A Protection Scheme For Microgrids Using Intelligent Relays

A Protection Scheme for Microgrids
Using Intelligent Relays
By Debi Prasad Mishra

Master of Engineering
Department of Electrical and Computer Engineering

McGill University, Montreal, Quebec
Dec 2014
A thesis submitted to McGill University in partial fulfillment of the requirements for the degree
of Master of Engineering in Electrical Engineering
© Debi Prasad Mishra, 2014
Abstract
Microgrids are emerging as an important part in modern power distribution systems due
to significant development in distributed generation technologies. Microgrids operate with a high
level of inverter interfaced distributed generation (IIDG) penetration such as fuel cells, solar cell,
and battery storage etc. The problem of microgrid protection becomes challenging as the fault
current varies widely depending upon the operating conditions such as grid-connected and
islanded mode in presence of IIDG. Fast and accurate fault detection coupled with a clearing
mechanism is required for a safe and secured micro-grid operation.
This thesis presents a microgrid protection scheme using intelligent relays. The proposed
intelligent relay is developed by combining a Wavelet Transform and Decision Tree models. The
relay detects and classifies faults using local measurements irrespective of the operating mode
of the micro-grid. The process starts by measuring and pre-processing the current signal at the
relaying point using the Wavelet Transform. Time-frequency features such as change in energy,
entropy and standard deviations are calculated using the wavelet coefficients. Cases
representing various faulted and normal conditions are simulated to generate the complete data
set containing the above mentioned features. This data set is used to train the decision tree for
fault detection. The fault classification data set contains line current negative and zero sequence
components along with the wavelet based features from current signals for the faulted cases
only. The detection and classification decision trees are extensively tested on a large unseen data
set and the test results indicate that the proposed relaying scheme can reliably protect the micro-
grid against faulty conditions under wide variations in operating conditions. The performance is
superior to that of instantaneous over current relays. The performance of the decision tree model
is compared with another data mining model known as the random forest model.
i|Page
Résumé
Les micro-réseaux pourraient jouer un rôle important dans les réseaux de distribution
modernes en raison de développements importants dans les technologies de production
décentralisée. Les micro-réseaux fonctionnent avec un haut niveau de pénétration de la
production décentralisée interfacée avec des onduleurs, telle que les piles à combustible, les
panneaux solaires, et le stockage de batteries. Le problème de la protection du micro-réseau
devient difficile étant donné que le courant de défaut varie largement en fonction des conditions
de fonctionnement, soit raccordé au réseau et en mode îlotage. Une détection de défaut rapide
et précise, couplée avec un mécanisme d’élimination de défaut, est nécessaire pour une
opération sûre et sécurisée du micro-réseau.
Cette thèse présente un système de protection des micro-réseaux utilisant des relais
intelligents. Le relais intelligent proposé est développé en combinant la transformée en
ondelettes et le modèle d'arbre de décision. Le relais détecte et classe les défauts, en utilisant
des mesures locales, quel que soit le mode d’opération du micro réseau. Le processus commence
par la mesure et le prétraitement du signal de courant au point de connexion du relais en utilisant
la transformée en ondelettes. Les caractéristiques temps-fréquence ainsi que les variations
d'énergie, d'entropie et les écarts types sont calculés en utilisant les coefficients de la
transformée en ondelettes. Les scénarios représentant diverses conditions avec et sans défaut
sont simulées pour générer un ensemble de données les caractéristiques choisies. Cet ensemble
est utilisé pour former le l’arbre de décision pour la détection de défaut. L'ensemble des données
de classification de défaut contient le courant de ligne, inverse et homopolaire et les
caractéristiques à base d'ondelettes des signaux de courant pour les cas de défauts. Les arbres
de décision pour la détection et la classification de défauts sont testés sur un grand ensemble de
données. Les résultats indiquent que l’approche proposée peut assurer la protection du micro-
réseau dans des situations de défaut, avec des conditions d'exploitation très variables et qu’elle
est supérieure aux performances des relais de surintensité instantanée. La performance du
modèle de l’arbre de décision a été comparée avec un autre modèle d'exploration de données
connu sous le nom de forêt d’arbres décisionnels.
ii | P a g e
Acknowledgments
I would like to acknowledge those people who have guided, supported and helped me
during the research related to MEng thesis. I would like to express my sincere gratitude to Prof.
Geza Joos for accepting me as his student. I would like to thank him for encouraging me to pursue
research in the area of microgrid. He has provided continuous supervision, guidance for this
research. It has immensely improved my knowledge of power electronics and power system.
I would like to acknowledge visiting Prof. Subhransu Ranjan Samantaray for helping and
guiding me to understand the power system protection. I would also like to thank Dr. Anthony J
Rodolakis for his assistance and support. I would like to acknowledge Diego Mascarella and
Martine Chela for their support in editing the French abstract.
I would also like to acknowledge the support and encouragement provided by my

colleagues Subhadeep Bhattacharya, Syed Qaseem Ali, Aboutaleb Haddadi, Michael Ross, Alexey
Sokolov, Fei Zhang, Harmeet Cheema, Shijia Li, Farah Awan, Qiushi Cui and Mike Quashie. I would
like to thank my parents, siblings, Ambika Prasad and Nibedita, for their unconditional support
and encouragement during the last two years.
The financial support provided through the NSERC/Hydro-Québec Industrial Research

Chair on the Integration of Renewable Energies and Distributed Generation into the Electric
Distribution Grid held by Prof. Geza Joos is acknowledged.
iii | P a g e
Table of Contents
ABSTRACT .............................................................................................................................. I
RÉSUMÉ ................................................................................................................................ II
ACKNOWLEDGMENTS........................................................................................................... III
TABLE OF CONTENTS ............................................................................................................ IV
LIST OF FIGURES ................................................................................................................... VI
LIST OF TABLES ................................................................................................................... VIII
LIST OF ABBREVIATIONS ....................................................................................................... IX
LIST OF ACRONYMS ............................................................................................................... X
CHAPTER 1 INTRODUCTION ................................................................................................... 1
1.1. BACKGROUND ........................................................................................................................... 1
1.2. CONVENTIONAL DISTRIBUTION SYSTEM VS MICROGRID ..................................................................... 2
1.3. PROTECTION ISSUES IN MICROGRID ............................................................................................... 3
1.3.1. Adaptive and Intelligent Relaying.................................................................................. 4
1.3.2. Unnecessary Operations of Protective Devices ............................................................. 4
1.3.3. High Impedance Fault .................................................................................................... 5
1.3.4. Relay Coordination in Microgrid .................................................................................... 5
1.3.5. Role of Communication System in Microgrid Protection ............................................... 5
1.4. LITERATURE REVIEW: PROTECTION SCHEMES FOR MICROGRIDS .......................................................... 6
1.5. PROBLEM STATEMENT ................................................................................................................ 8
1.6. OBJECTIVES .............................................................................................................................. 9
1.7. CLAIMS OF ORIGINALITY .............................................................................................................. 9
1.8. ORGANIZATION OF THESIS ......................................................................................................... 10
CHAPTER 2 MICROGRID MODEL ........................................................................................... 11
2.1 INTRODUCTION TO CERTS MICROGRID ........................................................................................ 11
2.2 COMPONENTS OF THE SIMULATED MICROGRID............................................................................... 13
2.2.1 Distributed Energy Resources ....................................................................................... 13
2.2.2 Three-Phase PI Line model ............................................................................................ 14
2.2.3 Loads in the Microgrid .................................................................................................. 15
2.2.4 Transformers................................................................................................................. 15
2.2.5 Current Transformers ................................................................................................... 15
2.3. MICROGRID OPERATION ........................................................................................................... 16
2.4 CURRENT CONTROLLED INVERTER MODEL ..................................................................................... 16
2.5 VOLTAGE CONTROLLED VSI SYSTEM.............................................................................................. 20
CHAPTER 3 FAULT CURRENT CHARACTERISTICS OF IIDGS ..................................................... 24
3.1 SIMULATION OF FAULT CURRENT CONTRIBUTION ............................................................................ 24
iv | P a g e
3.2 FAULT CURRENT DUE TO GRID-FREQUENCY IMPOSED VSIS ................................................................ 24
3.3 FAULT CURRENT DUE TO VOLTAGE CONTROLLED VSIS ..................................................................... 26
CHAPTER 4 PROPOSED PROTECTION SCHEME ...................................................................... 29
4.1 PROPOSED PROTECTION SCHEME USING AN INTELLIGENT RELAY ........................................................ 29
4.2 OVERVIEW OF WAVELET TRANSFORM........................................................................................... 30
4.2.1 Continuous Wavelet Transform .................................................................................... 30
4.2.2 Discrete Wavelet Transform ......................................................................................... 31
4.2.3 Energy of a Signal ......................................................................................................... 32
4.2.4 Shannon Entropy........................................................................................................... 32
4.2.5 Standard deviation ....................................................................................................... 33
4.3 SIGNAL PRE-PROCESSING USING WAVELET TRANSFORM .................................................................. 33
4.3.1 Haar Wavelet ................................................................................................................ 34
4.3.2 Selection of the Wavelet Coefficient ‘d3’ ...................................................................... 35
4.4 SEQUENCE ANALYZER ................................................................................................................ 37
4.5 DATA MINING TECHNIQUES USED FOR THE PROTECTION SCHEME ...................................................... 38
4.5.1 Decision Trees ............................................................................................................... 39
4.5.2 Random Forest .............................................................................................................. 40
4.6 METHODOLOGY FOR FAULT DATA GENERATION ............................................................................. 40
4.7 METHODOLOGY FOR NO FAULT DATA GENERATION ........................................................................ 42
4.8 GENERATING DATASET CONTAINING FEATURES .............................................................................. 42
4.9 BUILDING DECISION TREE MODEL FOR INTELLIGENT RELAY ............................................................... 43
4.10 RESULTS AND ANALYSIS ........................................................................................................... 44
4.10.1 Results for Fault Detection ......................................................................................... 44
4.10.2 Results for Fault Classification .................................................................................... 46
4.11 COMPARING INTELLIGENT RELAY WITH OVERCURRENT RELAY .......................................................... 47
CHAPTER 5 CONCLUSION ..................................................................................................... 50
5.1 SUMMARY ............................................................................................................................... 50
5.2 CONCLUSIONS .......................................................................................................................... 50
5.3 SUGGESTIONS FOR FUTURE WORK ............................................................................................... 51
APPENDIX A ........................................................................................................................ 52
(a) Case 1: Fault with 0.01 Ohm vs No fault condition in grid Connected Mode. ................. 52
(b) Case 2: Comparison of fault current with 0.01 ohm and 2 ohm Fault resistance in Grid
connected mode. ................................................................................................................... 52
(c) Case-3: Comparison of fault current between Loop and Radial topology ........................ 53
(d) Case 4: Fault current comparison for SLG fault having 0.01 Ohm Resistance for Both Grid
Connected and islanded mode .............................................................................................. 54
(e) Case 5: Fault current comparison SLG fault in Grid Connected mode with 40% vs. 10% DG
penetration ............................................................................................................................ 55
APPENDIX B......................................................................................................................... 56
REFERENCES ........................................................................................................................ 57
v|Page
List of Figures
Figure 2.1 CERTS Single Line Diagram 12
Figure 2.2 Block Diagram of a DER using Solar Power 17
Figure 2.3 Block Diagram for DC-bus voltage Based Current Controlled Inverter 19
Figure 2.4 Schematic Diagram of Voltage Controlled VSI 21
Figure 2.5 Diagram of the control Block for a voltage controlled VSI 22
Figure 3.1 (a) Voltage (b) Current from current controlled inverter in pu for AG fault 25
Figure 3.2 (a) Voltage (b) Current in pu from current controlled inverter for fault ABG 25
Figure 3.3 (a) Voltage (b) Current in pu from current controlled inverter for fault ABCG 26
Figure 3.4 (a) Voltage (b) Current in pu from voltage controlled VSI for AG fault 27
Figure 3.5 (a) Voltage (b) Current in pu from voltage controlled VSI for fault ABG 27
Figure 3.6 (a) Voltage (b) Current in pu from voltage controlled VSI ABCG fault 28
Figure 4.1 Proposed Protection Scheme for CERTS Microgrid 29
Figure 4.2 (a) Single decomposition step (b) Decomposition upto level 3 of a signal ‘x’ 31
Figure 4.3 Frequency bands related to each co-efficient of the wavelet transform 34
Figure 4.4 Haar Wavelet 34
Figure 4.5 Wavelet detail Coefficients Fault and No Fault Signal 36
Figure 4.6 Decision Tree for Classification 40
Figure 4.7 Decision Tree for Fault Detection 48
Figure 4.8 Decision Tree for Fault Classification. 49
Figure A1 Current for Phase A -AG Fault with 0.01 Ohm vs No fault 52
Figure A2 D3 Coefficient for Phase A -AG Fault with 0.01 Ohm vs No fault 52
Figure A3 D3 Coefficient for Negative Sequence –AG Fault with 0.01 Ohm vs No fault 52
Figure A4 D3 Coefficient for Zero Sequence - AG Fault with 0.01 Ohm vs No fault 52
Figure A5 Current for Phase A - AG Fault with 0.01 Ohm vs 2 Ohm 52
Figure A6 D3 Coefficient for Phase A - AG Fault with 0.01 Ohm vs 2 Ohm 53
Figure A7 D3 Coefficient for Negative Sequence – AG Fault with 0.01 Ohm vs 2 Ohm 53
Figure A8. D3 Coefficient for Zero Sequence - AG Fault with 0.01 Ohm vs 2 Ohm 53
Figure A9 Current for Phase A - AG Fault with 0.01 Ohm in Radial vs Loop 53
Figure A10 D3 Coefficient for Phase A - AG Fault - Radial vs Loop 53
Figure A11 D3 Coefficient for Negative Sequence – AG Fault - Radial vs Loop 54
Figure A12 D3 Coefficient for Zero Sequence- AG Fault - in Radial vs Loop 54
Figure A13 Current for Phase A - AG Fault - Grid-connected vs Island 54
Figure A14 D3 Coefficient for Phase A - AG Fault - Grid-connected vs Island 54
Figure A15 D3 Coefficient for Negative Sequence – AG Fault - Grid-connected vs Island 54
Figure A16 D3 Coefficient for Zero Sequence- AG Fault Grid-connected vs Island 55
Figure A17 Current for Phase A - AG Fault with 10% and 40% DG penetration 55
Figure A18 D3 Coefficient for Phase A - AG Fault - 10% and 40% DG penetration 55
vi | P a g e
Figure A19 D3 Coefficient for Negative Seq –AG Fault-10% and 40% DG penetration 55
Figure A20 D3 Coefficient for Zero Sequence - AG Fault - 10% and 40% DG penetration 55
vii | P a g e
List of Tables
Table 2-1 Generation capacity of DERs in microgrid .................................................................... 14
Table 2-2 Line impedance ............................................................................................................. 14
Table 2-3 Load details in the microgrid ........................................................................................ 15
Table 2-4 Transformer details in the microgrid ............................................................................ 15
Table 2-5 Base values used for per unit calculation ..................................................................... 17
Table 2-6 Boost converter parameters ......................................................................................... 17
Table 2-7 Parameters for the voltage source inverter ................................................................. 18
Table 2-8 Parameters for voltage controlled VSI .......................................................................... 23
Table 2-9 Parameters for inner current control loop ................................................................... 23
Table 2-10 Parameters of outer current control loop .................................................................. 23
Table 4-1 Detailed coefficient frequency range details ................................................................ 35
Table 4-2 Number of cases simulated – different types of faults in the network........................ 41
Table 4-3 Parameters for simulating different condition for no fault cases in microgrid ............ 42
Table 4-4 Features used for detection and classification of faults ............................................... 43
Table 4-5 Different classes used for detection ............................................................................. 44
Table 4-6 Different classes used for fault classification ............................................................... 44
Table 4-7 (a) Error Matrix (%) for fault detection using Decision Tree ........................................ 45
Table 4-8 Fault detection summary for DT and RF ....................................................................... 45
Table 4-9 (a) Error (%) matrix for fault classification using DT ..................................................... 46
Table 4-10 Classification result summary using data mining methods ........................................ 47
Table 4-11 Performance comparison for various relays tested in the microgrid ......................... 47
viii | P a g e
List of Abbreviations
CERTS Consortium for Electric Reliability Solutions
CHP Combined Heat and Power
CT Current Transformer
CWT Continuous Wavelet Transform
DER Distributed Energy Resource
DG Distributed Generation
DT Decision Tree
HIF High Impedance Faults
IIDG Inverter Interfaced Distributed Generation
LL Line-to-Line
LLG Line-to-Line-to-Ground
LLLG Line-to-Line-to –Line-to-Ground
MPPT Maximum Power Point Tracking
PCC Point of Common Coupling
PLL Phase-Locked Loop
PMU Phasor Measurement Unit
RF Random Forest
SLG Single Line-to-Ground
SPWM Sinusoidal Pulse Width Modulation
STFT Short time Fourier Transform
VSI Voltage Source Inverter
ix | P a g e
List of Acronyms
ω Angular frequency
θ Phase of the grid from PLL
m Modulation index
Ψ Mother wavelet
μ Mean
σ Standard deviation
α Complex operator
φ Phase angle
Ω Impedance in ohm
p Energy probability distribution
a Dilation parameter
b Translation parameter
IB Current supplied by VSI to grid by controlled frequency VSI
Idref Id reference
Iqref Iq reference
IDG Current passing thought the inductor using current controlled VSI
Kp Proportional gain
Ki Integral gain
VB Voltage across capacitor C of the controlled frequency VSI
Vdcref Reference voltage of the dc bus
Vdc Measured voltage of the dc bus
VBqref Reference voltage for q-axis component of VB
VBdref Reference voltage for q-axis component of VB
x|Page
Chapter 1 Introduction
1.1. Background
Microgrids are emerging as an important part of the modern power distribution
infrastructure. Microgrids have high penetration of distributed energy resources (DERs) along
with communication, control and protection devices [1, 2]. Microgrids operate in
semiautonomous manner and cover a small geographical area within a power distribution
system.
Distributed energy resources or DERs, also known as distributed generation (DG), are
small scale energy resources such as a fuel cell, gas turbine, micro turbine, diesel generator,
photovoltaic cell, wind turbine, battery storage etc. These DGs are embedded within the
distribution network forming the microgrid. Other than electrical power needs, DERs producing
waste heat such as micro turbines, gas engines, and small scale combined heat and power (CHP)
units can be placed closed to the heat loads in order to fulfil the thermal need of customers.
Thus, the presence of DERs in proximity to the loads makes the power delivery more reliable as
it reduces power transmission loss and increases the energy efficiency of the power system by
waste heat utilization. Efficient use of energy from DERs with a higher proliferation of renewable
energy resources within a microgrid also helps to reduce the carbon footprint of the utility
company and of the nation [1]. Thus, the microgrid enhances the reliability of the power network
with both economic and environmental benefits.
A microgrid must have the ability to operate independently when there is any disturbance
in the utility grid such as faults and other contingencies. Microgrids operating without any
connection to the utility grid are known to be operating in islanded mode. It can also operate in
grid-connected mode when it is connected to the utility grid. To improve the efficiency and
reliability of the microgrid, the DERs in the microgrid are connected in a plug-and-play model and
they operate in peer-to-peer manner. The plug-and-play property of the microgrid allows the
DERs to be placed at any point in the electrical system without reengineering the existing control
method and the protection strategy. Peer-to-peer operation mode tries to ensure that there is
no master controller unit, for co-ordination and communication within the microgrid, used to
1|Page
assist the control and protection scheme. A microgrid can operate with the loss of any operating
DERs. Due to the recent development in power electronic interfaces and renewable energy
technologies, the microgrid is embedded with various types of DERs. However, microgrids with
high level of DG penetration still encounter many challenges related to operation, control and
protection.
1.2. Conventional Distribution System vs Microgrid

