Lappeenranta University of Technology

Faculty of Energy Technology

Master Degree Program in Electricity Markets and Power Systems

Master’s Thesis

Arun Narayanan

ENERGY MANAGEMENT
SYSTEM FOR LVDC ISLAND
NETWORKS

Examiners: Prof. Jarmo Partanen

D.Sc. (Tech.) Pasi Peltoniemi
ABSTRACT

Lappeenranta University of Technology

Faculty of Energy Technology
Master Degree Program in Electricity Markets and Power Systems

Arun Narayanan

Energy Management System for LVDC Island Networks

Master’s thesis, 2013

108 pages, 17 figures, 6 tables

Examiners: Professor Jarmo Partanen

D.Sc. (Tech.) Pasi Peltoniemi

Keywords: low-voltage DC distribution; energy management system; load discon-

nection; prioritization; master/slave; sizing; operation modes

Recent developments in power electronics technology have made it possible to de-

velop competitive and reliable low-voltage DC (LVDC) distribution networks. Fur-
ther, islanded microgrids—isolated small-scale localized distribution networks—
have been proposed to reliably supply power using distributed generations. How-
ever, islanded operations face many issues such as power quality, voltage regula-
tion, network stability, and protection. In this thesis, an energy management system
(EMS) that ensures efficient energy and power balancing and voltage regulation has
been proposed for an LVDC island network utilizing solar panels for electricity pro-
duction and lead-acid batteries for energy storage. The EMS uses the master/slave
method with robust communication infrastructure to control the production, stor-
age, and loads. The logical basis for the EMS operations has been established
by proposing functionalities of the network components as well as by defining ap-
propriate operation modes that encompass all situations. During loss-of-power-
supply periods, load prioritizations and disconnections are employed to maintain
the power supply to at least some loads. The proposed EMS ensures optimal energy
balance in the network. A sizing method based on discrete-event simulations has
also been proposed to obtain reliable capacities of the photovoltaic array and bat-
tery. In addition, an algorithm to determine the number of hours of electric power
supply that can be guaranteed to the customers at any given location has been de-
veloped. The successful performances of all the proposed algorithms have been
demonstrated by simulations.
Acknowledgements
This thesis was written at an important juncture in my life, and it has been an en-
riching and fulfilling experience; in particular, the actual research work, ideation,
and software coding have been an adventurous and fun-filled journey. This journey
could not have been successfully completed without the assistance, contributions,
and encouragement of several people who supported me throughout the period of
the research work. I am grateful to Ensto for giving me the opportunity and as-
sistance to conduct research on this topic. In particular, the social relevance of this
project—“replacing the candle” in small villages without electricity—made it a par-
ticularly special one; I had always wanted to participate in projects that attempt to
make a practical difference to human lives, whether small or big, and I am truly
thankful for this chance. Moreover, my work was made immensely easier by the
invaluable feedback that I received during the several meetings with representatives
from Ensto, Mr. Aki Lahdesmaki, Mr. Mika Luukkanen, Mr. Tommi Kasteenpohja,
and Mr. Dai.

Pasi Peltoniemi, my supervisor, guided me throughout the thesis, listened patiently

to my questions, and gave me very good suggestions and directions. I am especially
grateful to him for allowing me to work at my pace without excessively controlling
my work, while at the same time giving me critical feedback when required and lis-
tening to all my questions—silly, basic, or important—with patience and attention.
Tero Kaipia’s outputs during meetings, feedback, and responses to my questions
were invaluable in shaping several crucial aspects of the energy management sys-
tem, and Jarmo Partanen also made useful suggestions during our meetings. I had
several discussions and consultations with my colleagues and friends Salman and
Javier who handled other issues of the LVDC network (protection and communi-
cation, respectively), and their viewpoints and approaches helped formulate and
clarify my own ideas.

Not a word of this thesis could have been written without the constant support of
my wife Amrita Karnik who uncomplainingly took my long absences from home
in her stride, encouraged me at every turn, and supplied dollops of lively conversa-
tions, happiness, peace, and fulfilment, the ideal ingredients for the success of any
undertaking. My parents and brothers, although very far away, enhanced my life
with their presence at every turn. My close friends Rahul Kapoor and Pallavi Jon-
nalagadda enriched my days with their stimulating discussions on all topics under
the sun, ranging from cricket to linguistics to game theory. Conversations over cups
of tea and coffee with Arvind Solanki and Victor Mukherjee served as important
distractions that enabled me to relax, regroup, and refocus on the task at hand.

Finally, I would like to acknowledge and pay tribute to the Open-Source Community
(whose passion, skill, attitude, and community-building never cease to astonish) and
especially the dedicated developers of LATEX, LYX, Octave, and Fedora; they made
it incredibly easy to write this thesis with complete freedom.

Lappeenranta, September 13, 2013

Arun Narayanan
Contents

List of Tables 9

List of Figures 10

1 INTRODUCTION 14

1.1 Low-voltage DC distribution networks . . . . . . . . . . . . . . . . 14

1.2 Microgrids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.3 Renewable energy . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.4 LVDC island networks . . . . . . . . . . . . . . . . . . . . . . . . 18

1.5 Thesis purpose and challenges . . . . . . . . . . . . . . . . . . . . 21

1.6 Energy management systems . . . . . . . . . . . . . . . . . . . . . 22

1.7 Thesis organization . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2 ELECTRICITY PRODUCTION, STORAGE, AND CONSUMPTION 26

2.1 Electric power production . . . . . . . . . . . . . . . . . . . . . . . 27

2.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 27

2.1.2 Solar power technology . . . . . . . . . . . . . . . . . . . 27

2.1.3 Solar irradiation . . . . . . . . . . . . . . . . . . . . . . . 31

2.1.4 Interfacing with LVDC island networks . . . . . . . . . . . 34

2.1.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2.2 Electricity storage . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2.2.2 Storage types . . . . . . . . . . . . . . . . . . . . . . . . . 35

5
 2.2.3 Lead-acid battery characteristics . . . . . . . . . . . . . . . 35

2.2.4 Battery energy management system . . . . . . . . . . . . . 37

2.2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

2.3 Electricity consumption . . . . . . . . . . . . . . . . . . . . . . . . 38

2.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.3.2 Types of loads . . . . . . . . . . . . . . . . . . . . . . . . 39

2.3.3 Typical load curve . . . . . . . . . . . . . . . . . . . . . . 41

2.3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

3 PRODUCTION CAPACITY: PLANNING and ANALYSES 45

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3.2 Capacity planning . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

3.2.1 Research background . . . . . . . . . . . . . . . . . . . . 46

3.2.2 Problem statement . . . . . . . . . . . . . . . . . . . . . . 48

3.2.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . 49

3.2.4 Results and discussion . . . . . . . . . . . . . . . . . . . . 53

3.2.5 Limitations and future studies . . . . . . . . . . . . . . . . 55

3.3 Capacity analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

3.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 55

3.3.2 Problem statement . . . . . . . . . . . . . . . . . . . . . . 56

3.3.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . 56

3.3.4 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

6
 3.3.5 Limitations and future studies . . . . . . . . . . . . . . . . 60

3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

4 ENERGY MANAGEMENT SYSTEM: PRINCIPLES AND FUNCTION-

ALITIES 62

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

4.1.1 Energy management systems . . . . . . . . . . . . . . . . . 62

4.1.2 General requirements and functions . . . . . . . . . . . . . 62

4.2 Energy and power balance principles . . . . . . . . . . . . . . . . . 64

4.3 Subnormal situations and emergencies . . . . . . . . . . . . . . . . 65

4.4 Functionalities of network components . . . . . . . . . . . . . . . . 69

4.4.1 PV array . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

4.4.2 Battery . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

4.4.3 Load division and prioritization . . . . . . . . . . . . . . . 72

4.5 Information and communication technology (ICT) systems . . . . . 75

4.5.1 Roles and requirements . . . . . . . . . . . . . . . . . . . . 75

4.5.2 Information and control signals . . . . . . . . . . . . . . . 76

4.5.3 ICT architecture . . . . . . . . . . . . . . . . . . . . . . . 77

4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

5 PROPOSED ENERGY MANAGEMENT SYSTEM 80

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

5.2 Power balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

5.2.1 Normal operations . . . . . . . . . . . . . . . . . . . . . . 81

7
 5.2.2 Subnormal operations . . . . . . . . . . . . . . . . . . . . 83

5.2.3 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

5.3 Energy balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

5.4 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . 91

5.5 Application to larger networks . . . . . . . . . . . . . . . . . . . . 95

5.6 Limitations and future studies . . . . . . . . . . . . . . . . . . . . 96

5.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

REFERENCES 102

8
List of Tables

1 Comparison between crystalline silicon and thin-film panel tech-

nologies for PV cells. . . . . . . . . . . . . . . . . . . . . . . . . 28

2 Input parameters required constantly by the control algorithm for

its operations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

3 Output parameters sent constantly by the control algorithm. . . . . 76

4 Normal Operations . . . . . . . . . . . . . . . . . . . . . . . . . . 83

5 Optimal-Operation Mode during Emergency Operations. . . . . . . 84

6 SubOptimal-Operation Mode during Emergencies with PV power.

CL: Critical Loads; LCL: Limited CL; CEL: Critical-Essential Loads;
LCEL: Limited CEL; LCENL: Limited Critical-Essential-Normal
Loads. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

9
List of Figures

1 Example of a unipolar LVDC distribution system (Salonen et al.,

2008). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2 Example of a bipolar LVDC distribution system; different customer

connection alternatives are also shown (Salonen et al., 2008). . . . 19

3 Example of an LVDC distribution network over an area (with sim-

plified topology); a PV array supplies power to randomly distributed
loads, while a battery bank is used as backup. In practice, the loads
would be distributed more unevenly and the cables would be more
circuitous. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

4 Total monthly irradiation data for the year 2004 at two locations—
latitudes 0◦ and latitude 60◦ , corresponding to the Equator and much
higher north. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

5 Load consumption (kW) for one Finnish customer for one year.
(Note: The load peak power is approximately 800 W.) . . . . . . . 41

6 Load consumption (kW) for one Finnish customer for one day in
January. (Note: The average load peak power is approximately 800
W). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

7 Load demand (kWh) versus solar irradiation (kWh) for one Finnish
customer for one year. This thesis proposes an algorithm to ensure
that they are balanced and that the two curves (blue and red) coin-
cide; to ensure this, a battery whose state of charge is controlled is
employed, and/or loads are disconnected depending on the condi-
tions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

8 Flowchart for sizing PV array and battery. . . . . . . . . . . . . . . 52

9 Battery capacity versus time for various initial battery capacities. . . 54

10 Latitude versus number of service hours. The service hours increase

from 0◦ towards the Tropic of Cancer 22.5◦ until 30◦ and then de-
creases. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

10
11 Load prioritization flow. The proposed EMS uses this mechanism
during subnormal situations or emergencies to maintain power sup-
ply to at least some loads. . . . . . . . . . . . . . . . . . . . . . . 74

12 Load prioritization handling: the main control algorithm handles

the top- and sub-level prioritizations, while each load converter han-
dles its internal prioritizations. This simplifies the control and re-
duces the computational burden on the main algorithm. . . . . . . . 75

13 Hierarchy of information and communication technologies (ICT)

structure; here, IED refers to intelligent electronic device. . . . . . 78

14 Sample network for simulations. . . . . . . . . . . . . . . . . . . . 91

15 Power balancing during normal operations. PV production, SOC

statuses, and total load demand are shown. Note that the PV power
production is deliberately reduced to zero and increased at random
intervals; in practice, 100 ms is too short a time for such drastic
variations (they would, instead, occur through the day). . . . . . . . 93

16 Voltage regulation during normal operations. The red line repre-

sents randomly generated voltages some of which are forced to be
outside the allowed limits ([750 − 10% × 750 750 + 10% × 750] =
[675 825]). The blue line represents regulated voltages. . . . . . . 94

17 Power balancing during normal and subnormal conditions. Note

that battery SOC now decreases to almost 50%, i.e., the battery is
forced to cycle much deeper. Moreover, not all load demands are
met when the battery SOC is close to 50%. The arrow points to
crisis situations where only Critical Loads 1 and 2 are connected
and the other loads are disconnected. . . . . . . . . . . . . . . . . 94

11
 List of Symbols and Abbreviations

Abbreviations

BEMS Battery Energy Management Systems

CL Critical Loads
CB Critical-Battery
CEL Critical-Essential Loads
DES Discrete-Event Simulation
DG Distributed Generation
DOD Depth of Discharge
EMS Energy Management System
HVDC Low-Voltage Direct Current
ICT Information and Communications Technology
LCEL Limited Critical-Essential Loads
LCENL Limited Critical-Essential-Normal Loads
LCL Limited Critical Loads
LOPS Loss of Power Supply
LOLP Loss of Load Probability
LVAC Low-Voltage Alternating Current
LVDC Low-Voltage Direct Current
MPPT Maximum Power Point Tracking
MV Medium Voltage
NADs Number of Autonomous Days
PSH Peak Sun Hours
SOC State of Charge
STC Standard Test Conditions

12
Symbols

BCh Battery Charge Rates

BDch Battery Discharge Rates
E Solar Irradiance
E Batt. Energy (Battery)
ELoad Energy (Load)
ELosses Energy (Losses)
E PV Energy (PV)
H Solar Irradiation
N Number of Customers
L Hourly Load Data for a Year for One Customer
LoadsC Number of Critical Loads
LoadsE Number of Essential Loads
LoadsN Number of Normal Loads
PBatt. Power (Battery)
PLoad Power (Load)
PLosses Power (Losses)
PPV Power (PV)
PPV max Maximum Power (PV)
PrC1 − PrCLoadsC Priorities for each Critical Load
PrCE1 − PrCLoadsE1 Priorities for each Essential Load
PrN1 − PrNLoadsN Priorities for each Normal Load
SOCmin Minimum Battery SOC
SOCmax Maximum Battery SOC
SOCleast Lowest Possible Battery SOC

13
1 INTRODUCTION

1.1 Low-voltage DC distribution networks

Electricity is the most important power source produced worldwide and it is inte-
gral to modern society today; without electric power supply, modern civilization
and technical progress are simply unimaginable. We have become increasingly re-
liant on electric power supply to sustain our economies, daily necessities, as well
as comforts. The use of electricity is so ubiquitous that it is practically taken for
granted, especially in countries with uninterrupted power supplies. However, the
supply of uninterrupted electric power remains a challenging problem across the
world, especially with increasing consumption and demand. Outages—short- or
long-term loss of electric power supply to an area—often occur in many parts of
the world, and they have severe effects on customers, thereby increasing the outage
costs. The need for reduced outages has led to demands for more reliable network
solutions than the traditional 3-phase AC distribution systems. The ageing of cur-
rently used AC distribution grids, as well as their sizes and complexity, have also
increased concerns about the reliable transmission and supply of electric power, es-
pecially across long distances to remote places. In this scenario, DC distribution
has been proposed as a viable alternative, especially at low voltages and in smaller
areas such as remote villages or buildings.

DC systems were used for transmission in the early days of electricity production,
but they were quickly replaced by AC systems for practical reasons such as, for ex-
ample, AC voltages could be easily changed using transformers and power could be
distributed more economically. Today, DC transmission technology is primarily fo-
cused on long-distance high-voltage DC (HVDC) transmission systems, industrial
distribution, and electric drives. However, technical and economic developments
during the last decade, especially in power electronics technology, have given the
opportunity to develop competitive distribution systems based on low-voltage DC
(LVDC) distributions (Salonen et al., 2008). The most important requirements of
distribution networks are cost effectiveness and system reliability. At low voltages,
the LVDC system is an economically feasible alternative that can enhance the re-
liability and energy efficiency of distribution systems as well as the power quality
experienced by the customers. The transmission capacity of LVDC distributions is
better than that of low-voltage AC (LVAC) distributions, leading to economic ben-
efits. Further, small-scale power generation systems called distributed generations

14
(DGs) can be connected to the LVDC distribution network. (Kaipia et al., 2006)
Today’s power electronics technology and the possibility to use 1.5-kV DC voltage
in the LV network make it possible to connect small-scale DGs and energy storages
to LV networks. The customer’s voltage quality can be improved by eliminating
voltage dips, fluctuations, and short-term voltage drops using power electronics de-
vices. The total costs of constructing and operating a distribution system can also be
decreased. The LVDC distribution system is thus an economically and technically
feasible alternative to conventional 3-phase AC distribution systems, and in fact, it
is one of the primary new technological innovations in electricity distribution today.
(Kaipia et al., 2008)

In an LVDC distribution system, the DC connection makes it possible to avoid the

construction of medium voltage (MV) branch lines and traditional LVAC networks
to supply a group of customers. In general, an LVDC distribution network com-
prises only a DC connection between the wide MV main line and the coupling
points of the customers, where the DC voltage is inverted back to AC voltage. The
DC distribution system can be constructed with unipolar or bipolar connections,
with the primary difference between these two connections being the number of
voltage levels (Kaipia et al., 2008).

1.2 Microgrids

A microgrid is defined as a small distribution system with distributed energy sources,

storage devices, and controllable loads, which operates in two steady states: grid-
connected and islanded. Microgrids are designed to be islanded and isolated from
the main grid, if and when required; typically, there is a single point of common
coupling with the macrogrid that can be disconnected. In other words, microgrids
are a localized grouping of electricity production equipment, energy storage de-
vices, and loads, with the ability to function autonomously and co-operatively. (Las-
seter, 2002; Kaplan et al., 2009) In general, the generation resources in microgrids
use small energy sources—microsources—placed at customer sites and interfaced
with the help of power electronics devices. Power electronics technologies provide
the control and flexibility required by the microgrids concept (Lasseter, 2002); of
course, power electronics interfaces have the disadvantage that they may increase
harmonic injections and they can be sensitive to system disturbances, but this is
not problematic in DC grids (Barnes et al., 2007). DGs, which are often renewable
sources such as solar or wind but can also be conventional, are typically used to

15
supply the power in microgrids. Microgrids using DGs basically represent small
electrical grids located closer to the demand location, thereby decreasing the power
failures caused by long-distance transmission grids. Power is generated locally, and
hence, the dependence on long-distance transmission lines is significantly reduced,
which, in turn, decreases the transmission losses as well. Since multiple DGs can
be used and the microgrid can be isolated from a larger network, the reliability of
the electric power supply is increased considerably. Further, microgrids can be de-
signed according to localized customer needs and any other special requirements as
well. (Rizzo, 2012)

Island networks refer to networks that are completely isolated from the main grid
and operate completely independently and co-operatively. Islanded operations lead
to various economic and technical issues such as power quality, voltage regulation,
network stability, harmonics, reliability, protection, and control; hence, the system
network must be extremely well planned. Today, partially islanded operations of
microgrids, which play a kind of temporary supporting role to the main grid opera-
tions, are more common than completely islanded networks and operations.

1.3 Renewable energy

Besides electrical transmission and distribution networks, the electric power pro-
duction industry also faces several challenges today. The availability and prices
of fossil fuels are unpredictable and highly volatile, leading to several economic
challenges and geopolitical risks. Fossil-fuel supplies are not only limited but also
expensive. Moreover, fossil fuel-based energy production leads to harmful environ-
mental consequences since fossil fuels cause pollution and increase the dangers of
adverse climate changes. Hydroelectric power is a clean and environment-friendly
source of energy, but it can also alter or damage the surroundings, for example, by
changing the water quality or by hampering aquatic life. Similarly, nuclear power
plants also have potentially damaging environmental and human consequences. In
general, the energy sector is one of the primary emitters of greenhouse gases. As a
result, it has become imperative to utilize renewable energy sources that are contin-
ually replenished, such as sunlight, wind, rain, tides, waves, and geothermal heat,
where possible. (Prof. Zervos et al., 2010; Turner, J., 1999) The deployment of
renewable energy technologies is a promising approach that can be used to mitigate
man-made climate changes, reduce hazardous pollution, and enhance local energy
independence. Energy production from renewable sources such as wind and so-

16
lar energy is an attractive alternative for solving energy crises. Renewable energy
sources have already been deployed worldwide and many countries have begun to
use them to generate electricity, especially wind and sunlight.

Since the energy production from renewable energy sources is variable, intermit-
tent, and (usually) at comparatively low voltages, they are practical for applications
to smaller grids with fewer loads, such as microgrids; this also helps in reducing
their implementation costs. Hence, the use of renewable energy sources for micro-
grids has been considered from a very early stage. Many studies have proposed
and analyzed the applicability of solar power, wind power, other renewable sources,
or their combinations to microgrids, and various real-world installations (primarily
on-grid) have been implemented across the world (Barnes et al., 2007). Most of
these studies have focused on using microgrids as a backup to the main grid and
completely islanded operations have rarely been considered due to the variability of
the production. Moreover, completely islanded microgrids using renewable energy
sources present numerous techno-economic challenges such as protection, commu-
nication, reliability, and power balance issues. Nevertheless, they are attractive be-
cause they can be employed to supply renewable-energy-based pollution-free elec-
tric power to isolated or remote locations comprising small loads, such as remote
villages that have never received electric power supply. By using such completely
islanded microgrids, such places can be supplied electricity in a cost-efficient and
effective manner; this has important social benefits since the quality of human life
will be improved.

