2013 Maxime MScThesis PDF

Fast Real-time Detection of
Sub-Synchronous Oscillations in
Power Systems using
Synchrophasors
MAXIME BAUDETTE
Master’s Degree Project

Stockholm, Sweden February 2013
TRITA-EE xxxxxxx
Fast Real-time Detection of
Sub-Synchronous Oscillations in Power
Systems using Synchrophasors
Master’s Degree Project
MAXIME BAUDETTE
Master’s Thesis at SmarTS Lab, Electric Power Systems

Supervisor: Luigi Vanfretti
Examiner: Luigi Vanfretti
TRITA-EE xxx yyyy-nn

Abstract
Recent concerns about the environment have resulted
into large investments in renewable energies for electricity
production, especially wind power. The integration of re-
newable sources of energy raises however several problems
which have not yet been completely understood nor stud-
ied. Oscillatory events around 13 Hz have been recorded in
the US by Oklahoma Gas & Electric (OG&E). Such a high
frequency is very different from the traditional and well
studied Inter-area oscillations, it is also beyond the mea-
surement capabilities of most of the existing measurement
equipments and monitoring tools.
This Thesis focuses on the development and implemen-
tation of algorithms for oscillation detection which can sup-
port real-time monitoring tools. It proposes a real-time
monitoring tool that exploits synchronized phasor measure-
ments from PMUs, which allow real-time analysis of higher
frequency events, filling the lack of such monitoring appli-
cation in the power systems area. This tool was built as a
prototype for real-time applications which utilize real-time
PMU data for ehanced monitoring and control of power
grids.
Acknowledgements
I would like to express my full gratitude to my supervisor
Dr Luigi Vanfretti, who all along this Thesis was present
to assist me. Providing me with stimulating conditions and
good advice, he guided me towards even more challenges to
spark my interest in the area.
I am also grateful to all my friends, who despite the
distance for the majority, remained available to me, sup-
porting me in my work and decision to stay in Sweden for
this Master’s Thesis. I would like thank especially Achour,
Émilie, Olivier and Simon, who stayed in Sweden the past
eight months and supported me every day. Together with
my other friends in Sweden, they made me enjoy my stay
here very much. A special thank for Olivier and his skills
in LATEX, which significantly contributed to improve the
readability of the report.
I am thankful to my family, who supported me finan-
cially all along my education and encouraged me to always
pursue my studies. They have always been present and
loving for me.
I would also like to add a thank to SmarTS Lab, who
was there to assist me during this these and was also very
welcoming. A special thank to Viktor and Shoaib who
shared my day to day work and turned it into a very fun
experience.
The financial support of KIC InnoEnergy through the
Smart Power: Action 2.6 - PMU-Based Power System Op-
eration Tools is greatly acknowledged. Also in KTH, all
the other employees I have met during this Thesis are ac-
knowledged for the nice working environment as well as the
welcoming atmosphere they provided.
I would also like to thank the engineers team I met at
IREC for their help and cooperation during the experiments
in their facility. Some results from the joint work are shown
in Chapter 6.
The support of Austin White at OG&E providing wind
farm measurements from Oklahoma Gas & Electric is greatly
acknowledged.
Finally, I would like to thank my french engineering
school Supélec, without which I would not have had the
opportunity to pursue my studies abroad to do this Mas-
ter’s Thesis
Contents
List of Figures
Acronyms
1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Wide-Area Systems: Phasor Measurement Units and Phasor Data
Concentrators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3.1 General Objectives . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3.2 Specific Objectives . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
I Developing the test models 7
2 Modelling for RT Simulation and Analysis 9

2.1 Wind Turbine (WT) . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.1 Type 3 Wind Turbine Operating Principle . . . . . . . . . . . 10
2.1.2 Wind Turbine Controllers . . . . . . . . . . . . . . . . . . . . 11
2.1.3 Single WT - Infinite Bus . . . . . . . . . . . . . . . . . . . . . 12
2.2 Power System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 General Structure . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.2 Wind Farms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.3 Load Variations . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3 Perturbation Analysis 19
3.1 Controller Sensitivity Analysis: WT-Infinite Bus . . . . . . . . . . . 20
3.1.1 Grid Side Controller . . . . . . . . . . . . . . . . . . . . . . . 21
3.1.2 Rotor Side Controller . . . . . . . . . . . . . . . . . . . . . . 21
3.1.3 Pitch Controller . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.1.4 Grid Injection . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2 Network Sensitivity to Injected Oscillations . . . . . . . . . . . . . . 25
3.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
II Monitoring Tool 27
4 Fast Real-Time Oscillation Detection Tool 29

4.1 Oscillation Detection and Frequency Estimation . . . . . . . . . . . . 30
4.1.1 Fast Oscillation Detection . . . . . . . . . . . . . . . . . . . . 30
4.1.2 Frequency Estimation . . . . . . . . . . . . . . . . . . . . . . 33
4.1.3 Data Pre-Processing . . . . . . . . . . . . . . . . . . . . . . . 34
4.2 Implementation in LabView . . . . . . . . . . . . . . . . . . . . . . . 38
4.2.1 Real-Time Monitoring Tool . . . . . . . . . . . . . . . . . . . 38
4.2.2 Replay Tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5 Instrumentation and Data Acquisition 47

5.1 Compact RIO Configuration as PMUs . . . . . . . . . . . . . . . . . 48
5.1.1 CompactRIO Configuration . . . . . . . . . . . . . . . . . . . 49
5.1.2 Configuration of the PMU settings . . . . . . . . . . . . . . . 50
5.2 Phasor Data Concetrators . . . . . . . . . . . . . . . . . . . . . . . . 53
5.2.1 OpenPDC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.2.2 SEL-5073 Synchrowave . . . . . . . . . . . . . . . . . . . . . 55
5.2.3 PMU Connection Tester . . . . . . . . . . . . . . . . . . . . . 57
5.3 Real-Time Data Mediators . . . . . . . . . . . . . . . . . . . . . . . 58
5.3.1 PRL Configuration . . . . . . . . . . . . . . . . . . . . . . . . 59
5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6 Experimental Testing and Validation 63

6.1 Experiment Design and Results . . . . . . . . . . . . . . . . . . . . . 64
6.1.1 Experimentation Design . . . . . . . . . . . . . . . . . . . . . 64
6.1.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.2 Experiment Validation . . . . . . . . . . . . . . . . . . . . . . . . . . 71
6.2.1 IREC’s Microgrid . . . . . . . . . . . . . . . . . . . . . . . . . 71
6.2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7 Conclusion 81
7.1 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Appendices 82
A Additional Figures 83
Bibliography 87
List of Figures
1.1 Comparison Example PMU and SCADA . . . . . . . . . . . . . . . . . . 2

1.2 Diagram of a Wide-Area System . . . . . . . . . . . . . . . . . . . . . . 3
2.1 Type 3 Wind Turbine Diagram . . . . . . . . . . . . . . . . . . . . . . . 10

2.2 Test System: One Wind Turbine . . . . . . . . . . . . . . . . . . . . . . 12
2.3 One-line Diagram of the Test System Power System . . . . . . . . . . . 13
2.4 Power System in MatLab/Simulink . . . . . . . . . . . . . . . . . . . . 14
2.5 One-line Diagram of the Wind Farms . . . . . . . . . . . . . . . . . . . 16
3.1 PSD with Perturbation to Grid Side Controller . . . . . . . . . . . . . . 21

3.2 PSD with Perturbation to Rotor Side Controller . . . . . . . . . . . . . 21
3.3 PSD with Perturbation to Pitch Controller . . . . . . . . . . . . . . . . 22
3.4 PSD with Perturbation via Injection in the bus voltage . . . . . . . . . 23
3.5 PSD with Perturbation via Injection in the bus voltage . . . . . . . . . 24
3.6 PSD with Perturbation via injections in the voltage at PCC . . . . . . 25
4.1 Diagram Original Oscillation Detection . . . . . . . . . . . . . . . . . . 31

4.2 Diagram Oscillation Detection . . . . . . . . . . . . . . . . . . . . . . . 32
4.3 Diagram Spectral Estimation . . . . . . . . . . . . . . . . . . . . . . . . 33
4.4 Example of Confidence Interval . . . . . . . . . . . . . . . . . . . . . . . 35
4.5 Diagram Outlier Removal . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.6 Screen shots Outlier Removal . . . . . . . . . . . . . . . . . . . . . . . . 37
4.7 Screen shot Monitoring Tool Main Window . . . . . . . . . . . . . . . . 39
4.8 Screen shot Monitoring Tool Configuration Tab . . . . . . . . . . . . . . 40
4.9 Screen shot Module Configuration Tab . . . . . . . . . . . . . . . . . . . 41
4.10 Screen shot Monbitoring Tool Module . . . . . . . . . . . . . . . . . . . 41
4.11 Screen shot Replay Tool Main Window . . . . . . . . . . . . . . . . . . . 43
4.12 Screen shot Replay Tool Options . . . . . . . . . . . . . . . . . . . . . . 44
4.13 Diagram Read CSV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.1 cRIO-9074 Assembled as PMU . . . . . . . . . . . . . . . . . . . . . . . 48

5.2 Screen shot PMU Configuration Utility . . . . . . . . . . . . . . . . . . 49
5.3 cRIO PMU Web Interface . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.4 Wiring scheme PMU . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.5 Picture Voltage Module wired . . . . . . . . . . . . . . . . . . . . . . . . 51
5.6 Picture at SmarTS Lab . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.7 OpenPDC Manager Main Window . . . . . . . . . . . . . . . . . . . . . 54
5.8 OpenPDC: New Device wizard . . . . . . . . . . . . . . . . . . . . . . . 54
5.9 OpenPDC: Create an Output . . . . . . . . . . . . . . . . . . . . . . . . 55
5.10 SEL Synchrowave Interface: Main Window . . . . . . . . . . . . . . . . 56
5.11 SEL Synchrowave Interface: Menus . . . . . . . . . . . . . . . . . . . . . 56
5.12 PMU Connection Tester Interface . . . . . . . . . . . . . . . . . . . . . . 57
5.13 PRL Interface Screen shots . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.14 Screen shot PRL Data Selector . . . . . . . . . . . . . . . . . . . . . . . 59
5.15 PRL subVI in LabView . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.16 Screen shot PRL connected . . . . . . . . . . . . . . . . . . . . . . . . . 60
6.1 Experiment in KTH - Load Variation and Wind Turbulences . . . . . . 66

6.2 Experiment in KTH - 10.83 Hz Beginning of Injection . . . . . . . . . . 67
6.3 Experiment in KTH - 10.83 Hz Increase of Oscillation Injection . . . . . 68
6.4 Experiment in KTH - Minor Fault . . . . . . . . . . . . . . . . . . . . . 69
6.5 Experiment in KTH - Minor Fault . . . . . . . . . . . . . . . . . . . . . 69
6.6 Experiment in KTH - 10.83 Hz End of Injection . . . . . . . . . . . . . 70
6.7 Diagram of IREC Micro Grid . . . . . . . . . . . . . . . . . . . . . . . . 72
6.8 Photo of a Cabinet at IREC . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.9 Photo of a cRIO at IREC . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.10 Experiment in IREC - 1Hz Beginning of Injection . . . . . . . . . . . . . 75
6.11 Experiment in IREC - 6Hz Beginning of Injection . . . . . . . . . . . . . 77
6.12 Replay Simulation Data - Beginning of Injection . . . . . . . . . . . . . 78
A.1 Experiment in IREC - 1Hz End of Injection . . . . . . . . . . . . . . . . 83

A.2 Experiment in IREC - 6Hz End of Injection . . . . . . . . . . . . . . . . 84
A.3 Replay Simulation Data - End of Injection . . . . . . . . . . . . . . . . . 85
Acronyms
OG&E Oklahoma Gas & Electric

PMU Phasor Measurement Unit
PDC Phasor Data Concentrator
IREC Institut de Recerca en Energia de Catalunya
TSO Transmission System Operator
GUI Graphical User Interface
cRIO National Instruments CompactRIO
DFIG Doubly Fed Induction Generator
VSC Voltage Source Converter
WT Wind Turbine
PCC Point of Common Coupling
SCADA Supervisory Control And Data Acquisition
BPA Bonneville Power Administration
LP Low-Pass
HP High-Pass
RMS Root Mean Square
PSD Power Spectrum Density
FFT Fast Fourier Transform
PRL PMU Recorder Light
FIR Finite Impulse Response
CSV Comma Separated Values
NI National Instruments
DHCP Dynamic Host Configuration Protocol
IP Internet Protocol
SEL Schweitzer Engineering Laboratories
Chapter 1
Introduction
With recent environmental concerns, renewable energies have seen increased their
share of the electricity production in most of the western countries. In most of
these countries the full potential of hydro-power has already been exhausted, which
resulted in wind energy being the fastest growing energy technology in the 1990s [1].
Moreover the expansion of the share of electricity production from wind power will
continue, especially in Europe where several countries have fixed high goals for the
coming years [2].
1.1 Background
One of the impacts noticed with the increased amount of intermittent power sources
was an effect on power balancing generation units. It has been shown that the
operation of wind farms, which are one kind of intermittent source, could involve an
increase in polluting emissions [3]. Real-time monitoring and control tools could help
mitigating this undesired behaviour, by providing software applications to operate
the system with more flexibility adapting to intermittent generation sources [4].
The increasing amount of wind power in a power system has also been studied
from the perspective of power system stability, and several problems regarding tran-
sient stability have already been explained [5]. It is however only very recently, that
some Transmission System Operators (TSOs) have noticed the presence of parasite
oscillations in the grid between wind farms in the range of 13-15Hz, thanks to the
use of PMUs [6]. The oscillations occurred during high wind penetration periods,
and the impact was observable even at the consumer voltage level. The reasons
for these oscillations are turbine-converter interactions [7] and they differ from the
more commonly observed inter-area oscillations below 1 Hz [8].
As the consequences can be observed at all voltage levels and because these wind
farm oscillations have not yet been specifically characterized, it could be supposed
that such oscillations could damage equipment on the transmission system or at the
customers end. It could also be supposed that the phenomenon could be amplified
by the increasing use of wind power. A better ability to monitor the behaviours of
1
2 CHAPTER 1: INTRODUCTION
wind farms connected to the bulk grid appears thus necessary.