There are differences between a microgrid and a conventional distribution grid. Many
differences are due to high penetration power electronically interfaced DERs and switching
between islanded and grid-connected mode depending on the microgrid or the utility grid
stability status. There are three main differences which are as follows:
 Inverter Interfaced Distributed Generation

 Bidirectional Power Flow
 Islanded and grid-connected Mode of Operation
Microturbines generate power at high frequency as compared to grid frequency.

Renewable energy resources such as photovoltaic cells generate dc power and wind turbines may
generate electricity using permanent magnet synchronous generator at a frequency which is
dependent on wind speed. The power recovery mechanism for different types of energy storages
devices such as battery storage cells, flywheels are different from synchronous generators. In
order to transfer power from various types of DERs in a controlled manner to the microgrid,
power electronic devices act as interfaces between the DERs and the microgrid. Therefore,
microgrids have a high level of inverter interfaced DG (IIDG) penetration. The operation of an
IIDG is different from a rotating machine. An IIDG has very low inertia as compared to a rotating
machine. The maximum current carrying capacity of the power electronic devices is limited.
Typically, the maximum current carrying capacity is 2 to 3 times of per unit rated current [3-5].
Some power electronic interfaces need to switch between the control modes (See Chapter 2 and
3 for more information regarding switching) when the microgrid switches from grid-connected
to islanded mode of operation. So, additional measures are needed to switch the control strategy
of the power electronics interfaces during islanding operation.
2|Page
The power flow in a microgrid is dynamic and bidirectional as the microgrid has the
capability to import and export power from the main distribution grid[6, 7]. Also, due to a high
penetration of DGs, the power flow can be bidirectional within a feeder. Microgrids may have a
mesh topology to enhance reliability which creates additional challenges in terms of power flow
as compared to the radial topology.
A microgrid operating in grid-connected mode has a higher short circuit current level as
the fault current is supplied by the utility grid. But, the fault current level in islanded mode of
operation is considerably smaller as compared to grid-connected mode for the same kind of fault,
as the power electronic interfaces have limited current carrying capacity depending on the power
converter rating. This creates a new challenge for protection devices operating with the over
current principle. Thus, a comprehensive relaying scheme needs to be developed which must
operate accurately and reliably in both grid-connected and islanded mode with different levels
of short circuit current for the same type of fault [5, 8-14].
1.3. Protection Issues in Microgrid

Relays used for conventional power distribution grids are not fully effective, in identifying
faults, with the increasing penetration of IIDGs into the microgrid. Knowledge of microgrid
operation, during the pre-fault and post-fault period, is necessary in order to understand the
protection challenges for a microgrid. Problems arising due to bidirectional power flow and
reduced level of fault current in islanded mode demands a new protection strategy for protecting
the microgrid. The protection issues involved are as follows:
1. Adaptive and intelligent relaying

2. Unnecessary operations of protective devices
3. High impedance fault identification
4. Relay coordination
5. Role of Communication system in microgrid protection
3|Page
1.3.1. Adaptive and Intelligent Relaying
The fault current level in an islanded microgrid is reduced considerably as compared to
the grid-connected mode. The adaptive relaying scheme detects the grid status (grid-connected
or islanded) by using islanding detection algorithm and then the tripping is selected as per the
updated relay setting. During the islanded mode operation of the microgrid, it might lose some
of the DGs due to faults and this might cause changes in current levels in the microgrid feeders.
The relays should be intelligent and adaptive to sense the changes and protect the feeder
accordingly. Furthermore, these relays should be able to identify the fault in the presence of
inverter based sources supplying limited fault currents for a microgrid operating in islanded
mode. Relays are required to employ intelligent techniques in order to detect faults without the
help of communication and global measurements of current and voltages.
1.3.2. Unnecessary Operations of Protective Devices

Power flow in a microgrid is bidirectional due to the presence of distributed generation.
This may cause accidental tripping of the relays in the grid. In the islanded mode, electrical
equipment such as motors, which are equipped with protection systems, may disconnect when
the voltage goes lower than a set value for a certain period of time. Similar situation might occur
in the case of power sources like wind turbine generators [14]. This may cause further imbalances
within the microgrid operating in islanded mode.
Microprocessor based relays using the information available at the PCC cannot always
determine the fault location given the extreme difference in energy capability between the utility
and the microgrid [15]. As very high speed separation is required to be completed at PCC, there
is a possibility of faulty operation of the switch causing nuisance tripping. For a microgrid in the
islanded mode, nuisance tripping can cause unacceptable power quality problems. Microgrids
which can operate with low power quality are designed to shed all non-critical loads upon
separation from the utility. Nuisance tripping can be addressed by using high speed
communication between PCC and utility side substation.
4|Page
1.3.3. High Impedance Fault
High impedance faults (HIFs) are likely to occur in the microgrid. It is difficult to detect the
HIFs as the fault current magnitude is very small containing harmonic components. Thus, the
conventional protection scheme might fail. The HIF current characteristics need to be studied for
the islanded mode. So, the detection of high impedance fault under islanded mode is an unsolved
problem[9].
1.3.4. Relay Coordination in Microgrid

Most of the distribution protection is based on sensing current. Synchronous DGs will
contribute a large amount of current to a fault. So, the current sensing relays can be used to
detect these faults. However, the reliability of the relays is one of the key performance
parameters, which is further enhanced by the relay co-ordination scheme. In fact, the
coordination between primary and back-up protection is an important issue. Once the
coordination is achieved, the reliability of the protection scheme improves. The selectivity of the
directional over current relays for their zone should be proper in both grid-connected and
islanded mode of microgrid operation [16].
1.3.5. Role of Communication System in Microgrid Protection

Microprocessor based relays which communicate with a central controller can also be
used for protecting the microgrid. In case of a fault in the microgrid, the microprocessor relay
alerts the central controller which in turn will open the circuit breaker at the PCC. In case of a
communication failure between the relay and the central controller an alert message will be
initiated by the microgrid controller. A communication based protection scheme for microgrid
has been described for a medium voltage microgrid [13, 17]. There is no exclusive communication
link required for the relays within the microgrid [13]. However, the microgrid may need time-
synchronized measurements using PMUs if the feeders are spread over distances greater than
18 miles. It is assumed that a microgrid within a small geographic area can use the
communication channels for grid operation. In the case of communication failure, all the relays
are alerted and an alternative (comparative voltage protection) protection scheme is employed
5|Page
until the communication link is restored. However, the communication scheme makes the
protection strategy expensive and adds risks to the microgrid in case of communication failure.
1.4. Literature Review: Protection Schemes for Microgrids

Protection schemes for microgrids must address the problems related to bidirectional
power flow and different levels of fault current in islanded and grid-connected modes. The
protection strategy should be fast and accurate in order to maintain the stability with the
microgrid operating in islanded mode and to protect sensitive loads [2, 18] . Several methods
have been proposed to protect microgrids [8-14] [17-24].
Protection strategies using zero sequence components for single-line-to-ground faults

and negative sequence components of the line current for line-to-line faults in a microgrid was
proposed by Nikkhajoei and Lasseter [9]. The high impedance fault detection issue was
highlighted as an unresolved protection issue for the microgrid. A differential relaying based
protection strategy was proposed for a microgrid with IIDGs [10]. A protection scheme based on
dq-components of voltage was proposed for a microgrid[11]. The dq-components of voltage
were compared with the voltage reference dq-component. It is not clear how the protection
scheme will perform in case there is an unbalanced load in the system or for a high impedance
fault.
Several protection schemes for a microgrid using communication have been proposed
[12, 13, 18-22]. A differential relaying technique has been proposed for a medium voltage
microgrid containing synchronous generators and IIDGs. This method has been tested for
balanced and unbalanced faults including high impedance faults. A communication assisted
adaptive protection method with a high penetration of DGs [19]. However, it does not take into
account the islanded mode operation and the proposed method has not been tested on IIDGs. A
communication(wireless local area network protocol) assisted relaying technique has been
proposed using microprocessor based relays for a medium voltage grid with IIDGs [18]. Most of
the concepts used with this relay was originally proposed in [17] which uses the over current
method and negative sequence current component for the relaying system inside a low voltage
microgrid. An additional technique to identify high impedance faults using the third harmonic
6|Page
component of current has been added to the enhanced relay and the communication method
helps to reduce the operation time of the relay [18]. It also advocates for a back-up protection
strategy without using communication. In [22], a relaying strategy using communication for a
medium voltage looped network has been proposed which uses a similar protection strategy as
proposed in [18] . The strategies proposed in [18] and [22], are independent of the location of
DERs, microgrid operational mode, microgrid size and fault current level. However, the
protection strategy needs communication. A centralized communication assisted over current
based relaying technique, using IEC-61850 protocol, has been proposed for the microgrid [20]. It
contains synchronous DGs and IIDGs in the test network. The centralized controller is connected
to all DERs and relays in the network. The current value used in each relay is calculated
dynamically using microgrid operational parameters and updated repeatedly. There is no
information regarding the performance of the protection scheme for different types of balanced
and unbalanced faults. This scheme will not be effective for high impedance faults. The
protection technique as described in [20] has been improved to include back up relays [21]. The
current value used in each relay is calculated dynamically using microgrid operational parameters
and the relay will operate as an instantaneous time grading relay with latest setting in case of
communication failure. This paper has considered balanced three phases to ground faults only
and there is no information related to the performance of the protection technique on different
types of faults within the microgrid.
A protection scheme for a microgrid with DERs was proposed[23]. This method was
proposed for an islanded grid only. It does not consider the grid-connected mode of operation.
There were no results to validate the proposed protection scheme. A differential energy based
protection strategy was proposed for the microgrid in [24]. Balanced and unbalanced faults and
high impedance faults were discussed in this study. A protection scheme for only IIDG based low
voltage microgrids was proposed [25]. This method uses current and voltage schemes to protect
the grid. But, the relay settings need to be switched between grid-connected and islanded
modes. There is no experimental data for the verification of the proposed strategy. A statistical
classifier based protection scheme has been proposed for an islanded microgrid [26]. This paper
concludes that the differential relaying technique is a better technique as compared to the
7|Page
technique using local measurements. It shows that the symmetrical components of differential
current are better features for fault identification as compared to line current, voltage and
harmonic components of current. This strategy was tested for an islanded microgrid only. To
overcome the problem of limited fault current in islanded mode, a flywheel based system was
designed to supply a higher fault current [27]. This method will help over current relays to
operate in islanded mode. However, this method requires additional cost for investing in a
flywheel. It increases the operational complexity of the microgrid as well.
1.5. Problem Statement