Modern electric grids have begun to use information and communications technol-
ogy (ICT) systems to collect power supply and load demand information, perform
relevant operations, and send control signals automatically in order to improve the
efficiency, reliability, economics, and sustainability of the network; such grids are
commonly referred to as smart grids. Real-time load information is recorded by
“smart” meters and communicated by ICT systems to a control center that also con-
siders production statuses before taking the appropriate decisions; the decisions are
then communicated to the appropriate network components in the form of control
signals. Electric grids equipped with ICT systems are typically referred to as smart
grids, and they are designed to respond automatically, quickly, and efficiently to
power supply challenges. (Amin and Wollenberg, 2005)

In this thesis, an islanded and smart LVDC microgrid network that uses renewable
energy sources for production and batteries for storage has been considered; as
discussed previously, this offers the advantages of low losses, pollution, and costs,

17
but the implementation is challenging because control, protection, communication,
and other technical issues must be adequately resolved.

1.4 LVDC island networks

In an LVDC distribution system, the power production sources, storage devices,

and loads are interconnected by DC buses at low voltages and interfaced by power
electronic converters that also have DC links between them. Further, island net-
works do not have a connection to the main grid, and the entire grid is completely
isolated and independent. The LVDC distribution network can be constructed with
two implementations: unipolar or bipolar. The unipolar system has one voltage
level and all the customers are connected to this voltage level. On the other hand,
the bipolar system comprises two unipolar systems connected in series (Salonen et
al., 2008). In the bipolar system, customers can be connected between voltage lev-
els in many ways, for example, between a positive pole, between a negative pole,
and between positive and negative poles. The two systems—unipolar and bipolar
LVDC networks—are shown in Figs. 1 and 2, respectively. In this thesis, bipo-
lar LVDC networks are primarily considered since they provide low transmission
losses, high power transfer capability, high flexibility, and high availability (Lago,
2011); however, the network may also have unipolar branches.

Figure 1: Example of a unipolar LVDC distribution system (Salonen et al., 2008).

The nominal voltage of the system must be selected such that the DC network trans-
mission capacity is as high as possible, but simultaneously within the boundaries

18
Figure 2: Example of a bipolar LVDC distribution system; different customer con-
nection alternatives are also shown (Salonen et al., 2008).

of the standards set by LV directives and cable standardizations; the acquisition

prices of converters are also a factor. The European Union directive 2006/95/EC
defines the DC voltage to be in the range 75–1500 V DC (EU Low Voltage Di-
rective, 2006). Further, the cable standardization allows LV cables to be used in
DC voltage networks; the defined maximum voltage between conductors is 1500
V DC and between earth and conductor is 900 V DC (Salonen, 2008; SFS4879;
SFS4880). Hence, the nominal voltage of the bipolar network in this thesis is ±750
V.

In this thesis, an islanded bipolar LVDC microgrid has been considered for a popu-
lated area having a diameter of approximately 6 km and 200 customers. The typical
peak power of a single customer is specified to be 200 W with a projected increase
up to 800 W; therefore, the power handling capability of the total network can be
considered to be at least 160 kW for 200 customers. In practice, the populated areas
consist of small villages that have spread randomly around the area of the island
network and are located in the equatorial or tropical regions. In addition, single
or three-phase DC/AC converters are used for the actual customer connections that
have a voltage level of 230 V AC, and DC/DC converters are used to obtain DC
voltages of 48 V that can then be used for lighting purposes.

It is not possible to predict the exact network topology since this significantly de-
pends on the location where the project is implemented, leading to considerable

19
variations. Typical topologies are linear, radial, looped, and other combinations
whose complexity increases as the usage area increases. Figure 3 illustrates the
LVDC network deployed with a simplified topology. Here, the power source is the
PV array (the choice of solar power for power production is discussed in Chapter
2); a battery bank is used for backup; the loads are distributed randomly over the
area; and converters are employed for all three components. In practice, the net-
work would be somewhat more complicated with more uneven load distributions;
however, complicated networks are beyond the scope of this thesis, and a linear net-
work has been considered for simplicity in the proposed algorithms. At the same
time, the adaptability of the proposed algorithms to larger and more complicated
networks has been kept in mind while developing the EMS, and moreover, reason-
able suggestions for network expansions have been proposed.

DC/AC
=
L
=
=
DC/DC

+750 V

-750 V
=
= PV ARRAY
DC/AC
=
L
DC/DC = =

= DC/AC
L
DC/DC =
=

=
=
+750 V =
DC/DC

-750 V BATTERY

DC/DC
= L
= =
DC/AC

Figure 3: Example of an LVDC distribution network over an area (with simplified

topology); a PV array supplies power to randomly distributed loads, while a battery
bank is used as backup. In practice, the loads would be distributed more unevenly
and the cables would be more circuitous.

20
1.5 Thesis purpose and challenges

As mentioned previously, the successful deployment of such islanded networks

faces is not easy and there are several challenges such as the achievement of ef-
ficient power balance, protection, communication, and reliable operations. This
thesis addresses the issue of network stability in terms of power balancing and volt-
age regulation. Real-time operations of any electric power system must ensure that
the system remains stable and protected while meeting the customer’s power re-
quirements. Hence, a precise balance between power production and consumption
is required at all times; if this balance is not maintained, the system can become
unstable and the voltage may exceed the allowed limits, leading to damaged equip-
ment as well as outages. The purpose of this thesis is to obtain methods to achieve
efficient power balancing between the supply and demand in the network under all
conditions, and further, to ensure that the voltage is regulated within the prescribed
limits; in this manner, network stability is maintained under all circumstances and
conditions. This thesis proposes an energy management system (EMS) that is ba-
sically a system of computer-aided tools and operations used to monitor, control,
and optimize the production and/or transmission efficiencies and overall network
performances (Lukszo, 2010); the EMS is facilitated by a robust communication
infrastructure that regularly communicates the network system and component sta-
tuses, thereby enabling the software to perform the appropriate control operations.

The development of such an EMS for islanded LVDC networks is challenging for
many reasons. Since the power production is based on renewable energy sources,
it is unreliable and can even reduce to zero. The load demand must nevertheless
be met. In order to enable this, energy storage devices such as batteries are used,
but battery capacities decrease, and it is crucial to recharge the batteries at suitable
time periods such as when the power production is more than the load requirements.
Since the network is completely isolated, the load demand must be met solely by
this combination of renewable energy-based power source and battery storage. The
primary challenge faced by the EMS is to achieve this delicate balance between
production, storage, and load during the entire operation period of the network by
making the correct control decisions. The EMS has several tools at its disposal
for achieving this balance, for example, load shedding, production and storage ca-
pacity planning, production control, battery state of charge (SOC) control, and the
intelligent use of forecasted data of both production and load.

21
1.6 Energy management systems

The development of efficient and optimal EMSs for various kinds of networks has
been discussed extensively in previous papers, and several approaches have been
proposed. However, most of the published literature so far have focused on micro-
grids that are connected to AC grids; in such grid-connected microgrids, islanding
operations have been typically considered as an alternative for deploying during
emergencies or for any other specific requirements. Researchers have also con-
sidered many types of power production combinations such as PV–windpower AC
microgrids, semi-autonomous microgrids, and hybrid AC–DC systems. The pri-
mary purpose of the proposed EMSs has been to ensure the stability of AC and
DC bus voltages as well as to establish voltage and frequency controls during grid-
connected operations, transitioning to islanded operations, and islanded operations.
Some researches have also considered DC microgrids that are connected to the AC
grid using bi-directional power converters, which typically operate either in the
grid-connected mode or islanded mode, depending on the requirements.

Lopes et al. evaluated the feasibility of control strategies to be adopted for the
operation of a microgrid when it becomes isolated. In particular, the need for stor-
age devices and load shedding strategies has been included in this paper. Bo et
al. introduced several strategies to maintain the power balance among renewable
microsources, storage systems, loads, and the utility grid during grid-connected,
islanded, and transition operations; their strategies involved the controlling of the
different converters present in a microgrid comprising wind turbines, PV panels,
and batteries. A three-phase inverter was used to maintain steady DC bus voltages
during grid-connected operations, whereas microsources and storage systems were
used in islanded operations; the magnitude and frequency of the AC bus were con-
trolled by the droop character of parallel inverters. The droop control method has
been the most frequently used technique in standalone AC microgrids for voltage
and frequency control and has been analyzed many times (Barklund et al., 2008;
Pogaku et al., 2007; Hu et al., 2011). Droop control essentially refers to active and
reactive power regulation, and it is a decentralized control strategy wherein the ac-
tive power output is adjusted according to the frequency deviation, and the reactive
power output is adjusted based on the voltage deviation (Xiu et al., 2011). Braban-
dere et al. (2007) proposed strategies for the efficient and fault-tolerant control of
microgrids with renewable energy sources, intelligent loads, and storage units; their
strategies are based on droop control applied at the production unit, and they sug-
gest that the control can be extended with appropriate communication infrastructure

22
to other parts of the network.

Fuzzy control has also been proposed as a control method for DC microgrid sys-
tems. Chen at al. (2013) presented fuzzy control to optimize the energy distribution
in a DC microgrid comprising solar and wind power production, lithium-ion bat-
tery, DC load, and AC/DC converters; fuzzy logic was used to establish the control
rules and to vary the battery state of charge (SOC) parameters. In a similar fashion,
Papadimitriou and Vovos (2010) proposed a fuzzy-based local controller for DGs
that are either integrated into or isolated from the main network, depending on the
requirements. Kakigano et al. (2011) adopted a gain-scheduling technique on the
basis is that it is difficult to achieve good voltage regulation and good load sharing
when the DC voltage is controlled by several converters. Their technique changes
the DC gain according to the output power in order to obtain better voltage reg-
ulation and load sharing simultaneously. Zhang et al. (2011) proposed the power
control of DC microgrids using DC bus signaling; the DC bus voltage level was em-
ployed as an information carrier to distinguish four different operation modes. The
power was controlled by controlling the modular PV converters, battery converters,
and grid-connected converters, and smooth switching was realized between constant
voltage operations and maximum power point tracking (MPPT) operations. Even
though their study basically involved grid-connected networks, the proposed control
method maintained the power balance of the DC microgrid even when islanded.

In this thesis, the scope of the EMS is restricted to the effective and efficient man-
agement of the power balance and voltage regulation in the LVDC island network
discussed previously (Fig. 3). Few researches have been conducted into the devel-
opment of an EMS for such completely islanded LVDC microgrids whose power
production is based on renewable (hence, variable) energy. Karlsson and Svens-
son (2003) suggested two methods for achieving power balance control in such
networks: the communication or master/slave method, which relies on fast com-
munication between the source and load converters, and droop control which does
not require any communication at all. In Liao and Ruan (2009), a power manage-
ment control strategy has been proposed for a stand-alone photovoltaic (PV) power
system comprising PV array, battery, and DC–DC converters. Their power man-
agement control strategy was to control the converters to operate in suitable modes
according to the PV power and battery statuses. However, their strategy is applica-
ble only to small loads and does not cover all the possibilities that can occur when
numerous customers have to be supplied power reliably. Moreover, they have not
considered emergency situations wherein there is insufficient or no power produc-
tion.

23
This thesis proposes a power management strategy in which the master/slave method
is used with robust communication infrastructure and modern power electronics in-
terfaces. Further, operation modes that encompass all conditions are derived, ana-
lyzed, and used as the basis for the EMS operations. The proposed EMS consists
of a control algorithm that ensures optimal power balance in the network and elicits
quick responses to emergencies. In addition, the concepts of load prioritization and
disconnection are employed to ensure that some loads continue to receive power
supply, even when there are power production issues. Load prioritization and dis-
connection are very widely known actions that are applied throughout the world,
especially to AC grids (load shedding). However, in the case of AC grids, entire ar-
eas are disconnected from the power supply based on some criteria, whereas, in this
thesis, only some loads are disconnected on the basis of prioritization. The EMS
is thus designed to be able to not only ensure reliable power supply, good power
balance, and fast voltage regulation during normal operations but also to ensure that
suboptimal power production situations are handled capably. Moreover, this the-
sis also addresses the problem of determining the production and storage capacities
that are required to supply power reliably at any location, since it is important for
developing and testing the control algorithm. A sizing method based on discrete-
event simulations (DESs) has been proposed to obtain reliable sizes of the PV array
and battery. Finally, an additional question has been examined in this thesis: given
the sizes of the network components, how many hours of electric power supply
can be guaranteed to the customers at any given location? All the algorithms were
implemented using MATLAB© , and all the simulations were also performed using
MATLAB© .

1.7 Thesis organization

This thesis is organized in the following manner. After this initial introductory chap-
ter, the basic concepts and choices for the power production system and storage de-
vices are briefly discussed; further, typical consumption behaviors are described and
illustrated. Subsequently, in the third chapter, first, the production capacity prob-
lem is examined and an algorithm to determine the optimum battery and PV sizes
is presented; subsequently, a method to determine the impacts of different hours of
sunlight on the number of service hours is presented. The two algorithms are im-
plemented and simulated with real-time parameters, and the simulations results are
presented to demonstrate their accuracy. Chapter 4 describes the basic principles of
general EMSs as well as the EMSs applied to LVDC island networks; suboptimal

24
situations that could occur along with methods to resolve them; the functionalities of
the network components; and the structure of ICT systems and their requirements.

Chapter 5 discusses the operation modes in further detail and presents the developed
EMS. The proposed EMS was implemented in a sample network, and simulations
were conducted assuming various situations and parameters; the obtained responses
were studied in detail. The simulation results are then presented, and they show the
smooth operation and efficiency of the algorithm. The results clearly demonstrate
that the LVDC network is adequately sized and well controlled using the given
strategies. Power and energy balancing were achieved along with voltage regula-
tion under all the considered circumstances. Nevertheless, several challenges still
remain, and the algorithm and methodology can be improved in many ways. The
limitations of the algorithm and future studies and pending researches are described
in detail at the end of the chapter. In the final concluding chapter, the thesis problem,
methods used, results obtained, and pending work are summarized.

25
2 ELECTRICITY PRODUCTION, STORAGE, AND
CONSUMPTION

An island network is fundamentally a localized grouping of electricity production

systems, energy storage devices, and loads, which function co-operatively and in-
dependently. In practice and in literature, microgrids in which islanded operations
are employed as an option are more common than completely isolated islanded net-
works; nevertheless, islanded networks have several advantages, as mentioned in
Chapter 1. In this thesis, an islanded LVDC network is considered.

During the planning, analysis, and implementation of an island network, it is impor-

tant to choose the type of production systems and storage devices carefully (other
components such as converters, cables, and protection and communication devices
must also be carefully chosen but they are not considered in this thesis) since they di-
rectly affect the network performance and control. Several factors affect the choice
of the appropriate systems and devices, including reliability, costs, and availabil-
ity. The features and performance characteristics of the considered systems must
be considered before making the appropriate choice; moreover, their impacts on the
network performance must also be analyzed. Their applicability and suitability to
the types of loads that can be anticipated must also be considered. The nature of
loads and their consumption characteristics strongly influence the performance of
the network. Load types and their demands may differ depending on the location,
nature of load, and customer profiles, and it is difficult to accurately anticipate their
behaviors. Nevertheless, a general overview of typical loads can be envisaged and
their impacts can be considered. In particular, (typical) load types, load require-
ments, load profiles, and load usages can be examined.

This chapter briefly outlines and discusses the three crucial elements of the LVDC
island network—power production, energy storage, and load consumption. This
thesis is focused on the energy management (and partial system sizing) of the net-
work, and discussions on production, storage, and consumption characteristics are
restricted to their impacts on the relevant analyses of LVDC networks. The elec-
tricity production system used in this study—PV systems—and the applicability of
their characteristics to this particular network are discussed first. Subsequently, the
energy storage device selected for the investigated project—lead-acid battery—and
their parameters of interest are described. The types of loads that can be expected
during this project implementation are first discussed from a general viewpoint and

26
general comments are made on the potential challenges and possible solutions; their
characteristics that may influence the network performances are also discussed sub-
sequently.

2.1 Electric power production

2.1.1 Introduction

Electric power is produced from sunlight by converting solar radiation into DC

electricity using semiconductors exhibiting the PV effect; this method is commonly
referred to as photovoltaics (PV) and is based on the effect that was discovered
by Edmond Becquerel in 1839. Among the various renewable energy technologies
available, solar power has been chosen in this thesis for several reasons. Solar PV
is known to be both a feasible and sustainable energy source (Pearce, 2002). Fur-
ther, PV systems are attractive for electricity production because they are noiseless,
emission-free, flexible, and have reasonably simple operations and maintenance
(Dinçer, 2011). Moreover, the cost of PV has declined at a steady rate in recent
years (Swanson, 2009), and the increased demand for renewable energy sources has
spurred the manufacture of solar cells and PV arrays. According to the Global Mar-
ket Outlook for Photovoltaics 2013-2017 published by the European Photovoltaic
Industry Association (Masson et al., 2013), solar energy is one of the fastest grow-
ing energy production sectors today, and it is expected to become a mainstream and
mature source of electricity in the near future. The world’s cumulative PV capacity
has surpassed 100 GW, which is equivalent to as much annual electrical energy as
that produced by 16 coal power plants or nuclear reactors of 1 GW each. Further,
the LVDC island network is intended to be deployed in equatorial and tropical re-
gions which receive fair amounts of sunlight. Hence, for all these reasons, solar
power is a natural choice for the power source.

2.1.2 Solar power technology

PV panel design PV power production systems employ solar panels that com-
prise numerous solar cells consisting of one or two layers of a semiconducting ma-
terial (Fraile et al., 2009). Light shining on a solar cell creates an electric field across
the semiconducting layers, thereby causing electricity to flow; moreover, the elec-

27
tric current increases with the intensity of the light. The most commonly used PV
material is silicon in various forms, such as monocrystalline silicon, polycrystalline
silicon, and amorphous silicon; depending on the technology, cadmium telluride
and copper indium gallium selenide/sulfide are also used (Jacobson, 2009).

Two broad categories of technologies have been employed for PV cells—crystalline

silicon and thin film (Fraile et al., 2009). Crystalline silicon technology is the most
popular technology in the market today (Fraile et al., 2009), but thin-film panels
have shown tremendous potential, and considerable investments are being made
into researches to improve them. A thin film is made by depositing extremely thin
semiconductor material layers (hardly 0.3–2 µm thick) onto glass, plastic, or stain-
less steel substrates. Since the semiconductor layers are thin, the costs of raw mate-
rial are considerably lower than the capital equipment and processing costs, thereby
making this technology cost-effective. Further, its efficiency, which used to be lower
than that of crystalline silicon PV modules, has increased in recent years (Doni et
al., 2010). A comparison of crystalline silicon and thin film panel technologies for
PV cells is given in Table 1. At an irradiation of 1000 W/m2 , a thin-film panel with
10% efficiency costs 0.5 C/W; in other words, 2 m2 produces 200 W and costs 100
C. On the other hand, a crystalline silicon panel with 20% efficiency costs 0.7 C/W;
that is, 1 m2 produces 200 W and costs 140 C.

Table 1: Comparison between crystalline silicon and thin-film panel technologies

for PV cells.
Crystalline Silicon Panels Thin Film Panels
Monocrystalline and multicrystalline Amorphous Amorphous
silicon silicon
(superstrate); (substrate);
Cadmium Copper indium
telluride (CdTe) (gallium)
selenide
(CIS/CIGS)
X Proven and reliable technology X Low price X Low price
X High efficiency X Can be flexible X Does not
installation necessarily need
a transformer
X Electrically the easiest X Still unreliable X Not as reliable
as crystalline
panel
X Numerous manufacturers X Needs X Low efficiency
transformers
X High price X Low efficiency
X Thick panels

28
Thin-film panel technology is advancing very rapidly, leading to increasing efficien-
cies and other improvements, and additionally, its costs are lower; hence, thin-film
panels have been selected for the PV array in the proposed network.

PV array and installation A PV-based electricity production system typically

consists of a string or an array of solar panels, called a PV array, which is con-
nected to increase the power that can be delivered. PV arrays are installed in many
ways, for example, ground-mounted or built into the roof or walls of a building.
Additionally, the angle at which a PV module (or array) is installed—called the tilt
angle—influences the power produced. A solar panel that is tilted perpendicular
to sunlight typically receives more light on its surface than an angled solar panel.
Moreover, the tilting is also dependent on the location; in general, the optimal orien-
tation without non-tracking is nearly horizontal near the Equator, toward the south in
the northern hemisphere, and toward the north in the southern hemisphere (Chang,
2009).

Numerous studies have explored the optimum tilt angles at various locations on the
basis of many factors such as predicted data, temperature, and seasons (Chang and
Yang, 2012; Zhao, 2010; Tang, 2010). Moreover, many solar panels and arrays
are equipped with mechanical systems called solar trackers that tilt a solar panel
throughout the day, thereby following and tracking the sun’s movement; this sig-
nificantly enhances early morning and late afternoon performances. Trackers are
especially effective in regions that receive a large portion of sunlight directly; on
the other hand, they have little value in diffuse light. In general, tracking is most
beneficial at sites between ±30◦ latitude, with the benefits reducing at higher lati-
tudes due to the sun dropping low on the horizon during winter months. However,
trackers are expensive and they require energy for their operations, which may not
always be available especially in such LVDC islanded networks; moreover, it may
not always be possible to use them due to location constraints. Trackers are thus
not always applicable (Benghanem, M., 2010). Hence, the solar panel that will be
used in this study will not have solar tracking, that is, they will not be self-directing
panels. The location constraints during the actual implementation of this project
are as yet unknown and cannot be reasonably predicted. Hence, in this thesis, it is
simply assumed that the PV panel will be placed at the optimum position with the
optimal tilt at the chosen location.