In this Thesis any oscillatory behaviours occurring at a frequency below the
frequency of the power system different from inter-area or local oscillations will be
referred to as as sub-synchronous oscillations. Those behaviours are moreover not
specifically associated with sub-synchronous resonances, described in [9].
1.2 Wide-Area Systems: Phasor Measurement Units and

Phasor Data Concentrators
The ability to monitor fast dynamic behaviours such as the sub-synchronous oscil-
lations phenomena, described above, is tied to the performance of the measurement
equipment used in power systems. The sampling rate has a strong impact on the
highest frequency observable from the measurements. Traditional equipment and
monitoring tools have an asynchronous sampling frequency of typically one sam-
ple every few seconds. While this was sufficient to monitor very slow steady state
phenomena, it fails to capture faster dynamic behaviours, see Fig. 1.1.
1.03
1.02
Voltage Magnitude (p.u.)
1.01
0.99
0.98
PMU
SCADA
0.97
0 5 10 15 20 25 30 35 40
Time (s)
Figure 1.1: Comparison example between PMU and traditional equipment measure-
ments(SCASA) during a fast dynamic event
Phasor Measurement Units (PMUs) are devices able to measure the voltage and
current phasors of the three phase network with a sampling frequency of typically 25,
30 and 50 samples per seconds. The frequency of the system is internally computed
at a higher sampling rate and reported also at 25, 30 and 50 samples per second,
giving the complete state of the power system at the point of measurements.1 This
allows the study of dynamic phenomena occuring up to a frequency of half the
reporting rate. The reference for the angle of the phasors is derived from a GPS
1
Instant active and reactive power can be computed from the voltage and current phasors
1.2. WIDE-AREA SYSTEMS: PHASOR MEASUREMENT UNITS AND
PHASOR DATA CONCENTRATORS 3
clock and all the measurements are coupled to a time stamp allowing the alignment
and synchronization of several PMUs.
Phasor Data Concentrators (PDCs) have the role of collecting and gathering
the data of several PMUs. Since all the measurements have a time stamp acquired
from the GPS reference, the PDC can align the measurements in time. PDCs can
be configured for several usages, the main being the ability to output a stream of
time-aligned measurements from several PMUs. It can, however, also be configured
to store and archive data or perform several calculations on the measurements.
The typical architecture of a monitoring system using PMUs is presented in
Fig. 1.2. Based on their location in the power system and their owner the PMU
measurements are gathered in a specific PDC server. All the PDC servers feed
up the measurements to a Super PDC, either with a direct connection or through
the nearest PDC server. The Super PDC serves as main source of PMU data for
the TSO or group of TSOs in its network control operations. The communication
between different devices follows the IEEE C37.118.2 protocol [10].
TSO CONTROL ROOM
SUPER
PMU PDC PMU
PDC PMU
PDC
PMU
PDC
PMU PMU
PMU PMU
PMU
Figure 1.2: Layout of the communication and computer architecture of wide-area moni-
toring systems
The usage of PMUs to collect data on wide-area power systems is however rather
recent. It has enabled system operators to monitor their transmission system with
more information as well as to conduct deeper studies of grid stability, particularly
for inter-area modes [8]. This technology offers a great potential for building mon-
itoring systems [11, 12]. This Thesis work at SmarTS Lab will contribute toward
the development of monitoring applications using PMU measurements.
4 CHAPTER 1: INTRODUCTION
1.3 Objectives
1.3.1 General Objectives
The main objective of this Master’s Thesis is to build a working prototype of a real-
time monitoring tool, which should be able to detect oscillations in a grid within a
wide range of frequencies. The tool has been developed in the LabView environment
and part of this objective will be to build a user friendly Graphical User Interface
(GUI).
This tool uses PMU measurements to feed the algorithms. It has been tested
to verify its correct functioning. The test experiment have been carried out at
SmarTS Lab, with a real-time hardware-in-the-loop (HIL) simulation of a power
system. The second objective comprises the development of a real time power
system model for the test experiments.
The third and final objective is to validate the monitoring tool with different
simulations of the power system with two wind farms with the real-time power
system simulator, and perform validation tests at another laboratory with different
physical equipment.
1.3.2 Specific Objectives

1. Selection and development of a model including two wind farlss as a test power
system.
2. Familiarization with the real-time simulator at SmarTS Lab.
3. Familiarization with Lab-View and its programming environment.
4. Implementation of oscillation detection tools in LabView.
5. Configuration of the National Instruments CompactRIOs for PMU usage.
6. Test the monitoring tool with pre-recorded data and real-time data.
7. Test of the monitoring tool in different simulation scenarios and testing facil-
ities.
1.4 Thesis Outline

The Thesis report is organized as followed :
In Chapter 2, the models used for real-time simulation in the lab are presented.
The first model presented is used for analysis purposes and the second model is used
for simulating an entire power system and measurements, using HIL simulation.
In Chapter 3, a perturbation analysis is conducted on the controllers of the wind
turbine. This analysis enables to choose the right perturbation scheme to recreate
the sub-synchronous power oscillations phenomenon in the power system model,
similar to that observed in OG&E.
1.4. THESIS OUTLINE 5
In Chapter 4, the algorithms which the Monitoring Tool uses, are introduced .
Their modification for implementation in LabView is also detailed. The Chapter
finishes with the presentation of the final GUI of the Monitoring Tool.
In Chapter 5, the whole chain of real-time data acquisition is presented in details.
It is a collection of the required components and their configuration for setting up
the experiments reported in this Thesis.
In Chapter 6, the results of the experiments are presented. Both the experiments
performed at SmarTS Lab at KTH and at IREC’s microgrid facility in Spain are
reported.
Chapter 7 summarizes the work reported in this Thesis and outlines future work.
Part I
Developing the test models
7
Chapter 2
Modelling for RT Simulation and

Analysis
The phenomenon described in the Introduction was observed in a power system con-
taining a substantial penetration of wind-power. Moreover the oscillatory events oc-
curred specifically during high-wind period and wind-farm responses were identified
as origin of this unwanted dynamic phenomena.
Part of the process of better understanding phenomena is to identify their origin.
This Chapter lays out the power system models utilized to recreate the observed
oscillatory phenomena at OG&E in a general form. Analysis on these models is
carried out in Chapter 3, where a perturbation analysis helps to identify the source
and interactions responsible for the oscillatory behaviour of interest.
The models presented are developed in MatLab/Simulink and adapted to real-
time execution using a real-time digital simulator from Opal-RT; which is capable
of performing real-time simulation power system models and has re-configurable
analog inputs/outputs for external measurement devices.
9
10 CHAPTER 2: MODELLING FOR RT SIMULATION AND ANALYSIS
2.1 Wind Turbine (WT)

There are several types of wind turbines, differing in their operating principle, which
are described in [1]. The most common wind turbines nowadays are doubly fed
machines and converter fed machines. They are referred to as Type 3 and Type
4 wind turbines [13]. The power system of OG&E contains several wind farms of
Type 3 WT, this type of wind turbine was therefore selected to be the baseline of
our models.
The model is taken from the example library from Opal-RT, they were originally
developed in MatLab/Simulink by Hydro Québec. More information on these
model is available in [14]. In this Section the focus is on describing the operation
principle of a Type 3 wind turbine.
2.1.1 Type 3 Wind Turbine Operating Principle

Type 3 wind turbines are often referred to as variable speed turbines. They are built
around a Doubly Fed Induction Generator (DFIG) with a back-to-back converter
connected to the grid and feeding the rotor windings. The diagram of the typical
layout of a Type 3 wind turbine is depicted in Fig. 2.1
DFIG
STATOR
ROTOR
GEAR
BOX
575 V
~ DC =
=
ROTOR SIDE
LINK
~
GRID SIDE
Figure 2.1: Diagram of a Type 3 wind turbine
The DFIG is an induction machine with both windings connected. The stator
windings are connected to the grid and the rotor windings are connected to the out-
put of a DC to AC converter externally controlled. The control scheme is explained
in the following Sub-section.
In this wind turbine the control of the currents in the rotor windings will regulate
the power factor of the power output to the grid. Where a traditional induction
generator can only draw reactive power from the grid, the reactive power output
is fully controlled in this kind of turbine, within a certain range defined by the the
capability of the converter.
2.1. WIND TURBINE (WT) 11
The model chosen is an average model, because the converters are modelled
with ideal voltage and current sources. A version of the wind turbine model with
detailed switching components is available, but requires higher computation effort
for a complete wind farm. The detailed model has been designed for the study
of harmonics, whereas the average model is sufficient for the study of the sub-
synchronous frequency content.
The final goal was to model two wind farms with sufficient electric power pro-
duction, without explicitly modelling too many wind turbines, which would require
a large computational effort. Each wind turbine model thus represents a certain
amount of wind turbines, leading to, micro aggregate models (between 4 and 6
turbines).
2.1.2 Wind Turbine Controllers

The wind turbine controller is characterized with several control schemes, which
each have a specific role. All together they will enable the turbine to operate at
variable speed within a certain range, maximizing the power output that can be
converted from the wind power.
In a real turbine, part of the control scheme is to orient the blades in the pri-
mary direction of the wind, which is done by a controlled yaw mechanism. This
mechanism has slow dynamics, as the wind does not change direction so quickly.
Thus for modelling purposes, this mechanism is neglected. This assumption implies
that the chosen models do not have any yaw mechanism and always receive front
wind.
In this Sub-section only the role of each controller will be explained. Further
information on the different controllers is available in the literature [1, 14].
Grid Side Converter

The rotor of the DFIG generator is connected to the main grid through a DC Link
that utilizes Voltage Source Converter (VSC) scheme. The operating principle of
the VSC requires the voltage of the DC Link to be constant to ensure stability of
the link.
The grid side converter is connected to the main grid as stated by its name. Its
role is to maintain the voltage of the DC Link constant while drawing power from
the grid is at unity power factor , as in most case.
Rotor Side Converter

In the description of the Type 3 wind turbines, it was mentioned that it was possible
to modify their active and reactive power output. In other words, the power factor
of the turbine can be controlled. This behaviour is achieved by controlling the rotor
currents with a vector control scheme of the torque. This is the role of the rotor
side controller.
Pitch Control
Each of the blades of variable speed turbines is mounted on a rotating base, allowing
the control of their inclination. This will allow the control of the efficiency of the
blade. At wind speeds lower than the rated speed, the blades must be oriented to
exert the largest torque possible on the generator. However, when the wind speed is
greater than the rated speed, the pitch control is active to increase the angle of the
blades and thus reducing the drag effect and the torque applied to the generator,
avoiding the generator to rotate with a greater speed than the rated speed.
2.1.3 Single WT - Infinite Bus

In the Thesis work several experiments of different complexity level were planned.
First a perturbation analysis on one turbine was conducted, this experiment was
conducted on a simple system with one turbine connected to an infinite bus. In the
model the infinite bus was modelled by a voltage source. The layout is depicted in
Fig. 2.2. The model also includes all the blocks used for the perturbation analysis
in Chapter 3.
Infinite Bus
WT
25 kV
Figure 2.2: One-line diagram of the single WT - Infinite Bus

2.2. POWER SYSTEM MODEL 13
2.2 Power System Model

The model presented in the previous Section will only serve for the perturbation
analysis carried out in Chapter 3. This system does not fully match the behaviour
of a large power system and all the interactions between its components, and thus,
another model is needed for testing.
This Section presents the model that has been developed to include more in-
teractions of a real power system. The model represents a full power system with
loads, transmission lines and power generating units.
2.2.1 General Structure

The power system modelled is based on the two-area Klein-Rogers model, presented
in [15] and its implementation for real-time in [16]. The model was modified by
replacing the synchronous machines from the area on the right with wind farms.
This model has been chosen for its ability to replicate the most dominant dynamic
responses of real power systems. The one-line diagram of the final layout of the
model is depicted in Fig. 2.3.
G1 1
3 4
5 101
345 kV
50
10 km 10 30 - PCC1 WF1
25 km
25 km 10 km 25 kV
20 kV 110 km 110 km 20 - PCC2
25 kV WF2
2 20 kV
G2
Figure 2.3: One-line diagram of the test power system with two wind farms
Opal-RT simulators require separation of MatLab/Simulink models into sub-

systems. Each of them is attributed to an available CPU of the simulator, increasing,
de facto, the maximum complexity of the models able to run in real-time. Each real-
time target in the lab has twelve CPUs, one of them is dedicated to the execution
of the operating system and the synchronisation of the eleven other CPUs. This
means that in a set-up with only one real-time target, as in this case, a maximum
of eleven subsystems can be used.
The model as implemented for real-time execution in MatLab/Simulink is rep-
resented Fig. 2.4. One block is assigned to the console on the client side and each
of the others is attributed to one CPU available in the target. The communica-
tion and synchronisation of different sub-systems is ensured by a set of Opal-RT
proprietary components added to each input of the sub-system. For the electrical
signals transmitted through three-phase line, the synchronism is ensured by the use
of Opal-RT three-phase lines with natural delay.
Figure 2.4: Screen-shot of the power system implemented for real-time simulation in Mat-
Lab/Simulink
The console includes several displays to monitor key parameters of the power
system, including electrical measurements as well as mechanical variables in both
the synchronous generators and the wind farms. It also gathers all the commands
sent by the user to interact with the running model. These commands are the wind
configuration, the variable load configuration, and settings for oscillation injection
and faults.
The controls of the synchronous generators are gathered in one independent
light-blue block. It includes the turbine governor regulators, power system stabiliz-
ers and fault generation commands.
The other sub-systems represent the power system itself. The big blue rectangle
contains the first area with the synchronous generators, the loads and the two Points
of Common Coupling (PCC) of the wind farms. The remaining blocks in black are
dedicated to the wind turbines and are explained in more detail next.
2.2.2 Wind Farms

The wind farm model used in this power system model is directly inspired from [17].
It has been modified to have control over all the wind turbines as average models,
since the study focuses on sub-synchronous phenomena.
The PCC of the wind farm has a voltage of 25 kV. It is connected to the grid
through a transformer, which increases the voltage from 25 kV to the ratings of the
grid or 345 kV in our case. Each wind turbine is connected to a sub-bus through a
transformer to increase the voltage from 575 to 25 kV. These sub-buses are connected
to the PCC with transmission lines.
The final layout for the two wind farms is depicted in Fig. 2.5.
Preparation for Real-Time Execution

The wind farms are comprised by sixteen and thirteen independent turbines respec-
tively. This induces high computational effort. The implementation takes advantage
of the sub-systems structure, the first farm is divided into five sub-systems and the
second into four sub-systems.
There are three kinds of sub-systems in the wind farms:
• Sub-system of three average wind turbines models.
• Sub-system of five average wind turbines models electro-mechanical part.
• Sub-system of ten average wind turbines models control part.
The models separated into two parts is referred to as decoupled models. The
two farms are combinations of those three kinds of sub-systems chosen to have the
most wind turbines given the number of remaining subsystems to create.
Each subsystem is connected to the PCC with transmission lines, which will be
used for model separation. Inside the sub-systems, each turbine is connected to a
sub-bus.
GRID
CONNECTION
230 kV
PCC
25 kV
DECOUPLED MODELS
Figure 2.5: One-line diagram of the wind farms. The part circled in red is not present in
the second wind farm.
Wind Generation
The wind farm model also includes a wind profile generator, which can be configured
by the user. There are thus two wind profiles to configure for the user in the power
system model, each dedicated to one wind farm. The wind can be set to one of the
following configurations:
• Constant wind speed.
• Time series.
• Both of the aforementioned with configurable turbulence.
Turbulence of the wind is generated by a set of Kaimal filters, which are further
described in [1]. The main filter is responsible for generating the instant wind,
modelling wind gusts. The other filters, located in each wind turbine, model local
turbulences due to the moving blades of the turbines.
2.2.3 Load Variations

The model as presented in this Section has two loads, one in each area. The loads
are constituted from a constant load set as constant impedance and a varying load
that can be configured by the user during the simulation.
The variation are performed independently in the active and reactive compo-
nents of the load and can be individually configured according to:1 .
I
G1 .r + G2 if Constant Load,
P (t) = (2.1)
G1 .r + G2 . sin (2ﬁf t) if Sinusoidal Load.
Where r is a random number generated at each time step (50 µ sec).