Three important issues need to be addressed in order to design an effective protection
scheme for a low voltage microgrid. First, the relay should operate with high and low level of
fault current in grid-connected and islanded mode respectively. Second, the relay should operate
correctly in case of bidirectional power flow in the microgrid. Third, the protection schemes need
to be fast enough in identifying the fault as the microgrids based on a high penetration of IIDGs
which are very sensitive to disturbances in the network when operating in islanded mode. Also,
the protection scheme should have minimal impact while adjusting to changes in the structure
of the microgrid between radial and mesh topologies and for varying level of DER penetration.
The aforementioned problems can be alleviated by relay co-ordination using

communication between different relays and a central processing unit. The central unit can
change the settings for different types of relays (voltage, over current) depending on the state of
the microgrid and help to identify faults as it has access to more number of measurements as
compared to an independent relaying unit with local measurements. However, the use of
communication increases the cost of the protection scheme. Also the risk of failure of the
microgrid protection scheme is high in case there is a loss of the communication infrastructure.
Therefore, it is essential to develop intelligent relays which can operate using local
measurements to detect a fault quickly in order to maintain the operational stability of the
microgrid and continuity of service to the customer.
8|Page
1.6. Objectives
In order to design an intelligent relay for a low voltage microgrid which can identify all
kinds of balanced and unbalanced shunt faults with the help of local measurements, an analysis
will be performed using the following roadmap.
 A low voltage distribution microgrid will be developed and simulated using IIDGs only.
 An extensive fault study will be carried out for balanced and unbalanced faults with
varying fault impedances for different operating conditions within the microgrid.
 Setting suitable features which are most affected during the fault conditions.
 Building data-mining based model using Decision Trees considering the most affected
features derived at the local end to detect and classify faults in microgrids.
The advantages of using an intelligent relay with local measurements are as follows:
 The protection strategy using intelligent relays with local measurements can result in
accurate fault detection in grid-connected mode and islanded mode as compared
instantaneous over current relays.
 The protection strategy needs to be effective when the microgrid topology changes from
radial to mesh and vice versa.
 The protection strategy does not need any communication infrastructure; hence it
increases the reliability of microgrid operation and reduces the operation cost.
1.7. Claims of Originality

This thesis presents a protection scheme using intelligent relays. There are some
contributions for microgrids with high level of IIDG penetration which can be termed original to
the best of the author’s knowledge:
 For fast and accurate detection of faults using intelligent relays, a set of features were
extracted from the fault current signal using wavelet transform coefficients. Features
obtained for all faulted and unfaulted conditions were used for training the decision trees.
The decision tree based intelligent relay does not need any external information related
9|Page
to microgrid operating status in order to perform fault detection and classification task.
This scheme has been tested for wide range of operating conditions in the microgrid.
1.8. Organization of Thesis

Chapter 2 of the thesis presents a detailed overview of the microgrid which has been
adapted from the CERTS microgrid. It explains the different types of DERs used in microgrids. An
overview of the control and operation of DERs is presented in this chapter. Also, a brief
description of different modes of operation of the microgrid and switching between them has
been discussed. Different types of grid-frequency imposed VSIs and controlled frequency VSIs
using PI controllers have been explained. This chapter describes the simulation of current and
voltage controlled inverters using MATLAB/Simulink.
Chapter 3 describes the response of voltage and current controlled inverters in case of a
fault when there is just one type of control used for an IIDG. This chapter gives an explanation
for the maximum current carrying capability of the IIDG. Then, it becomes easier to understand
the varying short circuit current levels during grid-connected and islanded mode.
Chapter 4 proposes a new protection scheme for protecting the CERTS microgrid
(presented in chapter 2) using intelligent relays. The steps needed to design an intelligent relay
which will be used with local measurements have been explained in this chapter. Different types
of faults were simulated under various conditions. A detailed description of the process for
building a decision tree based intelligent relay has been presented. The performance of the
decision tree based intelligent relay as compared to the random forest algorithm and over
current relay has been studied in this chapter.
Chapter 5 concludes the thesis with a brief discussion regarding the advantages of the
intelligent relay and lists out research topics in order to build a complete protection scheme for
the microgrid.
10 | P a g e
Chapter 2 Microgrid Model
2.1 Introduction to CERTS Microgrid
There are microgrid projects in many countries in the world such as Canada, Japan, and
USA etc. The Consortium for Electric Reliability Solutions (CERTS) Microgrid Laboratory test bed
was developed in USA which operates with the help of modern technologies. The CERTS
microgrid tried to improve the integration small distributed energy resources into microgrid [28].
The CERTS microgrid is a low voltage subsystem in a conventional power distribution system. The
control and operation of the microgrid is different from a conventional power distribution grid.
CERTS microgrid uses advanced technologies to allow seamless transition between grid-
connected and islanded system. It employs a protection strategy as compared to a conventional
distribution grid which does not depend on high fault current and the modern microgrid control
helps to maintain voltage and frequency stability during islanded mode of operation.
CERTS microgrid has the ability to deliver high quality power to loads within the microgrid
[29]. But, the cost of interconnection protection increases with an increase in the number of DGs
[29].
This thesis proposes a protection strategy for a microgrid using power electronic
interfaced Distributed Energy Resources (DER). The microgrid model used is based on the
topology of the CERTS microgrid. Since the focus of this thesis is on power electronic interfaced
DGs, the 2 CHP units were replaced by photovoltaic based DGs interfaced using power electronic
converters. The battery storage based Distributed Energy Resource (DER), also interfaced using
a power electronic converter, remains in place. The 3 units are operated in current control mode
when the microgrid is connected to the grid. When disconnected, the battery based DER is
operated in voltage control mode.
The change in the nature of the DERs will affect the fault characteristics of the microgrid,
given that the fault behavior of power electronic converters is dictated by the control loops rather
than the inherent characteristics of the rotating machines, as in conventional generators. Figure
2.1 shows single line diagram of CERTS microgrid.
11 | P a g e
The CERTS microgrid uses conventional relays such as under-voltage and negative/zero
sequence current components. The conventional approach may not be suitable in power
electronic interfaced DGs, given the limited short circuit capacity and the fact that the short
circuit current is dictated by the current control loops. This work therefore investigates the use
of alternative protection strategies based on intelligent relaying technique.
Figure 2.1 CERTS Single Line Diagram
12 | P a g e
Given that the nature of the DGs has been changed in the microgrid under study, the only
comparison that can be made between the protection approach used in the original CERTS
microgrid, conventional generator based, and the power electronic interfaced DERs used in this
work is in the way the conventional relaying design approach works in the modified context. A
comparison of the reliability of the two approaches will yield different results, because of the
different short circuit behavior of power electronic based interfaces. The comparison between
the conventional relay design approach and the intelligent relay approach is discussed in Section
4.11.
2.2 Components of the Simulated Microgrid

The microgrid under study is a low voltage microgrid operating at 480 volts and 60 Hz.
The microgrid has been simulated using MATLAB/SIMULINK software. The distribution lines were
modelled using PI line model in SIMULINK. The DERs are connected to the microgrid using ∆/Y
transformers. The power in this microgrid is generated by photovoltaic and battery storage
based resources.
2.2.1 Distributed Energy Resources

The power generation in this microgrid is done by photovoltaic cells using solar power
and battery based storage resources. All the distributed energy resources in the microgrid are
connected to the microgrid using voltage source inverters. These inverters employ different types
of control depending on design of the microgrid and the type of DERs attached connected to it.
There are three sources generating power in the microgrid.
One battery based storage source and two sources based on solar power are present in
this simulated microgrid. Each photovoltaic resource is connected to the microgrid using a
current controlled inverter and the control method for the inverters remains unchanged whether
the microgrid operates in grid-connected or islanded mode. The battery storage based inverters
operate as current controlled inverters in grid connected mode, but the type of control changes
to voltage control method when the microgrid operates as an islanded grid. Current controlled
inverters always operate at unity power factor. Voltage controlled inverters are responsible for
supplying the required reactive power for all the inductive loads in islanded mode and control
13 | P a g e
the voltage in the islanded mode. It also sets up the microgrid frequency in islanded mode. A
detailed description of the control methods used for current controlled and controlled frequency
inverters have been presented in section 2.4 and section 2.5 respectively.
Table 2-1 Generation capacity of DERs in microgrid

DERs Name Generating Capacity
DG-A1 100KW
DG-A2 100KW
DG-B1 100KW
All the DGs used in this grid are interfaced with the microgrid using voltage source
inverters. DG-B1 and DG- A2 are powered by photovoltaics cells and always operate in current
controlled mode. The current controlled DGs operate with unity power factor.DG-A1 is powered
by a battery storage cell and the control action for this inverter switches from current control
mode to voltage control mode when the microgrid state is changed from grid-connected to
islanded mode and vice-versa. This DG can supply both active and reactive power as per the
demand in the microgrid. This DG is used to control the frequency and voltage of the microgrid
in islanded mode.
2.2.2 Three-Phase PI Line model

The three-phase PI line model lumps the line capacitance at the end of the line. The line
parameters such as resistance, inductance and capacitance are specified for zero sequence and
positive sequence. These parameters take inductive and capacitive coupling between 3-phase
conductors and ground as well. The bus numbers have been numbered from 1 to 8. The
impedances between two buses within the microgrid is given in the Table 2-2.
Table 2-2 Line impedance

Between Length R in L in Zero Zero
Nodes in Ohm/km Ohm/km Sequence R Sequence L
Meters in Ohm/km in Ohm/km
1-2 68.58 0.3772 1.206e-3 0.7637 3.5106e-3
3-4 68.58 0.3772 1.206e-3 0.7637 3.5106e-3
5-6 68.58 0.3772 1.206e-3 0.7637 3.5106e-3
6-7 22.8 0.3772 1.206e-3 0.7637 3.5106e-3
7-8 68.58 0.3772 1.206e-3 0.7637 3.5106e-3
14 | P a g e
2.2.3 Loads in the Microgrid
There are 4 loads in the microgrid diagram shown in the Figure2.1 .One load is outside
PCC. All the loads consume power during the grid-connected mode, however, during the islanded
mode, only three load banks are supplied by the sources within the microgrid. All the load models
used in this study are balanced in nature. The loads have star configuration. Details of power
consumed by each load in the microgrid are given in the Table 2-3.
Table 2-3 Load details in the microgrid

Load Name Resistive Load In KW Inductive Load in Capacitive Load in
KVAR KVAR
L-3 90 45 0
L-4 90 45 0
L-5 90 0 40
L- 6 90 0 20
2.2.4 Transformers
The microgrid is connected to a 13.8KV utility grid using a transformer ‘T1’. This
transformer is a step down transformer. Information related to all the transformers is given in
Table 2-4. Transformer ‘T3’ has higher capacity as the DG connected to this transformer has the
ability to supply higher amount of power in islanded mode as compared to the other two DGs
present within the microgrid.
Table 2-4 Transformer details in the microgrid

Transformer Primary Voltage Secondary Rating R Primary = R L in H
Name in Volts Voltage in in KVA Secondary in Primary =
Volts in pu Secondary
in pu
T1 13800 480 1500 0.01 0.03
T2 480 480 120 0.01 0.03
T3 480 480 220 0.01 0.03
T4 480 480 120 0.01 0.03
T5 480 480 120 0.01 0.03
2.2.5 Current Transformers

The positions of the current transformers are shown in Figure 2.1. There are four CTs for
measurement in the grid. The line current measured by the CTs will be given as input to the relays
used for protection.
15 | P a g e
2.3. Microgrid Operation
The microgrid changes the operating mode from grid-connected to islanded mode in case
of short circuit on the grid-side or due to intentional islanding. The microgrid islanding is done
intentionally in case of a disturbance in voltage and frequency is detected on the utility grid. The
islanding process involves change in the control and operation within the microgrid. The islanding
operation occurs very fast in order to protect the microgrid. Appropriate changes in operation
and control are done for all IIDGs in the microgrid in islanded mode. Problems related to stability
of microgrid and reliable power supply arises in islanded mode as power electronic interfaces
have insignificant inertia.
The simulated microgrid operates in both grid-connected and islanded mode. Under grid
connected mode, as shown in Figure 2.1, all the DERs operate in current controlled mode and
the voltage and frequency of the grid is set by the utility grid. In grid-connected mode, if the loads
within the microgrid exceed the microgrid generation capacity, they can be served within the
microgrid by importing power from the utility grid. The microgrid needs a dispatchable power
source in islanded mode in order set the voltage and frequency within the microgrid and the
dispatchable power source should be capable of changing the output power quickly. Battery
storage based sources can be used as dispatchable resources which will operate in a load
following manner during islanded condition only. In islanded mode of operation, the power
balance between generation and load needs to be maintained and non-critical load shedding is
done to maintain the power balance within the microgrid in case of overloading.
2.4 Current Controlled Inverter Model