29
Solar cell efficiency In general, the conversion efficiency (η) of a solar cell is
given by the following equation:

Pm
η= , (1)
E × Ac

where Pm is the maximum power from the cell (W); E, the input light irradiance
(W/m2 ); and Ac , the surface area of the solar cell. Moreover, solar cell efficiencies
are measured and specified, by convention, under standard test conditions (STC)—
temperature of 25◦C and an irradiance of 1000 W/m2 with an air mass 1.5 (AM1.5)
spectrum. Thus, for example, a solar panel with 15% efficiency and area of 1 m2
will produce approximately 150 W of power under STC. It is important to note that
in practice, commercial PV panel module ratings are typically given for STC. The
efficiency of a solar cell is further dependent on several factors such as the wave-
length of light, temperature (output decreases as temperature increases), dust and
debris accumulation (decreases the output), air mass variations, shading (decreases
the output), and reflection.

In this thesis, for simplicity, the effects of temperature, dust, reflection, and other
factors are ignored; it is also assumed that the PV panels are not shaded.

Calculation of solar power output In this thesis, the PV output has to be calcu-
lated for two different purposes, and hence, two different methods have been used
accordingly. Firstly, in the sizing algorithm, the predicted PV array output must
be known in order to calculate the required PV array and battery sizes. To obtain
the PV output, the peak-sun-hours (PSH) method has been used (Weixiang et al.,
2005). PSH, or just sun hours, refers to the number of hours in a day in which
the standard solar irradiance of 1 kW/m2 is experienced; more precisely, it is the
length of time in hours at a solar irradiance level of 1 kW/m2 needed to produce
the daily solar radiation obtained from the integration of irradiance over all daylight
hours (IEEE, 2007). The PSH for a day can be determined simply by taking the
average solar irradiation for the day and dividing by 1000; note that since the inte-
gration is not performed accurately, this gives an approximate value but it has been
considered sufficient for the purposes of this thesis. The PV output is then simply
PSH × PV array size since the PV array size is given under STC—1 kW/m2 —just
like PSH.

Secondly, the control algorithm, in practice, does not need the PV output to be cal-

30
culated; in a practical scenario, the real-time PV output is given as an input into the
algorithm (for example by the PV converter through a communication interface).
However, in this study, the PV output has to be calculated for the simulation of the
control algorithm. This calculation has been performed simply by taking the man-
ufacturer’s nameplate rating (which is given for STC) and applying the following
formula:

Rated Out put × Irradiance Data

PV Power Out put = W
1000

The PV power output calculated in this manner has then been used for the simula-
tions of the control algorithm presented in this study.

2.1.3 Solar irradiation

Historical irradiation data Solar irradiance E is the amount of solar power strik-
ing a given area and is given in W/m2 (Prof. Dr. Quaschning, 2013). The irradiance
measured over a period of time, or, its integral over a time period is, referred to
as solar irradiation or insolation H (Wh/m2 ). A nearly constant 1.36 W/m2 (called
solar constant) of solar irradiance strikes the earth’s outer atmosphere but this in-
cludes all wavelengths. Silicon PV modules use only the part of the spectrum from
0.3–0.6 µm. Additionally, the irradiance striking the earth is decreased by various
atmospheric factors as well as the climate and location. Only a part of the extrater-
restrial beam irradiance reaches the earth’s surface directly; it is estimated that the
total irradiance striking the earth on a sunny day is approximately 1000 W/m2 (the
basis for the PSH method).

Solar irradiation data provide information on the amount of energy striking a surface
at a location on the earth during a particular time period, H (Wh/m2 ). Many organi-
zations have and continue to obtain irradiation data based on satellite measurements,
such as (Helioclim, 2013) and National Weather Service, United States. Moreover,
several databases of irradiation data are available, and quite a few websites and
authoritative organizations provide time series of historical solar irradiation data
for free and open access, for example, the SoDa service, which is supported by
the European Commission, and National Renewable Energy Laboratory, which is
owned and funded by the US government. Where solar irradiation measurements
are not easily available for various reasons (high equipment costs and calibration re-

31
quirements), estimations of solar irradiation data by using models have often been
proposed (for example, Muzathik et al. (2010)).

The amount of solar radiation received is highly variable due to weather patterns
and the changing position of the sun. Many other factors also influence the amount
of solar radiation reaching the earth’s surface, such as clouds, local geographical
features, the time of day, season, and pollution. Solar irradiation data reflect this
variability, and its availability can be used to design and analyze practical PV-based
systems.

The global solar radiation on a horizontal surface has two components—direct

beam radiation and diffuse radiation. Direct beam radiation refers to the direct
radiation from the sun, while diffuse radiation is scattered out of the direct beam
by molecules, aerosols, and clouds. On clear days, the diffuse radiation is ap-
proximately 10% to 20%, while it is as high as 100% for cloudy skies. Further,
tilted planes have another component—radiation reflected from the ground (ap-
proximately 20% of the global irradiance (Prof. Dr. Quaschning, 2013)). The
sum of the direct beam, diffuse, and ground-reflected radiation arriving at a tilted
plane is called total or global solar radiation. Depending on the type of the mea-
suring station, either all the components are measured, or some are measured and
the remaining calculated (several methods have been presented in the literature, for
example, Liu and Jordan (1960).

Figure 4 shows the total monthly irradiation data for the year 2004 at two locations,
latitude 0◦ and latitude 60◦ , that is, corresponding to the Equator and much higher
north, for example, Helsinki (the Arctic Circle is at 66◦ ), respectively; the effects
of season and location can be clearly seen. At latitude 60◦ , the monthly irradia-
tion decreases to almost 0 during the winter months—December and January—and
increases to as high as 6000 Wh/m2 , which is higher than the highest amount for
the Equator during summer. These trends clearly demonstrate the effects of long
hours of darkness in winter and daylight in summer at latitude 60◦ . In contrast, the
irradiation never decreases to less than 4000 Wh/m2 at the Equator and is much
more consistent; the small variations in the levels can be attributed to the effects of
rain and cloud cover. These trends lend credence to the intuitive idea that the power
output can be expected to be more reliable near the Equator than near the poles. A
more detailed analysis will be presented in Chapter 3.

32
 Total Monthly Irradiation; Year: 2004; Latitudes: 0 and 60 degrees
6000

Latitude 0º
Latitude 60º
5000

Irradiation (Wh/m2)
4000

3000

2000

1000

0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Month

Figure 4: Total monthly irradiation data for the year 2004 at two locations—
latitudes 0◦ and latitude 60◦ , corresponding to the Equator and much higher north.

Forecasted irradiation data In addition, solar data forecasts are also an invalu-
able resource for both sizing and control. However, solar data forecasting is not an
easy problem since it is weather-dependent, and several researchers have attempted
to improve its accuracy and reliability. Many types of forecasting methods have
been proposed in the literature, including statistical methods such as ARIMA and
neural network-based methods (Yona et al., 2008; Huang et al., 2012; Heinmann
et al., 2006). Such forecasting algorithms and methods typically forecast solar ir-
radiation for very short periods (called now-casting and typically for a few hours),
short periods (short-term forecasts for up to 7 days), or long periods (long-term
forecasts that give monthly or annual estimates). Beyond 7 days, the reliability de-
creases tremendously and any generated data may not be of much value. Note that
forecasted solar irradiation data are typically not available for free.

Solar irradiation data collection All the solar irradiation data used in this thesis
was obtained from online resources (via the SoDa service) which gives time series
data of daily, monthly, and yearly solar irradiation data for the period 1985–2005
for free and open access. In the simulations in this thesis, only the data for inclined
planes has been considered, for consistency; this also ensures to a certain extent
that the values are neither under-estimated nor over-estimated. Concentrating solar
systems use only direct irradiation, whereas non-concentrating systems also use dif-
fuse irradiation; further, tilted planes are much more common in practice. Hence,
the data of the global component—that is, the sum of the diffuse, direct, and re-

33
flected components—is used, unless otherwise mentioned. In the case of inclined
planes, data is available for free for only one year—2005. Additionally, the data
has not been recorded in some cases (given as -999), but irradiation values on top
of the atmosphere are available. In all such cases, the unrecorded values have been
replaced by the corresponding values for the top of the atmosphere (this does not
result in serious simulation errors).

2.1.4 Interfacing with LVDC island networks

PV arrays can be connected or interfaced to the grid with or without a converter.

However, converters (or inverters) are commonly implemented because they can be
used to draw maximum DC power from the array. Most of the existing researches
have been conducted into optimum inverter connections with AC grids, while a few
have focused on converters and DC connections. In the LVDC islanded network
that is being considered, the PV array will be connected to the network through a
DC/DC converter that will also be used to enable maximum power point tracking
(MPPT; a method to extract maximum possible power). Further, the PV converter
is able to communicate the current power, voltage, and current statuses reliably
and quickly and also accept control instructions and act accordingly. The nominal
voltage should be as close as possible to the voltage of the DC network voltage
as possible, 750 V. PV power production is not reliable and can vary significantly
depending on the climatic conditions; hence, an important functionality of the PV
array for maintaining the power balance in the network is that its production should
be controllable, that is, it should be possible to maximize it or constrain the produc-
tion.

2.1.5 Summary

The renewable power source used in this study is solar energy, and the PV array
uses thin-film technology. The solar irradiation dataset used in this study for test-
ing and simulation purposes consists of the global component of solar irradiation
falling on an inclined plane over the year 2005. The irradiation values given in the
solar dataset are used to calculate the PV array outputs required for conducting sim-
ulations and to determine power output variations with locations. Further, the PV
arrays are connected to the DC mains via a DC/DC converter that is not only able
to communicate its power statuses quickly but also receive control commands and

34
act on them reliably.

2.2 Electricity storage

2.2.1 Introduction

Due to the high variability of power production from PV arrays and the isolation of
the network, backup power supplies are required to meet load requirements reliably.
Such backup power is typically supplied by batteries, which, naturally, should be
rechargeable. The basic function of the battery is to supply power when the system
load exceeds the power output from the PV array (IEEE, 2007). In addition, batter-
ies are either recharged by the PV array or discharged to supply load, depending on
the energy and power balance statuses of the network during the time period under
consideration.

2.2.2 Storage types

Today, many types of rechargeable batteries with different chemical combinations

are available, such as lead-acid, nickel cadmium (NiCd), lithium ion (Li-ion), and
nickel metal hydride (NiMH). In general, lead-acid batteries are popular for their
low costs, mature technological levels, high discharge rates, low maintenance re-
quirements, and recyclability. Stationary lead-acid batteries, which are designed for
deep discharges, are commonly used as large backups to power supplies. Such deep-
cycle lead-acid batteries are designed to be regularly deeply discharged, typically to
50% to 70% of its capacity. Due to the unreliability of PV power production, deep-
cycled batteries are important to enable the LVDC island network to supply power
reliably. Hence, deep-cycle lead-acid batteries are used as the additional backup
power supply source.

2.2.3 Lead-acid battery characteristics

Battery capacity refers to the amount of (usable) energy that a battery can store at
the nominal voltage, and it is usually expressed in terms of the current that can be
supplied by the battery under normal conditions (Ah). The state of charge (SOC)

35
expresses the current capacity of the battery as a percentage of the total capacity,
while the depth of discharge (DOD) is a measure of how deeply a battery has been
discharged. It is important to note that the higher the DOD—or, the deeper the bat-
tery discharge—on an average, the shorter the battery cycle life; simply put, deeper
discharges shorten battery life. As a result, it is not advisable to discharge batteries
too deeply, and the commonly recommended DOD is not more than 50%. Controls
are often employed to protect the battery from being over- or under-charged, which
typically include power conversion subsystems (inverters or converters). A charge
controller may also incorporate additional functions such as discharge termination,
regulation voltage, and status indication. (IEEE, 2008)

Batteries have specific charge and discharge rates, which must be considered in the
sizing and control algorithms. The discharge rates, expressed as C-rate, measures
the rate of discharge of a battery relative to its maximum capacity; a discharge rate
of 1C implies that the entire battery will be discharged in 1 h. For a 200 Ah bat-
tery, a discharge rate of C/2 means that the battery will be discharged in 2 h with a
discharge current of 100 A. Other parameters of interest in this thesis are the nom-
inal voltage (V) and the charge voltage (the voltage that the battery is charged to
when charged to full capacity). It is also important to understand that very rapid
discharging and charging rates, although attractive, can damage the cells; the maxi-
mum continuous discharge and charge currents (recommended) are typically set by
the manufacturer as preventive measures.

Lead-acid batteries also self-discharge at a rate that typically depends on the stor-
age or operating temperatures. However, the effects of self-discharges have been
ignored in this thesis in order to simplify the analysis. Further, sulfation (the crys-
tallization of lead sulfate) occurs when a battery is not used or charged for long
periods of time, leading to reductions in its capacity. This may not be a significant
risk in this network, because batteries are expected to be required to charge regularly
given the variable nature of the PV power supply. Nevertheless, it must be noted
that undercharging—not allowing the charger to restore the battery to full charge—
limits battery life, and continuous operations at a partial SOC can lead to sulfation.
The control algorithm must attempt to prevent undercharging and must attempt to
restore the SOC as soon as possible; this implementation has been considered in this
thesis. Note that sulfation can be a significant problem since it affects the charging
cycle, resulting in longer and less efficient charging and higher battery tempera-
tures; at the same time, desulfation methods can be used to reduce the sulfation that
may have occurred.

36
Further, there is an additional complication. The capacity of a lead-acid battery is
not fixed; it varies with the discharge rate in accordance with Peukert’s law (Doerffel
and Sharkh, 2006) that gives an empirical relationship between the discharge rate
and capacity. Nevertheless, in this thesis, the capacity is considered to be an input
from the battery management system which manages charge balance issues, and,
for simplicity, these variations are not considered in the simulations.

2.2.4 Battery energy management system

Modern batteries are equipped with battery management systems (BEMSs) that
monitor, manage, control, protect, and communicate the state of the battery (Pop
et al., 2008). The basic task of BEMSs is to ensure that the battery energy is op-
timally used and any risks of damage are prevented. Typically, the charging and
discharging processes are monitored, controlled, and communicated. Parameters
such as voltage, SOC, DOD, current, temperature are monitored, and additional
calculations based on these parameters may also be performed. Moreover, recharg-
ing is managed efficiently and the battery is protected from surges and other unsafe
operating conditions. In this thesis, it is assumed that a BEMS exists and that it
communicates the required parameters to the control software on a regular basis.
Batteries can be connected to the network with or without DC/DC converters. If a
converter is not used, the battery charges depending on the voltage in the network
and explicit control instructions are not required. However, in this network, DC/DC
converters are used (see Fig. 3), and further, the battery converter, with the EMS,
communicates the current battery statuses reliably and quickly, while also being
able to accept control instructions and act accordingly.

2.2.5 Summary

The LVDC island network uses solar power for power production, and this causes
problems of reliability of power supply. Hence, a lead-acid battery is used as the
storage device to either supply or absorb balance power, depending on the situation.
The battery capacity and management have a huge impact on the system availabil-
ity and reliability. High battery capacities can dramatically increase the number
of hours of power supply, and efficient management of the resources can reduce
losses and improve network performances. Larger batteries will cycle deeper less
frequently, thereby increasing the system availability and battery life. However,

37
the larger the battery, the greater the risk of sulfation. Nevertheless, in general,
increasing the battery capacity increases system availability more cost-effectively
than increasing the PV array size. (IEEE, 2008)

The battery is also important from the network control point of view, since control-
ling the battery SOC by charging or discharging is an important method to maintain
power balance as well as to ensure voltage regulation. In the algorithms introduced
in this thesis, battery parameters such as battery capacity and charge and discharge
rates are considered as input parameters entered by the network operator before be-
ginning the network operations. The battery bank is connected to the LVDC bus
with DC/DC converters, and it incorporates a management system that commu-
nicates the current battery statuses reliably and quickly, while also being able to
accept control instructions and perform the required actions.

2.3 Electricity consumption

2.3.1 Introduction

Consumption behavior is an important aspect of network planning, analysis, and

control since the entire purpose of the network is to meet the consumption demand.
Moreover, the quality of the sizing as well as control ultimately depends on the na-
ture of the customer connections and the resulting customer behaviour. The load
demand of customers in a rural area, who may never have used electricity before,
may be much lower than the load demand of customers in an urban area, who may
be regular and dependent users of electricity. Similarly, the reliability expectations
may be completely different—some customers may require electricity for running
their businesses, while some customers may not be as dependent and more tolerant
to losses in power supply. The type of load being supplied may also differ dramati-
cally; for example, some customers may require connections for refrigerators, some
others may desire internet connectivity, while some customers may be satisfied with
just lighting loads.

This project is meant to be implemented in populated areas comprising small vil-

lages that are spread randomly around the area of the island network. It is difficult
to anticipate the exact consumption behaviors and characteristics of the customers
who will use the network since it depends on the exact location where the project
will be ultimately implemented, the nature of consumption and consumers, the type

38
of load requirements, and the ground conditions at that location and period of time.
Therefore, due to the variability in the load type, it is challenging to implement the
sizing and control perfectly and in such a manner that all possible loads are encom-
passed. Clearly, some kind of a generalized approach is preferable, especially at this
preliminary stage; this approach can then be adapted according to the requirements.

Hence, in this thesis, certain general load types and their logical classifications have
been proposed, which have then been used to control the network; this is discussed
in more detail in Chapter 4. The basic intention has been to build the algorithms
such that variations in the load profile do not impact the efficiencies, operations,
and results as far as possible.

2.3.2 Types of loads

From the viewpoint of the nature of electric supply, two types of loads have been
considered in the thesis:

1. AC Loads: These loads comprise most household appliances and equipment.

A single or three-phase converter will be used for the DC/AC conversion and
the voltage levels will be approximately 230 V.

2. DC Loads: Mostly lighting loads with a voltage level of 48 V.

Typical loads Typical loads refer to the most common loads that may be used by
the customers. At the individual level, the customers of this network are most likely
to have requirements for lighting, fans, internet, television, and refrigeration. At
the community level, the typical loads may be doctor’s clinics or hospitals, schools,
street lighting, and loudspeakers. Note that some of these loads could be further
classified as the most critical or essential loads of the network.

Flexible loads An important aspect of the load usage in this study is load flexibil-
ity—the idea that not all loads need to be supplied regularly and that there can be
some flexibility in the load usage. Such flexibility may differ for different customers
and the loads may have to be chosen accordingly using some kind of pricing-based
user-side prioritization (user takes the decision to disconnect loads, typically on the
basis of pricing) or power-supply-availability-based operator-side prioritization (the

39
network operator disconnects loads depending on the power supply availability).
For example, some customers may not require the regular usage of refrigerators,
but they may need lighting and fans at night; similarly, other customers may wish to
prioritize television sets over washing machines. Loads such as refrigerators, tele-
vision sets, and washing machines are examples of flexible loads whose usages are
not critical and can be prioritized.

Load prioritization Due to the variable nature of the power supply, lowered reli-
ability, and the nature of the loads, it may not be possible to supply all the connected
loads. In such a situation, supplying some loads is more preferable than supplying
none. The loads that can be disconnected are determined by prioritizing the loads.
In other words, when the total load energy demand cannot be met, some loads must
be disconnected from the power supply and the choice of the loads for disconnec-
tion is based on prioritization. Prioritization works only if there are loads in the
network that are not critical and can be disconnected; since load flexibility has been
assumed, in this network, load prioritization is a meaningful method. Such prioriti-
zations may be online and real-time by the customer or may be determined and fixed
prior to the network operations (with the option of changing it at regular periods),
if required; this thesis considers the latter.

Another alternative is to perform load shedding wherein a part of the network, com-
prising many loads, is deliberately disconnected in order to ensure that the entire
system does not collapse. Load shedding is common in many developing coun-
tries since demand often exceeds supply especially in the summer season; it is also
common to communicate load shedding periods in advance to the customers. If
forecasts of power production and balance can be made with reasonable accuracy,
the communication of such information must be considered during the operations
of the LVDC islanded network as well. The actual classification of loads and the
mechanism of load prioritization used in this thesis are discussed in detail in subse-
quent chapters.

It is assumed that the loads can be disconnected quickly, reliably, and accurately,
as and when required. Further, while the load profile is given as an initial input to
the algorithm for sizing, the real-time load parameters are regularly communicated
by the load converters to the control algorithm. In addition, forecasted load profiles
must be given as inputs to the control algorithm so that it can make the necessary
tweaks required for optimal operations; this aspect is not considered directly in this
thesis and must be addressed in the future.

40
2.3.3 Typical load curve

In this thesis, the average load peak power is considered to be approximately 800
W. For simulation purposes, a typical load profile scaled from a typical Finnish
customer’s load curve has been considered. Figure 5 shows a practical load demand
curve of a single Finnish customer living in row-and high-rise apartments without
heaters for 1 year. Figure 6 shows a plot of the same load demand for 1 day. The
load curve data is taken from real data given as hourly indexes and with the peak
power as approximately 800 W; further, a confidence interval of 95% has been
considered. Note that this load data has been used in all the simulations presented
in this thesis, but simulations with other loads were also performed for verification
purposes.

Figure 5: Load consumption (kW) for one Finnish customer for one year. (Note:
The load peak power is approximately 800 W.)

Figure 5 somewhat approximates the possible load behaviour of customers closer to

the Equator, but with the months inverted! Note that the total load demand increases
during the winter months (the two ends in the figure) and decreases toward the
summer (middle of the year)! This happens in Finland because less lighting and
heating are required in summer. For a customer in the equatorial or tropical regions,

41
Figure 6: Load consumption (kW) for one Finnish customer for one day in January.
(Note: The average load peak power is approximately 800 W).

the situation would be somewhat reversed, with greater load demand in the summer
than in the winter. Figure 5 shows the load demand of the Finnish customer for
one day; the load demand is low at night, but it increases in the morning with
a perceptible spike during the morning hours, and there is a subsequent further
increase toward the evening before it decreases again.