The varying load presents several advantages for the simulation. It will first
simulate a system with a stochastic load and therefore create a behaviour closer
to reality concerning the evolution of the grid frequency in the simulated power
system. The possibility of configuring the load as sinusoidal will be used to excite
the system at low frequencies, close to the inter-area modes of the system.
1
The equation is written for active power, reactive power is modelled in a similar way
2.3 Summary
In this Chapter, the important parts of the model were presented, as well as the
assembled system, which will be used for the simulation. The ratings adopted for the
power system model, notably the total amount of wind power (nearly 200 MW with
two farms), was chosen to represent approximately ten percent of the generation of
the modelled power system.
Chapter 3
Perturbation Analysis
The phenomenon of high frequency oscillations in the Oklahoma Gas& Electric

power system as presented in Chapter 1 was identified as originating from wind-
farms through measurement analysis [7]. These oscillations have been widely at-
tributed to wind turbine controls. However, no specific analysis has been made
to determine which of the wind turbine controls are involved in the emergence of
these oscillations. In the description of Type 3 wind turbines given in Chapter 2,
the controls are presented together with an explanation of they primary role in the
operating principle of the wind turbine.
In this Chapter a wind turbine controller sensitivity analysis is performed to aid
in determining which wind turbine controller is involved in the dynamic interaction
giving rise to high frequency oscillations observed in the grid.
To this aim the effect of injected oscillations to the reference value of different
controllers will be studied, to determine which one has the strongest impact on the
wind turbine output. This study is conducted in the single WT-infinite bus model
introduced in Chapter 2.
The second focus of this Chapter is the effect of such perturbation on the power
system model presented in Chapter 2. The goal is to be able to mimic a behaviour
consistent with that observed from the real measurements from OG&E.
19
20 CHAPTER 3: PERTURBATION ANALYSIS
3.1 Controller Sensitivity Analysis: WT-Infinite Bus

The oscillatory events occurred in the power system operated by OG&E, which is a
system with a strong penetration of wind power. The wind farms have been iden-
tified as the origin of the phenomenon. Further confidential studies commissioned
by OG&E have shown that the controllers of the wind turbine were causing this
behaviour.
In this Section the behaviour of the wind turbine model is analysed when subject
to different perturbations successively applied to each of the controllers described in
Chapter 2. The model used for this study is thus developed with several additions
to generate different perturbations.
The original oscillations observed at OG&E were occurring at 13.4 Hz. This
frequency is slightly below the quarter of the UE Eastern interconnection system
frequency, 60 Hz. In Europe, the system nominal frequency is 50 Hz. Thus, for
this analysis it was decided to scale down the frequency of the oscillations injected
the study network, in order to reflect the system frequency in European networks;
leading to a perturbation frequency of 10.83 Hz.
The first step of the study is to probe each of the existing controllers in the wind
turbine. The model used for this purpose is the simplest model presented, with just
one wind turbine connected to an infinite bus. The simulations are run successively
with a different perturbation selected during 300 seconds. The data recordings
begin only after the first 60 seconds of simulation to avoid any unwanted dynamic
behaviour due to the setting of the simulation with imperfect initial conditions.
The perturbation applied is a sinusoidal signal at the reference value of the
controllers. The frequency of the signal is set to 10.83 Hz and the amplitude of the
sinus is set to 0.05 p.u. in all cases. The bus voltages and currents, the active and
reactive powers and the voltage in the DC Link are recorded during the simulation
with a frequency of 200 Hz in order to capture the whole spectrum of interest of
[0 50] Hz without aliasing.
The primary interest of this study is the resulting oscillations at network in-
terface variables, for this reason the frequency content of the active and reactive
power outputs of the wind turbine are presented as results. The other reason for
not checking the frequency content of the voltage is that the infinite bus cancels
all the effect of oscillations in this variable. The frequency content of the voltage
in the DC Link is also presented. The results presented are the frequency spectra1
which are determined by using Welch’s method, described in Section 4.1.2, with the
following parameters:
• Window size : 8000 samples.
• Window : Hanning2 .
• Overlap : 75 %.
1
The spectra are normalized in dB with the same offset in all cases.
2
MatLab implementation of Welch’s method.
3.1. CONTROLLER SENSITIVITY ANALYSIS: WT-INFINITE BUS 21
3.1.1 Grid Side Controller

A perturbation to the grid side controller was injected on the reference of the DC
link voltage of the turbine. The frequency content is presented in Fig. 3.1.
360 250 180
Active Power Reactive Power Vdc
340 160
320 140
200
300
120
Amplitude (dB)
Amplitude (dB)
Amplitude (dB)
280
100
150
260
80
240
60
220 100
200 40
180 20
50
160 0
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
Frequency (Hz) Frequency (Hz) Frequency (Hz)
(a) Active Power (b) Reactive Power (c) Voltage in DC Link
Figure 3.1: Power Spectrum Density (PSD) of the active and reactive powers and the
voltage at DC link when subject to a perturbation to the grid side controller
The active power output is greatly influenced by the perturbation with an am-
plitude of the 10.83 Hz component measured at 355 dB, see Fig 3.1a. The reactive
power is impacted to a lesser extent, this was expected since the grid side controller
is not responsible the reactive power output regulation.
It can also be seen that the voltage in the DC link is greatly impacted by the
perturbation, which is expected since the perturbation has been applied to the
voltage reference of the DC link.
3.1.2 Rotor Side Controller
220 350 30
200
25
180 300
160 20
Amplitude (dB)
Amplitude (dB)
Amplitude (dB)
140 250
120 15
100 200
10
80
60 150
5
40
20 100 0
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
Figure 3.2: PSD of the active and reactive powers and the voltage in the DC link when
subjected to a perturbation to the rotor side controller
A perturbation to the rotor side controller was injected on the reactive power
output reference. The frequency content is presented in Fig. 3.2.
In this case the reactive power is the most influenced by the perturbation with
an amplitude of the 10.83 Hz component measured at 310 dB, see Fig. 3.2b. The
active power and the voltage in the DC link are less affected than in the previous
case. This result was expected, as the rotor side converter is regulating the reactive
power output of the turbine.
3.1.3 Pitch Controller

The last perturbation test is performed on the pitch controller at the rotor speed
reference. The frequency content is presented in Fig. 3.3.
300 140 180

280 160
120
260 140
240 100
120
Amplitude (dB)
Amplitude (dB)
Amplitude (dB)
220
80 100
200
60 80
180
60
160 40
140 40
20
120 20
100 0 0
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
subject to a perturbation to the pitch controller
The frequency content of all the quantities is affected in a different way than
in the two previous examples. It can be first noticed that the voltage of the DC
link is not affected, see Fig. 3.3c, which can be explained as the pitch controller
does not act upon any component of the DC link. Changing the pitch of the blades
will however modify the torque applied to the shaft of the generator, influencing
therefore the active and reactive power outputs.
The pitch mechanism has an intermittent operating principle: up until the nom-
inal wind speed the mechanism is not active and the torque is regulated by the rotor
side controller. When the wind speed increases above the nominal speed, the elec-
trical torque can no longer be increased as it has reached the limit of the turbine
ratings. The speed of the generator thus increases, going over the nominal speed
triggering the pitch mechanism.
The perturbation in the rotor speed reference actually triggers the pitch mech-
anism artificially, generating a collateral effect with a very slow dynamic, that can
be observed on the frequency content in Fig. 3.3a and 3.3b.
3.1. CONTROLLER SENSITIVITY ANALYSIS: WT-INFINITE BUS 23
3.1.4 Grid Injection

In the previous Sub-sections the results of adding a perturbation on each of the
WT controllers were presented. Perturbations to the grid side controller showed
the biggest impact on the active power output. This controller is therefore the most
likely to trigger oscillations between wind turbines and the grid.
The grid side controller is connected to the three phases on the grid side, as
stated in its name; it is therefore very likely to react and /or even amplify os-
cillations present in the power system. This additional test was also motivated
by confidential studies commissioned by OG&E, which suggested the injection of
oscillations directly at the bus level.
The injection of oscillations is performed directly at the bus level with pro-
grammable voltage sources. The voltage sources are fed with a balanced three
phase signal, which frequency and amplitude are configurable by the user. The
frequency content is presented in Fig. 3.4.
400 400 200
180
350 350 160
140
Amplitude (dB)
Amplitude (dB)
Amplitude (dB)
300 300 120
100
250 250 80
60
200 200 40
20
150 150 0
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
subject to injection of oscillations in the bus voltage
The results obtained show a strong frequency component at exactly 39.17 Hz,
which was not expected. This frequency is however exactly 50 Hz minus the injected
frequency of 10.83 Hz. This is due to the fact that the grid side controller uses dq0 -
quantities and the Park transformation of a signal composed of the fundamental at
50 Hz and an additional frequency f1 gives a frequency component of 50 ≠ f1 Hz.
An additional test was performed with the injected frequency set at 39.17 Hz, the
results are presented in Fig. 3.5.
The results show a strong frequency component at 10.83 Hz in both the active
and reactive power outputs, measured at 370 dB, see Fig. 3.5a and 3.5b. The voltage
in the DC link is also impacted, see Fig. 3.5c. The oscillation injected in the grid
voltage are the origin of the WT dynamic response and lead to an oscillation in
the dq-quantities, which are used to control the grid side converter, responsible for
notably the voltage control in the DC link.
The results of this Section indicate that the phenomenon of strong oscillations in
400 400 200

180
350 350 160
140
Amplitude (dB)
Amplitude (dB)
Amplitude (dB)
300 300 120
100
250 250 80
60
200 200 40
20
150 150 0
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
subject to injection of oscillations in the bus voltage
the active power generation with oscillations also present with smaller amplitudes
in the reactive power output of the turbine originate from the grid side controller.
The behaviour observed in the real measurements is recreated by either perturbing
the grid side controller or by injecting oscillations in the bus voltages. Provided
that the oscillations are injected at 50 ≠ f1 Hz the resulting behaviour is satisfactory
in the meaning it matches the actual power system measurements. It is also much
easier to scale up, which is the main reason for continuing with its integration in
bigger models.
The results also show an important fact: the WT controller is actually respond-
ing to higher frequency dynamics (39.17 Hz) that on-set those observed oscillations
in the network interfaces. the origin of such high frequency dynamics in actual grids
(such as operated by OG&E), is subject for further research.
3.2. NETWORK SENSITIVITY TO INJECTED OSCILLATIONS 25
3.2 Network Sensitivity to Injected Oscillations

The previous Section showed that the injection of oscillations to the network inter-
face bus will drive the wind farms control systems to interact with the network by
creating high frequency oscillations.
In this Section the response to a perturbation in the grid voltage at the PCC
is analysed in the power system model introduced in Section 2.2. This model does
not include any infinite bus and is equipped with loads, both the voltage and the
current in the power system will therefore be impacted. The results are presented
in Fig. 3.6.
The impact of grid injection on voltage is verified from the presented result. A
strong frequency component at 10.83 Hz can be observed on Fig. 3.6a. The active
power and reactive power show a strong frequency component at 10.83 Hz measured
at 418 dBm for both. This shows also that moving the injection of oscillations to
the PCC excites all the turbines of the farm, leading to power oscillations of greater
amplitude.
250 420 420
Voltage Active Power Reactive Power
400
400
200 380
380
360
Amplitude (dB)
Amplitude (dB)
Amplitude (dB)
360
150 340
340 320
100 300
320
280
300
50 260
280
240
0 260 220
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
(a) Voltage (b) Active Power (c) Reactive Power
Figure 3.6: PSD of the active and reactive powers and the voltage at the PCC when
subject to injection of oscillations in the voltage at the PCC
The overall behaviour of the power system during the injection of oscillations
bears a strong resemblance to the real measurements obtained from OG&E. This will
allow for adequate modelling and integrated testing using the real-time hardware-
in-the-loop environment of the Monitoring Tool, presented in Chapter 4.
3.3 Summary
In this Chapter a perturbation analysis of the wind turbine controls was performed,
probing each controller with similar types of perturbations. The perturbation ap-
plied to the grid side controller led to results similar those observed with the real
measurements obtained from OG&E.
Another approach to perturb the grid side controller was investigated with in-
jection of oscillations directly in the bus voltage. This approach proved to be more
realistic and adequate for scaling up the phenomenon to an entire wind farm. The
power system model with these injections will be used for testing the Monitoring
tool in its real-time version.
This approach also highlighted the fact that higher frequency dynamics create
the observed oscillations. The origin of those higher frequency dynamics would
require further research, that is the most important conclusion of this analysis.
Part II
Monitoring Tool
27
Chapter 4
Fast Real-Time Oscillation Detection

Tool
The main focus of the Thesis work is the phenomenon of power system oscil-
lations at higher frequencies than the traditional ones (inter-area modes, local
modes). As these frequencies are much higher, the traditional measurement equip-
ment (SCADA) is unable to detect it. Therefore the tools for detecting and analysing
this kind of phenomenon are nearly non existent.
In this Chapter, an adaptation of existing work is proposed and implemented in
LabView, in a process of building a proof of concept for detecting such oscillatory
phenomena by using PMU measurements.
29
30 CHAPTER 4: FAST REAL-TIME OSCILLATION DETECTION TOOL
4.1 Oscillation Detection and Frequency Estimation

Working with real-time measurements always implies a compromise between the
precision of the information extracted and the quantity of measured data. Therefore,
one key aspect to consider is the time required to acquire the measurements and
process them. Since the effects caused by the phenomenon of oscillation in the
frequency range of 10 Hz to 15 Hz have been reported in [6], the main goal is to
build a detection tool with a fast response time.
Regarding the traditional power system oscillations, the important parameters
for a system operator to monitor are the frequency and the damping for each oscilla-
tory mode. The strategy generally adopted is composed of two separate algorithms,
each of them dedicated to the estimation of one parameter. This is the strategy,
which is adopted in this case, with one algorithm dedicated to the estimation of the
amount of energy in the oscillations and the other dedicated to frequency estimation.
4.1.1 Fast Oscillation Detection

Given the higher frequency of the studied oscillations, the purpose of the algorithm
remains to detect the presence of oscillations and their "activity". A literature study
revealed that an algorithm for detecting oscillations fulfilling the set criteria had
been developed by Hauer [18]. The first section describes this algorithm and the
second emphasizes the changes performed to adapt it to the specific requirements
from the implementation choices.
An Algorithm for Fast Detection of Oscillations

The original algorithm presented in detail in [18] was developed as a complementary
mean of detection for low frequency detection in the Bonneville Power Administra-
tion (BPA) power system in 1987. At that time the existing monitoring devices
failed to trigger fast enough recordings of some oscillatory events in the power sys-
tem.
The basic principle of the algorithm is to detect the size of the envelope of the
signal that gives an approximation of the amplitude of the oscillations. The complete
algorithm is depicted in Fig. 4.1 and the processing steps are the following:
1. Low-pass filter #1 removes unnecessary content from the signal (unwanted

higher frequencies). It may also provide anti-aliasing if the signal is down-
sampled.
2. The high-pass filter formed by low-pass filter #2 and a delay allows to extract
the fast varying part of the signal. Low-pass filter #2 estimates the trend of
the signal to be removed from the delayed copy of the signal.
3. The absolute value provides the magnitude of the variations.