A three-phase grid-connected sinusoidal pulse width modulated current controlled
inverter has been simulated for this study [30]. In this model, the inverter receives power from
photovoltaic cell via one dc to dc boost converter. The dc-dc converter operates using MPPT
controller to extract maximum power from the photovoltaic unit. The photovoltaic cell is capable
of generating approximately 100KW (irradiance at 1000 Watt per square meter) power using
MPPT algorithm. The current controlled inverter operates at unity power factor by setting the
reactive reference current to zero. The current control technique has been implemented using
16 | P a g e
PI control in dq-frame. The phase-locked loop (PLL) plays an important role in synchronizing the
inverter with the grid by sensing the grid frequency. So, the current controlled inverters are also
known as grid-imposed frequency controlled inverters. The current controlled inverter can be
implemented either by using PQ-control or dc-bus voltage control. Constant dc-bus voltage
based current control technique has been explained in this section as the dc-bus is fed using a
power source i.e. solar cell. The schematic diagram of the inverter with solar cell has been shown
in the diagram Figure 2.2.
V, I MPPT
Control
Duty Cycle
T VB
L
C Grid
IDG
Solar DC-DC Inverter
Figure 2.2 Block Diagram of a DER using Solar Power
Table 2-5 Base values used for per unit calculation

Base Values for Per Unit Calculations
Base Power 100KVA
Base voltage 480V
Base Impedance 2.304Ohm
Table 2-6 Boost converter parameters
Boost Converter Parameters

Switching Frequency 5000Hz
Inductor on PV side 5mH
Capacitance on PV side 100e-6F
Diode Resistance 0.1mOhm
17 | P a g e
The solar cells used in this study operate using the maximum power point tracking
algorithm. The solar cell produces 100 KW using 41 parallel strings where each string has 8
modules. Each module contains 96 cells. These solar cells operate using the maximum power
dP
tracking control algorithm (incremental conductance method  0 ). The parameters used for
dV
the boost converter are given in Table 2-6. The specifications of the inverter connected to the
grid have been specified in Table 2-7.
Table 2-7 Parameters for the voltage source inverter

Inverter Parameters
DC-LINK voltage 900V
DC-LINK capacitance 11mF
Output filter inductance( L) 0.3mH
Inverter rating 100KVA
Inverter ‘on’ resistance 0.6mOhm
Switching frequency 8000Hz
There are three inputs to the control system. First input is VB which is the voltage sensed
after the inductive filter, second is the line current IDG passing through the inductor and the third
one is the dc bus voltage Vdc.
The control system can be divided into two loops. The inner loop is a current control loop
and the outer loop is the dc-bus voltage control loop. The outer loop works to stabilize the dc-
bus voltage around Vdcref and it produces a current reference signal Idref for the inner current
control loop which is responsible for the active power transfer done using the inverter. The
maximum and minimum limit of Idref has been set to 1.2 times of rated inverter current and zero
respectively. In order to achieve unity power factor control, the reactive power supplied by the
inverter is controlled by setting Iqref to zero.
The schematic diagram of a current controlled VSI has been presented in Figure 2.3 . The
symbols in the control diagram are explained as follows:
PLL = Phase locked loop,
IDG = Current passing thought the inductor,
18 | P a g e
Vdcref =Reference voltage of the dc bus,
Vdc= Measured voltage of the dc bus,
C= DC link capacitance,
L= Inductance of the filter,
m = modulation index,
φ = Phase angle,
f=frequency of the grid
VBq
abc-to-dq
VBd
ϴ
ωt
PLL
VB(abc)
Phase-to-ground
ϴ Iqref VBq
m=+1
+ Vqref
Iq - +
Kp +Ki/s
+ m
+ m=sqrt
m=0
(Vdref^2+
IDG(abc) Vqref^2)
ωL 2/Vdc
abc-to-dq & SPWM
ωL
Φ= atan2
- (Vqref/
Vdref Vdref) φ
Kp +Ki/s
Id - +
+
+
Idref = +1.2 Idref
VBd
Vdc
Kp +Ki/s
- +
Idref = 0 To inverter
Vdcref
Figure 2.3 Block Diagram for DC-bus voltage Based Current Controlled Inverter
ω = 2πf,
ϴ = Phase of the grid from PLL,
SPWM= Sinusoidal pulse width modulation,
abc-to-dq = dq transformation of 3 phase ac quantities.
19 | P a g e
Idref =Id reference,
Iqref =Iq reference,
Kp= Proportional Gain,
Ki= Integral Gain
2.5 Voltage controlled VSI system

In this section, a frequency and voltage controlled voltage source inverter is presented.
The inverter is responsible for controlling the frequency and the voltage within the microgrid. It
supplies both active and reactive power. This type of inverter is fed by a dispatchable energy
source. In the test system, power for this inverter is supplied by a battery storage system. This
inverter needs a capacitor in the filter bank in order to measure the voltage which is used as
feedback to the voltage control.
The VSI system is modelled and controlled in dq rotating frame [30]. The frequency of the
VSI is set by a virtual PLL at 60Hz. The control loop can be divided into two parts. The inner
current control loop and the outer voltage control loop.
The outer voltage control loop takes the voltage sensed across the filter capacitor as an
input signal after converting it to the dq-frame. The q-axis voltage reference Vqref is set to zero
and d-axis reference Vdref is set to 1 pu. The outputs Idref, Iqref of the outer voltage loop act as
reference signal to the inner loop. The current measured after the filter capacitor is given as input
to the inner current control loop. Finally, the switching signals are fed to the inverter. A schematic
diagram of a controlled-frequency VSI has been presented in Figure 2.4.
20 | P a g e
Control
VB
L
DC Source To Grid
IDG IB
Inverter C
Figure 2.4 Schematic Diagram of Voltage Controlled VSI
The detailed control diagram of a controlled-frequency VSI has been presented in Figure
2.5. The additional symbols in the control diagram are explained as follows:
VB = Voltage across capacitor C,
VBqref= Reference voltage for q-axis component of VB,
VBdref = Reference voltage for q-axis component of VB,
IB= Current supplied by VSI to grid,
IDG = Current passing thought the inductor,
Vdc= Measured voltage of the dc bus
21 | P a g e
IBq
IBd
IB (abc) VBqref IBq
+
Iq=+1.5
VBq - +
Kp +Ki/s
+
Iqref
abc-to-dq + Iq=-1.5
VB(abc) ωC
Phase-to-ground
ωC
Kp+Ki /s + - Id=+1.5
Idref
VBd - + + Id=-1.5
VBdref IBd
ϴ
ωt
PLL
Iqref VBq
ϴ
Vqref
m=+1
+
Iq - +
Kp +Ki/s
+ m
+ m=sqrt
m=0
(Vdref^2+
IDG(abc) Vqref^2)
ωL 2/Vdc
abc-to-dq & SPWM
ωL
Φ= atan2
- Vdref
(Vqref/
Vdref) φ
Kp +Ki/s
Id - + +
+
Idref VBd
To inverter
Figure 2.5 Diagram of the control block for a voltage controlled VSI
It uses PI control to control for the inner and the outer loop. PI gains and other parameters
have been given in Table 2-9 and Table 2-10. The per unit calculation was done as per the data
specified in Table 2-5.The specification of various components used with the voltage controlled
inverter are mentioned in Table 2-8.
The maximum and minimum level of reference currents, generated in dq-frame by the
outer voltage control loop, has been limited to 1.5 and -1.5 times of the rated current capacity
respectively as shown in Figure 2.5. Thus the maximum current magnitude for this VSI is limited
to approximately 2.1 times of the rated current.
22 | P a g e
Table 2-8 Parameters for voltage controlled VSI
Parameters for Voltage Controlled VSI
Battery Voltage 900V
Output Filter inductance 0.3mH
Filter Capacitance .08mF
Inverter Rating 100KVA
Inverter ‘on’ resistance 0.6mOhm
Switching Frequency 8000Hz
Table 2-9 Parameters for inner current control loop

Parameters for Inner Current Control Loop
Kp 1.5
Ki 3
Table 2-10 Parameters of outer current control loop

Parameters for Outer Voltage Control Loop
VBqref 0pu
VBdref 1pu
Kp 0.16
Ki 0.36
23 | P a g e
Chapter 3 Fault Current
Characteristics of IIDGs
3.1 Simulation of Fault Current Contribution
The fault current contributed by a DG employing grid-frequency imposed VSI current
control or controlled frequency VSI has been shown in this chapter. This chapter explains how
the IIDGs current contribution characteristics are responsible for different fault current levels in
the grid-connected and the islanded mode. All the locations used for fault simulation in the
microgrid has been shown in Figure 2.1.Balanced and unbalanced faults were simulated at F4
with fault resistance 0.01 Ohm for all cases shown in this chapter. The fault duration was for 20
cycles and it starts at 4.5 seconds after the simulation starts.
3.2 Fault current due to Grid-frequency Imposed VSIs

The simulation of fault current contribution by grid–frequency imposed VSIs were done
in grid-connected mode. The DG connected at bus number 3 was chosen for the representing
the VSI response due to fault. The voltage and the current measurement were taken at the
terminal of transformer ‘T5’ which is connected to bus 3. The current contributed by these VSIs
is observed for various types of faults.
Figure 3.1 shows the voltage and current, for a single line-to-ground fault at F4, at the bus
where DG-B1 has been connected. The current output by the current controlled inverter is 1.2
times of the pre-fault current.
Figure 3.2 shows the voltage and current for a line-to-line-ground fault involving phase-a
and phase-b at the bus where DG-B1 has been connected. The current output of the current
controlled inverter is around 1.2 times of rated current
Figure 3.3 shows the voltage and current for a three phase-to-ground fault involving
phase-a and phase-b at the bus where DG-B1 has been connected. The maximum current output
of the current controlled inverter is around 1.2 times of rated current.
24 | P a g e
Voltage Per Ph in pu
Phase a Phase b Phase c
1
-1
4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5

Time (s)
(a)
2
Line Current in pu
-1
-2 Phase a Phase b Phase c

4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Time (s)
(b)
Figure 3.1 (a) Voltage (b) Current from current controlled inverter in pu for AG fault
2

1
-1
-2
4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Time (s)
(a)
2
Line Current in pu
-1
-2
4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Time (s)
(b)
Figure 3.2 (a) Voltage (b) Current in pu from current controlled inverter for ABG fault
25 | P a g e
2
1
-1
-2
4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Time (s)
(a)
2
Line Current in pu
-1
-2
4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Time (s)
(b)
Figure 3.3 (a) Voltage (b) Current in pu from current controlled inverter for ABCG fault
3.3 Fault Current due to Voltage Controlled VSIs

The simulation of the fault current contribution by voltage controlled VSIs were done in
islanded mode. The DG connected at bus 5 represents a frequency controlled VSI in islanded
mode. The voltage and the current were measured at the terminal of the transformer ‘T5’ which
is connected to the bus 5. The maximum current contributed these VSIs for this study will be
around 2.1 times of the rated current.
The voltage and current measured for voltage controlled VSI DG-A1, for a single line-to-
ground fault at F4, is shown in the Figure 3.4. The current contributed by the DG is around 2.1
times of the rated current. Figure 3.5 and Figure 3.6 show voltage and current for DG-A1 for line-
to-line fault and three phase-to-ground faults respectively.
26 | P a g e
4
2
-2
-4
4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Time (s)
(a)
4
Line Current in pu
-2
-4
4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Time (s)
(b)
Figure 3.4 (a) Voltage (b) Current in pu from voltage controlled VSI for AG fault
4

2
-2
-4
4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Time (s)
(a)
4 Phase a Phase b Phase c
Line Current in pu
-2
-4
4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Time (s)
(b)
Figure 3.5 (a) Voltage (b) Current in pu from voltage controlled VSI for ABG fault
27 | P a g e
4
2
-2
-4
4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Time (s)
(a)
4
Line Current in pu
-2
-4
4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Time (s)
(b)
Figure 3.6 (a) Voltage (b) Current in pu from voltage controlled VSI for ABCG fault
This chapter shows the current contribution by current controlled and voltage inverters
during different types of faults. The maximum current contribution by the IIDGs during the fault
condition depends on the type of control as explained in Chapter 2. The fault current in the
islanded mode for this microgrid is the summation of the currents supplied by IIDGs only.
Therefore, the fault current magnitude in the islanded mode is lower than the fault current
magnitude in the grid-connected mode.
28 | P a g e
Chapter 4 Proposed Protection
Scheme
4.1 Proposed Protection Scheme Using an Intelligent Relay
Decision Fault
Tree
Wavelet
For
Transform
Current Fault No
Signal Detection Fault
Microgrid I (abc)
Fault LG
Classification LL
Sequence Wavelet
Decision
Analyzer Transform LLG
Tree
LLLG
Figure 4.1 Proposed protection scheme for CERTS microgrid
A new protection scheme has been proposed for the CERTS microgrid as shown in Figure
4.1 . The objective of this protection scheme is to provide primary protection using intelligent
relays. The intelligent relay uses local measurements to make decision regarding fault detection
and classification. The sequence analyzer produces sequence components by analyzing input
current. This scheme includes signal pre-processing using wavelet transform. The line current
signal was measured for 3904 conditions i.e. different faulted and healthy conditions within the
microgrid. The current signal is the input signal to the intelligent relay. This current signal is
processed by the sequence analyzer to produce negative and zero sequence components.
Wavelet coefficients for the measured current signal and for the sequence components were
extracted. Features using the wavelet coefficients were calculated. These features will be used
29 | P a g e
by the decision tree for detection and classification of faults. This scheme uses 15 different
features based on wavelet coefficients. These features are the inputs for data mining models
used for fault detection and classification. Nine out of fifteen features were used to train the
decision tree for fault detection and the decision tree used for fault classification was trained
using uses all the features. The output of the intelligent relay will identify and classify different
types of faults.
4.2 Overview of Wavelet Transform

The Wavelet transform helps to analyze a signal in time-frequency domain and gives time
localized frequency information [31]-[32]. The Wavelet transform is a powerful tool for the
analysis of non-stationary signals as the frequency content of a non-stationary (transient) signal
changes with time. The Wavelet transform is a better tool than the Fourier transform as the latter
is used for analyzing a signal in frequency domain only. The Fourier transform does not give any
information related to time-frequency domain. The short time Fourier transform (STFT) has
limitations in resolution due to window size. The Wavelet transform is a better tool as compared
to the Fourier transform or STFT, providing both time and frequency information which is
required for the analysis non-stationary fault transient signals.
4.2.1 Continuous Wavelet Transform

