Professional Documents
Culture Documents
A Dissertation in
Mechanical Engineering
by
Peter M. Austin
Doctor of Philosophy
December 2015
The dissertation of Peter M. Austin was reviewed and approved∗ by the following:
Thomas G. Hughes
Senior Research Associate Emeritus of Applied Research Laboratory
Dissertation Co-Adviser and Co-Chair of Committee
Susan W. Stewart
Research Associate and Assistant Professor of Aerospace Engineering and
Architectural Engineering
Dissertation Co-Adviser and Co-Chair of Committee
Hosam K. Fathy
Assistant Professor of Mechanical Engineering
Jeffrey S. Mayer
Associate Professor of Electrical Engineering
Horacio Perez-Blanco
Professor of Mechanical Engineering
James A. Turso
Senior Research Associate of Nuclear Engineering
Special Member
Karen Thole
Mechanical and Nuclear Engineering Department Head
∗
Signatures are on file in the Graduate School.
ii
Abstract
Modern microgrids with microsources and energy storage are dependent on power
electronics for control and regulation. Under certain circumstances power electron-
ics can be destabilizing to the system due to an effect called negative incremental
impedance. A careful review of the theory and literature on the subject is pre-
sented. This includes stability criteria for both AC and DC systems, as well as a
discussion on the limitations posed by the analysis. A method to integrate stability
assessment with higher-level microgrid architectural design is proposed. Crucial
to this is impedance characterization of individual components, which was accom-
plished through simulation. DC and AC impedance measurement blocks were
created in Matlab simulink to automate the process. A detailed switching-level
model of a DC microgrid was implemented in simulink, including wind turbine
microsource, battery storage, and three phase inverter. Maximum power point
tracking (MPPT) was included to maximize the efficiency of the turbine and was
implemented through three rectifier alternatives and control schemes. The sta-
bility characteristics of each was compared in the final analysis. Impedance data
was collected individually from the components and used to assess stability in the
system as a whole. The results included the assessment of stability, margin, and
unstable operating points to demonstrate the feasibility of the proposed approach.
iii
Table of Contents
List of Tables xi
Acknowledgments xii
Chapter 1
Background 1
1.1 Recent Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Microgrids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Power Electronics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Negative Incremental Input Impedance . . . . . . . . . . . . . . . . 12
1.6 Design Challenge . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.7 Proposed Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.8 Dissertation Objective . . . . . . . . . . . . . . . . . . . . . . . . . 16
Chapter 2
Stability 18
2.1 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.1.1 Linear Time-Invariant (LTI) Assumption . . . . . . . . . . . 19
2.1.2 Internal and External Stability . . . . . . . . . . . . . . . . 22
2.2 DC system Stability . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.1 Electrical System Transfer Function . . . . . . . . . . . . . . 24
2.2.2 Nyquist Stability Criteria . . . . . . . . . . . . . . . . . . . 26
2.2.3 Stability Criteria . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.4 Passivity-Based Stability Criteria . . . . . . . . . . . . . . . 30
2.2.5 Generalized Admittance Space Analysis . . . . . . . . . . . . 31
2.3 AC bus Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.3.1 Generalized Nyquist . . . . . . . . . . . . . . . . . . . . . . 32
iv
2.3.2 Stability Criteria . . . . . . . . . . . . . . . . . . . . . . . . 33
2.4 Proposed Stability Assessment Method . . . . . . . . . . . . . . . . 34
2.4.1 Minimizing Conservatism . . . . . . . . . . . . . . . . . . . . 34
2.4.2 Stability Margins . . . . . . . . . . . . . . . . . . . . . . . . 37
2.4.3 Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.4.4 Operating Points . . . . . . . . . . . . . . . . . . . . . . . . 40
Chapter 3
Simulated Impedance Measurement 43
3.1 Nyquist Contour and Impedance . . . . . . . . . . . . . . . . . . . 43
3.2 DC Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2.1 Current Perturbation Injection . . . . . . . . . . . . . . . . 48
3.2.2 DC Impedance Measurement Block . . . . . . . . . . . . . . 49
3.2.3 Results and Verification . . . . . . . . . . . . . . . . . . . . 52
3.3 AC System Impedances . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.3.1 The dq0 Synchronous Reference Frame . . . . . . . . . . . . 56
3.3.2 DQ Impedance Matrix . . . . . . . . . . . . . . . . . . . . . 56
3.3.3 AC Impedance Measurement Block . . . . . . . . . . . . . . 57
3.3.4 Results and Verification . . . . . . . . . . . . . . . . . . . . 59
Chapter 4
Notional Microgrid 64
4.1 Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.2 Notional Microgrid Layout . . . . . . . . . . . . . . . . . . . . . . . 68
4.3 Wind Turbine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.4 Rectification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4.1 Passive Diode Bridge . . . . . . . . . . . . . . . . . . . . . . 73