4.1. OSCILLATION DETECTION AND FREQUENCY ESTIMATION 31
4. Low-pass Filter #3 smoothens its input and therefore provides an indication

of persistence of the activity in a given frequency range.
5. The trigger level comparator acts as a flag indicating an activity higher than
a set threshold.
For reliability purposes, this algorithm was designed with Finite Impulse Re-
sponse filters (FIR), which only perform a computation of the current output as the
weighed sum of the previous inputs. This ensures an intrinsic stability of the filters
and therefore provides reliability to the algorithm.
SUMMING
DELAY
LP FILTER + ABSOLUTE LP FILTER
INPUT
INPUT #1
- VALUE #3 LEVEL
COMPAR
TRIGGER
TRIGGER
LP FILTER FLAG
FLAG
#2 ISON
TRIGGER
TRIGGER LEVEL
LEVEL
HP FILTER
Figure 4.1: Diagram of the algorithm for oscillation detection proposed by Hauer [18]
Adaptation of the Algorithm

As mentioned in the previous Section the algorithm was designed for detecting ac-
tivity among low frequency oscillation, which is not the objective of this Thesis.
There are however examples of adaptations to faster dynamics oscillations in the
literature in [19], which are directly inspired from the algorithm mentioned previ-
ously. Considering the existing examples, an new adaptation is proposed, that takes
into account the equipment of the lab and the length of the Thesis project.
The spectrum of frequency of interest with potential oscillatory activity starts
from 0.1 Hz and the maximum frequency, which can be observed, is set by the
sampling frequency of the PMUs. In the set-up of the lab, the PMUs were set to
measure 50 samples per second, the highest observable frequency is thus 25 Hz.
To cover such a broad span of frequencies, four instances of the algorithm can be
executed in parallel. Thus, the algorithm is duplicated in four copies, each of them
being independently configured to monitor one particular range of frequency. The
following ranges are suggested for the configuration:
• 0.10 Hz - 1 Hz: Inter-area modes, e.g. system-wide electromechanical swings.
• 1 Hz - 5 Hz: Local-area modes, e.g. internal plant swings.
• 5 Hz - 15 Hz: High frequency oscillations, e.g. wind farms controller interac-

tions.
• 15 Hz - 25 Hz: Other sub-synchronous oscillations, e.g. sub-synchronous

resonance.
In the original algorithm, the first filter was a low-pass filter, which role was
to remove unnecessary content from the input signal. In that case the targeted
frequency range was all the frequencies below 1 Hz. In order to accommodate each
frequency range previously mentioned, the first filter was replaced by four band-pass
filter set to the boundary frequencies of each range.
SUMMING
BANDPASS DELAY + LP FILTER
FILTER  
2
MOVING  
0.1 – 1 Hz
LP FILTER
- AVERAGE
LEVEL
COMPA
TRIGGER
TRIGGER
FLAG
FLAG
TRIGGER LEVEL RISON
HP FILTER
SUMMING
FILTER  
2
MOVING  
1– 5 Hz
LP FILTER
- AVERAGE
LEVEL
COMPA
TRIGGER
TRIGGER
FLAG
FLAG
TRIGGER LEVEL RISON
INPUT
INPUT HP FILTER
PRE
VOLTAGE
VOLTAGE PROCESS SUMMING
PHASOR
PHASOR
FILTER  
2
MOVING  
5 – 15 Hz
LP FILTER
- AVERAGE
LEVEL
COMPA
TRIGGER
TRIGGER
FLAG
FLAG
TRIGGER LEVEL RISON
HP FILTER
SUMMING
FILTER  
2
MOVING   LEVEL
15 – 25 Hz
LP FILTER
- AVERAGE COMPA
TRIGGER
TRIGGER
FLAG
FLAG
TRIGGER LEVEL RISON
HP FILTER
Figure 4.2: Diagram of the adaptation of the algorithm for fast real-time oscillation de-
tection
The output of the high-pass filter remains the "activity" of the selected frequency
range. The Root Mean Square (RMS) value of this output is used for energy com-
putation that implies the following computations sequentially: squaring, averaging
and finally computing the square-root of the signal. The low-pass moving average
filter is used to extract the main trend of the squared signal. This is necessary
so that a persistent and stable signal is provided to the forthcoming trigger level
comparison. In this adaptation the algorithm is fed by voltage measurements from
PMUs and the RMS calculation provides a good approximation of the RMS energy
of the oscillations for each given frequency range. Finally a trigger level comparison
indicates if the computed energy exceeds a pre-set level.
A step of preprocessing is added, which is necessary when using PMU data to
ensure data quality.1 The complete adaptation of the original algorithm proposed
1
Down-sampling is not performed in this algorithm in order to avoid lowering the reaction
times of the algorithm.
to fit the purpose of fast real-time oscillation detection is shown in Fig. 4.2.
4.1.2 Frequency Estimation

The frequency estimation algorithm, depicted in Fig. 4.3, comprises two different
algorithms running in parallel, leaving to the user the choice to activate one of
them or both simultaneously. The reason is that a non-parametric method, which
is described below, can be used when the frequency of the oscillations within a
certain range is unknown, whereas a parametric method, which is described later,
is appropriate when the knowledge about the number of possible oscillations within
a range is known.
PARAMETRIC
AR SPECTRAL
ESTIMATION (YW)
INPUT
INPUT PRE BAND-PASS
PHASOR
PHASOR PSD
PSD
PROCESS FILTER
SIGNAL
SIGNAL
NON-PARAMETRIC
WELCH’S METHOD
Figure 4.3: Diagram of the algorithm for Spectral Estimation
Non-Parametric: Welch’s Method

Welch’s power spectrum estimation is a method based on the standard Periodogram.
This method is mostly used in fields of applied sciences, where the length of the
signals are not infinite, and thus, spectral estimation is subject to noise. The aim
is to increase the readability of the Power Spectrum Density (PSD), reducing the
noise while decreasing the frequency resolution.
The method works as follows:
1. The input signal is split into overlapping segments of length M (the overlap-
ping rate is set by the user).
2. Each segment is windowed (the window is chosen by the user).
3. The Fast Fourier Transform (FFT) is computed for each windowed segment.
4. The PSD is obtained by averaging all the resulting spectra, thereby reducing
the final variance.
As mentioned above, several parameters for this method are left for the user
to choose, allowing wide tuning perspectives to adapt the PSD calculation to any
signal quality.
However since this method is based on FFT computation, it will highlight all
the content of the power frequency spectrum, which is an intrinsic property of non-
parametric estimation. This property is especially useful for signals, which actual
frequency content is unknown or only partially known.
For a complete description of the method, refer to Peter Welch’s original arti-
cle [20].
Parametric: Auto-Regressive Methods

Parametric methods for estimating power spectral density use a auto-regressive
mathematical model of the input signal. In this way they integrate available knowl-
edge about the input signal to improve spectral estimation.
Auto-regressive models have been applied for estimating power system frequency
content, as described in [21]. For this specific reason this model is chosen in the
Monitoring Tool. The specific method utilized is, however, left to the choice of
the user. All of the implemented methods work on the principle of model fitting.
They diverge however on the method used to optimize the fitting process. For
further details on the parametric methods, refer to [22] and for further details
on their application to power systems, refer to [23]. It is worth mentioning that
during experimentations, the Burg Lattice method, available in LabView, has been
particularly efficient.
The order of the model is a very important parameter. It’s value cannot be fixed
in advance because it is mainly dependent on the number of peaks to be identified,
and thus, the frequency content itself. In the case of the Monitoring Tool where the
frequency spectrum of interest is divided into four frequency ranges, the order might
be different in each of the ranges. It is worth mentioning that too small orders will
lead to a smoothen spectrum and some modes might be left unidentified. On the
other hand, too large orders might lead to the identification of artificial modes and
their appearance on the estimated spectrum.
4.1.3 Data Pre-Processing

The algorithms for estimating the energy and frequency of an oscillatory phe-
nomenon described in the two preceding Sections involve filtering, averaging and
spectral estimation processes. Filtering and averaging processes use past samples
for computing the output, thus, errors in the input measurements may be propa-
gated. Hence these processes require reliable data and no missing measurements.
The spectral estimation process uses techniques coming from signal analysis theory
and have been developed to provide results from ambient data.2 It is then natural to
assume that better results will be achieved with a stochastic signal for the spectral
estimation.
This particularly introduces requirements on the original data, which in this
case are measurements acquired either in the lab or recorded by system operators.
2
Ambiant data is considered as a stochastic noise resulting from random load variation
In the case of PMU, the measurements are performed at a rather high frequency
(generally 25, 30, 50 samples per second) and transmitted over IP networks leading
to wrong or missing measurements. The pre-processing of input data thus appears
necessary to satisfy the requirements imposed by the methods used in the detection
tool.
The pre-processing implementation was inspired from former work on PMU data
pre-processing available in [24] , however some changes were performed, considering
that the original pre-processing algorithm used archived data and the Monitoring
Tool processes the measurements as they are received.
Outlier Removal - Interpolation

The outlier removal algorithm suggested is the result of an adaptation of the original
algorithm [24] to the LabView environment and therefore is tied to the LabView
language. The basic principle is explained in this Section.
In general, the variation of frequency in a power system is the result of the in-
teraction between varying loads and the generation that follows the same variations
with an inertial delay. It should thus be a rather smooth process that implies that
any value is expected to be within a Confidence Interval, as shown in Fig. 4.4. It
can be determined from neighbouring values and intrinsic system properties. The
performance of the outlier removal is determined by the level of confidence of the
interval used.
50.1
50.08
50.06
50.04
Frequency (Hz)
50.02
50
49.98
49.96
49.94
Lower limit
49.92 Upper limit
Frequency
49.9
1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8
Time (s)
Figure 4.4: Example of a Confidence Interval for the frequency
The Confidence Interval suggested here is built according to the following steps
(depicted as a diagram in Fig. 4.5)
1. The input signal: (f ).
2. Create one copy (A) of the input signal delayed by 10 samples: Ai = fi+10 ,
where i is the time index.
3. Filter one copy (Mq

) of the input signal with a moving average filter of order
f
equal to 20: Mi = 0<t<20 i≠t
20 .
4. Subtract (M ) from (A). The combination of a delayed copy (A) and (M ) is

equivalent to a having a moving average filter, where the average is computed
on the 10 preceding samples and 9 following samples. Compute the standard
deviation ‡ of the resulting signal.
5. Create the upper (U ) and lower (L) limits of the Confidence Interval by
adding/subtracting k‡ to (M ), k the sensitivity factor, is set by the user.
INPUT
INPUT MOVING
INTERPOLATE
INTERPOLATE
FREQUENCY
FREQUENCY AVERAGE +
+ SAMPLE
SAMPLE
SIGNAL
SIGNAL F: 10, B:9
-
+ σ X
+
-
NO
kk INTERVAL WITHIN?
YES
KEEP
KEEP
TEST SAMPLE
SAMPLE
Figure 4.5: Diagram of the algorithm for Outlier Removal
Apart from the algorithm to build a reliable Confidence Interval, the Outlier
Removal operation is quite straight forward. Each element of (A) is compared to
(U ) and (L). If Ui > Ai > Li the element is preserved, else it is dropped and will be
replaced by an interpolated value. The interpolation is done by a built-in function
of LabView, which takes as inputs the elements of the signal, the indexes of theses
elements and all the indexes at which an element will be returned. Each returned
element is either an input element if its index is on both lists or an interpolated
value if its index is only on the second list. The chosen method of interpolation
is linear interpolation, as it gave the best results during testing. Testing results
showed that it has no divergence risks compared to methods involving higher order
polynomials.
An example of the Outlier Removal algorithm in action on real-time PMU data
is shown in Fig. 4.6. Because the screen shots were done on real-time data, there
is a small delay between the two plots. The data points removed are single points
far from the neighbouring values. Outlier Removal effectively performs the tasks
described above.
(a) Outlier Removal not active (b) Outlier Removal active
Figure 4.6: Screen shots of the Frequency signal with/without Outlier Removal, the com-
mon part of the input signal is highlighted in red
Filtering - Down-sampling
After proceeding with Outlier Removal, the signal is sufficiently pre-processed for
the oscillation detection step. It is thus directly sent to the detection algorithm,
which, according to Fig. 4.2, will take care of removing unnecessary frequency con-
tent by filtering. The frequency estimation algorithm require however further pre-
processing, which are carried out directly at the level of the frequency estimation
algorithms.
As mentioned earlier, the frequency estimation algorithm requires stochastic
data to work properly. That is the reason for not considering the frequency content
resulting from the more deterministic actions of the governors of the generators.
Such frequency content is located very close to 0 Hz and lower than any electro-
mechanical mode, so they will be attenuated by mean removal and low-pass filtering.
Furthermore, according to Shanon’s theorem, all the frequency content below fre-
quency f1 can be restored if the sampling frequency is at least 2 ◊ f1 . The typical
sampling frequencies of PMUs are 25, 30 or 50 Hz, which give the highest observ-
able frequencies 12.5, 15 or 25 Hz. The observed frequency in the tool can be much
lower, a step of down-sampling is added to remove redundant data. This step is
preceded by a low-pass filter acting as an anti-aliasing filter.
The implementation was done using a band-pass filter in a Filtering and Down-
sampling algorithm added to each of the Oscillation Detectors in the aforementioned
frequency ranges. The down-sampling factor is calculated from the cut-off frequency
of the band-pass filter, in regards to Shanon’s criterion. An upper limit equal to 10
was added to the down-sampling factor in the implementation. This was motivated
by the limited size of the buffered data, which could lead to too few samples for the
frequency estimation to perform well after down-sampling by a larger factor.
4.2 Implementation in LabView