The continuous wavelet transform (CWT) can be defined as follows:

1 t b
X a ,b (t ) 
a
 x(t ) 

dt
a 
(4.1)
Here, x(t ) is the signal and  (t ) is the mother wavelet. The translated and dilated version
1 t b
of the wavelet is given by  a ,b (t )    where, a, b represent the dilated and
a  a 
translated parameters respectively. The translation and dilation are responsible for time and
frequency resolution respectively.
30 | P a g e
4.2.2 Discrete Wavelet Transform
The practical application of wavelet transform is performed using the discrete wavelet
transform. Any time series signal x(t ) can be completely decomposed into approximations by
using a scaling function  j (t ) which is also known as father wavelet. The signal can be
decomposed into wavelet coefficients using mother wavelet j (t ) as shown in Figure 4.2 [33].
Aj+1 D1
Low Pass Filter 2
x
D2
Aj
Dj+1 A1 D3
High Pass Filter 2
A2
A3
(a) (b)
Figure 4.2 (a) Single decomposition step (b) Decomposition upto level 3 of a signal ‘x’
 jk (t )  2  j / 2  (2  j t  n) (4.2)
 jk (t )  2  j / 2 (2  j t  n) (4.3)
where n  Z , j and k are integers
j
As j and k are integers and the basis function is scaled by a factor 2 and translated by n
units of time. The scaling function is associated with the low pass filters with filter coefficients H=
{ hn } and the wavelet function is associated with the high pass filter with filter coefficient G= {
gn }. It is important to note that G is the reverse of H interleaved with sign changes. The two
functions give rise to filters:
 (t )   h(n) 2 (2t  n) (4.4)

n
 (t )   g (n) 2 (2t  n) (4.5)

n
So for a signal x(t ) having length 2 M , then there are maximum of ‘M’ levels of
decomposition. The expression for the signal using the wavelet transform is given as follows:
31 | P a g e
2 M  j 1 2 M  j 1
x(t )  
k 0
a jk  jk (t )  
j
d
k 0
jk  jk (t ) (4.6)
where, a jk and d jk are known as the approximation and detail coefficients of the signal
at level j respectively. The signal is passed through high and low-pass filters. Then the outputs of
both the filters are down-sampled by a factor of 2 and the approximation and detailed
coefficients for level-1 A1 and D1 are obtained. This process is repeated for the approximation
coefficient till the desired level of detailed coefficient is retrieved
4.2.3 Energy of a Signal

Using the wavelet coefficient, the formula for energy content of a signal can be written
as follows [34]-[35]:
n 2
E i   d ij (4.7)
j 1
where, i =1, 2,…,I stands for the scale and j=1,2,…,n denotes the number of points used
for each coefficient. So, change in energy using the wavelet coefficient shows the change in
energy of a signal due to some disturbance. Change in energy is calculated by subtracting the
energy between two successive cycles.
Change in Energy  EB  EA (4.8)
where, EB =Energy of latest cycle and EA=Energy of last cycle.
4.2.4 Shannon Entropy

The Shannon entropy is a tool for analyzing and comparing probability distribution of a
data series. It is a tool to measure the disorder or imbalance or uncertainty in a system. The
disorder is associated with system represented by states and probability. The sampled values of
the line currents under fault conditions are independent at different time instances, given the
randomness inherent in the behavior of the system under fault conditions. The formula for
calculating Shannon’s entropy is given below [36].
32 | P a g e
n
E i   p j log( p j ) (4.9)
j 1
where, p j is the energy probability distribution of the wavelet coefficients d ij for scale i.
2
n d ij
p
j 1
j  1 and p j log p j  0 if p j  0 ; p j 
Ei
[37, 38], E i is the energy as per equation 4.7.
Parameter p j is calculated based on the wavelet coefficients used in the wavelet
transformation presented in Section 4.2.2. It is therefore dependent on the type of wavelet used
and the frequency band used.
4.2.5 Standard deviation

Standard deviation is a statistical measure of a disturbance present in a signal. It
measures the dispersion of the data from the average. The standard deviation for a series of data
points is calculated as follows:
N
1

N
 (x
i 1
i  )2 , (4.10)
N
1
where,  
N
x
i 1
i and xi represents the ith data element of a data series and i=1,…,N;
σ, μ represents standard deviation and mean of the data series.
4.3 Signal Pre-Processing using Wavelet Transform

The line current signal is sampled at 6.66 KHz. The measurement was taken using per unit
values. The base values used for per unit measurement were 480V, 100KVA. Line current data is
processed through ‘db-1’ or Haar wavelet and the wavelet coefficients are generated. Further,
different features like “Change in Energy”, “Shannon Entropy”, and “Standard deviation” of the
level “d3” coefficients are derived which are used as input for data-mining model. This pre-
processing by the wavelet transform was performed using MATLAB. Figure 4.3 shows the
frequency bands related to the wavelet coefficients.
33 | P a g e
d1(1.66-to- 3.33)KHz
(0 -to- 1.66)KHz
a1
d2 (0.833-to-1.66)KHz
7th,11th,13th
Harmonics
(0 -to- 833)Hz
a2
d3(416-to-833)Hz
(0-to-416 )Hz
a3
Figure 4.3 Showing frequency bands related to each coefficient of the wavelet transform
4.3.1 Haar Wavelet

There are a number of implementations of the wavelet transform, depending upon the
function used to process the data. The Haar wavelet is frequently used for power signal analysis,
for example for processing voltage and current signals [39, 40] as it can effectively retrieve time-
frequency information from non-stationary signals. It is therefore an appropriate choice for
extracting useful information from transients, such as those that occur in a faulted system. Figure
4.4 shows the basic function used in the wavelet transform. Six such functions were used to
generate the coefficients.
The choice of the wavelet function has an impact on the on the accuracy of detection and
classification. However, the choice of the window over which the data is analyzed by the
waveform transformation is more important than the actual function for low frequency signals,
voltages and currents, such as are found in power systems.
 (t )
+1
1
0 t
-1
Figure 4.4 Haar Wavelet
34 | P a g e
4.3.2 Selection of the Wavelet Coefficient ‘d3’
There were 6 detailed coefficients (d1 to d6) on obtained after analyzing the input signal
using the wavelet transform. The information related to frequency range for all the detailed
coefficients obtained using the wavelet transform has been mentioned in Table 4-1.
Table 4-1 Detailed coefficient frequency range details

Wavelet Coefficient Frequency Range in kHz
d1 1.66-3.33
d2 0.833-1.66
d3 0.416-0.833
d4 0.208-0.416
d5 0.104-0.208
d6 0.052-0.0104
In the Figure 4.5 , a comparison of wavelet coefficients of the current signal obtained for
a “SLG fault with 2 ohm fault resistance” in grid connected mode and another current signal for
a “no fault” condition within the microgrid. The figure shows all the coefficients obtained by
processing the two current signals. We can see that the coefficient “d3” is a superior indicator of
the fault as compared to other wavelet coefficients. Appendix A lists many other cases of
comparison for ‘d3’.
35 | P a g e
Figure 4.5 Wavelet detail coefficients for fault and no fault signal
4.3.3 Features using Wavelet Coefficient “d3”

Fifteen different features were calculated using the wavelet coefficient “d3”. The wavelet
coefficients of line currents and sequence components were used in the features calculation
process to calculate the features listed below. All the features are independent of each other,
because they provide information based on a different concept, see Section 4.2. Although these
features are normally considered off-line features and are calculated off-line, there are special
36 | P a g e
algorithmic implementations of the equations defining these features that can yield results in
real time. The description of each feature is as follows:
F1 – Change in Energy of Phase A (Post Fault - Pre fault for one cycle)
F2 – Change in Energy of Phase B (Post Fault - Pre fault for one cycle)
F3 – Change in Energy of Phase C (Post Fault - Pre fault for one cycle)
F4 – Entropy of Phase A one cycle post fault
F5 – Entropy of Phase B one cycle post fault
F6 – Entropy of Phase C one cycle post fault
F7 – Standard deviation of Phase A after the fault occurs for one cycle
F8 – Standard deviation of Phase B after the fault occurs for one cycle
F9 – Standard deviation of Phase C after the fault occurs for one cycle
F10 – Change in Energy of Negative Sequence Component
F11 – Entropy of Negative Sequence after the fault occurs for one cycle
F12 – Standard deviation of Negative Sequence after the fault occurs for one cycle
F13– Change in Energy of Zero Sequence Components
F14 – Entropy of Zero Sequence after the fault occurs for one cycle
F15 – Standard deviation of Zero Sequence after the fault occurs for one cycle
4.4 Sequence Analyzer

The sequence components are also known as symmetrical components. They are used to
analyze unbalanced three-phase power systems. They are also used for analyzing different types
of faults in the power system. The Sequence Analyzer analyzes the there-phase signal and
produces positive, negative and zero sequence components of a three-phase signal. Let I a , I b
and I c represent the current flowing through phase a, b and c respectively. The positive
sequence component, negative sequence component and the zero sequence component can be
37 | P a g e
denoted by I 1 , I 2 and I 0 .respectively. The expressions for sequence components are given
below:
1
I1  (I a    I b   2  I c ) (4.11)
3
1
I 2  (I a   2  I b    I c ) (4.12)
3
1
I 0  (I a  I b  I c ) (4.13)
3
where,   1120 is the complex operator.

4.5 Data Mining Techniques Used for the Protection Scheme

There is a need to understand system behavior of complex systems in order to solve the
problems associated with it. Data mining models aim to create a comprehensive structure by
analyzing a dataset which can be further used for predicting or understanding the system
behavior. Data mining is a method to model the relationship or dependencies between system
variables of a dataset containing data pertaining to the system. The data is gathered under
different circumstances or conditions within the system. The analysis of the system is performed
by combining human and machine intelligence or human intelligence alone. It has become a
widely used tool to improve decision making process for business organizations, predictive and
analysis tool for scientists analyzing complex data e.g. analyzing atmospheric changes, helping to
identify trends and gathering intelligence for governments .Data mining is a useful to search for
new valuable information from a large dataset with the help of computers. The models created
by data mining are used for prediction and description of system outputs [41]. Data mining
methods has its roots in artificial intelligence, machine learning, and statistics. Predictive data
mining tries to produce a system model as per the input dataset and this model can be used for
classification, estimation, prediction etc. Descriptive data mining aims to discover patterns and
relationships in large data sets.
The data mining method used for the designing the intelligent relay falls into predictive
category i.e. classification. The objective of the intelligent relay proposed in this document is to
detect and classify faults. So, the data mining method would be used to build a classifier which
will predict the output value for a set of input features. To build a classifier model, a dataset
38 | P a g e
consisting of system data is assigned to different type of groups as defined by the user and fed
to a data mining algorithm to produce a classifier model. Each group of input data is known as a
class. There are different methods for data mining such as decision tree, random forest, support
vector machine and neural networks etc. The proposed study uses two effective data-mining
models. The transparent data-mining model is decision tree model and the black-box model is
random forest model. However, there is a trade-off between transparency and accuracy between
these two data-mining models which are discussed in the results and analysis section. Next few
paragraphs discuss regarding two data mining methods used for classification.
4.5.1 Decision Trees
The Decision Tree (DT) is a highly popular and efficient way to build classifiers. The
decision tree algorithms are based upon supervised learning algorithms. It resembles an inverted
tree structure and works on a series of conditional (if-then) logic. A decision tree consists of
‘decision node’ where the input attributes are evaluated. The first node is referred as ‘root node’.
The ‘branches’ originating from the nodes correspond to the outcomes of the ‘decision node’.
The branches may lead to a class which is known as ‘leaf’ node or to another ‘decision node’
which is the root for another sub-tree within the DT. So, every ‘non-leaf’ or a ‘decision node’
has two branches and a ‘leaf’ node has no branch. Any node in the DT except ‘leaf’ node and
‘root node’ is known as ‘internal node’ [42]. DT is also known as binary tree if every ‘non-leaf’
node has two branches. A path starting from `root node` and ending at a `leaf` node represents
the set of evaluations done to identifying a class. These set of evaluations can be grouped to
create a classification rule. DTs can be converted to a set of classification rules [33] .DTs are easy
to implement and transparent in nature as it shows the possible alternatives for any ‘decision
node’.
An example of a decision tree has been shown in Figure 4.6. This decision tree is used to
classify a system into 3 classes depending on the attributes of the system. There are two input
attributes A, B. There are three output classes (class1, class 2 and class 3). The ‘root node’
evaluates the attribute A. If the attribute A is greater than zero the branch leads to a ‘leaf’ node
and the system is classified as class 1, else the branch from the root node proceeds to another
39 | P a g e
decision node to evaluate attribute ‘B’. When B is equal to 5 the branch leads to the ‘leaf’ node
class 2, otherwise it leads to class ‘3’.
A>0
s No
Ye
Class 1 B=5
s No
Ye
Class 2 Class 3
Figure 4.6 Decision Tree for classification
So the decision rules from this decision tree can be listed as follows.
 When A is greater than zero the output is ‘Class 1’.
 When A is less than or equal to zero and B is equal to five then the output is ‘Class
2’.
 When A is less than zero or equal to zero and B is not equal to five the output is
‘Class 3’.
4.5.2 Random Forest

Random forest (RF) is an ensemble learning method to increase the classification accuracy
[43]. It is created by combining many classifiers. It uses a collection of trees acting on same set
of input attributes. The DTs used for the random forest method are independent of each other.
The individual classifiers vote for a class and the most popular class is selected. The output of a
single tree might be noisy. Therefore, RF method counts on an average decision using more
number of DTs.
4.6 Methodology for Fault Data Generation