2.3.4 Summary

The nature of loads and the consumption behavior are very important since they
determine the power balancing requirements; however, they differ with the location
and customers. Nevertheless, some generalizations can be made about them and
a general control algorithm can be built. Further, both DC/AC and DC/DC load
converters communicate all relevant information to the control algorithm at regular
intervals. Since the power supply is variable, load prioritizations and disconnections
are important tools that are used by the control algorithm proposed in this thesis.
Further, forecasts of load consumption are critical data that can be used to optimize

42
the power and energy balancing.

The basic problem addressed by the thesis is illustrated in Fig. 7. During many
days of the year, especially during the winter months, the solar irradiation (before
converting to electricity) is considerably less than the load demand. This gap is par-
tially met by using large PV arrays that can increase the amount of available power,
but this is an expensive and sometimes impractical solution. A more reasonable
approach is to also use battery to supply the balance power. Another advantage of
using a battery is that when the produced power is more than the load demand, it
can absorb the excess power and hence balance it. The problem addressed by this
thesis is the control of the battery SOC, PV power supply, and loads in such a way
that power balance is achieved and maintained at all times.

Load Demand (kWh) versus Solar Irradiation for One Finnish Customer for One Year
11

Load Demand (kWh)

10
Solar Irradiation (kWh)
Load Demand, Irradiation (kWh)

0
0 50 100 150 200 250 300 350

Day of the Year

Figure 7: Load demand (kWh) versus solar irradiation (kWh) for one Finnish cus-
tomer for one year. This thesis proposes an algorithm to ensure that they are bal-
anced and that the two curves (blue and red) coincide; to ensure this, a battery whose
state of charge is controlled is employed, and/or loads are disconnected depending
on the conditions.

2.4 Conclusions

The renewable power source used in this study is solar energy and thin-film PV tech-
nology is used for the PV array. The PV arrays are connected to the DC mains via
a DC/DC converter that is not only able to communicate its power statuses quickly
but also receive control commands and act on them reliably. A lead-acid battery
is used as the storage device to either supply or absorb balance power, depending

43
on the situation. The battery capacity and management have a huge impact on the
system availability and reliability. Controlling the battery SOC by charging or dis-
charging is an important method to maintain power balance as well as to ensure
voltage regulation. The battery bank is connected to the LVDC bus with DC/DC
converters, and it incorporates a management system that communicates the cur-
rent battery statuses reliably and quickly, while also being able to accept control
instructions and perform the required actions.

The nature of loads and the consumption behavior are very important since they
determine the power balancing requirements. However, they differ with the loca-
tion of the network and the nature of the customers’ requirements and practices.
Both DC/AC and DC/DC load converters communicate all relevant information to
the control algorithm at regular intervals. Since the power supply is variable, load
prioritizations and disconnections are important tools used by the control algorithm
proposed in this thesis.

In this manner, this thesis considers a bipolar 750-V LVDC network that is powered
by thin-film PV panels with rechargeable lead-acid batteries to supply loads that
can be flexible and prioritized, if required. Further, the basic problem addressed
by the thesis is the control of the battery SOC, PV power supply, and loads in the
network in such a manner that power balance and voltage regulation are achieved
and maintained at all times.

This chapter has laid the foundations for subsequent discussions into the develop-
ment of the EMS itself and related concepts. The three main elements of the LVDC
microgrid network—power production, energy storage, and load consumption—
have been discussed with particular focus on their applicability to the LVDC island
network being considered. In the next chapter, the efficient and effective sizing of
the components in the network will be discussed.

44
3 PRODUCTION CAPACITY: PLANNING and ANAL-
YSES

3.1 Introduction

In this thesis, production capacity planning and analyses refer to investigations into
the ability of the LVDC island network—basically, the power production system—
to reliably supply a certain amount of power to a certain number of customers for
a certain time period. Capacity planning involves solving the problem of the power
production capacities needed to reliably supply power for a certain time period,
while capacity analysis involves an investigation into whether the sized system can
supply power, and if yes, for how many hours. In other words, capacity planning
addresses the question: What should be the sizes of the PV array and battery for
the given network parameters and location? On the other hand, capacity analysis
attempts to answer the question: Given PV array and battery sizes, how many ser-
vice hours can be reliably guaranteed at a specific location? Such capacity planning
may not seem to directly affect the control algorithm, but, nevertheless, it is im-
portant because the answers impact network reliability and functionality; accurate
sizing implies smoother and more optimal control. Moreover, reasonably accurate
sizing improves the testing of the control algorithm. Similarly, capacity analyses
are important from the network operator’s viewpoints, for example, in the case of
the service guarantees that can be made to the customers of the network.

In the considered LVDC island microgrid, the two largest system components are
the production and storage components, and their sizes, or capacities, must be de-
termined reasonably accurately first; it is also useful to know the sizes of the cables,
converters, and other component, but the PV array and battery bank are the most
important since they typically dictate the sizes of the other components. These two
components must be sized carefully, optimally, reliably, and cost-effectively; in ad-
dition, it is important to perform the sizing early since not only control but also
other aspects of network planning, such as protection and communication planning,
can then be carried out effectively. Reasonably accurate estimates of their capac-
ities can be crucial from the viewpoint of other network planning aspects such as
costing, power, and reliability as well.

In this chapter, first, previous research approaches to sizing are described. Subse-
quently, an algorithm for accurately determining the PV and battery capacities is

45
presented; the accuracy of the proposed algorithm is then supported by simulation
results which are described and discussed. In the final section, an algorithm to de-
termine the number of service hours that can be guaranteed by the network at any
location in the world is presented along with the obtained results. The chapter is
then concluded with a brief discussion on the limitations of the approach and the
potential areas of further research.

3.2 Capacity planning

3.2.1 Research background

Numerous researches have been conducted on the sizing of PV-battery standalone

systems in the last few decades, and several analysis methods have been proposed
and developed. Early analysis methods (Solar Power Corp., 1978) focused on the
concept of the number of autonomous days (NADs) in order to ensure that the power
supply was completely reliable; however, this somewhat simple approach oversizes
the system and is uneconomical, especially in the case of large systems. Ofry and
Braunstein (1983) proposed the loss of power supply (LOPS) probability concept
as a method to optimize the sizes of the PV array and battery storage system, espe-
cially from the viewpoint of cost. Instead of using the NAD concept, they defined
reliability as the total number of hours in which the consumer’s power demand is
more than the power supply.

Chapman (1987) employed the loss of load probability (LOLP) concept, which is
a widely used reliability measure in capacity planning (Wang et al., 1977), to pro-
pose a noncomputerized sizing technique that enables sizing for the required LOLP,
given the load demand and insolation data. Gavanidou & Bakirtzis (1992) con-
sidered LOLP as another design objective to be minimized, along with the capital
investments, and employed the tradeoff/risk method to design an autonomous sys-
tem. Borowy & Salameh (1994) considered a standalone hybrid wind/PV system
and calculated the optimum size of a PV array; subsequently, they extended their re-
search to battery banks by using the LOPS probability concept (Borowy & Salameh,
1996). Shrestha & Goel (1998) conducted studies into the optimal sizing of a PV
system on the basis of statistical models. Conti et al. (2002) proposed a multi-
objective optimization approach that uses fuzzy logic to avoid the disadvantages of
both NAD and LOPS concepts. In 2006, Markvart (2006) used the time series of
irradiation data in place of LOLP in order to size PV systems.

46
Markvart et al. (2006) also succinctly summarized the objectives and procedures for
sizing. According to them, the basic objective of sizing procedures is to consider
the relationship between the sizes of the PV array and battery which deliver energy
to the load with a certain reliability of supply that can be tolerated by the user: the
results of the sizing procedure are often summarized in the form of a sizing curve.
Since the key parameter is supply reliability, it must be satisfactorily defined, and
LOLP is most often used for this purpose. IEEE (2008) and IEEE (2007) provided
methods for properly sizing the PV array and battery, respectively, in PV-battery
standalone systems. In their methods, the sizing ensures that the load demand is
met in a cost-effective manner and the system performances and operating lifetimes
are improved. The PSH method of sizing is used, and the systems are sized based
on the worst-case solar radiation, load consumption, and system losses. In addition,
the basis for the system sizing is the current of the PV module. The guide recom-
mends the use of LOLP. Khatib (2010) listed and discussed the general procedures
and guidelines for designing the system components of standalone PV systems; in
addition to recommending the LOPS concept, current was proposed as a suitable
parameter for sizing the PV array since it does not depend on temperature, unlike
voltage and power.

In this thesis, the concept of NADs has not been used; instead, it has been assumed
that there should be reliable power supply for 24 h. In addition, the DES method
has been used. In DESs, the operation of a system is modeled as a discrete sequence
of events in time; each event is supposed to occur at a particular instant in time and
marks a change of state in the system. (Banks et al., 2009) DESs can be event-based
or time-based; in time-based simulations, the system is simulated at predefined time
steps even if nothing happens, whereas in event-based simulations, the system is
updated only when an event occurs. In this thesis, the state of the network must be
checked at regular time intervals, and hence, time-based DESs have been employed.
There are no changes in the system between consecutive time intervals that may
be as high as 1 h for energy balancing and as low as <1 s for power balancing.
Basically, the system states are updated after each chosen time interval, and the
situation and conditions are checked.

The output data from a simulation directly corresponds to the outputs that can be
recorded from the real system. Simulation models contrast with optimization mod-
els in that they are “run” rather than solved. The model is run using the given
particular set of input and model characteristics, and the simulated behavior is ob-
served; a set of scenarios is then evaluated and a good solution is recommended for
implementation. Simulations have several advantages, and the relevant ones among

47
them are as follows: (1) New ideas and hypotheses can be explored and tested for
feasibility without disrupting the ongoing operations of the real system; (2) Testing
is easy; (3) Time can be compressed or expanded, as per the requirement; (4) In-
sights into the interaction of variables and the importance of variables to the system
can be obtained; and (5) “What if” questions, which are particularly useful in the
design of new systems, can be answered. In this manner, simulation is a reason-
able approach because it mimics what happens in a real system (or the perception
at the design stage). Further, a DES is ideal for analyzing this network since it
operates over a definite time period and experiences changes in its state and condi-
tions at discrete time intervals. Of course, simulation modeling and analyses can be
time-consuming and expensive, but this is not a serious disadvantage in this type of
relatively smaller DC networks. (Banks et al., 2009)

In order to properly size the PV array and battery in a standalone PV system, two
pieces of information are critical—accurate load and solar irradiation data. The
more accurate the load and solar irradiation data, the more accurate is the system
sizing; in contrast, inaccuracies may lead to over- or under-sizing. In this thesis, it is
assumed that accurate load and solar irradiation data are available and are accurate
(Chapter 2). Moreover, since the use of MPPTs is assumed and not considered
explicitly in this thesis, it is not independently sized; it is simply assumed that the
MPPTs will be sized based on the array wattage. These systems also commonly
employ controls to protect the battery from being over- or undercharged as well as
power conversion subsystems (inverters). However, in this thesis, the sizing of any
system controllers, inverters, wiring, or other system components is not considered.

In this thesis, the PV array and battery are sized for 24-h service times; this is not a
(serious) restriction and the algorithm can be modified to allow the user to change
the number of required service hours (this has been left for a future study). Further,
the effects of losses have been neglected.

3.2.2 Problem statement

To optimally size the two largest components of the given LVDC Island Network—
PV Array for Production and Lead-acid Battery Bank for Storage—in order to im-
prove the performance, cost-effectiveness, and lifetimes of the network

48
3.2.3 Methodology

Battery The battery is sized first because it is a critical component and its sizing
has to be reasonably accurate for optimum reliability. This is because the battery’s
backup essentially decides the NAD, that is, the number of autonomous days in
which the load demand can be satisfied by the battery alone. Moreover, the PV
array must not only meet the load demand but also charge the battery, and since
these two operations must ideally be performed simultaneously, it is preferable to
know the battery size before the PV array size.

The simplest approach is to consider the annual solar data and then size the battery
to meet the load demand for the maximum continuous time period for which the PV
array cannot supply power. So, for example, assume that all 365 days have some
hours without sunlight or with insufficient sunlight. The value of the maximum
uninterrupted non-sunlight period can then be determined and the battery can be
sized accordingly. However, if such a simplistic approach is employed, the battery
will be under-sized; the battery’s recharge capability may not be sufficient for it
to recharge quickly after the discharge, and this will cause problems if the next
non-sunlight period is too close. Of course, at the same time, some loads could
be disconnected and critical loads could be given more autonomy than non-critical
loads. The amount of compromise that is required can be made by the control
algorithm. Nevertheless, this is not a good solution because emergencies leading
to the lack of sufficient PV production are not predictable, and for reliability, it is
preferable to avoid under-sizing, especially in the case of batteries.

In many approaches, as discussed previously in Section 3.2.1, the NAD is given as

a design criterion and the battery is sized accordingly. However, in this study, the
initial approach is to consider that the power supply is always available for all loads;
hence, the NAD has not been considered, or, rather, the NAD has been considered
to be the time period being investigated. Further, since reliability is a key parameter,
it is assumed that the PV array will be sized sufficiently to at least meet the average
daily power requirements of the load under normal conditions and charge the battery
simultaneously. At the same time, the battery sizing methodology assumes that no
power is available from the array, and hence, the resulting battery capacity should
be adequate to meet the typical load requirements.

The inputs to the algorithm are the network voltage; the load profile data (includ-
ing the number of customers) and hourly solar irradiation data for the considered
time period; battery charge and discharge rates; and the minimum SOC of the bat-

49
tery. The idea used here is to assume a minimum battery capacity as a starting point;
either a reasonable assumption can be made for this starting point, or it can be deter-
mined analytically. Beginning with this battery capacity, the network is simulated
for the time period in question and it is verified if the battery SOC decreases to less
than the minimum SOC SOCmin . As soon as SOC < SOCmin , the battery size is in-
cremented and the simulation is restarted. The battery size is then chosen to be the
minimum battery capacity for which the SOC never decreases to less than SOCmin .
Note that the event can be updated at time periods specified by the user, and the
only potential restriction is the availability of irradiation data at those intervals. In
this thesis, 1-h intervals have been used.

PV array The PV array has two basic functions: (1) Supply power to loads and
(2) Charge the battery. Charging the battery is not required once the battery is
charged to its maximum point, and even otherwise, complete charging from the
minimum level may not be required in many situations. Nevertheless, for reliability,
the worst-case scenario should be considered, and the sizing should consider the
power required for fully charging the battery as well as for supplying the load, even
though this may lead to over-sizing. Subsequently, the optimum size can be revised
by real-time simulations, if needed (not done in this thesis).

The PV array is sized to replace the ampere-hour (Ah) in the battery consumed by
the load and to supply the load demand. This is done by simply adding the average
load demand for a day and the maximum battery SOC that can be charged.

Algorithm

Aim To determine the optimum capacity of the battery and PV array

User-input data Hourly solar irradiation H kW/m2 for the time period; number
of customers N; hourly load data for a year for one customer, L Wh, for the time
period (or, all the loads separately); network voltage level V ; battery charge and
discharge rates (BCh and BDch , respectively); and minimum and maximum SOCs of
battery (SOCmin and SOCmax , respectively).

The algorithm proposed in this thesis is given as follows (Algorithm 1), and the
flowchart is given in Fig. 8:

50
Algorithm 1 Algorithm for sizing PV array and battery
Battery
1. Obtain user-input data.

2. Initialize SOCmax .

3. Begin Simulations from time = 1 to timeperiod in steps of 1.

(a) If SOC(time) = SOCmax , do not charge battery.

(b) If H > 0 and SOC(time) < SOCmax , charge battery.
(c) If H = 0, discharge battery to meet load demand.
(d) If SOC(time) = 0, SOCmax = SOCmax + 5. Repeat from Step 3.

4. Else, battery capacity required = SOC(time).

PV Array

1. Determine the average load demand for an hour.

2. PV capacity = Load demand + Battery capacity.

51
 START

Obtain Is
User-Input SOC(time) Do not charge
Data = SOCmax? battery

Initialize
SOCmax

Is H > 0 and Charge battery

SOC(t) < SOCmax?

Yes
Is time >
time period?
No
Discharge
Is H battery to meet
= 0? load demand

Is SOC(t) = 0? SOCmax = SOCmax+ 5

Battery Capacity = SOC(t)

STOP

Figure 8: Flowchart for sizing PV array and battery.

This algorithm was implemented in Matlab© , and the results are given in the next
section.

52
3.2.4 Results and discussion

For the simulations, the following data was chosen: time period = 1 year, or, 8760 h
(8760 simulation steps); number of customers = 20; the hourly solar irradiance for
an entire year for latitude 60◦ was taken as per Section 2.1.3 ; the load profile was a
typical Finnish load profile, as discussed earlier in Section 2.3.3; SOCmin = 0; and
BCh and BDch = C/10.

Figures 9a–9b show the variations in the battery capacities with time. In Fig. 9a
(left), the initial battery capacity is low, and hence, by the 80th day of operation,
the SOC becomes less than SOCmin : SOC(80) < SOCmin . In Fig. 9a (right), the
battery capacity is increased, and now, the SOC becomes less than SOCmin only by
the 8142nd day. In Fig. 9b (left), SOC(8624) < SOCmin . Finally, in Fig. 9b (right),
the battery capacity is at the level where it never goes below zero throughout the
operation period.

53
 Battery Capacity (Ah) versus Time (h) Battery Capacity (Ah) versus Time (h)
500 400

200

Battery Capacity (Ah)

Battery Capacity (Ah)

−500 −200

−400

−1000

−600

−1500
−800

−1000
0 1000 2000 3000 4000 5000 6000 7000 80008142 9000
−2000 Time (h)
0 80 1000 2000 3000 4000 5000 6000 7000 8000 9000
Time (h)

(a) SOC(80) < SOCmin = 0 and SOC(8142) < SOCmin = 0.

Battery Capacity (Ah) versus Time (h)
Battery Capacity (Ah) versus Time (h) 350
350

300 300

250

250
Battery Capacity (Ah)

Battery Capacity (Ah)

200

150 200

100
150

50

0 100

−50
0 1000 2000 3000 4000 5000 6000 7000 8000 8624 9000
Time (h) 50
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
Time (h)

(b) SOC(8624) < SOCmin = 0 and SOC(time) > SOCmin = 0.

Figure 9: Battery capacity versus time for various initial battery capacities.

For the given parameters, the obtained battery capacity is 300 Ah (225 kW h), and
the PV array size is 28.657 kW . If it is assumed that the peak power 800 W h is
demanded by the load at every hour, the energy demand per hour would be equal to
16 kW h
800 × 20 = 16 kW h. The battery capacity per hour would then be 750 V = 21 Ah.
However, the obtained required battery capacity is as high as 300 Ah (225 kW h).
Why is it so much higher? The reason is that the battery is not charged immediately
after discharging, which means that if there were 10 hours of continuous darkness,
21 × 10 = 210 Ah (157.5 kW h) would be needed to be charged. At 60◦ latitude,
there are often 15–20 hours of darkness, and hence, the abovementioned result of
300 Ah (225 kW h) seems quite reasonable for the latitude. In contrast, if the same
parameters are applied at 0◦ latitude (solar irradiation data is changed), the required
battery capacity decreases to 95 Ah (71.25 kW h). In this manner, the obtained re-

54
sults seem quite reasonable.

3.2.5 Limitations and future studies

This algorithm is somewhat over-simplified and requires some modifications to im-

prove the accuracy of the estimation; in particular, not all concepts and ideas have
been considered here. For example, consider the second condition in the battery
sizing: “If H > 0 and SOC(time) < SOCmax , charge battery”. Here, it is assumed
that the power production is sufficient to supply the load and charge the battery,
but this may not be true; there may be enough power production to meet the load
demand but not to charge battery. Similarly, in the case of the condition “If H = 0,
discharge battery to meet load demand”, the battery SOC may not be at a sufficient
level to meet the load demand. These factors may cause some inaccuracies, which
may perhaps not be significant, but, nevertheless, should be kept in mind. More-
over, the use of daily averages of the load demand leads to inaccuracies in the PV
array capacity; this can be improved by conducting DESs for the PV array as well.
Further, the energy required to overcome system losses and inefficiencies has been
neglected in this study. However, this is very important and must also be included
in future studies. Other system components—cables, system controllers, inverters,
wiring, and protection devices—must also be correctly sized in the future.

3.3 Capacity analysis

3.3.1 Introduction

Capacity analysis, as mentioned previously, refers to the ability of the planned ca-
pacity to reliably meet the load demand for the required service hours. There are
many ways of approaching this problem, such as, for example, addressing the fol-
lowing two questions: (1) How does the guaranteed service hours change as the
distance of the network location from the Equator increases (or, as the PV power
availability changes)? (2) How many customers can be supported by the network at
a given location? In this thesis, the question of service hours at the given location
is considered, since knowledge of tolerances to the loss of load can improve the
system design, especially with respect to the cost.

In this section, the effects of the geographical location on the service hours have

55
been investigated. Service hours is defined as the minimum period of time per day
for which uninterrupted power supply can be guaranteed to the customers. Many
variables influence the service hours, such as the number of customers, PV array
and battery capacities, load data, weather conditions, solar irradiance levels that
depend on the location, functioning of other network components, network losses,
and other network parameters. Here, a general method to determine the minimum
number of service hours at any location has been derived. The general algorithm
gives an average answer. Subsequently, practical solar irradiation data at different
latitudes were obtained (in the manner described in Section 2.1.3) and the effects of
the location on the service hours were examined.