The Monitoring Tool has been implemented as a real-time graphical tool, which
could be used by system operators or wind farm owners. The choice of LabView
was made partly because of the existing real-time data mediators, see Section 5.3
and partly because designing a graphical user interface is a straightforward process
in this programming language.
The original idea was to build a tool consuming real-time data from PMU/PDC
streams. At this time no recorded measurements from TSOs were available, but
later during the project some recorded measurements were received from OG&E,
leading to the development of the Replay Tool, in order to exploit them. First the
Monitoring Tool is presented in details, followed by the Replay Tool.
4.2.1 Real-Time Monitoring Tool

The Monitoring Tool has been developed as a real-time tool to detect oscillations
occurring in the power system. It is fed with PMU measurements, which usually
have a sampling rate of either 25, 30 or 50 samples per second, allowing monitoring
of oscillatory events at a frequency up to 25 Hz. Such a frequency range contains
different categories of phenomena, which are mentioned in Section 4.1.1. The Mon-
itoring Tool therefore presents a redundant interface with some elements in four
instances, grouped under the name Module.
As mentioned earlier, the Monitoring Tool has been developed with the objective
of providing an intuitive visual tool. The resulting GUI is depicted in Fig. 4.7.
The graph on the top-right of the interface is a simple representation of the
buffered input signals received by the Monitoring Tool. The signals are displayed
one at a time, and the current signal can be chosen from the list box on top of the
graph in Fig. 4.7 (the voltage magnitude is displayed in this case). This graph is
for example useful for verifying the tuning of the outlier removal algorithm.
The top left part of the GUI is dedicated to the configuration of the tool and
divided in five tabs. The first tab "Options" gathers the general configuration of
the tool as well as state LEDs, see Fig. 4.8. The remaining tabs "Module 1 to 4 "
are four identical tabs dedicated to the configuration of the processing modules, see
Fig. 4.9.
The general configuration should, if needed, be changed at launch of the tool.
The following elements are available, together with their action:
• PSD Data Source: Choosing the data source for all the frequency estimation
algorithms.
• Fs: Specifying the sampling frequency of measurements.
• Buffer Parameters: Changing the parameters of the buffer.

-Size: Number of elements in the buffer.
-On?: Activating the outlier removal.
4.2. IMPLEMENTATION IN LABVIEW 39
Figure 4.7: Screen shot of the interface of the Monitoring Tool

-Sensitivity: Changing k, see Section 4.1.3.

-Method: Choosing the method for interpolation, linear is preferable, refer
to Section 4.1.3.
• Activate PRL Read: Activating the delivery of measurements from the PRL.
• Data Selector: Displaying the Data Selector front panel, mainly used at initial
launch.
• Activate Filters: Activating the oscillation detection algorithm.
• Threshold: Configuring the three trigger levels, see Section 4.1.1.
• Refresh Rate: Changing the refresh rate of the interface.
• Status: Messages received from the PRL.
• PRL Visible: Displaying PRL front panel.
Figure 4.8: Screen shot of the Configuration tab of the Monitoring Tool
After launching and initializing of the tool the user should configure the modules.
As explained in Section 4.1.1, the tool has been implemented with the purpose of
monitoring four frequency ranges simultaneously. There are therefore four modules
on the main window (see Fig. 4.7) each having its own configuration tab as shown
in Fig. 4.9.
Figure 4.9 presents an interface with two separated blocks. The block on the left
is dedicated to the configuration of the band-pass filter, the parameters are common
for both the Oscillation Detection and the Frequency Estimation algorithms. The
block on the right is dedicated to the Frequency Estimation algorithm and contains
the parameters for each of the methods mentioned in Section 4.1.2. It also includes
Figure 4.9: Screen shot of the configuration tab of one module Monitoring Tool
the parameter for the spectral averaging. Finally the user is able to choose in the
list box if both methods are to be used simultaneously or just one of them.
Once the configuration is set, the outputs of the algorithms are displayed in a
Module, depicted in Fig. 4.10. The graph on the right presents the power spectrum
density, which is the output of the Frequency Estimation algorithm. It is scaled
according to the parameters of the band-pass filter, highlighting the content of
interest.
The graphical display on the left has several components, the most important
are the three LEDs Low Activity, High Activity and Danger!!!, corresponding to the
trigger flag of the three thresholds defined in the general configuration tab. The
comparison is made on the latest sample, which value is mentioned by the field
Energy. The graph provides a history to easily corroborate the energy computed
with the input signal displayed. Given that the input signal is filtered three times
by FIR filters, the output signal is shorter than the input signal. Moreover the order
of the filters have a strong influence on the length of the output signal. This can be
changed by the user to adapt to their needs, measurement features and particular
network.
Figure 4.10: Screen shot of one Module of the Monitoring Tool

4.2.2 Replay Tool

In the previous Section, the Monitoring Tool has been presented and several con-
figuration options have been highlighted. While some options do not require any
knowledge of the power system studied and can therefore be tuned directly, others
have to take in consideration the properties of the power system, especially for the
Frequency Estimation parameters. The tool thus needs calibration. There are sev-
eral ways to perform calibration for this tool. One could tune the parameters using
prior knowledge, however it could be convenient to make use of archived data.
The original idea was to replay archived data into a PDC, broadcast it as an
Output Stream and use the Monitoring Tool. The diversity of archive techniques
and file structures raised the issue that a PDC, such as OpenPDC, was not flexible
enough and more specifically not adapted to the archived data format available.
The Replay Tool has been developed using the same code as the real-time tool.
However, additional software has been developed so that it can use archived data
instead of using live PMU streams. The interface is almost identical to the Mon-
itoring Tool, as shown in Fig. 4.11. The processing algorithms are identical. This
tool has however some specific features. For example it allows to scroll along the
replayed data, which can be a useful feature to get a quick overview of the content
of the selected file.
The general configuration tab is also different, because most of the options were
related to the PRL and its control. Among the remaining options only one is new
namely Path, which lets the user fill in the path of the file to be replayed and select
a particular archive, see Fig. 4.12.
The Replay Tool works otherwise exactly like the Monitoring Tool. During
this Thesis work, this tool has been used to replay data from OG&E presenting
oscillatory events at a frequency of approximately 13 Hz. This served as a test to
verify that the algorithms worked properly. The screen shot in Fig. 4.11 has been
taken while replaying one of the OG&E archives. The beginning of the oscillatory
event can be seen, as well as the red flag on the Module dedicated to frequencies
between 10 Hz and 15 Hz (bottom right on the figure). Additionally the oscillatory
frequency is identified to 13.4 Hz by the Frequency Estimation algorithm.
CSV File Reader
The files received from OG&E are CSV files (Comma Separated Values), or in other
terms a text file where a two-dimensional array is stored line by line, separated by
a return character. In one line the columns are delimited by a comma. This type of
file does not provide any information on its content, neither a structured content. It
has however the advantage of being very easy to read and convert into processable
data.
The received files contain the time stamps in the first column; this column is
thus treated separately as a string and sent to an algorithm to convert strings into
the LabView Timestamps data-type, while the other columns are decoded as real
Figure 4.11: Screen shot of the interface of the Replay Tool

Figure 4.12: Screen shot of the Configuration tab of the Replay Tool
numbers. The columns corresponding to the voltage angle and the current angle are
converted from degrees to radians to build the voltage phasor and current phasor.
Finally the data is sent as an output containing several arrays, each corresponding
to one measurement (time stamps, voltage phasor, current phasor, frequency). The
algorithm is depicted in Fig. 4.13.
SELECT
DECODE AS
FIRST
TIME STAMP
COLUMN
INPUT
INPUT READ OUTPUT
OUTPUT AS
AS
CSV
CSV FILE
FILE FILE ARRAYS
ARRAYS
SELECT IF ANGLE
DECODE AS BUILD
OTHER CONVERT
DOUBLE PHASORS
COLUMNS TO RADIAN
Figure 4.13: Diagram of the algorithm to read CSV files
It is important to mention that this CSV File Reader was developed to read the
files with the data format used by OG&E. The algorithm would therefore require
changes to be adapted to other data format.
File Reader Enhancements

As mentioned in the previous Paragraph, the CSV File Reader has been developed
to suit the data format used by OG&E and can therefore only read one type of data
format. Some adaptations have been implemented to be able to decode American
or European time stamp strings, for example. However, the code lacks flexibility
and requires the time stamps to use one of the three formats considered during the
implementation.
The Replay Tool has gained a strong interest for its ability to replay archived
data and brought new ideas for future development. The file reader therefore has
to be improved, and several options are still under consideration.
Statnett SF, the Norwegian TSO, which is a strong partner of the lab, uses
intensively LabView for its research and development activities. The PMU mea-
surements they record are saved using the LabView TDMS data storage system,
which provides a flexible database system. Given that Statnett SF has provided
data for analysis to the lab, a future version of the file reader will certainly support
the TDMS file structure.
4.3 Summary
In this Chapter, all the algorithms used in the Monitoring Tool were introduced.
Their modification, necessary for an implementation in LabView, were also ex-
plained . Finally the final graphical user interface of the Monitoring Tool was
presented, as used for experimentation, testing and validation, which are presented
in Chapter 6.
Chapter 5
Instrumentation and Data Acquisition
The models presented in Chapter 2 are simulated using a real-time target with
analog voltage outputs configured. The Monitoring Tool presented in Chapter 4
detects oscillations using PMU measurements. In order to connect both of the
previously mentioned systems, a chain of real-time data acquisition must be installed
and configured.
While installing and configuring this chain of data acquisition is not the primary
objective of the Thesis work, it does play a crucial role. In this Chapter, all the
elements of the chain of real-time data acquisition are presented and the important
steps for a functioning configuration are detailed.
47
48 CHAPTER 5: INSTRUMENTATION AND DATA ACQUISITION
5.1 Compact RIO Configuration as PMUs

The CompactRIO (cRIO) is a reconfigurable input and output system from National
Instruments (NI), which combines a traditional controller and reprogrammable in-
puts/outputs via a Field Programmable Gate Array (FPGA). Each input/output is
accessible via a serial port in which a National Instruments’ module can be inserted,
offering a wide range of possibilities, considering all the modules available.
The reconfigurable aspect of the device was important in the choice of the sys-
tem for building a PMU, because the purchased equipment can then be re-used in
another project if needed. It also enables the research team to develop it’s own
PMU software to install on the cRIO, leading to a full knowledge of the utilized
PMU algorithm. For the purpose of this Master’s Thesis a pre-compiled version of
the NI PMU software was used.
The following equipment was used to build a PMU (as shown on Fig. 5.1):
• NI Compact RIO 9074.
• NI Voltage measurement module 9225.
• NI Current measurement module 9227.
• NI GPS receiver module 9467.
• Compiled PMU software image from NI.
Figure 5.1: NI Compact RIO 9074 with modules inserted in PMU a set-up
As mentioned previously, the PMU software was provided by NI as a pre-

compiled software image. The installation process consisted in downloading the
image onto the cRIO and rebooting the device to run the freshly installed software.
The remaining configuration is done on the web interface of the PMU software,
therefore it is important to ensure that the network connection settings are correctly
set in the cRIO.
5.1. COMPACT RIO CONFIGURATION AS PMUS 49
5.1.1 CompactRIO Configuration

The newly assembled PMU must be integrated into an existing data network to
ensure communication with the computer used for configuration, as well as com-
munication with the existing PDC servers of the lab. The cRIO is equipped with
automatic network settings (DHCP mode), as required by the network environment
at KTH.1
The following steps are needed to set/refresh the IP settings of cRIO (see
Fig. 5.2):2
1. Open NI PMU Configuration Utility executable.
2. Refresh the list of available NI cRIO on the network.
3. Select the desired unit.
4. Query the current settings of the selected unit.
5. Check that DHCP/Link-Local is selected in the field IP Adress Request Mode.
6. Press Change IP Settings to apply and refresh the IP settings of cRIO.
1) Refresh the list of 3) Query the current 5) Press ”Change IP settings” to refresh the
available equipment settings of the selected unit network configuration of the unit selected
2) Select a Unit 4) Check that ”DHCP/

Link-Local” is selected
Figure 5.2: Steps for configuring cRIO with DHCP
1
Automatic network settings will suit most of the network environments, but for specific cases
refer to your network administrator about the necessary settings for static IP configuration.
2
A reboot of cRIO can help refreshing the IP settings if the previous steps were not sufficient.
5.1.2 Configuration of the PMU settings

The cRIO PMU is now reachable through standard network communication and
the configuration can be performed through the web interface of the PMU software
available at: http://x.x.x.x:8000/PMURemoteFrontPanel.html, were x.x.x.x is the IP
address of cRIO (using Internet Explorer is highly recommended for configuration).
Once the URL is validated, the control of the remote front panel is acquired
by right clicking on the panel and selecting Request Control of VI. The settings of
the PMU communication are in the tab /Configuration/Base Config/. The station
name and PMU ID must be set to unused values in the lab, but the other parameters
can be set to the same values as in Fig. 5.3a. The two channels available will allow
two separate outputs to two different PDC severs.
The phasors transmitted according to the outputs selected in the tab /Config-
uration/Phasors/. For this project the following were selected (see Fig. 5.3b):
• Positive Sequence Voltage V+.
• Positive Sequence Current I+.
• Voltage Phase a, b, c, respectively Va, Vb, Vc.
• Current Phase a, b, c, respectively Ia, Ib, Ic.