In the grid-connected mode, the DG penetration was set to three different values 100%,
40%, and 10% of the total DG capacity. The load was equally distributed among the 3 generators.
In the grid-connected mode, when the generation capacity of the DG within the microgrid
40 | P a g e
decreases, then additional power flows from the grid. For the islanded mode, the 3 DG operate
at a 100% of capacity. The DG penetration remains at 100 %, in islanded mode, in order to have
maximum grid current. Load shedding is needed in islanded mode if the generation capacity is
less than the demand of the microgrid. The faults were simulated at the middle of the distribution
line between two buses. The measurements were taken at a bus closest to the fault. Fault data
was generated by simulating different types of conditions such as different types of fault
resistance, fault location, changing fault inception angle, various levels of DG penetration within
different topologies of microgrid. Table 4-2 presents a summary of the conditions used for
simulating faulted conditions within the microgrid. As per this table there are a total of 2880
cases for grid connected mode, 960 for islanded mode for a total of 3840 different cases
corresponding to different types of fault conditions.
Table 4-2 Number of cases simulated – different types of faults in the network
Mode of Topology Type Types of No. of No. of DG Total no. of
Operation of Fault Fault Fault Penetration Cases
Fault Resistance Location Inception Level
Angle
Grid 2 10 3 4 4 3 2880
Connected
Islanded 2 10 3 4 4 1 960
The values used for simulating different types of faults are as follows:
 Different balanced and unbalanced faults were simulated. They can be referred as AG,
BG, CG, AB, BC, CA, ABG, BCG, CAG, and ABCG (10 types).
 Different types of fault resistance 0.01 Ω, 0.5 Ω and 2 Ω were used for all the faults
simulated (3 cases).
 Faults were simulated in the network at location F1, F2, F3 and F4 as shown in Figure 2.1.
These faults are simulated at the mid-point on the distribution line between two adjacent
buses. (4 cases)
 Three different DG penetration levels were used for the simulation were 100%, 40%, 10%
for simulating every fault with microgrid operating in grid-connected mode. It applies to
IIDGs operating as a grid-frequency imposed current control mode only. It does not apply
41 | P a g e
to frequency controlled VSI generation capacity. In islanded mode, the IIDGs operating
with grid-frequency imposed current control mode operate at 100% of its capacity.
 Different fault inception angles[44] used in the simulation such as: 0 degree, 45 degree,
90degree, 180 degree (4 cases).
 The microgrid was operated in loop and radial configurations.
4.7 Methodology for No Fault Data Generation

The line current measurements for no fault conditions were made under heavy load and
light loading conditions within the microgrid. Three loads within the microgrid were changed for
adjusting the total microgrid load. Those three loads were load bank 3, load bank 4 and load bank
5. The measurements were taken at each bus. During the heavy load condition within the
microgrid, the microgrid load is: 2 x [90kw, 45Kvar (inductive load)] + [90 Kw, 40Kvar (capacitive
load)]. The description for lighting loading condition is: 2 x [20kw, 5Kvar (inductive load)] + [20
Kw, 5 Kvar (capacitive load)]. The DG penetration was also varied as per the description given in
the last Section 4.6. The microgrid was operated with both radial and loop topologies for these
simulations. Table 4-3 presents a summary of all the 64 conditions used for simulating no-fault
conditions within the microgrid.
Table 4-3 Parameters for simulating different condition for no fault cases in microgrid
Mode of Topology Types of DG Relay Total no.
Operation Load penetration Points of
Levels cases
Grid Connected 2 2 3 4 48
Islanded 2 2 1 4 16
4.8 Generating Dataset Containing Features

There are total 3904 measurements of current signals were taken at various buses as per
Table 4-2 and Table 4-3. Then the sequence components were generated from these signals.
Discrete samples of current signals of each phase and the sequence components for one cycle
duration were pre-processed using wavelet transform. The detail coefficient ‘d3’ obtained from
wavelet transform for each phase and for the sequence components corresponding to the
42 | P a g e
current signal is used for generating features. These features are change in energy, standard
deviation and Shannon’s entropy as explained in section 4.2.Each phase of the three-phase
current has three features. So, the line current produces nine features. Negative and zero
sequence components produce another six features. Fifteen features are studied for building the
intelligent relay. These features are stored in one row within a comma separated file i.e. ‘.csv’.
Each row is assigned a class depending on the classification logic. This class represented by one
extra column in the feature dataset.
4.9 Building Decision Tree Model for Intelligent Relay

The line current signals were collected and processed using the wavelet transform.
Fifteen different features which were generated using the detail coefficient from the wavelet
transform will be used by data mining algorithms to build decision tree. Table 4-4 lists all the
features used for fault detection and classification. ‘R’ programming software was used to build
the decision tree for the intelligent relay [45]-[46]. The intelligent relay needs one decision tree
model for fault detection and another one for fault classification.
Table 4-4 Features used for detection and classification of faults

Model Objective No of Input Features Name of the Input Features
Fault Detection 9 F1,F2,F3,F4,F5,F6,F7,F8,F9
F1,F2,F3,F4,F5,F6,F7,F8,F9
Fault Classification 15
F10,F11,F12,F13,F14,F15
For building a decision tree model to detect fault, nine features related to line currents
were given as inputs along with the class to which they belong. It contains features extracted for
faulted as well as no-fault conditions. All the features related to related to fault conditions
were classified as ‘1’ and data generated under no fault condition was classified as class ‘2’. Table
4-5 presents a summary of the classes used for detection. The input dataset for building the fault
detection DT contains 3904 cases (refer Table 4-2 and Table 4-3) and 16 columns which include
15 columns for input features and one column for output class.
43 | P a g e
Table 4-5 Different classes used for detection
Situation Class
Fault 1
No-Fault 2
For training a decision tree model to classify a fault, fifteen features related to the line
current were given as inputs along with the class of fault. Ten types of faults mentioned for this
study were classified into four categories. All single line-to-ground(LG) faults were bundled into
one class ‘1’ , the line-to-line(LL) faults are classified into class ‘2’ , the line-to-line-to-ground(LLG)
faults were classified as class ‘3’ and the last category representing three-phase-to-ground(LLLG)
faults were classified as class ‘4’. So the dataset contains 3840 rows for fault measurement and
16 columns which include 15 columns for features and one column for class. Table 4-6 shows the
classes used in the classification process. Appendix B shows a sample dataset used for this study.
Table 4-6 Different classes used for fault classification

Fault-Type Class
LG 1
LL 2
LLG 3
LLLG 4
4.10 Results and Analysis

4.10.1 Results for Fault Detection
The decision tree (DT) for fault detection was generated using the open source data
mining package ‘R’ [46]. The nine input features (F1 to F9) and two output classes were included.
The dataset was partitioned into two parts. The partitioning was done randomly by the
software package ‘R’. The first part of the dataset consists of 70% of the data and it was used for
training the DT. The second part consists of the remaining 30% of the data and it was used for
testing the operation of the DT model. The two datasets are therefore independent.
The datasets used for building the DT for fault detection and classification were also used
for building the random forest model.
44 | P a g e
Table 4-7 presents the error matrix for decision tree model and random forest model.
Figure 4.7 shows the decision tree model obtained for the fault detection.
 Error Matrix- An error matrix or confusion matrix can be defined as a matrix that
summarizes the classification performance of a classifier with respect to some test
data. It is a two dimensional matrix. One dimension of the matrix represents actual
class of an object and in other dimension the class predicted by the classifier.
The performance summary given in Table 4-8 compares the performance of decision tree
and random forest method. The definition of the terms used in Table 4-8 are as follows [47]:
 Misdetection - Those faulted conditions which were reported as no-fault condition
by the relay.
 False Alarm - Those no-fault conditions which were reported as fault by the relay.
 Accuracy - Accuracy of a relay is calculated by subtracting misdetection and false
alarms from all the conditions.
As per Table 4-7(a), the results obtained by the DT used for fault detection. The actual
class `1` which represent fault condition has been successfully predicted by the DT for 96% of the
test data. The DT has classified 1% of the faulted data as no-fault condition which is a
misdetection. The DT has classified 2% of the no-fault data as fault i.e. false alarm.
Table 4-7 (a) Error Matrix (%) for fault (b) Error Matrix (%) for fault detection using
detection using Decision Tree Random Forest
Predicted Predicted
Class Class
Actual Actual
1 2 1 2
Class Class
1 96 1 1 97 0
2 2 1 2 1 2
Table 4-8 Fault detection summary for DT and RF

Data Mining Accuracy in % Misdetection in % False Alarm in %
Method
Decision Tree 97 1 2
Random Forest 99 0 1
45 | P a g e
Similarly as per Table 4-7 (b), the random forest method has predicted 97% of fault test
data correctly and there is no misdetection using the RF model. RF model has classified 1% of
data as false alarm. So, the DT model has 97% accuracy and RF model has 99% accuracy as per
the fault detection summary Table 4-8.
4.10.2 Results for Fault Classification

The decision tree generated for fault classification using the data mining package ‘R’
resulted in a tree containing five features out of fifteen input features and four output classes.
The features selected by the DT are related to sequence components only. Those features were
F10, F11, F12, F13 and F14. Figure 4.8 represents the fault classification DT.
Table 4-10 represents fault classification summary after analyzing the data shown in the
error matrix given in Table 4-9 (a) and (b). As per Table 4-9, 12 % of LLG faults are improperly
classified as LG faults and 3 % of LG faults are improperly classified as LLG faults. In case of the
random forest (RF) method, 4 % LLG faults are improperly detected as LG faults and 2 % of LG
faults are being classified as LLG faults. It is observed that DT and RF models are able to classify
type of fault with 85 % and 94 % accuracy respectively. The random forest method has superior
performance in classification as compared to DT method.
During the fault classification, the problem arises when LG faults are classified as LLG
and vice versa. As per Table 4-10, the fault classification accuracy for RF based relay is 9 % higher
than the DT based relay.
Table 4-9 (a) Error (%) matrix for fault (b) Error (%) matrix for fault classification
classification using DT using RF
Predicted Predicted
Class Class
Actual Actual
1 2 3 4 1 2 3 4
Class Class
1 28 0 3 0 1 28 0 2 0
2 0 29 0 0 2 0 29 0 0
3 12 0 18 0 3 4 0 26 0
4 0 0 0 10 4 0 0 0 11
46 | P a g e
Table 4-10 Classification result summary using data mining methods
Data Mining
Accuracy in % Misdetection in %
Method
Decision Tree 85 15
Random Forest 94 6
4.11 Comparing Intelligent Relay with Overcurrent Relay

The performance of intelligent relays was compared with conventional overcurrent
relays. These relays are the more common means of protecting lines and devices. The relay is an
instantaneous over current relay. The threshold setting for the overcurrent relay for the lines
within the microgrid was set at 1.5 times of the maximum load current value. They were used to
detecting faults in the test microgrid for all the fault conditions described in Table 4-2.
In more sophisticated conventional protection designs, other relays are used in
conjunction with overcurrent relays, including inverse time overcurrent relays and differential
relays. However these were not considered in this work, because they are very much dependent
upon the configuration and layout of the distribution system and microgrid.
As per Table 4-11, the fault detection performance of the overcurrent relay is poor as it
fails to detect 44 % of the fault occurrences. This is due to the fact that the fault contribution of
inverter based DGs is limited and the fault impedances limit the fault currents. The fault detection
performance of the decision tree (DT) and random forest based intelligent relays is better than
of the conventional overcurrent relays as shown in the Table 4-11. The misdetection reduces to
1 % for the DT based relay and 0 % percent for the random forest based relay. False detection in
the case of DT is 2 % compared to 1 % for the random forest approach.
Table 4-11 Performance comparison for various relays tested in the microgrid
Data Mining
Accuracy % Misdetection % False Alarm %
Method
Decision Tree 97 1 2
Random Forest 99 0 1
Conventional
Method
Overcurrent Relay 56 44 Not Applicable
47 | P a g e
F2 >= 0.0043
s No
Ye
Fault F3 >=0.0091
s No
Ye
Fault F1 >= 0.0044
s No
Ye
Fault F4 < 0.057
Yes No
F1 >= 0.0019 F6 >= 0.021
s s
Ye N Ye
N
o
o
Fault F8 >= 0.0031 Fault No Fault
N
s
o
Ye
F4 < 0.018 No Fault
Yes No
F1 >= 750e-6 F3 >= 0.0025
s No s No
Ye Ye
Fault F8 >=0.0046 F9 < 0.013 F5 >= 0.041
s s
No Ye
N
s Ye
N
o
Ye
o
Fault No Fault Fault F8 < 0.0081

Fault F2 <-48e-6
s
s
Ye
Ye
N
o
o
F3 < 0.002 No Fault

Fault No Fault
N
s
o
Ye
F7 >= 0.0069 No Fault

s
N
Ye
F9 >=0.0058 No Fault
s
Ye
N
o
Fault No Fault
Figure 4.7 Decision Tree for Fault Detection
48 | P a g e
F13 >= 413e-9
No
Yes
F10 < 68e-6 F13 >=847e-21
No
s
Ye
s No
Ye
Class 2 Class 4
Class 1 F14 >= 806e-6
No
s
Ye
F12 < 0.0082 Class 3
s
No
Ye
F12 < 0.0023 Class 3

s
Ye
No
Class 1 F14 >=0.0046

s
No
Ye
F13 < 0.01 Class 3

s
No
Ye
F11>=0.0051 Class 3
s
No
Ye
Class 3 Class 3
Figure 4.8 Decision Tree for Fault Classification.
49 | P a g e
Chapter 5 Conclusion
5.1 Summary
In this thesis, a microgrid protection scheme has been proposed using intelligent relays
based on data mining models. This scheme addresses the possible problems of fault detection
arising due to the wide variation in fault current magnitude in grid connected and islanded mode
of a microgrid having a high penetration of inverter interfaced distributed generation (IIDG). The
proposed approach includes a comparative study of two extensively used data mining models
such as the decision tree (DT) and random forest (RF). Furthermore, a comparison between
traditional instantaneous overcurrent relays and the intelligent relays are presented, showing
the improvement in reliability and accuracy of the proposed scheme.
5.2 Conclusions
It is observed that the instantaneous over current relay cannot detect faults with a high
accuracy in a microgrid with IIDGs. The accuracy problem has been addressed by using intelligent
relays which are modelled by two types of data mining methods such as DT and RF. The accuracy
of fault detection improves to 97% for DT based relay and 99% for the RF based relay as compared
to 56% accuracy of the instantaneous over current relay. The improvement in accuracy of the
intelligent relay is due to the use of time-frequency information of the current signal used to
build the data-mining models.
Fault classification task can be dealt effectively by the proposed data mining models. It is
observed that the fault classification accuracy of the RF based relay is 9% higher than the DT
based relay. Similar observations are made for fault detection also where RF outperforms DT.
However, RF being a black box data-mining technique may face challenges for real
implementation on DSP or FPGA boards while DT being transparent, is easier to implement.
Though the proposed scheme has been extensively tested on simulation platform (MATLAB),
however, the response time response time within two cycles from the fault inception time. This
two cycle response time includes 1.7 cycle for signal processing and another 0.1 cycle for decision
making by the DT. This shows the potential of the proposed scheme for online application. Thus,
50 | P a g e
DT based data mining relays are recommended for building commercial relays for micro-grid
protection.
5.3 Suggestions for Future Work