4.4.2 Boost Converter . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.4.3 Switching Bridge . . . . . . . . . . . . . . . . . . . . . . . . 84
4.5 Battery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.6 Inverter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.6.1 "Stiff" grid P plus resonant Inverter . . . . . . . . . . . . . . 94
4.6.2 "Stiff" grid dq0 Inverter . . . . . . . . . . . . . . . . . . . . . 99
4.6.3 "weak" grid dq0 Inverter . . . . . . . . . . . . . . . . . . . . 102
Chapter 5
Results and Conclusions 106
5.1 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.1.1 P Resonant Inverter and Stability Margins . . . . . . . . . . 106
5.1.2 Notional Wind Microgrid and Operating Point Stability . . . 109
v
5.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Chapter 6
Recommendations 118
Bibliography 121
Appendix A
Grid Control 129
A.1 Grid vs. Microgrid . . . . . . . . . . . . . . . . . . . . . . . . . . 129
A.2 Electrical Grid Control . . . . . . . . . . . . . . . . . . . . . . . . 130
A.2.1 AGC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
A.2.2 CERTS microgrid control . . . . . . . . . . . . . . . . . . . 131
A.3 High Level Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
Appendix B
Matlab Code 135
B.1 Matlab Impedance Fitting Code . . . . . . . . . . . . . . . . . . . 135
B.2 Impedance Data Interpolation . . . . . . . . . . . . . . . . . . . . 136
B.3 Matlab Nyquist Plotter . . . . . . . . . . . . . . . . . . . . . . . . 137
Appendix C
Voltage Source Converters 140
C.1 Two Level Voltage Source Three Phase Converter . . . . . . . . . 140
C.2 dq0 Inner Current Control Loop Design . . . . . . . . . . . . . . . 141
C.3 PMSG control Design . . . . . . . . . . . . . . . . . . . . . . . . . 143
C.3.1 dq0 Outer Voltage Control Loop Design . . . . . . . . . . . 145
C.3.2 Symmetrical Optimum Method . . . . . . . . . . . . . . . 147
C.4 Component Sizing . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
C.4.1 Current Ripple Design . . . . . . . . . . . . . . . . . . . . 148
C.4.2 Capacitor Sizing . . . . . . . . . . . . . . . . . . . . . . . . 148
vi
List of Figures
vii
3.9 Diode Bridge Circuit Test Diagrams . . . . . . . . . . . . . . . . . . 54
3.10 "Source" type Diode Bridge Impedance Results . . . . . . . . . . . . 54
3.11 "Load" type Diode Bridge Impedance Results . . . . . . . . . . . . . 55
3.12 dq0 Transformation of Three Phase Voltage . . . . . . . . . . . . . 57
3.13 Injector Block . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.14 AC Current Perturbation . . . . . . . . . . . . . . . . . . . . . . . . 58
3.15 Simple AC LTI Circuit Diagram . . . . . . . . . . . . . . . . . . . . 60
3.16 Simple AC LTI Circuit Impedance Magnitudes . . . . . . . . . . . . 61
3.17 Simple AC LTI Circuit Impedance Angles . . . . . . . . . . . . . . 61
3.18 Six Pulse Diode Bridge Diagram . . . . . . . . . . . . . . . . . . . . 62
3.19 Six Pulse Diode Bridge Input Impedance Magnitudes . . . . . . . . 62
3.20 Six Pulse Diode Bridge Input Impedance Angles . . . . . . . . . . . 63
viii
4.28 Switching Bridge Rectifier: Variable Wind . . . . . . . . . . . . . . 88
4.29 Switching Bridge Rectifier Diagram . . . . . . . . . . . . . . . . . . 88
4.30 Switching Bridge Rectifier Impedance Magnitudes . . . . . . . . . . 89
4.31 Switching Bridge Rectifier Impedance Angles . . . . . . . . . . . . . 90
4.32 Battery Equivalent Circuit . . . . . . . . . . . . . . . . . . . . . . . 92
4.33 Battery Nominal Impedance Plot . . . . . . . . . . . . . . . . . . . 92
4.34 Battery Equivalent Circuit Impedances . . . . . . . . . . . . . . . . 93
4.35 Two-Level Three Phase Voltage Sourced Converter Diagram . . . . 94
4.36 P Resonant Controller . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.37 P Resonant Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.38 P Resonant Time Domain Simulation . . . . . . . . . . . . . . . . . 97
4.39 P Resonant DC Impedance Magnitudes . . . . . . . . . . . . . . . . 98
4.40 P Resonant DC Impedance Angles . . . . . . . . . . . . . . . . . . 98
4.41 P Resonant AC Impedances: P = 2.5 MW Q = 0.5 MVAR . . . . . 99
4.42 "Stiff" dq0 Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.43 "Stiff" dq0 Time Domain Results . . . . . . . . . . . . . . . . . . . 101
4.44 "Weak" dq0 Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.45 "Weak" dq0 Time Domain Simulation . . . . . . . . . . . . . . . . . 103
4.46 "Weak" dq0 DC Impedance Magnitudes . . . . . . . . . . . . . . . . 104
4.47 "Weak" dq0 DC Impedance Angles . . . . . . . . . . . . . . . . . . 105
4.48 "Weak" dq0 AC Impedance . . . . . . . . . . . . . . . . . . . . . . . 105
ix
A.2 Simple Hybrid Wind System . . . . . . . . . . . . . . . . . . . . . 133
A.3 Simple Hybrid Wind System Time Domain Simulation Example . . 134
x
List of Tables
xi
Acknowledgments
First I would like to thank my dissertation advisor, Dr. Tom Hughes who has over
the years taken considerable time to meet with me weekly and provide guidance
on this research as it evolved. This project would simply not have been possible
without his support. Also there is co-advisor Dr. Susan Stewart, present since the
very beginning, who has consistently supported me through rough patches. She
also introduced me to some of the interesting aspects of wind energy and encour-
aged me to pursue work and opportunities associated with it.