None of the analog values were selected in the tab /Configuration/Analog/.
Other more advanced parameters, like active and reactive power computation, are
available in the remaining tabs, but can be left to default settings.
(a) Base Config tab (b) Phasor tab
Figure 5.3: Screen shot of the Web interface of the PMU
The cRIO is now fully configured as PMU and ready to be integrated to a

PDC. The configuration of the outputs can be check with a software such as PMU
Connection tester, see Section 5.2.3. The integration of the PMU with the reminder
of the existing lab equipment requires the following steps:
• Wiring of the voltage measurement module (see Fig. 5.4).
5.1. COMPACT RIO CONFIGURATION AS PMUS 51
• Wiring of the current measurement module (see Fig. 5.4).
• Wiring of the GPS receiver to a GPS antenna.
• Adding the PMU to the PDC servers of the lab, see Section 5.2.
Figure 5.4: Wiring scheme of voltage and current measurement modules
For the purpose of the experiments performed in Chapter 6, the modules are
connected to the outputs of the simulator as shown in Fig. 5.5
Figure 5.5: Voltage Module 9225 wiring for connection to the simulator
Figure 5.6 presents the complete set-up during an experiment involving HIL
simulation. The following elements are highlighted on the figure:
1. The real-time target from Opal-RT.
2. An oscilloscope connected to an analog output of the target, showing the

three-phase signal produced by one wind farm.
3. A cRIO PMU connected to the analog output aforementioned.
4. The connection cable of the PMU for voltage measurements, connected to the
back of the simulator, as shown in Fig. 5.5.
5. The network connection of the PMU to stream the data to the PDC server of
the lab.
6. The PDC server of the lab receiving the PMU data and broadcasting an
Output Stream to itself for this photo.
Figure 5.6: Photo of the set-up of the experiment at SmarTS Lab

5.2. PHASOR DATA CONCETRATORS 53
5.2 Phasor Data Concetrators

The concept of Phasor Data Concentrator (PDC) was introduced in Section 1.2.
There are however several vendors of PDC server software, leading to a wide variety
of graphical user interfaces. The lab is equipped with OpenPDC and SEL-5073
Synchrowave server. Even if both of them have the same role, their graphical
interface differ strongly, a step-by-step guide is provided for both of them, describing
the tasks required for the experimentation set-up:
1. Add a PMU to the PDC.
2. Create an Output Stream containing the measurements of the input PMU.
For the following examples, the PMU with the configuration presented in Table 5.1
will used.
Table 5.1: Example Configuration of a PMU

Name PMU-KTH
PMU ID 6
IP Address 130.237.53.75
Port TCP: 4712(openPDC)
or 4718(SEL)
Data Rate 50 Samples/s
5.2.1 OpenPDC
OpenPDC is an open-source software maintained by Grid Protection Alliance. It
is available at http://openpdc.codeplex.com/, additionnal documentation can also
be found on the website, especially the installation procedure. Once installed, the
configuration is done with OpenPDC manager, see Fig. 5.7.
Add a PMU to the concentrator
The prior steps required for the configuration of a new input to OpenPDC is to
create a connection file and a configuration file with PMU Connection Tester, as
explained in Section 5.2.3 with the connection information from Table 5.1.
A new PMU is added by clicking on Add New Device in the Quick Links rubric.
It will launch a wizard, see Fig. 5.8. In the first step the connection file must be
provided, see Fig. 5.8a, then the configuration file, see Fig. 5.8b. The last step is
simply to acknowledge the configuration of the new PMU added. The PMU is now
ready to be used for an Output Stream.
Figure 5.7: Screen shot of the main window of OpenPDC Manager
(a) Select connection file (b) Select configuration file
Figure 5.8: Screen shots of the wizard for adding a new PMU
Create an Output Stream

From the Home screen of the main window, the quick link Manage Output Streams
will lead to the window to create a new Output Stream. Pressing Clear will wipe
the form and allow the creation of a new Output Stream. The information required
are to be set by the user, see Fig. 5.9a. The Port number and ID Code can not be
values already used by the other Output Streams configured on the server.
The measurements are added to the stream by clicking on Device Wizard, where
the user is prompted for devices to include in the stream, see Fig. 5.9b. By default
in OpenPDC, all the measurements associated with one device are added to the
stream, it is preferable to leave it as such. Finally the stream is initialized by
pressing the button initialize.
(a) Create an Output Stream (b) Add measurements to the Stream
Figure 5.9: Screen shots of the wizard for creating an output
5.2.2 SEL-5073 Synchrowave

SEL-5073 Synchrowave is a commercial product from Schweitzer Engineering Labo-
ratories (SEL). It is a very stable and robust PDC, with a very simple and intuitive
user interface, see Fig. 5.10.
Add a PMU to the concentrator

The Input menu is accessible in the menu on the left of the main window by a
simple click. A new unit can be added by clicking on Add PMU, that will add a
new unit ready to be configured by filling in all the fields, see Fig. 5.11a, with the
configuration information from the PMU /see Table 5.1). Any modification in the
configuration of SEL PDC will trigger a change of state of the PDC Sync box, see
Fig. 5.10, indicating that a synchronisation is required. The configuration of the
new unit is saved by pressing the button Send Settings.
The server will now initiate the connection to the PMU. It can be verified in
the Real-time menu. Once the connection established the tags of the PMU can
Figure 5.10: Screen shot of the SEL-5073 Synchrowave interface
(a) Input Menu (b) Output Menu
Figure 5.11: Screen shots of the sub menus of SEL-5073 Synchrowave
be queried, back on the Input sub-menu and the configuration must be saved once
again. The PMU is now fully configured in SEL PDC.
Create an Output Stream

Output Streams can be configured in the Output menu, on the left of the main
window. Similarly to the Input menu a new Output Stream is created by clicking
Add Output, which needs to be configured, see Fig. 5.11b. The parameters are left
for the user to fill in. The Port number and PDC ID, can not be similar to values
already in use in other Output Streams. The desired quantities to make available in
the stream are added by clicking the Edit button. The frequency corresponding to
the added measurements will be automatically included in the stream.
5.2.3 PMU Connection Tester

PMU Connection Tester is a software also maintained by Grid Protection Alliance,
its purpose is to test a connection to either a PMU or a PDC Output Stream. It
is available at http://pmuconnectiontester.codeplex.com/. It has been very useful
for several steps of the configuration of the experiment, but also has a key role in
the configuration of OpenPDC, simplifying the procedure for adding a PMU.
The connection information must be entered in the main window of the software
and then the connection can be initiated by pressing connect, see Fig. 5.12a. Once
the connection is initiated, the connection file and the configuration file can be saved
through the File menu, see Fig. 5.12b. Those file are used when adding a PMU into
OpenPDC, see Section 5.2.
(a) Main window (b) Save menu
Figure 5.12: Screen shots of the interface of PMU connection Tester

5.3 Real-Time Data Mediators

PMU Recorder Light and BableFish
The implementation of the Monitoring Tool was carried out in LabView for its
simplicity when creating graphical user interfaces. There is however no standard
LabView VI to communicate with PMUs or PDC servers using the IEEE C37.118.2
protocol [10] and bring PMU measurements in the programming environment. PMU
Recorder Light (PRL) from Statnett SF [25] and BableFish from SmarTS Labs [26]
are two real-time data mediators filling this gap. BableFish is under development
stage, therefore only a beta version of the adaptation of the Monitoring Tool using
BabeFish was developed. Also because BableFish may be subjected to several
changes in its LabView working scheme, the Monitoring Tool was mainly developed
using the PRL.
(a) Main window (b) Configuration window
Figure 5.13: Screen shots of the PRL
The PRL has it own interface in LabView, allowing configuration of the con-
nection to the output stream of the PDC, see Fig. 5.13b, and a quick control of its
functioning behaviour through several LED indicators and gauges, see Fig. 5.13b.
The data available as input comes from the Phasor Data Concentrator (PDC) as
a concentrated and time-synchronized stream of data from one or several PMUs
(see Section 5.2 for further description of the PDC server). A PMU or PDC can
provide Phasor data (voltage, current phasors), Analog data (frequency, active and
reactive power) and Digital data (user defined triggers). In most cases, several sets
of measurements are available for each data type. They are decoded from the IEEE
C37.118.2 protocol [10] and presented as a list of sets by the PRL, according to
their data-type, see Fig. 5.14. Once selected in that list, they are made available as
output from the PRL SubVI, which can be integrated into a LabView application,
e.g. the Monitoring Tool. The PRL SubVI is presented in Fig. 5.15, it can be
seen that the outputs are 2D-Array data-types, each row being a different set of
measurements.
PRL provides also a remote control feature, allowing its complete and transpar-
5.3. REAL-TIME DATA MEDIATORS 59
ent integration into any application needing its services. The remote control is used
for starting PRL, initiating the connection to the PDC stream, opening the front
panel, etc.
Figure 5.14: Screen shot of the PRL Data selector
Figure 5.15: Screen shot of the PRL SubVI
5.3.1 PRL Configuration

Real-Time Data Mediators are a key part of the real-time measurement chain. These
need to be configured to connect the Output Stream used for the experiment. The
required information is the following:
• IP address of the host-computer with PDC server.
• Port dedicated to the Output Stream of interest.
• ID code of the Output Stream of interest.
As mentioned in Section 5.3, the Monitoring Tool was mainly developed with
the PRL, therefore only the configuration procedure of PRL is described. Note that
the BableFish has a similar configuration procedure.
Once opened and launched within LabView, the PRL is presented with its main
window, see Fig. 5.13a. The configuration is edited by pressing the Config button
on the main window. The configuration windows appears, and the connection de-
tails need to be entered in each field, see Fig. 5.13b. The configuration must be
validated.3
3
A restart of the PRL may be needed to take into account the latest changes.
The connection can now be initiated by pressing the Startup button, see Fig. 5.16.
Once the connection successfully established, data selection must be performed by
opening the data selector VI. Each element of the lists can be selected, one can use
Ctrl key for selecting several element of one list, see Fig. 5.14. For the experiment
the voltage phasor, current phasor (if available) and the frequency are selected.
Figure 5.16: Screen shot of PRL after successful connection to the PDC Output Stream
5.4. SUMMARY 61
5.4 Summary
In this Chapter the complete chain of real-time data acquisition was presented with
all its elements as well as their configuration. The chain starts with measurements
from outputs of the simulator and ends with delivery of the measurements in Lab-
View via the PRL.
Chapter 6
Experimental Testing and Validation
The Thesis work focused on a specific type of interactions in a power system, which
occur at rather high frequency compared to the traditional power system oscillations
in the range of 0.1 Hz to 1 Hz. It is only with the increasing usage of PMUs that
they could be detected. In Chapter 2 the models used during the Thesis work were
introduced. A perturbation analysis on theses models is performed in Chapter 3 in
order to better understand the origin of the oscillatory phenomenon.
In Chapter 4 a tool for detecting oscillations was presented. The Monitoring Tool
developed for the Thesis work has been proven able to detect correctly the presence
of oscillations at higher frequencies by using recorded data from real oscillation cases
from OG&E and the use of the Replay Tool.
The Monitoring Tool utilizes PMU measurements provided by a chain of real-
time data acquisition, completely described in Chapter 5. The PMUs can be hooked
to the simulator, the next step is then to assemble all of the previous work and build
a proof of concept by using HIL simulation technique.
The power system behaviour is simulated in real time with preconfigured outputs
on the bus of interest in the power system. A PMU is connected to each of the
analogue outputs and reports the measurements to the PDC server of the lab.
Finally the Real-Time Monitoring tool is connected to an output stream of the
PDC via PRL, containing the measurements of the real-time simulation.
In this Chapter the focus is on the experiments and their results. Both experi-
ments at SmarTS Lab and validation tests at IREC’s Microgrid lab are reported.
63
64 CHAPTER 6: EXPERIMENTAL TESTING AND VALIDATION
6.1 Experiment Design and Results

The experiment carried out to test the Monitoring Tool uses the model described in
Section 2.2 and depicted in Fig. 2.4. In Section 3.2 different ways were explored to
recreate the behaviour of power oscillations at higher frequencies. In this Section the
design of the experiment will be specified to highlight the features of the Monitoring
Tool and especially its ability to detect power system oscillations within a high
frequency range.
6.1.1 Experimentation Design

The Monitoring Tool, as described in Chapter 4, has been developed to monitor the
frequency range up to 25 Hz, assuming a European system and PMU with a rate
of 50 samples per second. Testing the full capacity of the tool requires to generate
several dynamics behaviour in the power system monitored.
The power system model, which is used for the experiment, is equipped with
varying loads, that can be configured by the user, as described in Section 2.2. These
loads will be configured with a sinusoidal profile to generate components in the low
frequency range, as well as random load variations. The model is also equipped
with a perturbation generator, whose effects have been analyzed in Chapter 3. The
model also incorporates transmission line faults which can be applied by the user.
These are used for exciting low frequency modes of the system.
The simulation configuration takes all of this into account and the timeline of
each step executed is summarized in Table 6.1.
Table 6.1: Timeline of the experiment