The proposed protection scheme concentrates upon primary protection or main
protection of the microgrid only. In order to improve the reliability of the protection scheme, a
comprehensive protection scheme including a backup protection needs to be designed. The
coordination between primary and secondary protection scheme also needs to be studied for the
accurate operation of the protection devices and to minimize the fault effect on the loads in the
microgrid. This proposed primary protection scheme can be verified using real time simulators.
51 | P a g e
Appendix A
(a) Case 1: Fault with 0.01 Ohm vs No fault condition in grid Connected Mode.
current in pu
20
FaultSLG
10 nofault
0
PhaseinApu 20
FaultSLG
-10
A A current
10 nofault
0 50 100 150 200 250
0 Sample Number
1
Phase
-10
Figure A1.
0
Current50for Phase100
A -AG Fault
150
with 0.01 Ohm vs 250
200
No fault
D3-Phase D3-Phase
0 Sample Number
1
A
-1
Neg Seq Current in pu
00
20 50 100 150 200 250
Sample Number FaultSLG
10 nofault
-10
0 50 100 150 200 250
Sample Number
-10
Figure A2. D3
0 Coefficient
50 for Phase
100 A -AG 150
Fault with200
0.01 Ohm250
vs No fault
Sample Number
Seq
0.01
in pu
20
D3 - Neg
0 FaultSLG
Zero Seq current
10 nofault
-0.01
0
0 50 100 150 200 250
-10 Sample Number
Figure A3. D3 Coefficient
0 for50Negative100
Sequence150
–AG Fault200
with 0.01250
Ohm vs No fault
Sample Number
0.01
D3-ZeroSeq
-0.01
0 50 100 150 200 250

Sample Number
Figure A4. D3 Coefficient for Zero Sequence - AG Fault with 0.01 Ohm vs No fault
(b) Case 2: Comparison of fault current with 0.01 ohm and 2 ohm Fault resistance
in Grid connected mode.
20
Current Phase A in pu
0.01-OhmFaultSLG
10 Two-OhmSLG
-10
0 50 100 150 200 250

Sample Number
Figure A5.
1 Current for Phase A - AG Fault with 0.01 Ohm vs 2 Ohm
0.5
D3- PhaseA
52 | P a g e
0
-0.5
-1
0 50 100 150 200 250
0.01-OhmFaultSLG
Current Phase A i
10 Two-OhmSLG
-10
0 50 100 150 200 250

Sample Number
0.5
D3- PhaseA
0

20
-0.5 0.01-OhmFaultSLG
10 Two-OhmFaultSLG
-1
00 50 100 150 200 250
Sample Number
-10
Figure A6. D3 Coefficient for Phase A - AG Fault with 0.01 Ohm vs 2 Ohm
0 50 100 150 200 250
Sample Number
in pu
0.01
NegSeq
20
0.01-OhmFaultSLG
Current
100 Two-OhmFaultSLG
Zero SeqD3-
0
-0.01
-100 50 100 150 200 250
Sample Number
0 50 100 150 200 250
Figure A7. D3 Coefficient for NegativeSample
Sequence – AG Fault with 0.01 Ohm
Number vs 2 Ohm
0.01
D3- ZeroSeq
-0.01
0 50 100 150 200 250

Sample Number
Figure A8. D3 Coefficient for Zero Sequence- AG Fault with 0.01 Ohm vs 2 Ohm
(c) Case-3: Comparison of fault current between Loop and Radial topology (0.01
ohm Fault resistance for SLG fault in Grid connected mode).
in pu
20
0.01-Ohm-Radial
0.01-Ohm-Loop
PhaseA inCurrent
10
pu
20
0
0.01-Ohm-Radial
0.01-Ohm-Loop
PhaseA Current
10
-10
00 50 100 150 200 250

Sample Number
1
-10
D3 -PhaseA
0 50 100 150 200 250

Figure A9. Current for Phase A - Sample
AG Fault with 0.01 Ohm in Radial vs Loop
Number
0
1
D3 -PhaseA
-1
00 50 100 150 200 250
Sample Number
-1
0 50 100 150 200 250
Sample Number
Figure A10. D3 Coefficient for Phase A - AG Fault with 0.01 Ohm in Radial vs Loop
53 | P a g e
20
0.01-Ohm-Radial
Neg Seq Current i

10 0.01-Ohm-Loop
-10
0 50 100 150 200 250
Sample Number
0.01
current in pu
Zero Seq D3-NegSeq
20
0.01-Ohm-Radial
0
10 0.01-Ohm-Loop
0
-0.01
-100 50 100 150 200 250

Sample Number
0 50 100 150 200 250
Figure A11. D3 Coefficient for Negative Sequence – AG
Sample Number Fault with 0.01 Ohm in Radial vs Loop
0.01
D3-ZeroSeq
-0.01
0 50 100 150 200 250

Sample Number
Figure A12. D3 Coefficient for Zero Sequence- AG Fault with 0.01 Ohm in Radial vs Loop
(d) Case 4: Fault current comparison for SLG fault having 0.01 Ohm Resistance for
Both Grid Connected and islanded mode
in pu in pu
20
0.01-GridConFaultSLG
A Current
20
10 0.01-IslandFaultSLG
A Current
10
Phase
0
-10
0 50 100 150 200 250
APhase
Sample Number
-10
20
Figure A13. Current for50Phase A -100
AG Fault
Sample
150 200
- Grid-connected
Number vs 250
Island mode
2
D3-Phase
A
0
D3-Phase
0
20
-2
100 50 100
0.01-IslandFaultSLG 150 200 250
Sample Number
-2
0 50 100 150 200 250
0 Sample Number
Figure A14. D3-10Coefficient for Phase A - AG Fault - Grid-connected vs Island mode
0 50 100 150 200 250
Sample Number
0.02
D3-NegSeq
-0.02
0 50 100 150 200 250
Sample Number
Figure A15. D3 Coefficient for Negative Sequence – AG Fault - Grid-connected vs Island mode
54 | P a g e
20
Zeo Seq Current in

-10
0 50 100 150 200 250
Sample Number
0.01
D3-ZeroSeq
0
-0.01
0 50 100 150 200 250

Sample Number
Figure A16. D3 Coefficient for Zero Sequence- AG Fault - Grid-connected vs Island mode
(e) Case 5: Fault current comparison SLG fault in Grid Connected mode with 40%
vs. 10% DG penetration
in pu
20
DG40Percent
Current
DG10Percent
10
pu
20
DG40Percent
CurrentA in
0
DG10Percent
10
Phase APhase
-10
0 50 100 150 200 250
0 Sample Number
0.5
-10
0 50 100 150 200 250
D3-PhaseA
Figure A17. Current for Phase A - 0.01 Ohm

SampleAG Fault with 10% and 40% DG penetration
Number
0
0.5
D3-PhaseA
20
-0.5
0
0 50 100 150 200
DG40Percent250
Sample Number DG10Percent
10
-0.5
00 50 100 150 200 250
Sample Number
Figure A18. D3
-10Coefficient for Phase A - AG Fault- 10% and 40% DG penetration
0 50 100 150 200 250
Sample Number
0.01
D3-NegSeq
in pu
20
0
DG40Percent
Zero Seq Current
10
-0.01 DG10Percent
00 50 100 150 200 250

Sample Number
Figure A19. D3 Coefficient
-10 for Negative Sequence –AG Fault - 10% and 40% DG penetration
0 50 100 150 200 250
Sample Number
0.01
D3-ZeroSeq
-0.01
0 50 100 150 200 250