I would like to thank all the committee members for taking the time to review
my work and provide feedback. Special thanks are in order for Dr. Jeffrey Mayer
and Dr. James Turso who have put in countless hours over the past year advising
me on the electrical engineering matters that my background lacked.
Finally, I must thank my wife who has been more than patient over the years
as I pursued this work. Whenever a difficult task was encountered or progress
stalled, she was there to provide the necessary encouragement and confidence to
power through.
xii
Epigraph
" The wise man doesn’t give the right answers; he poses the right questions."
- Claude Levi-Strauss
xiii
Chapter 1 |
Background
One could argue that it is the largest machine ever created. Colloquially known
as "the grid", it is a large network of electrical devices that despite its staggering
complexity is so ubiquitous and familiar to the average person it is practically a ne-
cessity. Generators, motors, and other devices are all electrically coupled together
in a network that crosses borders and continents. Although the technologies and
fuels it uses change with the times, the basic structure has been largely unchanged
since its creation. Electricity is generated by large centralized power plants using
synchronous generators. It is distributed across regions using a high voltage alter-
nating current (AC) three phase transmission system. At substations the power
steps down to lower voltages and is distributed over a local area. Finally the elec-
tricity makes its way to our homes and businesses in the (relatively) low voltage
distribution system that lines our streets with poles, lines, and transformers. This
structure, shown in Figure 1.1, was developed early in the twentieth century and
with the conclusion of the current wars, in favor of Tesla’s AC power, has persisted
since [1].
1
Figure 1.1: Structure of the Grid
growth, like long distance transmission lines and large centralized plants. A pref-
erence for more flexible solutions and technologies has appeared. The following are
some technologies and ideas that have been used to delay traditional investments
and allow the grid to supply an increasing customer base.
• Demand Response (DR): To better balance and stabilize the grid some
customers voluntarily decrease consumption in exchange for financial com-
pensation. The ubiquitous use and availability of fast digital communication
allows DR to be quickly implemented when the grid is stressed. This allows
the transmission system to be more fully utilized.
2
• Distributed Generation: Placing small scale generation and combined
heat and power (CHP) at the substation level can overcome economies of
scale by achieving high efficiency and deferring infrastructure upgrades [4]
Some of these solutions can have negative effects as well as positive. This is
especially true for renewables and distributed generation, which in small amounts
have negligible adverse effects. However as penetration grows, conflicts can arise
between DG and the grid due to the existing grid’s rigid structure. Renewables
and distributed generation produce (from the utility’s perspective) highly variable
and uncontrolled power. As stated previously in Section 1, the basic structure of
the grid; one directional flow of power from centralized plants to end users through
a hierarchical AC transmission system, was not designed for two way power flow
or for accommodating scores of independently controlled generators. Although
its structure has remained largely unchanged in the face of new technologies and
fuels, there is reason to believe change is possible. Advancements in computing,
communications, and especially power electronics technology, has made decentral-
ized approaches to grid operation more feasible. Microgrids have been proposed
3
as an alternative, or at least an augmented, approach to modernizing the grid and
addressing some of its limitations.
1.2 Microgrids
What exactly is a microgrid? Although there is no agreed upon exact definition,
the general consensus states they are autonomous electrical networks with power
capacities in the kW to MW range. This definition would technically include
several related applications not usually labeled as microgrids, such as remote sites
and vehicles.
• Remote Sites: In instances where grid power is not available (cabin, cell
tower, rural community) electricity needs are generated and distributed on
site. Commercial-off-the-shelf (COTS) generators mimic properties of the
grid so that the user can connect regular loads and devices. These systems
are expensive to run as small generators are typically less efficient and require
a higher quality fuel than their larger scale counterparts [7]. Also modularity
and scalability are limited. Generator control is simple for a single unit
system. Connecting many units together, however, requires coordination to
operate smoothly and efficiently.
In this work the modern microgrid concept of an electrical network with dis-
tributed generation will be assumed. Such systems are expected to contain renew-
ables, energy storage and be able to work independent of or in concert with the
regular grid. Incorporating microgrids into the main grid is expected to provide a
number of potential benefits to both the utility and user, such as:
4
• Reliability: A microgrid invariably must contain some generation sources.
In the case of a grid failure, the microgrid can island itself and operate
independently until power is returned. A microgrid differentiates itself from
a simple backup system in that it is expected to smoothly transition between
modes, run parallel to the grid, and sustain itself for longer periods of time [4].