Start Oscillation In- Additional Major Fault End
jection Perturbations
Random load Set pertur- Generating Three phase End of the os-
variation bation at minor faults Fault and line cillation injec-
10.83 Hz opening tion
6.1.2 Results
The experiment is carried out with a real-time simulation and the Monitoring Tool
running in parallel. Obtaining all the screen shots was not a straight forward task
and required to repeat the experiment several times. This explains the difference
in time stamps on the screen shots. The screen shots are shown following the order
presented in the time-line, see Table 6.1.
The simulation of the power system model is started with both wind farms
receiving an average wind speed 12 m/s with 10 % turbulence. The loads are also
configured to have a sinusoidal profile at different frequencies, according to the
following parameters:
6.1. EXPERIMENT DESIGN AND RESULTS 65
• Load in area 1:
Active power 1 MW random load variation and 2 MW of sinusoidal load
at 0.4 Hz.
Reactive power 1 MW random load variation and 2 MW of sinusoidal
load at 0.8 Hz.
• Load in area 2:
Active power 1 MW random load variation and 2 MW of sinusoidal load
at 1 Hz.
Reactive power 1 MW random load variation and 150 MW constant load.
The resulting frequency of the power system model can be observed in Fig. 6.1a.
It can be observed that unlike the response from traditional simulation environment,
appropriate modelling results are obtained in a response close to real power systems.
The processing algorithms of the Monitoring Tool detect the slow dynamic activity
resulting from both load variations and wind turbulences, see Fig. 6.1b. Frequency
estimation also highlights both frequency components at 0.4 Hz and 0.8 Hz.
Forced oscillations are then injected at first with 0.05 p.u. amplitude at the
PCC of the first wind farm. They can be observed in the frequency graph of the
Monitoring Tool, see Fig. 6.2. It can also be noticed that, while most frequency
ranges remain inactive (no trigger flags on), the frequency range containing 10.83 Hz
is active with the flag Danger!!!. This shows the fast reaction of the tool and its
ability after a few seconds to estimate the level of energy in these oscillations. The
frequency estimation algorithm does not update as quickly as the energy detection
algorithm, the frequency of the injected oscillations is thus not detected as quickly.
It can however be noticed that the parametric method starts to show distinctively
a peak at the right frequency, see Fig. 6.2.
After the beginning of the injection with 0.05 p.u. amplitude the injection
is increased up to 0.07 p.u. amplitude, see Fig. 6.3. The resulting oscillations
have a bigger amplitude, as shown in Fig. 6.3a, and the energy detection algorithm
identifies an increase in the energy level in the oscillations. The frequency estimation
algorithm also detects very precisely the frequency at which the oscillations are
occurring as shown in Fig. 6.3b.
The injection of oscillations is maintained and the power system model stabi-
lizes with these high frequency power oscillations still active. Additional (minor)
perturbations performed on the system are:
• A brief step increase in the mechanical power input of one of the synchronous
generators (0.05 p.u.).
• A three-phase fault at the bus of one of the sub-system of the first wind-farm
(Three wind turbines affected).
(a) Frequency
(b) Module [0.1-1] Hz

Figure 6.1: Screen shots of the Monitoring Tool at the beginning of the simulation with
only load variations and wind turbulences
The perturbations are performed sequentially, see Fig. 6.4. The first perturba-
tion is performed at an instant approximatly at the middle of the scope and the
second at the end of the scope.
The first perturbation induces a small ripple in the frequency of the system,
leading to the detection of activity in almost every frequency range. While the most
affected are the lower frequency ranges, which is expected. The detections mostly
result in discontinuities in the voltage measurements. The phenomenon is amplified
in the case of the second perturbation, which induces a greater discontinuity in
the voltage measurements and therefore triggers the Danger!!! flag in almost all
frequency ranges, see Fig. 6.4.
The major fault performed is a three phase fault on one of the lines between the
two sectors of the power system model. After the fault is applied, the line is opened
Figure 6.2: Screen shots of the Monitoring Tool at the beginning of the oscillation injection
at 10.83 Hz
(a) Frequency
(b) Module
Figure 6.3: Screen shots of the Monitoring Tool during the increase of oscillation injection
at 10.83 Hz
and re-closed after clearing the fault, eleven second after the application of the fault.
The effect on the power system model is strong, but stability is maintained after
re-closing the line, see Fig. 6.5.
The Monitoring Tool is however strongly affected by discontinuities of the volt-
age measurements, triggering temporarily a Danger!!! flag in all frequency ranges.
This behaviour should be eradicated and the tool should be less sensitive to such
behaviours. The only information that can be retrieved from the tool during such
events is the lower frequencies that are the most affected by such perturbation. This
is noticed by the trigger flag staying on for a longer time in this frequency range.
The frequency estimation algorithm is also vastly affected by the discontinuity of
the measurements, making it very difficult to read anything during such events.
At the end of the injection of oscillations, the high frequency power oscillations
Figure 6.4: Screen shot of the Monitoring Tool during the oscillation injection at
10.83 Hz and minor faults
Figure 6.5: Screen shot of the Monitoring Tool during the oscillation injection at
10.83 Hz and a major fault
stop very quickly and the frequency of the power system returns to a normal range,
only driven by the wind turbulences and load variations, see Fig. 6.6a. The energy
detected in the oscillations at 10.83 Hz decreases very quickly and the trigger flags
are turned off. The frequency estimation algorithm still shows a distinctive peak
at 10.83 Hz, because the estimation is performed on the whole buffered signal, see
Fig. 6.6b.
(a) Frequency
(b) Module
Figure 6.6: Screen shots of the Monitoring Tool at the end of the oscillation injection at
10.83 Hz
6.2. EXPERIMENT VALIDATION 71
6.2 Experiment Validation

The experiment carried out in the lab uses HIL real-time simulations and the chain of
real-time data acquisition described in Chapter 5. During the project SmarTS Lab
collaborated with IREC, a research institute in Barcelona, Spain, equipped with
a microgrid with controllable power emulators. A collaboration was initiated to
design and carry out a validation experiment involving the Monitoring Tool and
PMU measurements from the microgrid during oscillating power injections.
6.2.1 IREC’s Microgrid

The microgrid at IREC is comprised by completely different equipments than that
used in the lab at KTH. The purpose of such an equipment is to replicate at a lower
scale a power system with one infinite bus (the connection point to the national
grid).
The microgrid is built with three independent bus-bars rated at 400 kV con-
nected to the national grid to which different equipments are connected. Some
equipment are actual generators or loads, such as an electric vehicle, batteries,
wind turbines, etc. The other connected devices are fully configurable cabinets
containing power converters in a back-to-back set-up, referred to as emulators, see
Fig. 6.8.
The power output in these cabinets can be fully configured, or in other words
they can be configured as a load or generating unit. This operating principle is
depicted in Fig. 6.7. The power flow is controlled by the converter named AFE and
the DC-bus Voltage is controlled by the converter named AR to ensure stability in
the DC-bus. Such devices will allow the emulation of any kind of generating unit and
will be used in this experiment to inject power oscillations. The PMU measurements
are taken at the output of the converter named AFE, which is connected to the bus-
bar.
Each converter is piloted by a controller, whose logic is to follow a reference value
set by the user. For the purpose of this experiment the engineers at IREC developed
an application to easily interact with the power emulators. The application handles
all the communication protocol required by the controller and generates periodical
reference values according to the profile set by the user. The profiles available are
either pure sinusoidal power injections at a frequency f , following Eq. 6.1 or a replay
mode, reading the values from recorded data (via external files).
pref (t) = Pconst + Pconst≠1 sin (2ﬁf t) (6.1)

A full description of the microgrid is available in [27].
Validation Set-up at IREC

The set-up of the experiment has been partly carried out be the team at IREC
for all tasks related to the emulator cabinets. The control application has been
Figure 1.2: Simplified circuit diagram of the three-phase voltage sources.
T
AF
out in out losses
DR
in out losses
out
Figure 6.7: Diagram of the operating principle of the microgrid from [27]
Figure 1.3: Power flow. (a) Generation emulation. (b) Consumption emulation.
developed and tested by IREC in advance, which uncovered some limitations of the
microgrid controls explained more precisely in the next Section.
The set-up of the experiment also involved duplicating the whole chain of real-
time data acquisition described in Chapter 5. Two cRIOs were configured at KTH
and sent to IREC ahead of our visit in Barcelona to be installed in the converters
cabinets, see Fig. 6.9. The rest of the software has been installed and configured
directly at IREC.
Once the cabinet was ready and the PDC server configured for receiving the
PMU stream, the Monitoring Tool was connected to the PDC Output Stream and
the good working order was checked. Initially, the replay of simulation record-
ings from the previous experiment with the power emulator, see Section 6.1, was
planned. The simulation contained injected oscillations at 10.83 Hz, which is above
the limits of the power converters in IREC. Instead it was decided to carry out
two experiments, the first one using the application developed by IREC to inject
constant oscillations and the second one replaying the simulation with oscillations
at half the frequency used during the experiment at KTH (10.83 Hz).
Figure 6.8: Photo of the cabinet enclosure of one power emulator
Figure 6.9: Photo of the set-up of the PMU cRIO-9074 at IREC
Practical Issues during the Validation Tests

The validation step has been done in a completely different facility. First the IREC
lab is located in Barcelona, Spain, without the possibility of remote-accessing the
computers there. The set-up in their lab has been duplicated during one visit.
Second the lab equipment available was completely different: at KTH power systems
are simulated, whether in IREC power systems are emulated with a microgrid and
programmable converters. Despite all that, successful validation experiments were
carried out for the Monitoring Tool. Two issues are however worth mentioning.
The first issue comes from the lab equipment specifications in IREC. The mi-
crogrid is connected to the main Spanish grid, the voltage magnitude and frequency
are thus completely unalterable given the power output limits of the microgrid. The
Monitoring Tool was designed for extracting its information from the voltage mag-
nitude for the oscillation detection and from the frequency for frequency estimation.
This led to necessary on-site work for taking this into account and modify all the
algorithms to use the current magnitude.
The second issue is linked to the fact that high frequency phenomena were
simulated, which the microgrid was not designed for. This led to extensive work from
the Spanish team to revise the control logic of the controllers of the microgrid to be
able to emulate power oscillations with frequencies up to about 6-7 Hz (the response
of the controllers to higher frequencies was not possible). Also the emulation is
controlled by a set of reference values sent periodically. A constant or stable delay
could however not be guaranteed, leading to the following behaviour:
• Shifted spectra, if the new reference values are taken into account too early
or too late.
• Spectra with more noise, not as distinct, if the time for new reference values
to be taken into account is not constant.
6.2.2 Results
As mentioned in the previous Section the validation is composed of two experiments.
In the first one IREC’s application was used to generate a power oscillation. In the
second one simulation data with oscillations at of frequency of 5.42 Hz was replayed.
For the following experiments one screen shot at the beginning of the injection
of oscillations is shown, an additional screen shot at the end of the injection of
oscillations is however available in Appendix.
It is important to note that the Monitoring Tool has been modified on-site to
take into account the characteristics of the microgrid, as explained in the previous
Section. There are therefore some larger offsets in the time stamps displayed in the
Monitoring Tool.
For visualisation purposes the PMU measurements were scaled up and the trig-
ger levels of the oscillation detection lowered. This is needed to counteract the
limited ratings of the microgrid and to adapt to lower values for the oscillation
detection. Current measurements instead of the voltage measurements are used for
oscillation detection. The global factor is 30 on powers (10 on the voltage and 3 on
the current).
Constant Oscillations
This experiment was performed twice with injections at 1 Hz and 6 Hz. The reason
for repeating twice the experiment is the attenuation of the signal by the controllers
of the converters. While oscillations at 1 Hz are certainly in the pass-band of the
controllers, oscillations at 6 Hz might not be.
For the first part of the experiment, the power emulator was started with con-
stant power injection of 500 W. The oscillations injection is then started with
Pconst = 2000 W, Pconst≠1 = 1000 W and f = 1 Hz, following the formula in
Equation 6.1. This part was used for the first calibration of the Monitoring Tool
and check that it could work based on current measurements.
(a) Active and Reactive Powers
(b) Module
Figure 6.10: Screen shots of the Monitoring Tool at the beginning of the oscillations
injection at 1 Hz
A partial screen shot of the Monitoring Tool is presented in Fig. 6.10. The screen
shot was taken a few seconds after the beginning of the injection of oscillations. This
is verified on the graph of the powers, see Fig. 6.10a, where the transition between
constant power and oscillating power is noticeable.
After the beginning of the oscillations, with the usual delay induced by the filters
in the algorithm, the oscillations are detected by the Monitoring Tool, see Fig. 6.10b,
provided that the trigger levels have been adjusted. Frequency estimation shows
a frequency around 1.15 Hz, which can be explained by the issues evoked in the
previous Section.
The Danger!!! flag is triggered shortly after the beginning of the oscillations,
only in the Module for the frequency range containing the frequency 1 Hz. It can
also be seen in Fig. A.1 that the trigger flag is turned off shortly after the end of
the oscillations injection. This test is then considered conclusive.
The second part of the experiment was conducted by injecting oscillations at
6 Hz, the new parameters for the injections are then Pconst = 2000 W, Pconst≠1 =
1000 W and f = 6 Hz.
The results of the second part of the experiment are however slightly different
than in the first part. A screen shot taken at the beginning of the oscillation injection
is presented in Fig. 6.11. As it can be seen in the power graph, see Fig. 6.11a, even
if the magnitude of the oscillations in the reference values is the same as in the first
part, the resulting oscillations are much attenuated and thus the flag Danger!!! is
not triggered. This behaviour is due to limitations in the control scheme of the
converters and leads naturally to less energy in the oscillations. The trigger levels
were therefore modified for the second experiment.
Frequency estimation shows a peak around 6-6.1 Hz, see Fig. 6.11b, which is
closer to the original 6 Hz, compared to the first part. The shift of frequency is
however still caused by the same issue explained in the previous Section.
Replay of Simulation Data