3.3.2 Problem statement

Given the production and storage capacities, to determine the number of service
hours that can be guaranteed at any location

3.3.3 Methodology

The number of customers, PV array output power under STC, battery capacity, bat-
tery charge and discharge rates, battery minimum and maximum SOC, load data,
solar irradiance data at the location, and network voltage are assumed to be input by
the user. The functioning of other network components, network losses, and other
network parameters are neglected. Moreover, for simplicity, load prioritization dur-
ing subnormal or emergency situations—or, in other words, unexpected situations
leading to lack of sufficient power supply—is not considered here, and only normal
operations are considered. In the analysis of the effects of the geographical loca-
tion on the service hours, all the factors and parameters (including air mass) other
than the solar irradiance data are assumed to be identical in order to ensure that the
comparisons are valid.

The PV array output is calculated using the PSH method mentioned in Section 2.1.2.
Subsequently, the average PV production and load demand for a day are calculated.
Simulations are then conducted for 1 day; these simulations are not real-time and are
conducted on the basis of average values. Note that this is purely for simplicity, and
the simulations can be extended to any time period, as per the requirements. Finally,
it is assumed that when the PV production is available, the first preference will be

56
for charging the battery as soon as possible and secondly to supply the load. Again,
this is also for simplicity and the analysis can be extended to load prioritization with
real-time simulations.

3.3.4 Algorithm

Aim To calculate the service hours for any location given user-input data

User-input data Hourly solar irradiation data H W/m2 for the time period; num-
ber of customers N; hourly load demand data for a year for one customer, L Wh,
for the time period (or, all the loads separately); network voltage level V ; battery
size (maximum SOC) SOCmax ; battery charge and discharge rates (BCh and BDch ,
respectively); minimum SOC of battery, SOCmin ; and PV array size PVP .

The algorithm proposed in this thesis is then given as follows (Algorithm 2):

57
Algorithm 2 Algorithm for calculating the service hours for any location
1. Obtain the user-input data for the location.

2. Determine the total peak sun hours for the entire year for the location, PSHY .

3. From PSH Y and PV P , determine the possible output energy from the PV array
for the entire year, PVE kWh: PVE = PV P × PSH Y .

4. Determine the average PV energy production for a day: PVE(day) =

(PVE )/365.

5. From the load data, determine the load energy for the entire year, LE kWh, and
then determine the average load demand for a day, LE(D) : LE(D) = LE /365.

6. Begin simulations for 1 day (1 to 24); counter = 1 : 1 : 24

(a) If PV energy is exhausted (PVE(day) (counter) = 0), end the loop and
record the counter.
(b) If the battery capacity is not full,
i. Charge battery: SOC = SOC + BCh .
ii. Supply load (as much load as possible without becoming negative).
iii. Obtain PV energy for next hour (= PV energy in this hour - energy
required to charge battery and supply load).
(c) If the battery capacity is full,
i. Supply load.
ii. Obtain PV energy for next hour (= PV energy in this hour - energy
required to supply load).
(d) Repeat from Step (a).

7. Service hours = counter + time for which battery SOC can supply power.

Results

Simulations using the general algorithm and real solar irradiation data were con-
ducted at various locations with different latitudes. Only the solar irradiation data
was changed and all the other parameters were kept constant in order to facilitate
effective comparisons. The solar irradiation dataset was obtained in the manner
described in Section 2.1.3—hourly data for the global component of the solar ir-
radiation dataset was obtained for the year 2005. The latitude was varied from 0◦
to 65◦ in steps of 5◦ . Additionally, the effects of natural landscape were taken into
account, at least to some extent. For example, areas that comprise deserts were
excluded from the analysis since rainfall is less in such regions, leading to less

58
cloudy weather and skewed results; moreover, Similarly, as far as possible, all the
areas were chosen such that they have the same elevation from sea level, thereby
avoiding altitudinal variations. These factors have resulted in deliberate longitudi-
nal variations in the locations. Since the SoDa service gives free data only for some
longitudes (−66◦ to 66◦ ), some latitudes (for example, 15◦ , 20◦ , and 25◦ , which
point to areas in African deserts or the Atlantic ocean) have been omitted.

A period of a year—365 days—has been considered, and hourly data has been taken
from a single year, 2005. Now, it is natural that the service hours would be different
for different days, and hence, for comparison, the average number of service hours
has been considered; in other words, the number of service hours is calculated using
total service hours
365 .

Figure 10 shows the results of the simulation. The service hours increase from the
Equator (0◦ ) towards the Tropic of Cancer (22.5◦ ) until 30◦ ; this is possibly due to
the higher amount of rainfall and cloud cover in the equatorial regions. The service
hours clearly decrease as the latitude increases beyond 30◦ , which, as expected,
is due to lesser sunlight hours. The slight jump between latitudes 55◦ and 60◦ is
probably due to slight inaccuracies in the solar irradiation data. Although there is
no solar irradiation data beyond 65◦ , the number of service hours is not 0 because
the battery supplies the load.

Variation in Service Hours with Amount of PV Irradiation

20
Service Hours (h)

15

10

0
0 10 20 30 40 50 60 70

Lattitude

Figure 10: Latitude versus number of service hours. The service hours increase
from 0◦ towards the Tropic of Cancer 22.5◦ until 30◦ and then decreases.

59
3.3.5 Limitations and future studies

For simplicity, load prioritization during subnormal operations (when neither the
PV array nor the battery can meet the load demand) is not considered here, and
only normal operations are considered. Further, for more accurate results, more
latitudes (perhaps in steps of 1) should be included.

3.4 Conclusions

In this chapter, two capacity issues—planning and analysis—have been considered.

In the first part, the problem of capacity planning—the determination of the re-
quired PV array and battery capacities—was examined and solved. The proposed
algorithm successfully obtained reasonably accurate battery and PV sizes under dif-
ferent load and sunlight conditions.

In the second part, the problem of capacity analysis—the number of service hours
that can be guaranteed at a location—was investigated. First, a general method to
determine the service hours at any location in the world, given accurate solar irradi-
ation data, was proposed and demonstrated. Subsequently, the method was applied
to different places located at various latitudes. The algorithm clearly demonstrated
the variations in the number of service hours that can be guaranteed to a customer
when the network is located at different latitudes.

The algorithms used were introduced, and the results obtained from their imple-
mentation in MATLAB© were subsequently presented and discussed. Finally, the
limitations of the approaches and the future studies that can be conducted to further
improve the results were proposed.

Overall, the sizing estimations are preliminary, and some work is still required to
improve the estimation accuracies and to eliminate the limitations of the proposed
algorithm. In addition to losses, the algorithm must also consider control concepts
such as load prioritization. Further, the results of the approaches, although reason-
able, have not been compared with those of other studies; such comparisons must be
made in the future in order to test the efficiency of the algorithms and to verify their
improvements and advantages and disadvantages. Moreover, the sizing of system
controllers, inverters, wiring, and other system components is a pending task.

Nevertheless, the estimations are reasonable and practical, and they provide a good

60
basis for proceeding to write algorithms to control the power and energy balances
of the network.

61
4 ENERGY MANAGEMENT SYSTEM: PRINCIPLES
AND FUNCTIONALITIES

4.1 Introduction

4.1.1 Energy management systems

As mentioned in Chapter 1, Introduction, an EMS is basically a system of computer-

aided tools and operations used to monitor, control, and optimize the power produc-
tion and/or transmission efficiencies and overall network performances, and they
are facilitated by a robust communication system that regularly communicates the
network system and component statuses. Efficient and optimal EMSs have been de-
veloped previously for various kinds of networks, and several approaches have been
proposed, but these have primarily focused on grid-connected systems that have the
option of islanding, if required: Very few researches have been conducted into the
development of an EMS for completely islanded LVDC microgrids whose power
production is based on renewable (hence, variable) energy.

This chapter begins with a description of the essential requirements that any EMS
for microgrids must typically fulfill as well as a brief listing of the key features of
EMSs; simultaneously, when relevant, the discussion is expanded to LVDC island
networks as well. The basic principles of operations of the EMS proposed in this
thesis are then elucidated. An important aspect of network control in the completely
islanded mode is the handling of subnormal and emergency conditions which must
be elucidated before proceeding further into the core of the EMS. Hence, the type
of emergencies that can be expected is now explained, and some of the approaches
to deal with them are mentioned. This is followed by a description of the func-
tionalities of the three elements of the network—PV array, lead-acid battery, and
load—and the manner in which their operations are considered and controlled by
the EMS. Finally, the chapter is concluded by summarizing the main points.

4.1.2 General requirements and functions

In this thesis, the term EMS refers to a collection of softwares and their paradigms
that have been designed and developed to enable the proposed LVDC island network

62
to function optimally. Optimal functioning of the proposed LVDC island network
implies that (1) Customers receive uninterrupted power supply (for the guaranteed
service hours); (2) There is no unbalanced power in the network; (3) The network
is adequately protected; (4) Power quality standards are complied; and (5) Energy
costs are optimized in the long term. The role of the EMS is to ensure that the
network functions optimally without, or with minimal, user intervention; in other
words, the network must function automatically and independently (Note that this
thesis does not consider protection of the network.).

In general, EMSs for DC microgrids—when off-grid—must be able to perform

independent operations such that the following requirements are fulfilled (adapted
from Pedrasa & Spooner (2006) and Bhaskara & Chowdhury (2012)):

• Autonomy: EMSs must be able to respond to events autonomously using only

local information.

• Power Quality: EMSs must be able to ensure that power quality requirements
are satisfied.

• Optimization: EMSs must be able to optimize the energy usage.

• Energy Balance: EMSs must be able to maintain energy balance between

supply and demand.

• Voltage Regulation: EMSs must be able to regulate voltages (and frequencies,

in AC microgrids) within acceptable limits.

• Response Speed: EMSs must be able to respond quickly.

• Emergencies: EMSs must be able to handle subnormal and emergency con-

ditions intelligently.

• Communication: EMSs must be able to accept data and send command sig-
nals (if communication infrastructure is present).

The EMS proposed in this thesis has been designed to satisfy all the abovemen-
tioned requirements. The methods adopted for satisfying these requirements will be
elucidated in this and the subsequent chapter. Further, reliable EMSs for microgrids
must comprise certain key features for enabling and ensuring optimal operations;
they are listed as follows:

1. Effective control strategy (algorithm)

63
 2. Efficient software implementation (of the strategy)

3. Reasonable handling of forecasting of production and load demand

4. Reliable communication

5. Reliably sized network components (only for testing purposes; in practice,

the EMS must be able to respond irrespective of the size)

This thesis attempts to incorporate as many of these features as possible. Typically,

reliable sizing of network components has been assumed to be an independent prob-
lem in most prior studies, but, in this thesis, the sizing of PV arrays and batteries
has been considered (Chapter 3). Forecasting, however, has been considered only at
a superficial level, and its efficient integration into the software has been left for fu-
ture studies. Further, communication-related issues have not been explicitly solved
in detail, but some basic ideas have been considered, for example, the nature of the
data used by the softwares as inputs and outputs, acceptable timeframes for obtain-
ing the data, and a basic communication structure; these are discussed separately in
greater detail in Section 4.5. Similarly, protection-related issues have not been con-
sidered at all, and it is simply assumed that protection devices exist and act quickly
and decisively; however, during the implementation of the project, it may be neces-
sary to consider protection issues in greater detail, and in particular, to analyze their
potential impacts on the control efficiency and reaction speeds.

4.2 Energy and power balance principles

In the operation of electric power systems, the main challenge is to ensure that the
“electrical” storage in the system is negligible. This implies that the supply and con-
sumption of electrical power must be balanced at all times. The load is changing all
the time in ways that cannot be perfectly predicted, and hence, the production must
follow the load in real time. Typically, the balance between supply and demand is
performed using a hierarchical control scheme, with crude matching at the longer
timescale (energy balance) and finer matching at the shortest timescale (power bal-
ance). (Kassakian and Schmalensee, 2011) Note that in the LVDC islanded network
being considered, the production also includes the storage.

Thus, the fundamental principles of both energy and power balancing are simple:
the power/energy production must meet and satisfy the load demand (losses should

64
be considered as well). Power balancing must be performed instantaneously while
energy (power × time) balance can be performed over longer periods of time (for
example, daily). This balancing takes place under some constraints, for example,
cost, efficiency, and reliability.

Mathematically, the balancing can be expressed by the following equation:

ˆT
min(Production + Battery − Load − Losses) > 0 (2)
0

Equation 2 expresses the idea that over a time period [0 T ], the total power supply
at any particular point in time must always be greater than the load demand (and
losses). However, as mentioned above, this balancing occurs under some constraints
such as efficiency, costs, reliability, protection, and communication infrastructure.
Hence, the balancing can also be expressed as follows:

(PPV + PBatt. )constraints = (PLoad + PLosses ) (3)

(EPV + EBatt. )constraints = (ELoad + ELosses ) (4)

In this thesis, the primary constraint considered is efficiency.

4.3 Subnormal situations and emergencies

Electricity production from renewable energy sources such as solar and wind tend
to be variable, and hence, inherently unreliable. Moreover, island networks are iso-
lated and cannot depend on other grids to compensate for the lack of power supply.
These two factors imply that there may be situations in which the load demand
cannot be met. Such situations are commonly referred to as subnormal situations
in this thesis, and if the situation worsens, as emergencies (the exact conditions
and classifications are given in the next chapter.). Energy storage devices such as
batteries are a solution to reduce the occurrence of such situations, but they also
have limitations such as cost, size, and maintenance. The occurrence and nature
of subnormal situations depend on the production and storage capacities as well as

65
the infrastructure selected for protection, communication, transmission, and con-
trol; for example, there will be no shortage of power production if the PV array
and battery capacities are very high. However, this is not an economically feasible
solution, and compromises may have to be made during practical implementations
of the network, keeping in mind techno-economic constraints. Further, there are
reasonable chances that losses in power supply will occur in the network, and the
control algorithm must take adequate measures to reduce their impacts. Hence,
before proceeding further, it is necessary to elucidate network situations wherein
power supply may be lost (subnormal situations, including emergencies) and the
possible approaches that can be implemented to mitigate their impacts.

In this thesis, the term “subnormal situation” refers to any situation in which the
normal or typical functioning of a network is constrained by external factors which
may be anticipated (for example, night-time, forecasted cloud cover or rainfall) or
unanticipated (for example, storms) and controllable (for example, faults) or un-
controllable (for example, equipment malfunction). “Emergency” refers to a sub-
normal situation that is close to collapse. Typically, most subnormal situations are
some variations of loss-of-power-supply situations but the root cause may often be
different; for example, it may be due to weather conditions or faults.

The term emergency is used to imply both subnormal and emergency situations in
the subsequent discussion in this chapter and practically used interchangeably. The
exact conditions that define normal, subnormal, and emergency situations are elab-
orated in greater detail in Chapter 5. Here, the emergency situations are considered
and are classified into situations depending on the root causes, and some proposed
approaches are discussed.

Production emergency Production emergency refers to emergency situations in

which, for some reason, the power production is not equal to the load demand (inde-
pendent of other network conditions or parameters). In the network being studied,
this implies that not only the PV array but also the battery cannot provide sufficient
energy to meet the power (or energy) demands of the network. The basic reason
for this emergency to occur is undesirable weather conditions at the location, es-
pecially over a period of time. Some emergency situations may occur regularly;
for example, regular hours of darkness daily can lead to emergency situations if
the sizing is inaccurate, for example, if the battery size is insufficient or it does not
get charged sufficiently during the periods of sunlight. Similarly, some emergency
situations may occur seasonally; for example, the production may be lowered dur-

66
ing the rainy season alone. Some emergency situations may occur unpredictably,
for example, unexpected rainfall, storms, and production equipment failures. The
failure of production equipment can cause the production capacity to decrease or
even stop; for example, batteries whose lifetimes may be approaching end-of-life
will have decreased SOCs and DODs.

Predictable emergency situations can be handled by accurate sizing. However, it

may be sometimes necessary to make compromises on the sizing, which may lead
to emergency situations occurring on a daily basis. Such situations as well as other
unpredictable situations (for example, inclement weather) can be handled by load
prioritization and load disconnection. Note that the customers must know and un-
derstand the implications of load prioritization and despatch and their expectations
must be set accordingly. Additionally, accurate solar forecasting data can be crit-
ical for anticipating and responding to emergencies. Battery end-of-life must be
checked regularly to prevent or respond to emergency situations; further, the PV
array must be inspected and maintained to sustain optimum power production.

Protection emergency Protection emergency refers to emergency situations in

which power imbalances are caused by faults occurring in the network. Faults can
also lead to secondary issues such as production and transmission failures. In gen-
eral, it is expected that the protection equipment itself will be able to handle such
emergencies and restore the network production and transmission. Hence, protec-
tion emergencies have not been addressed in this thesis. At the same time, the EMS
may have to quickly respond at some level to ensure that the operations continue to
be optimal in the remaining parts of the network. The impact of protection emer-
gencies on the control of the network and how the algorithm can respond should be
addressed in a future study.

Communication emergency Communication emergency refers to failures in the

communication systems as a result of which there is either a time delay or infor-
mation loss. The control system is significantly dependent on the communication
infrastructure since it requires inputs from network components such as convert-
ers and other devices for calculations whose results determine the decisions and
control signals that need to be returned to the network components (or elsewhere).
A communication emergency leads to two problems: (a) The control algorithm
receives/sends incomplete or no information, and (b) The control algorithm re-
ceives/sends incorrect information. In such a situation, the control algorithm may

67
use such information and send incorrect control signals that may then have a cumu-
latively large effect on the network, leading to severe imbalances.

In the case of (a), if the control algorithm does not receive all the required infor-
mation within the acceptable timeframe, it can assume that there is an emergency
situation and start operating in the emergency mode. At the same time, it can start
sending signals to the part of the network from which the communication is not re-
ceived, till it receives a response. In the case of (b), the control algorithm can have
some inbuilt error-check mechanism, which it can then use to verify the communi-
cation received. However, such an error check cannot cover all possible situations
and is not a perfect solution; determining a more acceptable solution can be a sub-
ject for future study.

Another question is what actions should be performed if there is no communication

for a long period of time, perhaps because of collapse in the communication infras-
tructure; arrangements for handling the network in the absence of communications
must be considered in a future study. Note that when the control algorithm be-
comes aware of an emergency situation, the network operator is also immediately
informed. Finally, here, it is assumed that the communication infrastructure and
planning will have some measures for handling communication emergency issues,
such as backup communication, and steps to restore communications will be taken
as and when necessary.

Network component emergency Network component emergency refers to fail-

ures in network components, other than production and storage, that may cause the
optimized control algorithm to lose efficiency; typical network components may be
converters and cables. In this thesis, methods to address network component emer-
gencies are not addressed directly, and this could be an important subject for future
studies.

Miscellaneous emergencies Miscellaneous emergencies refer to any emergency

situations that have not been listed previously; typically, these are unexpected un-
foreseen network conditions. Such emergencies may have to be handled in a differ-
ent manner, and their analyses have not been included in this thesis.

68
4.4 Functionalities of network components

Most of the power in the network being studied is either supplied or absorbed by
three main elements—the PV array, battery, and load—and power and energy bal-
ancing require their inputs and outputs to be analyzed and controlled. In general,
the PV array always supplies power, while the load always absorbs power; in con-
trast, the battery both supplies as well as absorbs power. Further, the power outputs
from the PV array and battery and the power input into the battery are controllable,
while the load power demand can only be adjusted by disconnection based on pri-
oritization.

Power flow and energy management problems have been studied previously, for ex-
ample, in Jun et al. (2011), Lagorse et al. (2010), Riffonneau et al. (2011), and
Locment et al. (2012). These researches have primarily considered various power
system operation modes (typically multi-source) and different strategies. However,
their studies are mostly related to power balance control without considering power
supply restrictions or limitations. Wang et al. (2012) extended their researches to
provide an interface for energy management that considers the possible limits im-
posed by the utility grid; limited PV production and load shedding have been con-
sidered to some extent. However, their research is also limited to grid-connected
supplies and only considers occasional islanding. Completely islanded networks
which lead to additional problems, especially those of reliability, service-hour re-
strictions, and optimization, and few researches have considered them. In such
islanded networks, correct battery usage and load prioritization become even more
important.

This thesis proposes power-system operation modes that encompass the various sit-
uations faced by an LVDC island network and considers power and energy manage-
ment strategies for all possible scenarios. The PV array, battery bank, and load have
different operation modes depending on the network conditions (normal operations
or emergency operations). The control algorithm is designed based on this logical
foundation to send control signals depending on the operation mode of the network
components.

69
4.4.1 PV array

In the network, the power output from the PV array is PPV , and it is within the range
0 ≤ PPV ≤ PPV max . PPV is not directly controllable in the sense that it depends on
the solar radiation; however, it can be limited if necessary. The PV array has the
following modes depending on its state of operation. These modes are dependent
primarily on the weather conditions and sometimes on the control signal.

Load Supply Mode: The PV array, in combination with the MPPT algorithm, pro-
duces as much power as possible and is able to meet all the loads: PPV = PPV max =
Total Load Demand

Limited Load Supply Mode: The PV array, in combination with the MPPT algo-
rithm, produces as much power as possible but the power supply is not sufficient to
meet all the loads: PPV = PPV max 6= Total Load Demand (But, PPV + PBat may be
equal to the total load demand.).

Constrained Mode: In this mode, the PV array is constrained to output less power
than the maximum power from the MPPT in order to balance power or energy:
0 ≤ PPV < PPV max

Switchoff Mode: PV array is switched off and there is no power production from the
PV array: PPV = 0 (or less than an irradiation threshold (Wang et al., 2012)). This
mode occurs either when there is no sunlight or when the control algorithm has to
deliberately force the PV array to stop production.