• Combined Heat and Power: CHP is highly efficient from an energy view-
point since "waste" heat is utilized to offset electricity demand. Heat, unfor-
tunately, is difficult to physically transport, necessitating a heat distribution
system of steam or chilled water pipes. Microgrids allow many smaller CHP
generators to be placed at load locations in lieu of a single generator with a
complex heat distribution network. [4]
• Power Quality: Microgrids can customize the power to meet the specific
needs of the load. Incoming power quality issues such as low displacement
power factor can be corrected for and loads supplied with the correct form
and quality of power that they require. Although, as will be discussed in sec-
tion 1.5, power electronics-based microgrids can introduce new power quality
problems of their own.
5
for the conventional grid, these techniques don’t transfer to microgrid design as
well as one would expect. The problem arises because microgrids, despite their
name, do not necessarily behave like small versions of the main grid. There are
a number of key differences between the main grid and a (modern) microgrid, as
described in Table 1.1.
Note that Table 1.1 is not absolutely definitive as microgrids can blur the lines
of distinction, containing some combination but not all of the qualities mentioned.
A more detailed breakdown and explanation of the properties is presented in Ap-
pendix A.1.
1.3 Operation
Microgrids are very versatile. They are not restricted to operate in the way the
main grid does, although there are advantages to doing so. They can be DC or
AC, they can adopt a convenient fundamental frequency and voltage, they can im-
plement new control techniques. They can be freely customized to meet the needs
specific to their location and purpose. This greater versatility comes at a cost as
it has made drafting design rules and standards very difficult [13], although it has
been attempted recently with the publishing of IEEE 1547 family of standards for
"Interconnecting Distributed Resources with Electric Power Systems".
Vehicle electrical networks and microgrids actually have a lot of cross over.
In vehicles the design problem may be challenging but designers have complete
control over the components to be included in the system. Also, the final prod-
6
uct is usually to be mass produced so the design costs can be sufficiently diluted
to allow a process of iteration and improvement. This (ideally) results in highly
performing cars and aircraft, that meet expectations of safety and reliability. An
exception in this category is ships. The scale, time, and cost of construction limits
the experiment and refinement possible before completion, incentivising designers
to pre-empt problems or solve them early. Much of the work in this dissertation
is based on research originally undertaken to solve ship based electrical network
problems.
• Plug and Play: System can be reconfigured or accept new hardware with-
out significant re-engineering required. Since microgrids must be adapted
to local resources, needs, loads, etc, there is no "optimal" design to be cre-
ated and mass produced. A microgrid must be capable of accepting new
generators and loads without the need to rework the whole system.
7
Appendix A.2.2. Although the best architecture and method of operation is open
to debate, the necessity of power electronics is not. They are the unifying common
component in modern microgrids and they present new challenges to the design
process.
8
Figure 1.2: Some Basic DC-DC Converter Topologies
The boost converter steps voltage up, the buck steps it down, and the buck-
boost can do both but inverts the polarity at the output. There are many versions
and types of power converter circuits, but the operating principles remain the
same. High power semiconductor elements create a nonlinear circuit tailored to
some electrical conversion operation. Unlike their electromechanical counterparts
in AC systems, these converters can be dynamically controlled and used to perform
many other tasks in addition to voltage transformations. For example, later in this
dissertation, see section 4.4.2, a boost converter will be designed and modeled in
Matlab to mechanically load a wind turbine, achieving maximum aerodynamic ef-
ficiency.
Critically, power electronics allow not just DC-DC voltage transformations but
also efficient power conversion between DC and AC networks. The fundamental
building block of conversion between three phase AC and DC is the half-bridge
circuit, Figure 1.3. When power is converted from AC to DC the device is called
a rectifier. When power is converted from DC to AC it is called an inverter. His-
torically most electrical loads (ex. motors) required AC power and the remainder
(ex. incandescent lighting) were indifferent to it. However today many modern
devices (virtually all electronics and computers) run principally on DC power. As
distribution of DC power is as yet not practical except in special circumstances,
the solution is to build rectifying power supplies into the loads [1, 9, 17]. Therefore
virtually all electronics that plug into an electrical outlet contain a power rectifier
of some kind.
The counterpart to the rectifier is the inverter. While not as ubiquitous, its use
is increasingly widespread. It has two primary uses, interfacing DC power sources
9
Figure 1.3: The Half Bridge Circuit
10
by the speed of the synchronous generators [19]. DC sources such as photovoltaics
(PV) and batteries produce voltage through electrochemical reactions and have no
moving parts. Since most loads, including DC ones, require a standard AC voltage
input, inverters are necessary whether connected to the grid or not.
System designers can balance the benefits and costs to build systems adapted to
11
their environment and specific circumstances. However there is a critical downside
that transcends the trade-off process; power electronics induced instability. An
unstable system will, at least under certain conditions, seriously malfunction. This
offsets possible efficiency or performance benefits.
dv v
rin = =−
di i
12
1.6 Design Challenge
In high level microgrid design, site specific information such as local resource and
demand data, is used to find Pareto optimal combinations of equipment. This
type of analysis can be accomplished through statistical means, as is done in
Hybrid2 [23], or via Monte-Carlo methods in the time domain, as is done in
HOMER [24]. This was explored in previous work modeling the economics of
microgrid designs under different scenarios [25]. An important convenience of such
programs is that they are adaptable to the level of information available. With
coarse data and information, the analysis will produce equivalently uncertain es-
timates. If more detailed data and information is available, results with a higher
confidence can be obtained. Work on custom models show such techniques are
limited only by the resolution of input data and modeling of dynamics, Appendix
A.3, making these analysis tools very powerful in the preliminary phases of design,
and potentially later on given enough information. A key facet of the analysis is
the combination of component modules in different quantities and architectures,
of which power electronics play a significant role, Figure 1.4.