This second experiment has been built with recorded data from simulation done in
the lab at KTH. The data had been recorded at the point of common connection of
the wind farm with injected oscillations. Initially the data was sampled at 200 Hz,
but was down-sampled according to the limitations of the microgrid converters.
Also as mentioned in the introduction of the Section the initial goal of replaying the
simulation files with oscillations at 10.83 Hz was changed, because of the limitations
of the control scheme of the converters. Instead a simulation with oscillations at
half this frequency or 5.42 Hz was replayed.
The data recorded from the simulation was in per-unit. This was useful espe-
cially for this experiment, because the ratings of the microgrid are much lower than
the ratings of the simulated power system in the lab at KTH. A scaling factor was
applied to scale output values safely.
The screen shot shown in Fig. 6.12 is taken at the beginning of the oscillation
injection. Power oscillations can be seen in Fig. 6.12a.
(b) Module
Figure 6.11: Screen shots of the Monitoring Tool at the beginning of the oscillations
injection at 6 Hz
Among the two modules shown, see Fig. 6.12b, one is dedicated to the frequency
range [3-7] Hz containing the frequency of the oscillations and a second one dedi-
cated to another frequency range. The second frequency range is not related to the
injected oscillations. This allows to verify that despite lowering the trigger levels,
the Monitoring Tool is still capable of detecting oscillations and also not trigger any
flags in inactive frequency ranges.
In this experiment the trigger levels were lowered, they could have been lowered
more, because only the High Activity flag was triggered. Despite this missing Dan-
ger!!! flag, given that in the other frequency ranges not even the Low Activity flag
was triggered, the Monitoring Tool is considered selective enough and the second
experiment is considered successful.
(b) Modules
Figure 6.12: Screen shots of the Monitoring Tool at the beginning of the oscillation injec-
tion at 5.42 Hz using IREC’s microgrid for simulation replay
6.3. SUMMARY 79
6.3 Summary
In this Chapter, the experiments performed during the Thesis were described. The
first experiment, described in Section 6.1, was executed with the equipment available
at SmarTS Lab, utilizing real-time HIL simulations. This first experiment was
designed to highlight the functioning and utility of the Monitoring Tool to detect
diverse oscillatory phenomena in a power system. It has been shown successful,
provided that it was correctly configured.
The second experiment, presented in Section 6.2, served as validation. The
measurements used were taken from a mircogrid injecting real power oscillations in a
bus connected to an actual distribution network. The validation was successful, the
Monitoring Tool needed to be configured according to the ratings of the microgrid.
The Monitoring Tool has however shown some drawbacks in its behaviour during
the application of faults in the system, removing these drawbacks will be part of
future work. One of the first functionalities to be improved would be to implement
outlier removal for the voltage and current measurements. Another improvement
would be to make the energy detection less sensitive to measurement discontinuities.
Finally it would be interesting to test the Monitoring Tool on real-time data
from a high voltage substation for example. It could show if the tool is fast enough
to detect oscillations present in an actual high voltage network.
Chapter 7
Conclusion
This Thesis work focused on the main objective of building a monitoring tool capable
of detecting oscillatory events in the frequency range of 10 - 15 Hz. At the beginning
of this work there were no "real" measurements data available in the lab, that led
to additionnal work in the Thesis with the necessity to model the phenomenon and
interface a real-time simulator with the equipment of the lab.
Recreating the phenomenon of sub-synchronous oscillation required the devel-
opment of a test power system model with configurable perturbations to trigger
behaviours close to the description of the phenomenon. In Chapter 2 the mod-
els assembled during the Thesis were presented and in Chapter 3 the perturbation
analysis performed. The perturbation analysis showed that among the controllers
of a Type 3 wind turbine, it was the grid side controller which was the most likely
responsible for the phenomenon. This has been later confirmed by additional infor-
mation received from OG&E. Together with additional information on the origin of
the phenomenon, real measurements from different events were received, which con-
firmed that the perturbation applied to the power system model triggered equivalent
behaviours, fulfilling the second objective of the Thesis work.
The Monitoring Tool has been implemented in LabView with the algorithms
described in Chapter 4. LabView has been proven very flexible for the task and made
it easy to build the final graphical user interface of the tool. Additionally the tool has
been developed in a replay version, capable of reading the recorded measurements
from OG&E. This replay version was very useful to test the algorithms for oscillation
detection and frequency estimation. The graphical user interface and features of
the tool have been under continuous development during this Thesis work, leading
to its final form as presented in Chapter 4, fulfilling the main objective.
The fulfilment of the third objective was however conditioned to the tool being
able to work with real-time PMU measurements, that brought additional work
during the Thesis to configure a chain of real-time data acquisition between the
real-time simulator and the Monitoring Tool. This step is described in Chapter 5,
which should serve as step-by-step guide for next students working at SmarTS Lab.
In Chapter 6 the experiments carried out during the Thesis were described.
81
82 CHAPTER 7: CONCLUSION
The first set of experiments took place in KTH and used real-time simulations with
the power system model presented in Chapter 2. This experiment was successful.
The second experiment took place at IREC and utilized emulation. It served as
validation, since the measurement were actual measurements, and only the ratings
were down-scaled for replay. Both of the experiments fulfilled the third objective
of the Thesis. They provided key information on the limitations of the Monitoring
Tool, which are explained with more details in Section 7.1.
Finally it motivated the author to pursue his studies at KTH and all of its
aspects will serve as a complete introduction and initiation to the research area of
SmarTS Lab.
7.1 Future Work

The Monitoring Tool is the main output product of this Thesis. It succeeds to
fulfil its purpose on several test cases, which were available in the lab. It has
been successfully tested on recorded data, real-time simulated data and real-time
emulated data. Its behaviour showed however some inconsistencies with difficulties
to handle strong discontinuities in measurements. Further work is therefore required
to make it more reliable.
It is also important to mention that this project developed in LabView, served
as introduction to the LabView environment. The code of the tool is therefore not
optimized. At the time of writing this Master’s Thesis, the author was following a
course on LabView programming, which included personal projects. These projects
will focus on enhancing the performances of the tool and implementing missing
features, taking advantage of the newly acquired knowledge. A feature that will be
implemented is a fault recording option to save the PMU measurements prior to the
oscillation detection and until the deactivation of the trigger flags. Another feature
that will be added, is an option to save and load a configuration file, to avoid the
need to reconfigure the tool at each launch.
Another feature will be to make it run on a CompactRIO to have a Monitor-
ing Tool fed with local measurements. This combination would be a monitoring
equipment exploitable at a remote location for example.
The code will be also fully documented to be used as a tutorial example on
how-to develop LabView prototype applications at SmarTS Lab.
The final validation of the tool will be envisaged after all the corrections and
additions have been implemented. This validation would be to use the tool on a
real power system, where the phenomenon of fast oscillation may appear.
Appendix A
Additional Figures
(b) Module
Figure A.1: Screen shots of the Monitoring Tool at the end of the oscillation injection
83
84 CHAPTER A: ADDITIONAL FIGURES
(a) Active and Reactive Power
(b) Module
85
(a) Active and Reactive Power
(b) Module
Bibliography
[1] J. F. Manwell, J. G. McGowan, and A. L. Rogers, Wind Energy Explained:

Theory, Design and Application. John Wiley & Sons, 2010.
[2] D. Milborrow, “Europe 2020 wind energy targets, eu wind power now and the
2020 vision,” Windpower Monthly Special Report, march 2011.
[3] P. Bennett, J. Lieskovsky, and B. McBee, “Impacts of Intermittent Genera-

tion,” Large Energy Storage Systems Handbook, pp. 17–50, CRC Press, 2011.
[4] M. Patel (RAPIR Chair), “Real-time application of synchrophasors improv-

ing reliability,” tech. rep., North American Elecricity Reliability Corporation,
Princeton, NJ, 2010.
[5] J. Wiik, J. O. Gjerde, and T. Gjengedal, “Impacts from large scale integration
of wind energy farms into weak power systems,” in Proceedings. PowerCon
2000. International Conference on Power System Technology, 2000, pp. 49 –
54, 2000.
[6] A. White and S. Chisholm, “Relays become problem solvers,” Transmission &
Distribution world, November 2011.
[7] A. White (OGE), S. Chisholm (OGE), H. Khalilinia (WSU), Z. Tash-

man (WSU), and M. Venkatasubramanian (WSU), “Analysis of Subsyn-
chronous Oscillations at OG&E.”
[8] L. Vanfretti and J. H. Chow, “Identification of Dominant Inter-Area Modes in

the Eastern Interconnection from PMU data of the FRCC 2008 Disturbance
: an Eigensystem Realization Algorithm Illustration,” Tech. Rep. 2010:053,
KTH, Electric Power Systems, 2010.
[9] P. Anderson, B. Agrawal, and J. Ness, Subsynchronous Resonance in Power

Systems. 1990.
[10] “IEEE Standard for Synchrophasor Data Transfer for Power Systems,” IEEE
Std C37.118.2-2011 (Revision of IEEE Std C37.118-2005), pp. 1 –53, 28 2011.
87
88 BIBLIOGRAPHY
[11] L. Vanfretti et al., “Smarts lab; a laboratory for developing applications for
wampac systems,” in Power and Energy Society General Meeting, 2012 IEEE,
pp. 1 –8, july 2012.
[12] L. Vanfretti, S. Bengtsson, V. Peric, and J. Gjerde, “Effects of forced oscilla-
tions in power system damping estimation,” in IEEE International Workshop
on Applied Measurements for Power Systems (AMPS), pp. 1 – 6, sept. 2012.
[13] A. Ellis, E. Muljadi, J. Sanchez-Gasca, and Y. Kazachkov, “Generic models for
simulation of wind power plants in bulk system planning studies,” in Power
and Energy Society General Meeting, 2011 IEEE, pp. 1 –8, july 2011.
[14] C. Dufour and J. Belanger, “A real-time simulator for doubly fed induction
generator based wind turbine applications,” in IEEE 35th Annual Power Elec-
tronics Specialists Conference (PESC), vol. 5, pp. 3597 – 3603, june 2004.
[15] M. Klein, G. J. Rogers, and P. Kundur, “A fundamental study of inter-area
oscillations in power systems,” IEEE Transactions on Power Systems, vol. 6,
pp. 914 –921, aug 1991.
[16] C. Dufour, J. Paquin, V. Lapointe, J. Bélanger, and L. Schoen, “PC-Cluster-
Based Real-Time Simulation of an 8-Synchronous Machine Network with
HVDC Link using RT-LAB and TestDrive,” in International Conference on
Power Systems Transients (IPST), Lyon, France, 2007.
[17] J.-N. Paquin, J. Moyen, G. Dumur, and V. Lapointe, “Real-Time and Off-
Line Simulation of a Detailed Wind Farm Model Connected to a Multi-Bus
Network,” in Electrical Power Conference, EPC 2007. IEEE Canada, pp. 145–
152, Oct 2007.
[18] J. Hauer and F. Vakili, “An oscillation detector used in the bpa power system
disturbance monitor,” Power Systems, IEEE Transactions on, vol. 5, pp. 74
–79, feb 1990.
[19] D. Trudnowski, “Fast real-time oscillation detection.” Proc. North Amer. Syn-
chroPhasor Initiative (NASPI) Work Group Meeting, Orlando, FL., February
29 - March 1 2012.
[20] P. Welch, “The use of fast fourier transform for the estimation of power spectra:
A method based on time averaging over short, modified periodograms,” IEEE
Transactions on Audio and Electroacoustics, vol. 15, pp. 70 – 73, jun 1967.
[21] L. Vanfretti, S. Bengtsson, V. Aarstrand, and J. O. Gjerde, “Applications
of spectral analysis techniques for estimating the nordic grid’s low frequency
electromechanical oscillations,” in 16th IFAC Symposium on System Identifi-
cation, no. 16 in IFAC Proceedings Volumes (IFAC-PapersOnline), pp. 1001–
1006, IFAC, 2012. Invited Paper, Special Session: “Development of System ID
Methods for Power System Dynamics”.
BIBLIOGRAPHY 89
[22] M. H. Hayes, Satistical digital signal processing and modeling. John Wiley &
Son, 1996.
[23] J. Sanchez-Gasca, “Identification of electromechanical modes in power sys-

tems.” Special Publication, TP462, IEEE Power & Energy Society, July 2012.
[24] L. Vanfretti, S. Bengtsson, and J. O. Gjerde, “Preprocessing Synchronized Pha-

sor Measurement Data for Spectral Analysis of Electromechanical Oscillations
in the Nordic Grid,” European Transactions on Electrical Power.
[25] L. Vanfretti, V. H. Aarstrand, M. Shoaib Almas, V. Peric, and J. O. Gjerde,

“A software development toolkit for real-time synchrophasor applications,” to
appear in Proceedings of the IEEE PowerTech 2013.
[26] I. Al Khatib and L. Vanfretti, “Real-Time Data Mediation for Synchrophasor

Application Development compliant with IEEE C37.118.2,” IEEE PES General
Meeting 2013.
[27] A. Elias-Alcega, M. Roman-Barri, A. Ruiz-Alvarez, I. Cairo Molins, A. Sumper,

and O. Gomis-Bellmunt, “Implementation of a test microgrid in barcelona,”
CIRED 2011, Conference proceedings, 21st International Conference and Ex-
hibitionon Electricity Distribution, Frankfurt, Germany, 2011.

2013 Maxime MScThesis PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2013 Maxime MScThesis PDF

Uploaded by

Copyright:

Available Formats

Fast Real-time Detection of

Master’s Degree Project

Master’s Degree Project

Master’s Thesis at SmarTS Lab, Electric Power Systems

TRITA-EE xxx yyyy-nn

I Developing the test models 7

2 Modelling for RT Simulation and Analysis 9

4 Fast Real-Time Oscillation Detection Tool 29

5 Instrumentation and Data Acquisition 47

6 Experimental Testing and Validation 63

1.1 Comparison Example PMU and SCADA . . . . . . . . . . . . . . . . . . 2

2.1 Type 3 Wind Turbine Diagram . . . . . . . . . . . . . . . . . . . . . . . 10

3.1 PSD with Perturbation to Grid Side Controller . . . . . . . . . . . . . . 21

4.1 Diagram Original Oscillation Detection . . . . . . . . . . . . . . . . . . 31

5.1 cRIO-9074 Assembled as PMU . . . . . . . . . . . . . . . . . . . . . . . 48

6.1 Experiment in KTH - Load Variation and Wind Turbulences . . . . . . 66

A.1 Experiment in IREC - 1Hz End of Injection . . . . . . . . . . . . . . . . 83

OG&E Oklahoma Gas & Electric

wind farms connected to the bulk grid appears thus necessary.

1.2 Wide-Area Systems: Phasor Measurement Units and

TSO CONTROL ROOM

1.3.2 Specific Objectives

1.4 Thesis Outline

Developing the test models

Modelling for RT Simulation and

2.1 Wind Turbine (WT)

2.1.1 Type 3 Wind Turbine Operating Principle

Figure 2.1: Diagram of a Type 3 wind turbine

2.1.2 Wind Turbine Controllers

Grid Side Converter

Rotor Side Converter

2.1.3 Single WT - Infinite Bus

Figure 2.2: One-line diagram of the single WT - Infinite Bus

2.2 Power System Model

2.2.1 General Structure

Opal-RT simulators require separation of MatLab/Simulink models into sub-

2.2.2 Wind Farms

Preparation for Real-Time Execution

• Sub-system of three average wind turbines models.

• Sub-system of five average wind turbines models electro-mechanical part.

• Sub-system of ten average wind turbines models control part.

• Constant wind speed.

• Both of the aforementioned with configurable turbulence.

2.2.3 Load Variations

Where r is a random number generated at each time step (50 µ sec).

The phenomenon of high frequency oscillations in the Oklahoma Gas& Electric

3.1 Controller Sensitivity Analysis: WT-Infinite Bus

• Window size : 8000 samples.

3.1.1 Grid Side Controller

(a) Active Power (b) Reactive Power (c) Voltage in DC Link

3.1.2 Rotor Side Controller

(a) Active Power (b) Reactive Power (c) Voltage in DC Link

3.1.3 Pitch Controller

300 140 180

(a) Active Power (b) Reactive Power (c) Voltage in DC Link

3.1.4 Grid Injection

350 350 160

(a) Active Power (b) Reactive Power (c) Voltage in DC Link

400 400 200

350 350 160

(a) Active Power (b) Reactive Power (c) Voltage in DC Link

3.2 Network Sensitivity to Injected Oscillations

(a) Voltage (b) Active Power (c) Reactive Power

2) Select a Unit 4) Check that ”DHCP/