Sample Number
Figure A20. D3 Coefficient for Zero Sequence - AG Fault - 10% and 40% DG penetration
55 | P a g e
Appendix B
A sample file data used for training and testing the DT model is given below.
F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 Class
3.385209 0.016677 0.002994 3.519972 0.083593 0.0582 0.174058 0.014627 0.012773 0.002376 0.016012 0.004576 0.002478 0.016199 0.004675 1
0.003793 3.20219 0.012709 0.059616 3.420563 0.071645 0.012941 0.169332 0.013292 0.00234 0.015798 0.004542 0.002451 0.016049 0.004649 1
0.01181 -0.00082 3.17885 0.068528 0.049875 3.34933 0.013163 0.011589 0.168725 0.002355 0.015907 0.004556 0.002468 0.016144 0.004665 1
3.908081 3.993003 -0.0022 2.468175 2.894825 0.019277 0.187689 0.18892 0.010918 0.005729 0.035817 0.007081 3.06E-17 0 5.22E-10 1
0.00401 3.897733 4.012361 0.030604 2.502422 2.93613 0.012671 0.187459 0.189408 0.00574 0.035877 0.007088 2.31E-17 0 4.54E-10 1
4.041184 0.002085 3.952688 2.903681 0.039513 2.487453 0.190048 0.012507 0.188766 0.005712 0.035742 0.00707 5.32E-17 0 6.89E-10 1
4.902225 4.367352 0.076584 2.029103 3.323369 0.205518 0.209945 0.19762 0.028568 0.010347 0.053365 0.009575 0.007322 0.034512 0.00807 1
0.082204 4.942175 4.099395 0.216838 1.997797 3.261481 0.029383 0.210817 0.191466 0.010268 0.053012 0.009538 0.007432 0.034824 0.008131 1
4.164376 0.083096 4.901569 3.291395 0.22033 2.017905 0.192988 0.029646 0.209958 0.010256 0.053003 0.009532 0.007354 0.034612 0.008088 1
1.838192 7.378457 2.409168 2.087798 3.075322 1.720127 0.129349 0.256534 0.146366 0.066367 0.212574 0.024326 1.87E-19 0 4.08E-11 1
9.731887 0.121845 -0.01261 1.78816 0.40091 0.097128 0.294777 0.034099 0.016245 0.005243 0.03303 0.00677 0.005762 0.035023 0.007099 1
-0.00737 9.640833 0.112741 0.105079 1.803676 0.388602 0.017212 0.293333 0.033021 0.005237 0.032992 0.006767 0.005745 0.034943 0.00709 1
0.113123 -0.01701 9.768353 0.385642 0.093987 1.838593 0.033674 0.016051 0.295273 0.005232 0.032958 0.006763 0.005732 0.034876 0.007081 1
5.05597 5.227025 -0.00095 2.385398 3.568725 0.06532 0.212501 0.215908 0.018541 0.007206 0.043371 0.007931 2.73E-17 0 4.91E-10 1
0.000688 5.085877 5.279681 0.072531 2.395563 3.59494 0.018725 0.213137 0.216956 0.007254 0.043584 0.007958 9.96E-18 0 2.97E-10 1
5.134615 0.002727 4.946714 3.56988 0.08074 2.363084 0.213967 0.019545 0.210235 0.007193 0.043323 0.007924 1.05E-17 0 3.05E-10 1
4.59796 7.405974 -0.01533 2.376383 3.628634 0.046597 0.203 0.256164 0.015318 0.010471 0.055396 0.009606 0.00268 0.015466 0.004882 1
-0.02066 4.591369 7.434374 0.027132 2.3502 3.644085 0.013179 0.202763 0.256684 0.010568 0.055595 0.009651 0.00268 0.015452 0.004882 1
7.36481 -0.02359 4.606195 3.724607 0.0355 2.393233 0.255531 0.013951 0.203105 0.01044 0.055253 0.009591 0.002687 0.015515 0.004889 1
6.544749 4.087005 0.600846 2.381831 3.291216 0.916426 0.241925 0.190679 0.075477 0.066928 0.214089 0.024429 3.46E-19 0 5.54E-11 1
8.759177 0.116809 -0.00609 1.023393 0.458228 0.071808 0.279681 0.036665 0.012998 0.005154 0.031648 0.006716 0.004715 0.029107 0.006422 1
-0.00764 8.56062 0.108305 0.064991 0.974739 0.437852 0.012472 0.276522 0.035588 0.00515 0.031643 0.006713 0.004702 0.029035 0.006414 1
0.113052 -0.00537 8.945499 0.454132 0.08527 0.899012 0.036572 0.01439 0.282664 0.005173 0.031716 0.006728 0.004714 0.0291 0.006422 1
1.696272 1.761272 -0.00113 1.421726 2.457843 0.049884 0.122711 0.126004 0.014643 0.001826 0.011739 0.004012 9.43E-17 0 9.11E-10 1
-0.00348 1.716517 1.784889 0.049105 1.424984 2.476596 0.01447 0.123366 0.126799 0.001828 0.011753 0.004014 3.31E-17 0 5.42E-10 1
1.766352 -0.00265 1.706843 2.456405 0.051489 1.439141 0.126146 0.014372 0.123036 0.001839 0.011788 0.004027 7.85E-17 0 8.33E-10 1
1.166484 5.489578 0.096556 2.104036 1.178662 0.331665 0.101996 0.220048 0.032939 0.003225 0.018822 0.005344 0.011198 0.056043 0.009941 1
0.084384 1.168631 5.45058 0.296831 2.079931 1.25067 0.031248 0.102163 0.219235 0.003231 0.018862 0.005349 0.011312 0.056507 0.009991 1
5.392376 0.090489 1.111682 1.247021 0.320847 2.025664 0.218103 0.03259 0.099632 0.003268 0.019022 0.00538 0.01119 0.056011 0.009937 1
5.952967 0.376419 5.126158 2.207003 0.978898 2.821024 0.230038 0.059684 0.214398 0.067236 0.214642 0.024485 1.68E-19 0 3.88E-11 1
3.007667 0.013382 -0.0016 2.152529 0.073368 0.065258 0.164128 0.014142 0.01079 0.002346 0.015847 0.004548 0.00245 0.016063 0.004649 1
-0.00309 3.184585 0.015988 0.069945 2.221933 0.07476 0.011495 0.168981 0.014216 0.002377 0.016016 0.004578 0.002457 0.016097 0.004655 1
0.011338 -0.00273 2.995945 0.070312 0.063062 2.196607 0.013614 0.010551 0.16394 0.002342 0.015831 0.004544 0.002445 0.016024 0.004643 1
4.035884 4.168698 -0.00434 2.742562 2.513675 0.080058 0.190861 0.193024 0.011481 0.005754 0.035942 0.007096 7.70E-18 0 2.62E-10 1
-0.00018 3.930439 4.098159 0.074636 2.641342 2.401335 0.011062 0.188177 0.19138 0.005802 0.036206 0.007126 8.77E-18 0 2.77E-10 1
4.071265 -0.0011 3.969795 2.456027 0.082335 2.693185 0.190725 0.011912 0.189153 0.005699 0.035668 0.007062 6.53E-17 0 7.63E-10 1
4.895093 4.064159 0.077542 2.406075 3.113156 0.305415 0.209792 0.190684 0.028639 0.010281 0.053044 0.009544 0.007396 0.034709 0.008111 1
0.069585 4.860877 4.062529 0.292391 2.466273 3.071843 0.027899 0.209145 0.190591 0.010286 0.053061 0.009547 0.007426 0.034815 0.008128 1
3.796508 0.068564 4.767905 3.055764 0.285594 2.422766 0.184318 0.027279 0.207168 0.010217 0.052758 0.009515 0.007423 0.034823 0.008126 1
1.62068 7.145126 2.415331 1.698976 2.081258 2.769825 0.121664 0.252488 0.146637 0.066941 0.213927 0.024431 1.14E-19 0 3.19E-11 1
56 | P a g e
References
[1] R. Lasseter, A. Akhil, C. Marnay, J. Dagle, R. Guttromson, A. Meliopoulous, et al., "The CERTS
microgrid concept," 2002.
[2] H. J. Laaksonen, "Protection Principles for Future Microgrids," Power Electronics, IEEE
Transactions on, vol. 25, pp. 2910-2918, 2010.
[3] J. Driesen, P. Vermeyen, and R. Belmans, "Protection Issues in Microgrids with Multiple
Distributed Generation Units," in Power Conversion Conference - Nagoya, 2007. PCC '07, 2007,
pp. 646-653.
[4] B. Kroposki, C. Pink, R. DeBlasio, H. Thomas, M. Simoes, and P. K. Sen, "Benefits of Power
Electronic Interfaces for Distributed Energy Systems," Ieee Transactions on Energy Conversion,
vol. 25, pp. 901-908, Sep 2010.
[5] J. Keller and B. D. Kroposki, Understanding fault characteristics of inverter-based distributed
energy resources: National Renewable Energy Laboratory, 2010.
[6] J. K. SHAH, "Dynamic Power Flow Control for a Smart Micro-grid by a Power Electronic
Transformer," PhD, UNIVERSITY OF MINNESOTA, 2011.
[7] M. Dewadasa, A. Ghosh, and G. Ledwich, "Protection of microgrids using differential relays," in
Universities Power Engineering Conference (AUPEC), 2011 21st Australasian, 2011, pp. 1-6.
[8] M. Brucoli and T. C. Green, "Fault behaviour in islanded microgrids," in 19th International
Conference on Electricity Distribution, 2007, pp. 1-4.
[9] H. Nikkhajoei and R. H. Lasseter, "Microgrid Protection," in Power Engineering Society General
Meeting, 2007. IEEE, 2007, pp. 1-6.
[10] H. H. Zeineldin, E. F. El-Saadany, and M. M. A. Salama, "Distributed Generation Micro-Grid
Operation: Control and Protection," in Power Systems Conference: Advanced Metering,
Protection, Control, Communication, and Distributed Resources, 2006. PS '06, 2006, pp. 105-111.
[11] M. A. Redfern and H. Al-Nasseri, "Protection of micro-grids dominated by distributed generation
using solid state converters," in Developments in Power System Protection, 2008. DPSP 2008. IET
9th International Conference on, 2008, pp. 670-674.
[12] E. Sortomme, G. J. Mapes, B. A. Foster, and S. S. Venkata, "Fault analysis and protection of a
microgrid," in Power Symposium, 2008. NAPS '08. 40th North American, 2008, pp. 1-6.
[13] E. Sortomme, M. Venkata, and J. Mitra, "Microgrid protection using communication-assisted
digital relays," in Power and Energy Society General Meeting, 2010 IEEE, 2010, pp. 1-1.
[14] P. Mahat, C. Zhe, B. Bak-Jensen, and C. L. Bak, "A Simple Adaptive Overcurrent Protection of
Distribution Systems With Distributed Generation," Smart Grid, IEEE Transactions on, vol. 2, pp.
428-437, 2011.
[15] D. C. D. William E. Feero, John Stevens. CERTS White paper on Protection Issues of The Microgrid
Concept [Online]. Available: http://certs.lbl.gov/pdf/protection-mg.pdf
[16] H. H. Zeineldin, E. F. El-Saadany, and M. M. A. Salama, "Protective Relay Coordination for Micro-
grid Operation Using Particle Swarm Optimization," in Power Engineering, 2006 Large Engineering
Systems Conference on, 2006, pp. 152-157.
[17] M. A. Zamani, T. S. Sidhu, and A. Yazdani, "A Protection Strategy and Microprocessor-Based Relay
for Low-Voltage Microgrids," Power Delivery, IEEE Transactions on, vol. 26, pp. 1873-1883, 2011.
[18] M. A. Zamani, A. Yazdani, and T. S. Sidhu, "A Communication-Assisted Protection Strategy for
Inverter-Based Medium-Voltage Microgrids," Smart Grid, IEEE Transactions on, vol. 3, pp. 2088-
2099, 2012.
57 | P a g e
[19] S. M. Brahma and A. A. Girgis, "Development of adaptive protection scheme for distribution
systems with high penetration of distributed generation," Power Delivery, IEEE Transactions on,
vol. 19, pp. 56-63, 2004.
[20] T. S. Ustun, C. Ozansoy, and A. Zayegh, "Modeling of a Centralized Microgrid Protection System
and Distributed Energy Resources According to IEC 61850-7-420," Power Systems, IEEE
Transactions on, vol. 27, pp. 1560-1567, 2012.
[21] T. S. Ustun, C. Ozansoy, and A. Ustun, "Fault current coefficient and time delay assignment for
microgrid protection system with central protection unit," Power Systems, IEEE Transactions on,
vol. 28, pp. 598-606, 2013.
[22] M. A. Zamani, T. S. Sidhu, and A. Yazdani, "A communication-based strategy for protection of
microgrids with looped configuration," Electric Power Systems Research, vol. 104, pp. 52-61, 2013.
[23] T. Loix, T. Wijnhoven, and G. Deconinck, "Protection of microgrids with a high penetration of
inverter-coupled energy sources," in Integration of Wide-Scale Renewable Resources Into the
Power Delivery System, 2009 CIGRE/IEEE PES Joint Symposium, 2009, pp. 1-6.
[24] S. R. Samantaray, G. Joos, and I. Kamwa, "Differential energy based microgrid protection against
fault conditions," in Innovative Smart Grid Technologies (ISGT), 2012 IEEE PES, 2012, pp. 1-7.
[25] M. A. Haj-ahmed and M. S. Illindala, "The influence of inverter-based DGs and their controllers on
distribution network protection," in Industry Applications Society Annual Meeting, 2013 IEEE,
2013, pp. 1-9.
[26] E. Casagrande, W. Wei Lee, H. H. Zeineldin, and N. H. Kan'an, "Data mining approach to fault
detection for isolated inverter-based microgrids," Generation, Transmission & Distribution, IET,
vol. 7, pp. 745-754, 2013.
[27] N. Jayawarna, M. Barnes, C. Jones, and N. Jenkins, "Operating MicroGrid Energy Storage Control
during Network Faults," in System of Systems Engineering, 2007. SoSE '07. IEEE International
Conference on, 2007, pp. 1-7.
[28] R. H. Lasseter, J. H. Eto, B. Schenkman, J. Stevens, H. Vollkommer, D. Klapp, et al., "CERTS
Microgrid Laboratory Test Bed," Power Delivery, IEEE Transactions on, vol. 26, pp. 325-332, 2011.
[29] P. Piagi and R. H. Lasseter, "Autonomous control of microgrids," in Power Engineering Society
General Meeting, 2006. IEEE, 2006, p. 8 pp.
[30] A. Yazdani and R. Iravani, Voltage-sourced converters in power systems: modeling, control, and
applications: John Wiley & Sons, 2010.
[31] S. Mallat, A wavelet tour of signal processing: Academic press, 1999.
[32] M. Sonka, V. Hlavac, and R. Boyle, Image processing, analysis, and machine vision: Cengage
Learning, 2014.
[33] J. Han, M. Kamber, and J. Pei, "8 - Classification: Basic Concepts," in Data Mining (Third Edition),
J. H. Kamber and J. Pei, Eds., ed Boston: Morgan Kaufmann, 2012, pp. 327-391.
[34] R. X. Gao and R. Yan, Wavelets: Theory and Applications for manufacturing: Springer, 2010.
[35] W. Gao and J. Ning, "Wavelet-based disturbance analysis for power system wide-area
monitoring," Smart Grid, IEEE Transactions on, vol. 2, pp. 121-130, 2011.
[36] Z. Liu, Z. Han, Y. Zhang, and Q. Zhang, "Multiwavelet Packet Entropy and its Application in
Transmission Line Fault Recognition and Classification."
[37] S. A. Hosseini, N. Amjady, and M. H. Velayati, "A Fourier Based Wavelet Approach Using
Heisenberg's Uncertainty Principle and Shannon's Entropy Criterion to Monitor Power System
Small Signal Oscillations," Power Systems, IEEE Transactions on, vol. PP, pp. 1-13, 2014.
[38] H. Heidari Bafroui and A. Ohadi, "Application of wavelet energy and Shannon entropy for feature
extraction in gearbox fault detection under varying speed conditions," Neurocomputing, vol. 133,
pp. 437-445, 6/10/ 2014.
58 | P a g e
[39] A. I. Megahed, A. Monem Moussa, and A. Bayoumy, "Usage of wavelet transform in the protection
of series-compensated transmission lines," Power Delivery, IEEE Transactions on, vol. 21, pp.
1213-1221, 2006.
[40] Y. Sheng and S. M. Rovnyak, "Decision Trees and Wavelet Analysis for Power Transformer
Protection," Power Engineering Review, IEEE, vol. 22, pp. 62-62, 2002.
[41] M. Kantardzic, Data mining: concepts, models, methods, and algorithms: John Wiley & Sons, 2011.
[42] N. Ye, Data Mining: Theories, Algorithms, and Examples: CRC Press, 2013.
[43] J. Han and M. Kamber, Data Mining: Concepts and Techniques: Concepts and Techniques: Elsevier,
2011.
[44] B. Das and J. V. Reddy, "Fuzzy-logic-based fault classification scheme for digital distance
protection," Power Delivery, IEEE Transactions on, vol. 20, pp. 609-616, 2005.
[45] G. J. Williams, "Rattle: a data mining GUI for R," The R Journal, vol. 1, pp. 45-55, 2009.
[46] R. Ihaka and R. Gentleman, "R: a language for data analysis and graphics," Journal of
computational and graphical statistics, vol. 5, pp. 299-314, 1996.
[47] I. Kamwa, S. R. Samantaray, and G. Joos, "Development of Rule-Based Classifiers for Rapid Stability
Assessment of Wide-Area Post-Disturbance Records," Power Systems, IEEE Transactions on, vol.
24, pp. 258-270, 2009.
59 | P a g e

A Protection Scheme For Microgrids Using Intelligent Relays

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

A Protection Scheme For Microgrids Using Intelligent Relays

Загружено:

Авторское право:

Доступные форматы

A Protection Scheme for Microgrids

Using Intelligent Relays

By Debi Prasad Mishra

Department of Electrical and Computer Engineering

I would also like to acknowledge the support and encouragement provided by my

The financial support provided through the NSERC/Hydro-Québec Industrial Research

1.2. Conventional Distribution System vs Microgrid

 Inverter Interfaced Distributed Generation

Microturbines generate power at high frequency as compared to grid frequency.

1.3. Protection Issues in Microgrid

1. Adaptive and intelligent relaying

1.3.2. Unnecessary Operations of Protective Devices

1.3.4. Relay Coordination in Microgrid

1.3.5. Role of Communication System in Microgrid Protection

1.4. Literature Review: Protection Schemes for Microgrids

Protection strategies using zero sequence components for single-line-to-ground faults

1.5. Problem Statement

The aforementioned problems can be alleviated by relay co-ordination using

1.7. Claims of Originality

1.8. Organization of Thesis

Figure 2.1 CERTS Single Line Diagram

2.2 Components of the Simulated Microgrid

2.2.1 Distributed Energy Resources

Table 2-1 Generation capacity of DERs in microgrid

2.2.2 Three-Phase PI Line model

Table 2-2 Line impedance

Table 2-3 Load details in the microgrid

Table 2-4 Transformer details in the microgrid

2.2.5 Current Transformers

2.4 Current Controlled Inverter Model

Figure 2.2 Block Diagram of a DER using Solar Power

Table 2-5 Base values used for per unit calculation

Table 2-6 Boost converter parameters

Boost Converter Parameters

Table 2-7 Parameters for the voltage source inverter

PLL = Phase locked loop,

IDG = Current passing thought the inductor,

Vdc= Measured voltage of the dc bus,

L= Inductance of the filter,

f=frequency of the grid

ϴ = Phase of the grid from PLL,

SPWM= Sinusoidal pulse width modulation,

abc-to-dq = dq transformation of 3 phase ac quantities.

Iqref =Iq reference,

Kp= Proportional Gain,

Ki= Integral Gain

2.5 Voltage controlled VSI system

Figure 2.4 Schematic Diagram of Voltage Controlled VSI

VB = Voltage across capacitor C,

VBqref= Reference voltage for q-axis component of VB,

VBdref = Reference voltage for q-axis component of VB,

IB= Current supplied by VSI to grid,

IDG = Current passing thought the inductor,

Vdc= Measured voltage of the dc bus

Table 2-9 Parameters for inner current control loop

Table 2-10 Parameters of outer current control loop

3.2 Fault current due to Grid-frequency Imposed VSIs

4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5

-2 Phase a Phase b Phase c

Phase a Phase b Phase c

3.3 Fault Current due to Voltage Controlled VSIs

Phase a Phase b Phase c

Figure 4.1 Proposed protection scheme for CERTS microgrid