13
Figure 1.4: HOMER Microgrid
Each component may function fine in laboratory, factory, and test settings
but fail when connected to the actual system.
• State Specific: Instability may only occur under a very specific set of rare
operating conditions, making even identification of the problem difficult in
some circumstances.
14
of a lower level stability assessment subroutine are identified as necessary to ac-
complish this. A technique that achieves these goals is expected to be useful to a
wide variety of microgrid design scenarios, although not all may not be necessary
for any given scenario.
15
Figure 1.5: High Level Design and Stability Analysis Relationship
No details about the higher level design process are assumed, requiring gener-
ality and versatility on the stability assessment end.
16
microgrid design program. A summary of theory and methods to assess
microgrid stability as well comments and ideas are presented in Chapter 2.
17
Chapter 2 |
Stability
2.1 Stability
Mathematical techniques for assessing stability can be very powerful but involve
more than a little subtlety. Although an intuitive concept, stability can be com-
plicated when viewed from a mathematically rigorous perspective. This makes it
important to be explicit with assumptions and the type of stability analyzed.
18
2.1.1 Linear Time-Invariant (LTI) Assumption
In reality true linearity is rare, yet analysis based on that assumption dominates
the field of control theory. This is primarily due to the powerful methods available
to analyze linear systems. To assess the stability of a microgrid we also use this
assumption, but there are two important limitations that must be considered.
First, any linear stability assessment is valid only in an operating region where
the system behaves approximately linearly. Components in the system are more
or less "black boxes", so no knowledge of their linearity can be assumed. Stability
assessment is therefore only valid in the neighborhood of the operating point lin-
earized around, hence the term small signal stability. This is overcome by carrying
out the stability assessment over all the operating points in a sufficient density. As
will be shown, this is a measurement and computational burden but establishing
stability for a linearized system makes the problem more tractable.
Second, the switching of the power electronics means the system is not time
invariant. This problem is solved by assuming averaged models of the power elec-
tronic converters. System states (voltage and current) are averaged over a switch-
ing cycle allowing them to take on smooth continuous trajectories. Averaging
the state space models of power electronics is commonly used to simplify analy-
sis [18, 26]. However in the frequency domain, the averaged model breaks down
when approaching the switching frequency of the components, which in turn in-
validates small signal stability analysis there. Since the the negative incremental
impedance effect of power electronics is due to the inner control loop, the danger
of instability is only present over frequencies at which active control occurs. The
results of stability analysis can therefore be trusted when the switching time is
much less than the time constant associated with control action. This is simple
to ensure for a single device but in a microgrid all components must be consid-
ered. Therefore technically for the analysis to hold, the fastest control action of
the fastest device must be slower than the slowest switching device, Figure 2.1
19
Figure 2.1: Switching and Control Frequency Issues
20
PMSG wind power systems [27–31], but slow turbines speeds lead to a correspond-
ingly low electrical frequency and therefore also a low switching frequency, Figure
2.3.
21
under quasi-steady state conditions. This restricts the validity of a small
signal stability assessment to at most the common set of operating point
equilibriums between interacting components. In practice this is further re-
stricted to the common operating points at which stability characteristics
can be measured, although some aspects of generalized admittance space
analysis seek to mitigate this, section 2.2.5 [32].
• Switching, Control, and Filters: Valid analysis requires that power elec-
tronics behave linearly, meaning an assumption of averaged operation is re-
quired. The degree to which a power electronic device obeys this is dependent
on the amount of filtering. In the case that stability needs to be assessed
at an interface with little filtering then the switching frequencies must be
greater than the bandwidth of the control (fastest) control system.
• Internal: Concerned with the response of the systems internal states, which
generally correspond to (or some combination of) modes of energy storage,
to a set of initial conditions. Notable is Lyapunov stability which ensures
internal states remain bounded around an equilibrium point when initially
displaced from it.
• External: Concerned with the response of the system’s outputs, which are
usually measurable variables of interest, to system inputs. Also applicable
to state space methods, but more easily understood as an assessment of a
system’s transfer functions.
22
ẋ(t) = A(t)x(t) + B(t)u(t)
BIBO Stability [33]: For any bounded input u(t) and initial condition x0 ,
there exists a finite constant N such that ky(t)k ≤ N for all t
In the case that the system is a LTI continuous system, section 2.1.1. A proper
transfer function can be constructed for a given set of inputs and outputs.
The system is BIBO stable for those inputs and outputs if the poles of the
transfer function are in the open left hand plane (LHP).
BIBO only guarantees that selected finite inputs do not result in selected out-
puts that go to infinity. System states or even the response of the system to no
input can theoretically be unstable for BIBO systems under certain conditions [33].
This deficiency is mostly overcome by testing stability using impedance techniques
at multiple interfaces in a system.