4.4.2 Battery

The energy storage device—the lead-acid battery in this case—is employed as the
backup power supply to compensate for the variable nature of the primary produc-
tion source. However, a battery has limited storage capacity as well as charge and
discharge constraints. In this thesis, three SOC limits have been defined: maximum
SOC, SOCmax , minimum SOC, SOCmin , and least possible SOC, SOCleast . SOCmax
refers to the maximum limit to which the battery can be charged; SOCmin refers
to the minimum limit to which the battery can be charged normally; and SOCleast ,
the lowest possible limit, beyond which the battery should not be discharged under
any conditions. The SOC is practically the most important parameter in network
control, and the battery charges and discharges are used for controlling the network

70
voltage and power balancing in this thesis.

Typically, the battery SOC will be in the range SOCmin ≤ SOC ≤ SOCmax during nor-
mal operations and SOCleast ≤ SOC ≤ SOCmin during subnormal operations. These
ranges are not fixed limits for the two operations since subnormal situations are de-
fined by the ability of the power supply to meet the load demand and not the SOC
alone; in other words, such situations may occur even when SOC ≥ SOCmin ! It is
assumed that all the three limits are specified by the manufacturer and entered into
the software by the network operator before operations. An important difference
between the PV array and battery is that the PV array power production is depen-
dent on the weather conditions, whereas the battery energy utilization is dependent
completely on the decisions taken by the algorithm.

Control signals are sent to the battery based on the decision taken by the algorithm,
and the battery enters one of the following operation modes:

Charge Mode: In this mode, the battery is charged at the maximum charge rate:

ChargeRate possible = ChargeRatemax

This happens when the PV production is optimum, that is, it can meet the load
demand as well as charge the battery.

Limited Charge Mode: Battery is charged, but the charge rate is lower than the
maximum charge rate because of non-optimum PV production:

ChargeRate possible 6= ChargeRatemax

Discharge Mode: Battery is discharged at the highest possible discharge rate:

DischargeRate possible = DischargeRatemax

Limited Discharge Mode: Limited discharge refers to emergency situations in which

it is preferable to charge the battery as quickly as possible (while supplying some
loads) rather than discharge it completely (and supply more additional loads). This
mode is important because the criticality of the battery backup makes it more im-
portant to charge the battery quickly rather than supply all loads; this action is also

71
better for the battery life. Here,

DischargeRate possible 6= DischargeRatemax

SwitchOff Mode: Battery is switched off and neither charging nor discharging takes
place. This happens when the SOC is already at the maximum limit (SOC =
SOCmax ); when SOC is at the lowest possible limit (SOC = SOCleast ; subnormal);
when SOC is at the minimum level (SOC = SOCmin ; normal); or when the PV array
output is sufficient to meet the load demand and losses (PPV = Load Demand +
Losses).

4.4.3 Load division and prioritization

Load division The basic function of any electrical network is to supply the power
demanded by the various connected loads. However, loads utilize power and it is not
always possible to meet the load demand. Load prioritization, load disconnection
and load shedding are potential solutions (as discussed in Chapter 2). In this thesis,
it has been proposed that the loads should be divided into three categories with
certain priorities and further sub-divided into sub-loads with sub-priorities. The
loads are then considered to have different operation modes that depend on whether
the total load demand can be met or not and the consequent prioritization. The
operation modes, load categories, and prioritizations are discussed below.

Normal Load Operations: All the loads receive power supply without any con-
straint.

Load-Constrained Operations: Only some loads can be serviced. The choice of the
loads that can be serviced is based on the load category and their prioritizations, and
this determines the operation mode as well.

1. Critical-Loads Mode: In this mode, only the critical loads are supplied. Crit-
ical loads are the most vital loads in the network, whose servicing may be
a life-and-death issue. Examples of critical loads are hospitals (or, doc-
tors’ clinics), critical lighting (highways or bus stations), emergency services,
triage centers, and disaster-response areas.

2. Essential-Loads Mode: Only essential loads are supplied in this mode. Es-

72
 sential loads refer to loads that are not vital, but, nevertheless, are important.
Typical examples of such loads are banks, ATMs, street lighting, schools,
houses, and water-pumping stations. Critical loads are typically community
loads, but essential loads can also be household loads such as lighting.

3. Normal-Loads Mode: Normal loads refer to any other loads besides critical
and essential loads. Typically, these loads may be household loads such as
fans, refrigeration, internet, and television.

Note that all these loads may or may not be present in the network application that
is proposed, that is, small populated villages; for example, ATMs and hospitals
may be absent in small villages. However, they have been mentioned because it
is preferable to assume that these loads do exist, to consider them as input data,
to build for them, and to then not use them in the calculations, depending on the
situation.

All the three load operations are prioritized according to the scheme given below,
which includes internal prioritization as well.

Load prioritization In the prioritization mechanism, the highest priority is given

to Critical Loads. The second priority is given to charging the battery. The battery
must be charged as soon as possible for two reasons: (1) The battery is not only the
primary but also the only backup; (2) This will prolong the battery life; the battery
is often cycled close to its least possible SOC during subnormal operations, and
hence, it is urgent to charge it. After the battery is charged till at least the minimum
SOC (and if possible, more), the third priority is given to Essential Loads and the
final priority to Normal Loads.

In addition, all the loads may or may not have sub-loads, which implies that there
is a need for internal prioritizations. To illustrate this, consider Critical Loads.
First of all, there may be many critical loads with different criticalities; hospi-
tals or medical services, for example, may be more critical than lighting loads.
Hence, each load type has a “sub-Load Mode” which is referred to as Limited Crit-
ical/Essential/Normal Load Mode when all the loads belonging to that load type
cannot be serviced. Moreover, each critical load may or may not have several loads.
For example, hospitals/doctors’ clinics may have several loads of varying impor-
tance (intensive care units are more critical than a waiting room!), whereas, in con-
trast, lighting loads could be grouped as one load that does not need any further
sub-divisions.

73
Therefore, the category “Critical Loads” has the highest priority, and it is then sub-
divided internally into several sub-loads with sub-priorities; each of these sub-load
is then divided again into super-sub-priorities. Figure 11 illustrates this mechanism.
Top-level loads refer to the three main loads—critical, essential, and normal—and
sub-loads refer to loads such as street lighting or households or hospitals, and super-
sub-loads are the smallest loads such as fans, intensive care units, or refrigerators.
Some super-sub-loads may be independent, for example, power supplies for com-
munication devices or a transformer (if required).

Figure 11: Load prioritization flow. The proposed EMS uses this mechanism dur-
ing subnormal situations or emergencies to maintain power supply to at least some
loads.

In this thesis, it is proposed that the main control algorithm handles the top prior-
ities and the sub-priorities related to each sub-level load, but the load converter of
each sub-load handles the super-sub-priorities (Fig. 12). This distributed control
(“master-slave”) approach simplifies the control mechanism and reduces the burden
on the main algorithm. Note that even if the load converter does not have provisions
for intelligent computing, it is not problematic to integrate the priority-assigning al-
gorithm into the main one.

74
 Top-Level Loads
Critical (Priority 1) n Sub-Loads n Super-Sub-Loads
Super-sub Load 1
(n Super-Sub-priorities)
Essential (Priority 3) (n Sub-priorities)
Normal Loads (Priority 4)

+
Battery Charging Handled by Sub-Load
Converters
(Priority 2)

Handled by Main Control Algorithm

Figure 12: Load prioritization handling: the main control algorithm handles the
top- and sub-level prioritizations, while each load converter handles its internal pri-
oritizations. This simplifies the control and reduces the computational burden on
the main algorithm.

Finally, each and every individual load prioritization must be input into the EMS
before the operations begin. Typically, this should be done by the network operator
on the basis of customer requirements, location-related issues, and technical or other
constraints.

4.5 Information and communication technology (ICT) systems

4.5.1 Roles and requirements

ICT systems play a critical role in ensuring smooth control and efficient power
and energy balancing of the network. The master-slave method employed in this
thesis for power balancing is based on high-speed information transfers, especially
during subnormal situations and emergencies. In particular, the ICT infrastructure
must enable information and control commands from various parts of the network
to be collected and transmitted quickly and reliably. Many previous researches have
investigated communication requirements, especially for managing grid-connected

75
microgrids during their islanding operations (Ding et al., 2009; Llaria et al., 2011;
Calderaro et al., 2008) and have given useful suggestions; Ding et al. (2009), for
example, suggested that the operation of microgrids should be implemented by the
cooperation of various controllers. In addition, they have also recognized that the
reliability and speed of ICT must be high during the operational control and energy
management.

In this thesis, the existence of high-speed and reliable ICT infrastructure is assumed.
Hence, in this section, only the aspects of communication that are relevant to the
optimal functioning of the EMS is discussed. In addition, some remarks are made
on the potential communication structure that will enable smoother distribution of
information as well as control signals.

4.5.2 Information and control signals

In the network, the ICT system has two roles: (1) It must supply certain information
to the control algorithm for its calculations, and (2) It must supply the appropriate
control signals to various parts of the network. Further, in all the operations, speed
and reliability are essential. The nature of the information required is listed in Tables
2–3. Note that information regarding power is strictly not required, since it can be
easily calculated by the control algorithm from the current; nevertheless, it has been
included for clarity.

Table 2: Input parameters required constantly by the control algorithm for its oper-
ations.
Parameters Input to Control Algorithm
PV converter Voltage Current Power –
Battery converter Voltage Current SOC SOH
Load converter Voltage Current Power –

Table 3: Output parameters sent constantly by the control algorithm.

Control Signals
PV converter Switchoff Constrained Load Supply –
Battery converter Charge Limited Charge Discharge SwitchOff
Load converter Disconnect Connect – –

An important question to consider is how quickly must these information inter-

changes happen. The basis for a solution is the requirement that the communication

76
from production, storage, and load must arrive simultaneously; this is important be-
cause the calculations depend on all three information, and data arriving at separate
times will lead to mixups and errors, and hence, incorrect control signals. Further,
the speed of information arrival should not lead to the situation that the control
system does not have enough time to perform all the calculations and send control
signals. Finally, if the network costing is based on an electricity market, the time
can also depend on how often the pricing is updated in the electricity market. In any
case, this is an important question that must be addressed in subsequent researches.

4.5.3 ICT architecture

The ICT structure is an important part of the communication network; how should
the various communication nodes be arranged for optimum communication? These
issues have been considered in another study, but a preliminary idea is to use a
master-slave approach wherein a master control delegates some control aspects to
various subunits that are located at other parts of the network; these subunits may
then delegate further to other subunits. Figure 13 shows the proposed ICT structure.
The master control is located centrally and obtains information that is communi-
cated by the local control devices. The local control devices, in turn, have delegated
their functions to intelligent electronic devices (IEDs) that gather basic information
and supply them to the more central local control. The master and local control
may also process the received information before communicating it further. Such
an arrangement is advantageous for the following reasons:

1. Even if one part of the network is disrupted, information flow to and from
other parts can continue;

2. Communication workload is shared, thereby increasing communication speed;

3. It becomes much easier to detect the location of communication disruption;

and

4. Emergencies can be handled better.

77
Figure 13: Hierarchy of information and communication technologies (ICT) struc-
ture; here, IED refers to intelligent electronic device.

4.6 Conclusions

In this chapter, the essential requirements of any EMS for microgrids were first elu-
cidated and its key features were listed; the discussion was also expanded to include
LVDC island network requirements, wherever there was a difference. The basic
principles of operations of the proposed EMS were then elucidated. Subsequently,
emergencies were introduced and explained in detail along with some approaches
to deal with them.

The primary methods used to control the power and energy balance in the net-
work, as well as to achieve voltage regulation, are the optimal control of the PV
array power production and battery SOC along with appropriate load disconnec-
tions. However, it is important to formulate the logic for performing these actions
in a systematic manner. Such a logical basis is presented subsequently in the form
of the functionalities and operation modes of the network elements. Depending on
the power balance status, these three elements of the network shift to the appropriate
operation mode, which ensures both power balance as well as voltage regulation.
In this manner, these three primary components can be controlled on the basis of
the given paradigm of operations modes. Finally, the importance of ICT systems as
well as the communication requirements were elucidated in detail. In addition, an
appropriate ICT structure based on the master-slave approach has been proposed.

The proposed EMS was developed on the basis of the operation-mode paradigm,

78
and its algorithm shall be introduced, along with the relevant simulation results, in
the next chapter.

79
5 PROPOSED ENERGY MANAGEMENT SYSTEM

5.1 Introduction

In any electric network, the power production and consumption must be balanced
continuously. Several methods have been proposed to achieve power and energy
balances in both AC and DC microgrids, and these have been discussed in Chapter
1. In addition, voltage regulation must be performed at the customer end and in
the DC bus in order to ensure the standard voltage quality. The network considered
in this thesis is an LVDC island network, and hence, only the active components
need to be controlled. The master-slave method (Karlsson and Svensson, 2003)
has been adopted in this thesis to achieve power balance. In this method, power
balance is achieved by controlling the vital elements of the island LVDC system—
production, storage, and consumption—and relies on fast communication between
the respective converters. The production is dependent on the weather conditions
and is highly variable, but it can be controlled; for example, the PV array can be
either constrained to produce less power than available, or its power output can be
maximized. The consumption behaviour is variable but it can be controlled by pri-
oritization and disconnection. Further, the storage elements can be controlled by
sending the appropriate control signals to their converters. In general, the approach
is to control the power flow by controlling the charging and discharging of the bat-
tery.

In this chapter, first, the various possible network conditions, the corresponding op-
eration modes, and the actions taken to achieve power balance control are described
in detail. Subsequently, the algorithm proposed for effective control, which follows
the general guidelines given in Chapter 4, is introduced. This algorithm is designed
to control the power balance under almost all circumstances. The algorithm was
simulated for a sample network, and the results obtained from the simulations are
presented and discussed. Finally, the chapter is concluded with a discussion on the
methodology by which the algorithm can be extended to larger networks.

5.2 Power balance

In this island LVDC network, power balancing is performed at regular time intervals
by collecting production, storage, and consumption data and making the appropri-

80
ate decisions. Further, voltage regulation is performed based on the information
from load converters. To perform these controls, two types of operations have been
defined depending on the status of the network: Normal and Subnormal operations.
The conditions that lead to the normal and subnormal operations as well as the con-
trol signals and responses determined by the algorithm are explained in the next
section.

5.2.1 Normal operations

Normal operations refer to the network control setting when the power production
and storage statuses are ideal—either the appropriately sized PV array, the battery,
or both together are capable of supplying the loads. In the ideal case, the PV array
can not only meet the load demand but also the battery charging requirements, if
any, whereas, in the non-ideal case, the battery assists the PV array to meet the
load demand, being allowed to discharge until the minimum limit. The battery
SOC is required to remain in the following range: Minimum SOC ≤ Battery SOC ≤
Maximum SOC; in other words, when the battery goes into Discharge Mode, the
discharging can happen only till minimum SOC is reached. In addition, the load
demand for that time interval cannot be greater than the power that can be supplied
by the battery, which is limited by the discharge rate (this can happen due to under-
sizing); in other words, the battery’s discharge rate must allow sufficient power to
be supplied to meet the load demand.

The control algorithm must first verify the following three conditions at the begin-
ning of each iteration:

1. Is there enough power in the PV array, battery, or their combination, to meet

the total load demand?

2. Will the battery SOC after the discharge be within the acceptable range (>
Minimum SOC)?

3. Is the battery’s discharge rate able to meet the required load demand?

If these three conditions are satisfied, the network can be said to be operating nor-
mally. Further, on the basis of the PV power production status, two operation modes
have been defined—Optimal-Operation Mode and Sub-Optimal-Operation Mode—
in this study as follows:

81
 1. Optimal-Operation Mode: PV production alone is sufficient to meet the load
demand AND to charge the battery completely1 (if required; charging, how-
ever, may not necessarily be at the maximum charge rate.).

2. Sub-Optimal-Operation Mode: PV production alone is not sufficient to meet

the load demand AND to charge battery completely (if required)2 . Note that,
nevertheless, PV production + battery discharge (SOC > SOCmin ) is sufficient
to meet the load demand and, maybe, to at least partially charge the battery.

In both these modes, certain conditions must be checked, and actions must be per-
formed on the basis of certain conditions and statuses. These conditions and actions
are described below and also listed, for convenience, in Table 4.

Optimal-Operation Mode

1. Battery SOC is at its maximum limit: the PV array goes into Load Supply
Mode and the battery goes into SwitchOff Mode.

2. Battery SOC is less than its maximum limit and PV production can charge
the battery at the maximum charge rate: the PV array goes into Load Supply
Mode and the battery goes into Charge Mode.3

3. Battery SOC is less than its maximum limit but PV production can charge
battery only at charge rates lower than the maximum charge rate: the PV
array goes into Load Supply Mode and the battery goes into Limited Charge
Mode.

SubOptimal-Operation Mode

1. There is some PV production and it is sufficient to meet load demand but not
to charge battery completely: the PV array goes into Load Supply Mode and
battery goes into Limited Charge Mode.

2. There is some PV production but it is insufficient to meet load demand, and of

course to charge battery as well: the PV array goes into Limited Load Supply
Mode and battery goes into Discharge Mode.4
1 Thiscorresponds to the abovementioned ideal case.
2 Thiscorresponds to the abovementioned non-ideal case.
3 During normal operations and Charge Mode, the charging stops when the maximum SOC is

reached.
4 The discharging can happen only till minimum possible SOC.

82
 3. There is no PV production: the PV array goes into SwitchOff Mode and bat-
tery goes into Discharge Mode.

Table 4: Normal Operations

Normal Operations PV Operation Sub-type SOC PV Mode Battery Mode
max Load Supply SwitchOff
= Battery + Load Optimal Normal min ≤ SOC < max Load Supply Charge
PV + Battery= min ≤ SOC < max Load Supply Limited Charge
Load + Losses 6= Battery + Load min < SOC ≤ max Load Supply Limited Charge
PV > 0 Sub-Optimal Normal min < SOC ≤ max Limited Load Supply Discharge
6= Battery + Load min < SOC ≤ max Switchoff Discharge
PV = 0
PV + Battery 6= Battery + Load min — —
Emergency
6= Load + Losses PV > 0
6= Battery + Load min — —
PV = 0

5.2.2 Subnormal operations

Fundamentally, the network is considered to be operating in the Subnormal setting

either when the total load demand is not met, or if the battery is forced to cycle
deeper than normal, that is, SOC ≤ Minimum SOC. This can happen for a variety
of reasons: neither the PV array nor the battery, singly or in combination, may be
able to supply all the loads; or the network or power production may be disrupted in
some manner; or the battery may have been cycled for too long in the previous time
periods. Emergency situations are a typical reason for subnormal operations; their
nature, impacts, and some resolutions have been discussed previously in Chapter
4. During subnormal operations, typically, the PV power production is zero (night)
or insignificant (dawn or dusk or cloud cover); the battery SOC is too low; or, the
battery’s discharge rate is too small to meet the load demand in that time period. In
addition, such situations can occur even when the PV power production and battery
SOC are high, if the load demand is abnormally high, although this would mostly
be a consequence of poor sizing.

In this case, the control algorithm does not have to verify any conditions at the be-
ginning of each hour; quite simply, if the abovementioned conditions for normal
operations (Section 5.2.1) are not met, the network operations revert to subnormal
setting. Load prioritization is a key tool used for power balancing during subnormal
operations and must be done carefully before the network operations begin; more-
over, the control depends on fast load connections and disconnections, and the ICT
systems must be able to handle this requirement capably. Once the subnormal op-
erations have begun, several scenarios have to be considered. Note that the battery
SOC will now remain in the following range: Least Possible SOC ≤ Battery SOC ≤

83
Maximum SOC; hence, when the battery goes into Discharge Mode, the discharging
will happen only till the least possible SOC is reached.

Unlike the case of normal operations, both the PV production and battery SOC
statuses are used to define the two operation modes—Optimal-Operation Mode and
Sub-Optimal-Operation Mode.

1. Optimal-Operation Mode: PV array and battery, singly or together, can meet

the total load demand, but only if battery is cycled to below the minimum
limit to the least possible limit; hence, Least Possible SOC ≤ Battery SOC ≤
Minimum SOC.

2. Emergency-Operation Mode: PV array and battery, singly or together, cannot

meet the total load demand, even if battery is cycled deeply, and some loads
have to be disconnected. Least Possible SOC ≤ Battery SOC ≤ Maximum SOC.

As in the case of normal operations, both these modes have several scenarios and ac-
tions that depend on certain conditions. These conditions and actions are described
below and also listed, for convenience, in Tables 5–6.

Optimal-Operation Mode (Table 5)

1. There is some PV production: the PV array goes into Limited Load Supply
Mode and battery goes into Discharge Mode.

2. There is no PV production: the PV array goes into Switchoff Mode and battery
goes into Discharge Mode.

Table 5: Optimal-Operation Mode during Emergency Operations.

Optimal-Operation PV Mode Battery Mode SOC Load Mode
Mode (Subnormal)
PV + Battery Limited Load Supply Discharge least ≤ SOC < min All Loads
= Total Load (PV > 0)
SwitchO f f (PV = 0) Discharge least ≤ SOC < min All Loads

Emergency-Operation Mode (Table 6)

1. There is some PV production: PV array goes into Limited Load Supply Mode
and battery goes into Discharge or Limited Discharge Mode depending on its
status (Table 6).

84
 (a) Critical Loads:
i. Limited CL (LCL) Mode: Only some critical loads can be supplied.
ii. Critical Loads (CL) Mode: All critical loads can be supplied.
iii. Critical-Battery Load (CB) Mode: All critical loads can be supplied
power, and if SOC < SOCmin , battery is charged preferentially over
supplying essential loads.
(b) Critical and Essential Loads:
i. Limited Critical-Essential Load (LCEL) Mode: All critical loads
and some essential loads can be supplied.
ii. Critical-Essential Loads (CEL) Mode: All critical loads and all es-
sential loads can be supplied.
(c) Critical, Essential, and Normal Loads:
i. Limited Critical-Essential-Normal Load (LCENL) Mode: All criti-
cal and essential loads can be supplied power, but, battery goes into
Limited Discharge mode if SOC < SOCmin (as before).
ii. Limited Critical-Essential-Normal Load (LCENL) Mode: All criti-
cal loads and essential loads and some normal loads can be supplied
power.

