu, P. V. Preckel, R. Carnegie, "Utility Scale Energy Storage Systems.
Benefits, Applications, and Technologies," State Utility Forecasting Group, 2013.
[90] F. Katiraei, J. Driesen, "Design for Distributed Energy Resources," IEEE, 2008.
[91] Investopedia. (2016, February) Investopedia. [Online]. http://www.investopedia.com/
[92] GAMS. (2016, March) GAMS. [Online]. https://www.gams.com/docs/intro.htm
[93] R. E. Rosenthal, "GAMS - A User's Guide," Washington, 2012.

90
[94] E. Castillo, R. Mínguez, R. García-Bertrand, A.J. Conejo, Decomposition Techniques in
Mathematical Programming.: Springer, 2006.
[95] Z. Caner, "Benders' Decomposition".
[96] M. Patriksson, "Benders Decomposition," 2009.
[97] P. Hilber, "Maintenance Optimization for Power Distribution Systems". Stockholm,
Sweden: KTH Royal Institute of Technology, 2008.
[98] M. A. Jafari, F. Farzan, "Closing the Loop between Short-term Operational Volatilities and
long-term Investment Risks in Microgrids," IEEE, 2014.
[99] Z. Wu, R. Bo, W. Liu, G. Zhou, W. Chen, Z. Wua, W. Gu, "Modeling, planning and
optimal energy management of combined cooling, heating and power microgrid: A
review," Elsevier, 2013.

91
TRITA TRITA-EE 2016:081
ISSN 1653-5146

www.kth.se