If transfer function H(s) is co-prime, then its state space realization is mini-
mal [34], and it completely characterizes all the internals states. BIBO stability
of the transfer function H(s) then indicates the more stringent bounded input
bounded state (BIBS) stability. This is stronger than both BIBO and Lyapunov
types and more rigorously establishes stability.
23
Since microgrid components in this work are "black boxes", explicitly establish-
ing a transfer function developed from measured data as co-prime is practically
impossible. Controllability and observability would need to be tested, requiring
the true state space realization. If such knowledge were available other simpler
techniques could be used to determine stability (ex. eigenvalue analysis). Being
electrical devices however, we can assume that the dominant energy storage modes
are functions of voltage and current. Impedance transfer functions, which account
for the relationship of these two variables, will likely capture the dynamics of these
modes, and therefore allow the presumption of actual stability from BIBO stability
(at least locally).
24
The transfer function is derived from the Thévenin equivalent circuit, see Fig-
ure 2.5.
Zl Voc vload
vlink = =
Zl + Zs hl
voc = hs vs
hs hl
vl = vs
1 + ZZsl
Disturbances in the source voltage are the input, resulting load voltage is the
output. Transfer function hs is a passive filter so it will never have right-hand
plane (RHP) poles. Transfer function hl will not have RHP if the converter was
designed to be stable with an ideal voltage source, which can be expected [20].
Zl Zs
vlink = vs + vl
Zs + Zl Zs + Zl
Ns
Zs =
Ds
Nl
Zl =
Dl
25
Nl Ds vs + Ns Dl vl
v=
Nl Ds (1 + ZZsl )
The link voltage is the output, while both source and load voltage disturbances
act as the input. If the source is designed to be stable when connected to a constant
load current and the load is designed to be stable when connected to a constant
voltage source then Nl and Ds do not have any RHP zeros.
1
1 + ZZsl
Zs
1+
Zl
26
vectors, divided by pole product vectors. By sweeping through the contour, a new
contour is created in the w-plane. The encirclements of the origin indicates the
relative number of poles and zeros encircled in the s-plane. By choosing a con-
tour, called the Nyquist contour, that encircles the entire RHP a Nyquist plot is
created. The Nyquist stability criterion states that the number of zeros is equal
to the number of critical point encirclements plus the number of poles.
Z =N +P
The critical point is the origin by default, but changes as the w-plane is shifted
to account for denominator constants in the transfer function. So in the derived
electrical system transfer function, and typical feedback systems, the desire is to
find zeros of 1 + G(s). The plane is shifted and the critical point becomes −1 + 0j.
27
has no RHP poles.
Ns
v(s) = Zs (s)i(s) = i(s) → Ds 6= RHP zeros
Ds
v(s) Dl
i(s) = = v(s) → Nl 6= RHP zeros
Zl Nl
Zs Ns Dl
= 6= RHP poles
Zl Ds Nl
The plotted transfer function has no poles. This by itself does not have a direct
effect on stability but prevents the zeros from "hiding" behind poles in the RHP.
This simplifies analysis because any unstable poles with manifest in encirclements
of the critical.
• GMPM [36]: Gain margin and phase margin criterion was developed in
1995 to be less conservative then the Middlebrook criterion by forbidding
less of the w-plane. It enforces a desired gain and phase margin but allows
the return ratio to inhabit the space outside of the region allowed by the
Middlebrook criterion, as long these margins are maintained.
28
• Opposing Argument Criterion [37]: Actually more conservative then
the preceding GMPM criterion, but more easily applied to parallel loads [38].
The forbidden region is denoted by a vertical line in the w-plane. The location
of this line is specified for individual loads and is in terms of their proportion
of power consumption and the source impedance.
Over time there has been a general trend of decreasing conservatism and in-
creasing computational requirement. This reflects the rapid strides made in com-
puter power since the Middlebrook criteria was proposed in 1976. Although more
computationally intensive, the less conservative criteria allow for less bulky pas-
sive filters to ensure stability [38]. A visual summary of the stability criteria is
presented in Figure 2.7.
29
Figure 2.7: Visual Summary of DC Stability Criteria
For electrical systems let the system input be voltage, output be current, and
the storage function be a measure of the energy stored in the system.
v(t)i(t) ≥ 0
Z t
v(τ )i(τ )dτ ≥ 0 for all t
0
The electrical system will be passive if it can only absorb energy. As derived
30
in [41] this reduces to a requirement in the frequency domain that
1Z∞
Re[Z(jω)]kI(jω)k2 dω > 0 for all I
π 0
This is true for a one port network if the Nyquist plot of Z(jω) lies entirely in
the closed RHP [41].
The passivity-based stability criteria has several properties that make it more
general then the other techniques presented.
There are some issues however, it can be difficult to ensure stability margin
using PBSC, although modifications have been offered to overcome this [43].
31
• Load Impedance Specification: The source impedance is applied to a
selected minor loop gain criteria, section 2.2.3, to specify an allowable range
for the load admittance (reciprocal of impedance). Thus making it design
oriented.