Table 6: SubOptimal-Operation Mode during Emergencies with PV power. CL:

Critical Loads; LCL: Limited CL; CEL: Critical-Essential Loads; LCEL: Limited
CEL; LCENL: Limited Critical-Essential-Normal Loads.
Emergency-
Operation Mode PV Mode Conditions Load Mode Battery Mode
(Subnormal)
PV + Battery < Critical Loads LCL Discharge
PV + Battery = Critical Loads CL Discharge
Limited Load
PV + Battery PV + Battery < (Critical + Essential) Loads CB Charge (may
Supply (PV > 0
6= Total Load SOC < SOCmin be Limited)
or PV = 0)
PV + Battery < (Critical + Essential) Loads LCEL Discharge
SOC ≥ SOCmin (may be
Limited)
PV + Battery = (Critical + Essential) Loads CEL Discharge
SOC ≥ SOCmin (may be
Limited)
PV + Battery < (Critical + Essential... LCENL Discharge
... + Normal) Loads
SOC ≥ SOCmin

2. There is no PV power production: The modes and operations are exactly as

before, except that the PV array is now in SwitchOff mode.

It is important to note that charging the battery to at least SOCmin is given higher
priority than the CEL and CENL modes; in other words, battery charging is assumed

85
to be more important than supplying essential or normal loads (for more on this,
refer Section 4.4.3 and Fig. 11).

5.2.3 Algorithm

In practice, the software is supplied data from battery, PV, and converters at regular
intervals (see Section 4.5 on Communications). It then calculates fresh information
and sends the appropriate control signals accordingly. Moreover, some data must
be entered into the software before the network operations begin.

User-input data

Before Operations

Required network voltage level Vnw V; Required DC load voltage VL(dc) V; Required
AC load voltage VL(ac) V; Number of Critical Loads, LoadsC ; Priorities for each
Critical Load, PrC1 − PrCLoadsC ; Number of Essential Loads, LoadsE ; Priorities
for each Essential Load, PrCE1 − PrCLoadsE1 ; Number of Normal Loads, LoadsN ;
Priorities for each Normal Load, PrN1 − PrNLoadsN ; Time interval t; Minimum SOC
SOCmin %; Maximum SOC SOCmax %; Least Possible Battery SOC SOCleast

During Operations

PV array power and voltage, SOC5 , Battery current and voltage, Load power and
voltage (from all loads)

Methodology The algorithm first checks the voltages from all the loads and ver-
ifies if the voltages are regulated with the acceptable limits (±10%, typically). If
any load, or loads, requires voltage regulation, the required power to regulate the
voltage is added to the load demand of that load. Subsequently, the SOC and power
production statuses are verified. Depending on these statuses and the type of load
that requires the voltage regulation, the algorithm makes a decision to supply more
power from the battery, constrain PV production, or to disable the loads.

Note that for voltage regulation, the power reference would be obtained by the fol-
lowing calculation:
5 This algorithm does not consider SOH and DOD presently, but they should be considered in the
future.

86
 Power = (Acceptable Voltage −Current Voltage) ×Current

Moreover, it may not be possible to supply the entire required power immediately
because it will be limited by the weather conditions and battery discharge rate.

Algorithm Preliminary

1. Obtain all user-input data

2. Determine total emergency, priority, and normal loads, and total loads.

3. Assign appropriate priorities.

4. Begin operations.

5. Receive data from converters.

6. Check all voltages. Does any (or many) load(s) need to be regulated?

(a) If yes,

i. VoltageFlag = 1. Note the loads that need to be regulated, including

their priorities and type (AC/DC).
ii. Add required power to the load that needs to be regulated and the
total load demand.
(b) If no, VoltageFlag = 0.

7. Check the following conditions:

(a) Is total power production (PV Power Production + Battery) >= than To-
tal Load Demand ( Load Demand + Losses)?
(b) If some load demand needs to be met by battery discharge, can the bat-
tery discharge meet it?
(c) Will meeting the load demand reduce the SOC to less than SOCmin ?

8. If all the answers to (7) are Yes, Normal Operations. If No, Subnormal Op-
erations.

Normal operations

87
1. Condition 1: If (PV Power Production≥ Total Load Demand) and (Current
SOC = Maximum SOC)

(a) Control Signal: SwitchOff Battery!

(b) Control Signal: Supply Load from PV!
(c) If VoltageFlag = 1, Supply appropriate power to the required load(s).
Power = (Acceptable Voltage - Current Voltage) × Current.

2. Condition 2: If (PV Power Production≥ Total Load Demand) and (Current

SOC < Maximum SOC)

(a) Control Signal: Charge Battery!

(b) Control Signal: Supply Load from PV!
(c) If VoltageFlag = 1, Supply appropriate power to the required load(s).
Power = (Acceptable Voltage - Current Voltage) × Current.

3. Condition 3: If (PV Power Production≥ Total Load Demand) and (Current

SOC < Maximum SOC) and (Discharge Rate < Maximum Discharge Rate)

(a) Control Signal: Limited Charge Battery!

(b) Control Signal: Supply Load from PV!
(c) If VoltageFlag = 1, Supply appropriate power to the required load(s).
Power = (Acceptable Voltage - Current Voltage) \times Current.

4. Condition 4: If (PV Power Production ≥ Total Load Demand) and (Current

SOC ≤ Maximum SOC)

(a) Control Signal: Limited Charge Battery!

(b) Control Signal: Supply Load from PV!
(c) If VoltageFlag = 1, Supply appropriate power to the required load(s).

5. Condition 5: If (PV Power Production < Total Load Demand) and (Current
SOC ≤ Maximum SOC)

(a) Control Signal: Discharge Battery!

(b) Control Signal: Limited Load Supply Mode!
(c) If VoltageFlag = 1, Supply appropriate power to the required load(s).

88
 6. Condition 6: If (PV Power Production = Total Load Demand) and (Current
SOC ≤ Maximum SOC)

(a) Control Signal: Discharge Battery!

(b) Control Signal: SwitchOff Mode!
(c) If VoltageFlag = 1, Supply appropriate power to the required load(s).

Subnormal operations (PV Power Production < Total Load Demand)

1. Condition 1: If (PV Power Production + Battery > Total Load Demand) and
(SOCleast ≤ SOCnow ≤ SOCmin )

(a) Control Signal: Discharge Battery!

(b) Control Signal: Supply Load from PV!
(c) If VoltageFlag = 1, Supply appropriate power to the required load(s).

2. Condition 2: If (PV Power Production = 0 and Battery > Total Load Demand)
and (SOCleast ≤ SOCnow ≤ SOCmin )

(a) Control Signal: Discharge Battery!

(b) Control Signal: SwitchOff!
(c) If VoltageFlag = 1, Supply appropriate power to the required load(s).

3. Condition 3: If (PV Power Production + Battery < Critical Loads) and (SOCleast ≤
SOCnow < SOCmin )

(a) Control Signal: Discharge Battery!

(b) Control Signal: Limited Load Supply: Some Critical Loads!
i. Assign Critical Loads according to priority.

(c) If VoltageFlag = 1, Supply appropriate power to only some critical load(s)

in order of priority.

4. Condition 4: If (PV Power Production + Battery = Critical Loads) and (SOCleast ≤

SOCnow < SOCmin )

(a) Control Signal: Discharge Battery!

(b) Control Signal: Limited Load Supply: All Critical Loads!

89
 (c) If VoltageFlag = 1, Supply appropriate power to as many critical load(s)
as possible in order of priority.

5. Condition 5: If (PV Power Production + Battery < Critical + Essential Loads)

and (SOCleast ≤ SOCnow < SOCmin )

(a) Control Signal: Limited Charge Battery!

(b) Control Signal: Limited Load Supply: All Critical Loads!
(c) If VoltageFlag = 1, Supply appropriate power to only critical load(s) in
order of priority.

6. Condition 6: If (PV Power Production + Battery < Critical + Essential Loads)

(a) Control Signal: Discharge Battery!

(b) Control Signal: Limited Load Supply: All Critical Loads and Some
Essential Loads!
i. Assign Essential Loads according to priority

(c) If VoltageFlag = 1, Supply appropriate power to all critical load(s) and

as many essential loads as possible in order of priority.

7. Condition 7: If (PV Power Production + Battery = Critical + Essential Loads)

(a) Control Signal: Discharge Battery!

(b) Control Signal: Limited Load Supply: All Critical and Essential Loads!
(c) If VoltageFlag = 1, Supply appropriate power to all critical load(s) and
as many essential loads as possible in order of priority.

8. Condition 8: If (PV Power Production + Battery < Critical + Essential +

Normal Loads)

(a) Control Signal: Discharge Battery!

(b) Control Signal: Limited Load Supply: All Critical and Essential Loads
and Some Normal Loads!
(c) If VoltageFlag = 1, Supply appropriate power to all critical and essential
load(s) and as many normal loads as possible in order of priority.

9. Iterate for next loop.

90
5.3 Energy balance

Energy balancing follows exactly the same principles mentioned above, but the en-
ergy balance is verified over larger time intervals, typically days. In addition, the
approach is to use forecasted load and production data and make preliminary ad-
justments to the supply statuses, control signals, and information to the user. Only a
few tweaks need to be made to revise the above algorithm to include energy balanc-
ing, and hence, the methodology has not been included in detail. Energy balancing
algorithms that will include forecasting will be investigated and written in the fu-
ture.

5.4 Results and discussion

Normal operations The simulations were conducted using the following data.
The total number of customer connections = 20; this was divided into 1 Critical
Load, 8 Essential Loads, and 11 Normal Loads. The peak power of the customers
= 800 W. A section of the network used for the simulations is shown in Fig. 14.

DC/AC DC/AC DC/AC DC/AC

L = L = L = L =

== = = =
= = =
DC/DC DC/DC DC/DC DC/DC

==
=
=

BATTERY
=
L =
= L =
= L =
DC/DC PV ARRAY DC/DC DC/DC
= = =
DC/AC DC/AC DC/AC

Figure 14: Sample network for simulations.

91
The load profile was taken, as mentioned in Section 2.3.3, from a Finnish cus-
tomer’s load profile. The same load profile was used for all types of loads; this is a
restriction, since, in practice, the load profiles for different loads would be different.
The solar irradiation data was taken for 60◦ latitude using the SoDa service in the
manner mentioned in Section 2.1.3.

The remaining input parameters are listed as follows:

Network Voltage = 750 V; Time interval = 100 ms; PV array size = 500 kW; Battery
capacity = 600 Ah (450 kW h) = Maximum SOC; Minimum SOC = 70% ; Least
Possible SOC = 50%; Maximum Charge Rate = C/10 = Minimum Charge Rate.

Under these conditions, the network operated normally, and no load disconnections
or prioritizations were required. Power and energy balancing and voltage regulation
were performed without any production problems. Figure 15 shows the simulation
results for a period of 100 ms. The PV power production, total load demand, and
battery SOC statuses are illustrated in the figure. In this figure, the PV production
is deliberately reduced to zero and then increased at random intervals in order to
demonstrate the battery SOC response; in practice, 100 ms is too short a time for
such drastic variations, which would, instead, typically occur through the day. The
battery SOC is given as a percentage of the battery capacity.

The figure shows that the PV power production is sufficient to charge the battery as
well as to meet the load demand during some periods. The battery SOC is charged
and increases to 100% when there is PV power production. When the PV power
production is low (or nil) and unable to meet the load demand, the battery is dis-
charged to SOCmin —70%—which is one of the criteria for normal operations. Note
that there are some situations where the PV power is more than total load demand
as well as battery charging requirements. In these cases, the PV production is con-
strained for power balancing.

92
 120
Control During Normal Operations

PV Prod. (kW) vs. SOC (%) vs. Total Load Demand (kW)
SOC Status (%)
PV Production (kW)
Total Load Demand (kW)
100

80

70

60

40

20

0
0 10 20 30 40 50 60 70 80 90 100

Time (ms)

Figure 15: Power balancing during normal operations. PV production, SOC sta-
tuses, and total load demand are shown. Note that the PV power production is
deliberately reduced to zero and increased at random intervals; in practice, 100 ms
is too short a time for such drastic variations (they would, instead, occur through
the day).

Figure 16 shows the voltage regulation during the normal operations. In LVDC
networks, acceptable voltages are approximately ±10% of the network voltage, and
the voltage should not be allowed to increase or decrease beyond these levels. Note
that in the proposed algorithm, the allowed range is an input by the network operator
(since the regulations may be different across the world.) In order to demonstrate
voltage regulation, random voltages were generated from one part of the network;
most of these voltages were maintained close to the network voltage of 750 V, but
some voltages were allowed to be outside the range [750 − 10% × 750 750 + 10% ×
750] = [675 825]. The randomly generated input voltages are shown in Fig. 16
as a red line. In the figure, the blue line shows the regulated voltages. The figure
clearly shows that the voltages outside the allowed limit are regulated and the LVDC
standards for voltage regulation are met.

93
 Voltage Regulation During Normal Operations
850

Voltage at One Part of Network

Regulated Voltage

800
Voltage (V)

750

700

650

600
0 10 20 30 40 50 60 70 80 90 100

Time (ms)

Figure 16: Voltage regulation during normal operations. The red line represents
randomly generated voltages some of which are forced to be outside the allowed
limits ([750 − 10% × 750 750 + 10% × 750] = [675 825]). The blue line represents
regulated voltages.

SubNormal operations In order to force the network to operate subnormally,

the battery capacity was reduced to 300 Ah (225 kW h) and the PV array size was
reduced to 300 kW. Moreover, the number of critical loads was increased to 3 and
the number of essential loads was reduced to 11. The simulations were conducted
again. The results are shown in Fig. 17.

Power Balancing During Normal and Emergency Operations

PV Prod. (kW) vs. SOC (%) vs. Total Load Demand (kW)

120
PV Production
SOC Status
Total Load Demand
100 Critial Load Demand
Crisis Essential Load Demand
Situation:
Only critical
80 loads 1 and 2;
Other loads
70
In such situations, where PV production is more than total load disconnected
demand, PV production is constrained.
60

50

40

20

0
0 10 20 30 38 40 50 60 69 70 78 80 90 100

Time (ms)

Figure 17: Power balancing during normal and subnormal conditions. Note that
battery SOC now decreases to almost 50%, i.e., the battery is forced to cycle much
deeper. Moreover, not all load demands are met when the battery SOC is close to
50%. The arrow points to crisis situations where only Critical Loads 1 and 2 are
connected and the other loads are disconnected.

94
In this figure, it is noticeable that the battery SOC is now cycling to 50% capacity
during some periods (SOCleast ). The network operates normally till 38 ms, but now,
the power available is too low and the battery is forced to cycle deeper, up to 60%.
However, the PV production increases suddenly, and the battery is able to quickly
recharge back. When the time reaches 69 ms, the situation is worse. Suddenly, there
is no PV production for a long time (63–77 ms), and the battery is forced to start
cycling deeper and deeper from 70% all the way to almost 50%. The battery is
restricted from cycling below 50% and there is no PV power; clearly, the available
power cannot meet all loads. The control algorithm sends signals to disconnect all
the loads, and only critical loads 1 and 2 are supplied power (Critical load 3 has
lowest priority). Luckily, at this time, the PV power production has restarted. This
means that all other loads can be reconnected in the order of prioritization. From
the figure, the battery SOC now begins to increase and reaches 100% again; the
emergency situation has passed and normal operations can begin again. Note that
there are some situations where the PV power is more than total load demand as well
as battery charging requirements. In these cases, the PV production is constrained.

In all these calculations, the peak power of a single customer has been considered to
be 800 W. This may, however, be higher in other circumstances. Higher peak power
would imply that the load demand is higher. As a result, there could be more losses
of power supplies in more parts of the network, which may have to be accounted
for in some manner, maybe by increasing the sizes of the PV array and/or battery or
by changing the number of service hours guaranteed to the customer.

5.5 Application to larger networks

This algorithm can be applied to larger networks with more customers by using a
master controller and many local controllers. The basic idea is to implement many
local control algorithms at various parts of the network, rather than having one
big algorithm that controls all the information. These local controllers handle the
individual power balancing and voltage regulations and report their statuses to a
master controller which verifies the information that all the situations are normal.
In the case of emergencies, the responsibility of the master controller increases and
it may have to take more decisions, for example, to prioritize and disconnect the
load. Such a master-slave approach is similar to the ICT architecture proposed in
Section 4.5.

95
The algorithms presented in this thesis primarily deal with power and energy bal-
ances. Hence, the algorithm considers power consumption by the loads at a general
level and does not address internal voltage or current requirements/variations within
the load unit; it is assumed that the load converters will ensure required voltage reg-
ulation within the load unit itself and will communicate power consumption details
on a regular basis to the main control algorithm. For example, if a household has
many loads such as lighting, fan, and refrigerators, the control algorithm requires
information regarding the total load demand of all these loads; individual loads are
then assumed to be distributed, controlled, and regulated by the load converter. This
decision affects the nature of prioritization, which will be discussed in a subsequent
chapter, but it simplifies the control algorithm and also makes it possible to expand
the network more easily.

5.6 Limitations and future studies

The successful operations and efficiency of the control and sizing algorithms pro-
posed in this thesis have been demonstrated by simulations and results. The re-
sults clearly demonstrate that the LVDC network can be reasonably well controlled
as well as sized using the outlined strategies. Power and energy balancing were
achieved along with voltage regulation under all the circumstances considered. Nev-
ertheless, several challenges still remain, and the algorithm and methodology can
be improved in many ways.

The control is not perfect because forecasted data have not been considered for both
the control operations and for simulations; the application of forecasted data will be
investigated thoroughly in the future. In the control algorithm, load prioritization
has not been assumed in the case of normal operations, but this could be assumed
in order to charge the battery quickly; this can be important in circumstances where
the battery needs to be charged really quickly. Since it may often be beneficial to
prioritize the fast charging of the battery over some loads, the concept of expendable
loads can be examined in a future study.

The algorithms presented here for the power balance control are based on techni-
cal control; in other words, the parameters being varied or changed are basically
technical in nature. Another higher-level control has been proposed previously:
price-based control. In price-based control, the customer’s energy consumption is
regulated by using price as the instrument of control. This is some kind of demand-

96
side management in which the load consumption and load prioritization can be con-
trolled by incentivizing the use of some loads and the non-use of some loads at
certain times of the day. For example, Shevchuk (2012) has suggested that as soon
as there is solar activity, consumption can be decreased and battery charged rapidly;
this would prolong the battery life. In addition, load prioritization during emer-
gencies can be achieved by incentivization. Further research is required, especially
in the case of the control aspect of the energy management system. In particular,
as discussed previously, price can be used as an instrument of control. However,
this is dependent on the development of an “electricity market”. In island LVDC
networks, it is important and practically inevitable to have a good pricing mecha-
nism and a smoothly functioning electricity market. Many researchers have argued
that demand response should be used for electricity pricing in electricity markets in
general. In future investigations, such a pricing mechanism could be developed and
linked to control by giving incentives for load consumption.

A network comprising 20 customers has been considered in the simulations. This

somewhat arbitrary figure is only for simplicity and does not severely restrict the
control or sizing accuracies since larger networks would only require scaling up
and expansion. The methodology for the network expansion has been suggested
in Section 5.5; however, the ideas are preliminary and control and other issues—
such as protection and load dispatch—must be explored in greater detail in a future
study with more in-depth analyses. For example, in very large networks, the con-
cept of load shedding—disconnection of entire areas—may be more critical than
individual load disconnections. The necessity, implications, and consequences of
such important emergency responses should be examined in detail in the future.

The typical peak power of a single customer connection is specified to be approx-

imately 200 W in the beginning. Further, it is assumed to grow to around 800 W
within a few years. This implies load growth which should, ideally, be considered
in the network calculations, especially in the case of cost or loss analysis. Load
growth, related cost parameters (such as interest and annuity), and other factors
have been neglected in this thesis. The typical peak power of a single customer has
been assumed to be 800 W in all the analyses. This may lead to overestimations,
for example in the case of sizing, and load growth must be factored before actual
practical implementation. Moreover, the same load profile was used for all types
of loads; this is a restriction, since, in practice, the load profiles for different loads
would be different.

Solar irradiation data is required and even critical in this thesis for planning, analyz-

97
ing, controlling, and optimizing the PV-based system. Further, the more accurate
the solar irradiation data, the more accurate the results. In all the simulations, illus-
trations, and examples throughout this thesis, it is assumed that the solar irradiation
data used is accurate and reliable. In other words, the software used for testing as-
sumes that all user-input data are accurate, complete, relevant, and in accordance
with the requirements. In other words, no error-checking code for incorrect data
has been built so far; this should be done in the future prior to real implementation.

Deep-cycle battery life and capacity are affected by temperature. Batteries per-
form best in moderate temperatures unlike PV modules that work better at lower
temperatures. As a result, it is often preferable to keep batteries indoors rather
than leaving them subject to outside temperatures. Moreover, the exact voltage-to-
battery-charge correlation is dependent on the battery temperature. Cold batteries
will show a lower voltage when full than hot batteries. In this thesis, the effect of
temperature on battery performance or life is not considered; however, this may be
an important aspect to consider in the future. Most BMSs also indicate the state-
of-health (SOH), which is the point that has been reached in the battery’s life cycle.
It reflects the general condition of a battery and its ability to deliver the specified
performance in comparison with a fresh battery (Pop et al., 2008). However, in this
thesis, battery life and SOH have not been considered; they should be considered in
future studies, especially when planning long-term network operations.