Additionally the method has been implemented and released open source as
a Matlab toolbox. The DC Stability Toolbox is coordinated by Dr. Scott Sud-
hoff and available online [44]. This practical tool automates many aspects of the
stability assessment and graphics, requiring only that impedance data be provided.
32
" #−1
Zdqs
vdq = I+ Hdq vdq
Zdql
The transfer function is analogous to that derived in section 2.2.1, except the
terms are now 2x2 matrices. Again the numerator transfer function Hdq is consid-
ered a priori stable if the source is stable by itself, which will always be the case for
a RLC filter no matter how complicated [43]. Stability determination is likewise
dependent on the the matrix transfer function minor loop gain.
[I + Ldq (s)]
D Channel Criterion [46]: In the case that the three phase system has high
power factor, stability can be determined exclusively by the dd channel impedances
values; Z(s)sdd and Z(s)ldd . The D channel can be constructed and analyzed using
the DC stability criteria, section 2.2.3. This forms a powerful duality with the well
33
developed area of DC stability criteria. Although it must be noted that translation
reduces the SISO Nyquist results to being sufficient but no longer necessary for
stability [47]. That is, there is a built in conservatism that is above and beyond
the margins already present in the DC criteria.
4. Check qualitatively for stability using Nyquist stability criterion (or GNC
for AC systems)
34
and sufficient condition. This is done similarly to the method proposed in [40],
except additional conservatism is removed because conditional stability is allowed.
This is accomplished using a simple algorithm that tracks zero crossings of the
boundary (−∞ + 0j, −1 + 0j), Figure 2.8.
X X
N= pn − np
This is applied to the simple DC LTI circuit in Figure 2.9, an example borrowed
from a Virginia Tech power electronics stability tutorial [48]. Although the system
a) is stable, as shown in time domain study, Figure 2.10, the resulting Nyquist plot
fails all the stability criteria listed in section 2.2.3. Application of the simple zero
crossing algorithm to the Nyquist plots correctly predicts the stability, figures 2.11
and 2.12
35
Figure 2.9: DC LTI Circuit Diagram
36
Figure 2.11: DC LTI example Stabe Nyquist Plot N = +1
37
return ratio Nyquist plot and the critical point, Figure 2.13. This is based on the
MPC and work in [40].
Note: that stability margin does not indicate stability. In fact as evidenced
in Figure 2.13 part b), a highly unstable system can have a margin better than a
stable one.
2.4.3 Interfaces
Describing the stability margin and assessing stability can be complicated by the
choice of measurement interfaces. The problem lies in the sensitivity of the return
ratio to component groupings. This can be demonstrated for the stable LTI DC
circuit in Figure 2.9. The circuit is simple so the return ratio can be analytically
determined and plotted for interfaces both pre and post capacitor. Let interface 1
be post-capacitor and interface 2 be pre-capacitor, then the return ratios will be
as below.
Zs1 Rs + sLs
L1 = =
Zl1 (1 + sRs C + s2 Ls C)(Rl + sLl )
As can be seen, the return ratio transfer functions differ significantly. The re-
turn ratios can be compared to each other in the w-plane, see Figure 2.14. Both
38
return ratio plots are of the same system, so despite their differences they describe
a system with the exact same stability margin.
This effect has two important implications. First, stability criteria with con-
servatism may fail at some interfaces even if the system is stable. Ostensibly this
is solved by considering a system to be stable if a stability criteria is met at any
interface. However, as described in section 2.1.2, the system states can only be de-
termined stable if the transfer function is a minimal realization of the system [20].
This is generally not the case and even if it were, it requires observability and
controllability to be confirmed, which is difficult to do using empirically obtained
impedance data. It does guarantee that interfaces with a common voltage can be
determined stable if either interfaces meets the stability criteria. Therefore inter-
faces separated only by a capacitor will be stable if either interface meets the given
stability criteria. Interfaces separated by other components, such as inductors or
resistors, will need to be checked if they are not part of a purely passive network.
The second implication is that stability margins cannot be used to describe the
39
system in an absolute way. Different interfaces result in very different values of
stability margin, making a quantitative comparison of different systems difficult.
Development of a new stability margin definitions to solve this issue remains an
open problem [40].
40
Figure 2.15: Military EMS Microgrid Concept
41
fast, the negative effects may be avoided. This would be an additional advantage
to adopting an intelligent EMS and to making microgrids more adaptable to future
components and situations that are not now anticipated.
42
Chapter 3 |
Simulated Impedance Measure-
ment
In chapter 2 it was shown that armed with detailed component impedance infor-
mation, the inter-component compatibility of power electronics can be determined.
In this chapter the details of impedance extraction for DC and AC systems are
covered. An impedance measurement tool built in Matlab’s SimPowerSystems
is presented. It is both a simulation of practical impedance measurement and a
practical tool for determining the operating point dependent impedances of devices
built in simulink. It determines the impedances of arbitrarily (subject to compu-
tational limitations) complex AC and DC power electronic converters. Tests and
verification of the simulated impedance measurement blocks are provided.
Z(s)s
L(s) = for s = σ + jω swept through Nyquist Contour
Z(s)l
Impedance is only considered in the context of frequency (jω) since practical
measurements have to be taken under steady state conditions, meaning σ = 0 [49].