98
5.7 Conclusions

The supply of uninterrupted electric power remains a challenging problem across

the world, especially with increasing consumption and demand. The need for higher
reliability has led to demands for more reliable network solutions than the tra-
ditional 3-phase AC distribution systems. Technical and economic developments
during the last decade, especially in power electronics technology, have given the
opportunity to develop competitive distribution systems based on low-voltage DC
(LVDC) distributions. Island networks that are completely isolated from the main
grid and operate completely independently and co-operatively have been consid-
ered for applications to remote areas that do not have electric power transmission
and supply. However, islanded operations lead to various economic and technical
issues such as power quality, voltage regulation, network stability, harmonics, relia-
bility, protection, and control. Further, environment friendliness is important today,
and renewable energy has been considered a viable alternative to conventional fuels.
The energy production from renewable energy sources is variable, intermittent, and
(usually) at comparatively low voltages, and they are practical for applications to
smaller grids with fewer loads, such as microgrids; this also helps in reducing their
implementation costs.

In this thesis, an LVDC islanded bipolar network using renewable energy sources
has been considered; a PV array supplies the power to typical domestic customer
loads, and a battery bank operates as the backup. The PV array uses thin-film
technology, and they are connected to the DC mains via a DC/DC converter that
is not only able to communicate its power statuses quickly but also receive control
commands and act on them reliably.

Solar power causes problems of reliability of power supply, and hence, a lead-acid
battery has been used as the backup. The battery can either supply or absorb bal-
ance power, depending on the situation. The battery bank is connected to the LVDC
bus with DC/DC converters, and it incorporates a management system that com-
municates the current battery statuses reliably and quickly, while also being able to
accept control instructions and perform the required actions. Further, both DC/AC
and DC/DC load converters communicate all relevant information from the loads to
the control algorithm at regular intervals; the loads are flexible and can be discon-
nected, if required.

In this manner, this thesis considers a bipolar 750-V LVDC network that is powered
by thin-film PV panels with rechargeable lead-acid batteries to supply loads that can

99
be flexible and prioritized, if required. Several challenges remain before the imple-
mentation of such an LVDC island network, such as protection, communication,
and power balancing. Many studies have proposed and analyzed the applicabil-
ity of solar power, wind power, other renewable sources, or their combinations to
microgrids, and various real-world installations (primarily on-grid) have been im-
plemented across the world. However, Most of these studies have focused on using
microgrids as a backup to the main grid and completely islanded operations have
rarely been considered due to the variability of the production. This thesis has fo-
cused on power and energy balancing and voltage regulation issues of completely
islanded LVDC networks. The aim of this thesis was to develop a methodology—
an Energy Management System—to control the power and energy balancing and
voltage regulation of the network.

In order to design the control algorithm, a method to reasonably size the PV array
and battery was first developed. This sizing is important because the EMS can use
the obtained values for its preliminary analyses. The developed sizing method was
based on DESs, and the proposed algorithm successfully obtained reasonable sizes
for PV array and battery under different load and sunlight conditions. Further, the
impact of the availability of sunlight hours at various locations was examined us-
ing another algorithm. First, a general method to determine the service hours at
any location in the world, given accurate solar irradiation data, was proposed and
demonstrated. Subsequently, the method was applied to different places located at
various latitudes (thus changing the number of hours of sunlight). The algorithm
clearly demonstrated the variations in the number of service hours that can be guar-
anteed to a customer when the network is located at different latitudes.

The proposed EMS is based on the master-slave/communication approach. There-

fore, it relies on robust ICT infrastructure; in this thesis, a probable ICT structure
has been proposed and the requirements from the ICT have been elucidated. In order
to logically establish the principles of EMS operations, this thesis defined function-
alities and operation modes of the network elements; these operations modes en-
compass all possible combinations of supply, storage, and load under all conditions
and situations. Using this logical basis as the foundation, these three primary com-
ponents were controlled on the basis of the given paradigm of operations modes.
The EMS functioned by optimally controlling the PV array power production and
battery SOC; in addition, the principles of flexible loads, load prioritization, and ap-
propriate load disconnections were used to ensure that at least some loads continue
to receive power at all times. The efficiency of the EMS was verified and demon-
strated by conducting simulations with a section of the entire network; the EMS

100
successfully and efficiently performed both power and energy balancing as well as
voltage regulation.

Nevertheless, some researches that can further improve the EMS are still pending.
Forecasts of load consumption are critical data that can be used to optimize the
power and energy balancing; they should be incorporated in the future. Price-based
demand-side management in which the load consumption and load prioritization
can be controlled by giving financial incentives to the customer can be considered.
Moreover, network expansion from smaller networks must be examined in greater
detail with extensive simulations.

The control algorithm presented here has been developed with the intention of ap-
plying it practically to LVDC network implementations in real-time situations. The
results presented here are for the Nordic situation since Finnish data has been used
for all the analyses. However, the potential for islanded microgrids supplied by PVs
will be considerably more in the southern regions, where the winters are not so prob-
lematic. Most of the practical implementations are intended to be in and around the
tropical regions. In the future, this control algorithm will be implemented in onsite
LVDC networks, and its efficiency in real situations will be demonstrated.

Completely islanded microgrids using renewable energy sources present numer-

ous techno-economic challenges such as protection, communication, reliability, and
power balance issues. Nevertheless, they are attractive because they can be em-
ployed to supply renewable-energy-based pollution-free electric power to isolated
or remote locations comprising small loads, such as remote villages that have never
received electric power supply. By using such completely islanded microgrids, such
places can be supplied electricity in a cost-efficient and effective manner; this has
important social benefits since the quality of human life will be improved. This
thesis has taken a small but important step toward “replacing the candle” in villages
by developing an efficient and robust EMS for LVDC networks.

101
References
Amin, S.M. & Wollenberg, B.F. 2005. Toward a smart grid: power delivery for the
21st century. IEEE Power and Energy Magazine, vol. 3, pp. 34–41.

Banks, J., Carson J.S., Nelson, B.L. & Nicol D.M. 2009. Discrete-event system
simulation. 5th ed. USA: Prentice Hall.

Barklund, E., Pogaku, N., Prodanovic, M., Hernandez-Aramburo, C. & Green T.C.
2008. Energy management in autonomous microgrid using stability constrained
droop control of inverters. IEEE Transactions on Power Electronics, vol. 23, pp.
2346–2352.

Barnes, M., Ventakaramanan, G., Kondoh, J., Lasseter, R., Asano, H., Hatziar-
gyriou, N., Oyarzabal, J. & Green, T. 2007. Real-world microgrids—an overview.
IEEE International Conference on System of Systems Engineering (SoSE ’07),
April 16–April 18 San Antonio, Texas, USA.

Benghanem, M. 2011. Optimization of tilt angle for solar panel: Case study for
Madinah, Saudi Arabia. Applied Energy, vol. 88, pp. 1427–1433.

Bhaskara, S.N. & Chowdhury, B.H. 2012. Microgrids—A review of modeling, con-
trol, protection, simulation and future potential. IEEE Power and Energy Society
General Meeting, July 22–July 26, California, USA.

Bo, D., Li, Y. & Zheng, Z. 2010. Energy management of hybrid DC and AC bus
linked microgrid. 2nd IEEE International Symposium on Power Electronics for
Distributed Generation Systems (PEDG), June 16–June 18, Hefei, China.

Borowy, S.B. & Salameh, M.Z. 1994. Optimum photovoltaic array size for a hybrid
wind/PV system. IEEE Transactions on Energy Conversion, vol. 9, pp. 482–488.

Borowy, S.B. & Salameh, M.Z. 1996. Methodology for optimally sizing the com-
bination of a battery bank and PV array in a Wind/PV hybrid system. IEEE Trans-
actions on Energy Conversion, vol. 11, pp. 367–375.

Brabandere, K.D., Deconinck G., Vanthournout, K., Driesen, J. & Belmans, R.

2007. Control of microgrids. IEEE Power Engineering Society General Meeting,
June 24–June 28, Tampa, USA.

Calderaro, V., Coppola, V., Galdi, V. & Siano, P. 2008. Assessing control and
management issues for micro wind generation integration into microgrids. The
14th IEEE Mediterranean Electrotechnical Conference (MELECON), May 5–May
7, Ajaccio, France.

Centre Energétique et Procédés of Ecole des Mines de Paris. Helioclim: Provid-

ing information on solar radiation, [online document]. [Accessed 6 May 2013].
Available at URL http://www.helioclim.org/index.html

Chang, T.P. 2009. The sun’s apparent position and the optimal tilt angle of a solar
collector in the northern hemisphere. Solar Energy, vol. 83, pp. 1274–1284.

Chang Y.-P. & Yang L.-D. 2012. Optimal tilt angle for PV modules considering the
uncertainty of temperature and solar radiation. International Conference on Renew-
able Energy Research and Applications (ICRERA), November 11–November 14,
Nagasaki, Japan

Chapman, R. 1987. Sizing handbook for stand-alone photovoltaic/storage systems.

USA: Sandia National Laboratories.

Chen, Y.-K., Wu, Y.-C., Song, C.C. & Chen. Y.-S. 2013. Design and implemen-
tation of energy management system with fuzzy control for dc microgrid systems.
IEEE Transactions on Power Electronics, vol. 28, pp. 1563–1570.

Conti, S. Tina, G. & Ragusa, C. 2002. Optimal sizing procedure for stand-alone
photovoltaic systems by fuzzy logic. Journal of Solar Energy Engineering, vol.
124, pp. 77–82.

Dinçer, F. 2011. The analysis on photovoltaic electricity generation status, potential

and policies of the leading countries in solar energy. Renewable and Sustainable
Energy Reviews, vol. 15, pp. 713–720.

Ding, M., Zhang, Y. & Mao, M. 2009. Key technologies for microgrids-a review.
International Conference on Sustainable Power Generation and Supply (SUPER-
GEN ’09), Nanjing, China. April 6–April 9, 2009.

Doerffel, D. & Sharkh, S.A. 2006. A critical review of using the peukert equa-
tion for determining the remaining capacity of lead-acid and lithium-ion batteries.
Journal of Power Sources, vol. 155, pp. 395–400.

Doni, A., Dughiero, F. & Lorenzoni, A. 2010 A comparison between thin film and
C-Si PV technologies for MW size applications. 35th IEEE Photovoltaic Specialists

103
Conference (PVSC), June 20–June 25, Honolulu, Hawaii, USA.

European Union Low Voltage Directive, LVD 2006/95/ec.

Fraile, D., Latour, M., Gammal, A.E. & Annett, M. 2009. Photovoltaic energy:
Electricity from the sun. Belgium: European Photovoltaic Industry Association.

Gavanidou, E.S & Bakirtzis, A.G. 1992. Design of a stand alone system with renew-
able energy sources using trade off methods, IEEE Transactions on Energy Conver-
sion, vol. 7, pp. 42–48.

Heinemann, D., Lorenz, E. & Girodo, M. 2006. Solar energy resource management
for electricity generation from local level to global scale (chapter), Forecasting of
solar radiation. New York: Nova Publishers.

Hu, J.F., Zhu, J.G., & Platt, G. 2011. A droop control strategy of parallel inverter-
based microgrid. International Conference on Applied Superconductivity and Elec-
tromagnetic Devices (ASEMD), December 14–December 16, Sydney, Australia.

Huang, R., Huang, T., Gadh, R. & Li, N. 2012. Solar generation prediction using
the arma model in a laboratory-level micro-grid. IEEE Third International Confer-
ence on Smart Grid Communications (SmartGridComm), November 5–November
8, Tainan, Taiwan.

IEEE recommended practice for sizing lead-acid batteries for stand-alone photo-
voltaic (PV) systems. 2007.

IEEE guide for array and battery sizing in standalone photovoltaic (PV) systems,
2008.

Jacobson, M.Z. 2009. Review of solutions to global warming, air pollution, and
energy security. Energy & Environmental Science, vol. 2, pp. 148–173.

Jun, Z., Junfeng, L., Jie, W. & Ngan, H.W. 2011. A multi-agent solution to energy
management in hybrid renewable energy generation system. Renewable Energy,
vol. 36, pp. 1352–1363.

Kaipia, T., Salonen, P. Lassila, J. & Partanen J. 2006. Possibilities of the low voltage
DC distribution systems. Proceedings of Nordic Distribution and Asset Manage-
ment Conference (NORDAC 2006), August 20–August 21, Stockholm, Sweden.

104
Kaipia, T., Lassila, J., Salonen, P., Voutilainen, V. & Partanen, J. 2008. A plan-
ning methodology for combined AC and DC electricity distribution networks. The
8th Nordic Electricity Distribution and Asset Management Conference (NORDAC
2008), September 8–September 9, Bergen, Norway.

Kakigano, H., Nishino, A. & Ise, T. 2011. Distribution voltage control for DC
microgrids using fuzzy control and gain-scheduling technique. IEEE Transactions
on Power Electronics, vol. 28, pp. 2246–2258.

Kaplan, S.M., Sissine, F., Wellinghoff, J., Kelly, G.S., Hoecke, J.J. & Abel A. 2009.
Smart grid: modernizing electric power transmission and distribution. USA: The
Capitol.Net Inc.

Karlsson, P. & Svensson, J. DC bus voltage control for a distributed power system.
2003. IEEE Transactions on Power Electronics, vol. 18, pp. 1405–1412.

Kassakian, G.J. & Schmalensee, R. 2011. The future of the electric grid. Mas-
sachusetts: MIT.

Khatib, T. 2010. A review of designing, installing and evaluating standalone photo-

voltaic power systems. Journal of Applied Sciences, vol. 10, pp. 1212–1228.

Lago, J., Moia, J. & Heldwein M.L. 2011. Evaluation of power converters to im-
plement bipolar dc active distribution networks—DC-DC converters. IEEE Energy
Conversion Congress and Exposition (ECCE), September 17–September 22, Ari-
zona, USA.

Lagorse, J., Paire, D. & Miraoui, A. 2010. A multi-agent system for energy man-
agement of distributed power sources. Renewable Energy, vol. 35, pp. 174–182.

Lasseter, R.H. 2002. Microgrids, IEEE Power Engineering Society Winter Meeting,
January 27–January 31, 2002, New York, USA.

Liao Z. & Ruan X. 2009. A novel power management control strategy for stand-
alone photovoltaic power system. IEEE 6th International Power Electronics and
Motion Control Conference (IPEMC ’09), May 17–May 20, Wuhan, China.

Liu, B.Y.H. & Jordan, R.C. 1960. The interrelationship and characteristic distribu-
tion of direct, diffuse and total solar radiation. Solar Energy, vol. 4, pp. 1–19.

Llaria, A., Curea, O., Jimenez, J., Mart, J.L. & Zuloaga, A. 2011. Wireless commu-

105
nication system for microgrids management in islanding. Proceedings of the 2011-
14th European Conference on Power Electronics and Applications (EPE 2011), Au-
gust 30–September 1, 2011, Birmingham, United Kingdom.

Locment, F., Sechilariu, M. & Houssamo, I. 2012. DC load and batteries control
limitations for photovoltaic systems. Experimental validation. IEEE Transactions
on Power Electronics, vol. 27, pp. 4030–4038.

Lopes, J.A.P., Moreira, C.L. & Madureira, A.G. 2006. Defining control strategies
for microgrids islanded operation. IEEE Transactions on Power Systems, vol. 21,
pp. 916–924.

Lukszo, Z., Deconinck, G. & Weijnen, M.P.C. (ed.) 2010. Securing electricity
supply in the cyber age: exploring the risks of information and communication
technology in tomorrow’s electricity infrastructure. Springer.

Markvart T., Fragaki A. & Ross J.N. 2006. PV system sizing using observed time
series of solar radiation. Solar Energy, vol. 80, pp. 46–50.

Masson, G., Latour, M. Rekinger, M., Theologitis, L.-T. & Papoutsi M. 2013.
Global market outlook for photovoltaics 2013–2017. Belgium: European Photo-
voltaic Industry Association.

Muzathik, A.M., Nik, W.M.N.W., Samo, K. & Ibrahim, M.Z. 2010. Hourly global
solar radiation estimates on a horizontal plane. Journal of Physical Science, vol.
21, pp. 51–66.

Ofry, E. & Braunstein, A. 1983. The loss of power supply probability as a technique
for designing stand-alone solar electrical (photovoltaic) systems. IEEE Transac-
tions on Power Apparatus and Systems, vol. 102. pp. 1171–1175.

Papadimitriou, C.N. & Vovos N.A. 2010. A fuzzy control scheme for integration
of DGs into a microgrid. In 15th IEEE Mediterranean Electrotechnical Conference
(MELECON 2010), April 26–April 28, Valetta, Malta.

Pearce, J.M. 2002. Photovoltaics—a path to sustainable futures. Futures, vol. 34,
pp. 663–674.

Pedrasa M.A. & Spooner, T. 2006. A survey of techniques used to control microgrid
generation and storage during island operation. Proc. Australian Universities Power
Engineering Conference. December 10–December 13, Melbourne, Australia.

106
Pogaku, N., Prodanovic, M. & Green, T.C. Modeling, analysis and testing of au-
tonomous operation of an inverter-based microgrid. IEEE Transactions on Power
Electronics, vol. 22, pp. 613–625.

Pop, V., Bergveld, H.J., Danilov, D., Regtien, P.P.L. & Notten P.H.L. 2008. Bat-
tery management systems. Accurate state-of-charge indication for battery-powered
applications. Springer.

Prof. Quaschning. The sun as an energy resource [online document]. [Accessed 2

March 2013]. Available at http://www.volker-quaschning.de/articles/fundamentals1/index.php

Prof. Zervos, A., Lins C. & Muth, J. 2010. Rethinking 2050—A 100% renew-
able energy vision for the European Union. Belgium: European Renewable Energy
Council.

Riffonneau, Y., Bacha, S., Barruel, F. & Ploix, S. 2011. Optimal power flow man-
agement for grid connected PV systems with batteries. IEEE Transactions on Sus-
tainable Energy, vol. 2, pp. 309–320.

Rizzo, R. 2012. An integrated power converter architecture for microgrids with

renewable energy sources. Journal of Electrical Systems, vol. 8, pp. 356–366.

Salonen, P., Kaipia, T., Nuutinen, P., Peltoniemi, P. & Partanen, J. 2008. An LVDC
distribution system concept. Nordic Workshop on Power and Industrial Electronics
(NORPIE), Espoo, June 9–June 11, Espoo, Finland.

SFS 4879—0.6/1 kV power cables. XLPE-insulated Al-cables. Construction and

testing., a.

SFS 4880—0.6/1 kV power cables. PVC-insulated PVC-sheathed cables. Con-

struction and testing., b.

Shevchuk, A. 2012, Operational requirements and control of island LVDC micro-

grids. Master’s thesis, Lappeenranta University of Technology, Department of Elec-
trical Engineering.

Shrestha, G.B. & Goel, L. 1998. A study on optimal sizing of stand-alone photo-
voltaic stations. IEEE Transactions on Energy Conversion, vol. 13, pp. 373–378.

Solar Power Corp. 1978, Solar Electric Generator Systems—Principal of Operation

and Design Concepts.

107
Swanson, M.R. 2009. Photovoltaics power up. Science, vol. 324, pp. 891–892.

Tang, R. & Liu, X. 2010. Installation design of solar panels with seasonal ad-
justment of tilt-angles. Asia-Pacific Power and Energy Engineering Conference
(APPEEC), March 28–March 31, Chengdu, China.

Turner J.A. 1999. A realizable renewable energy future. Science, vol. 285, pp.
687–689.

Wang, B., Sechilariu, M., Locment, F. 2012. Intelligent DC microgrid with smart
grid communications: control strategy consideration and design. IEEE Transac-
tions on Smart Grid, vol. 3, pp. 2148–2156.

Wang, L. 1977. The effects of uncertainties in forced outage rates and load forecast
on the lossof-load probability (LOLP), IEEE Transactions on Power Apparatus and
Systems, vol. PAS-96, pp. 1920–1927.

Weixiang S., Ehsan S.D., Bin, A.S.K. & Seng, O.K. 2005. A study on standalone
photovoltaic system with real meteorological data at Malaysia. Proceedings of
the Eighth International Conference on Electrical Machines and Systems (ICEMS),
September 27–29, 2005, Nanjing, China.

Xiu, Y., Xiang, Z., Fei, Y. & Hai-Yang, Z. 2011. A research on droop control
strategy and simulation for the micro-grid. International Conference on Electrical
and Control Engineering (ICECE), September 16–September 18, Yichang, China.

Yona, A., Senjyu, T., Saber A.Y., Funabashi, T., Sekine, H. & C.-H. Kim. 2008. Ap-
plication of neural network to 24-hour-ahead generating power forecasting for PV
system. IEEE Power and Energy Society General Meeting—Conversion and Deliv-
ery of Electrical Energy in the 21st Century, July 20–July 24, Pittsburgh, Pennsyl-
vania, USA.

Zhang, L., Wu, T., Xing, Y., Sun K. & Gurrero, J.M. 2011. Power control of dc
microgrid using DC bus signaling. 26th Annual IEEE Applied Power Electronics
Conference and Exposition (APEC), March 6–March 11, Fort Worth, Texas, USA.

Zhao, Q., Wang, P. & Goel, L. 2010. Optimal PV panel tilt angle based on solar
radiation prediction. IEEE 11th International Conference on Probabilistic Methods
Applied to Power Systems (PMAPS), June 14–June 17, Nanyang, Singapore.

108