This makes it impossible to construct the Nyquist contour in the s-plane as pre-
43
scribed by the Nyquist stability criterion, section 2.2.2. Fortunately we can exploit
the properties of a strictly proper transfer function.
This property is very useful as it guarantees that the return ratio Nyquist
plot will close through zero if it is a strictly proper transfer function. This can be
ensured by selecting appropriate interfaces to measure. Although the impedance
transfer functions will be unknown, grouping capacitive elements and filters with
the source will increase the return ratios denominator order, section 2.4.3. With
the assumption that the return ratio is strictly proper, impedance measurements to
an arbitrarily high frequency will close the Nyquist plot through zero. Thus effect
of the positive real part sections of the Nyquist contour becomes unimportant,
Figure 3.1.
Z(jω)s
L(s) for s = Nyquist contour ≈ L(jω) = for ω = (−∞, +∞)
Z(jω)l
This section provides justification for applying impedances to the Nyquist sta-
bility criteria as well as further encouragement to choose measurement interfaces
wisely.
44
Figure 3.1: Reduced Nyquist Contour
3.2 DC Systems
Impedance can be defined using generalized Ohm’s law.
V = IZ
Ẑ(ω, xop )
The choice of variables to include in xop and the ranges to measure ω and xop
are dependent on the system characteristics and to some degree the investigator’s
judgment.
Ideally impedance information would be available over the full frequency range,
45
Figure 3.2: Impedance Fitting Results
(a) Raw Data (b) Fitted Data
46
It may be enticing to simply select an extremely large value for ω, but there
are factors in opposition to this.
1
ωs > ωmax >
τ
where τ is fastest connected controller’s time constant and ωs is the device’s
switching frequency. Note this requirement is soft for "source" type power elec-
tronic components because their stability characteristics are assumed to based on
their output filter.
The selection of operating test points X̂op and ranges is even more nuanced
than selection of ω. In this work operating points are separated into two categories;
general and specific.
47
for convenience. Temperature, age, and manufacturing variability are excel-
lent examples of exogenous variables to be generalized since they effect the
impedance but can’t be easily controlled. In generalized admittance space
stability analysis, section 2.2.5, component loading ranges are also general-
ized [39].
The range covered by each variable, whether general or specific, should be repre-
sentative of the system’s steady operating conditions. Some engineering judgment
is required to differentiate between transient operating points and quasi-steady
ones.
V(jω)
Z(jω) =
I(jω)
Where ω is the perturbation frequency and V and I are the resulting phasors
measured at that frequency.
In this work shunt (parallel) current injection is chosen as the means to disturb
the system, Figure 3.3, series voltage injection being the alternative. In practical
implementations voltage injection can allow simpler sensors while current injec-
tion requires much less equipment to create the disturbance [50]. In this work
measurement is via simulation so those attributes are irrelevant. Current injection
is chosen because it convenient for testing loads, which most components in the
microgrid will be.
48
Figure 3.3: DC System Current Injection
6. Terminate simulation when voltage and current phasors are within bounds
for n cycles
7. Record impedance and save to data file with operating point information
49
Figure 3.4: DC Impedance Measurement Block
The block was made to be user friendly, allowing custom source and load cir-
cuits to be easily connected, Figure 3.4. The block can also run for a single set of
operating conditions and frequency to troubleshoot problems or optimize speed.
Diagnostic results are automatically generated and opening the block will yield a
number of useful plots and measurement values, figures 3.5 and 3.6.
50
Figure 3.5: Frequency Sweep Diagnostics
51
Figure 3.7: Simple LTI Circuit Test Diagrams
(a) Circuit (b) Simulink Implementation
Zl (jω) = Rl + jωLl
1 Rs + jωLs
Zl (jω) = (Rs + jωLs )k →
jωC jωRs C + 1 − ω 2 Ls C
Plots of the source and load impedances for the analytical results and those
measured with the impedance measurement block is presented in Figure 3.8. The
parameter values are presented in Table 3.1.
Parameter Value
Vs 100 V
Rs 1Ω
Ls 5mH
C 200 µF
Rl −0.1Ω
Ll 2mH
For a more challenging test with a nonlinear device, the impedance measure-
ment block was applied to the output of a six pulse diode bridge. In [51] the
analytical impedance of a filtered diode bridge is presented. The diode bridge
from this work is modified to create two versions, a "source" type and "load" type
52
Figure 3.8: Simple LTI Impedance Results
3
R∗ = Lac ωs + R + 2Rac
π
L∗ = L + 2Lac
53
Figure 3.9: Diode Bridge Circuit Test Diagrams
(a) "source" type (b) "load" type
3
Zl = Lac ωs + 2Rac + j2ωLac
π
The "source" type has a capacitive output filter, making it ideally paired with
a constant current load. The "load" type is stripped of this output filter making it
ideally paired with a constant voltage source. The results can be found in figures
3.10 and 3.11, and the parameters used are presented in Table 3.2.
54
Figure 3.11: "Load" type Diode Bridge Impedance Results
Parameter Value
Vrms 120 V
Rac 30mΩ
Lac 0.22mH
L 0.9 mH
R 60mΩ
Rc 30mΩ
C 1.05 mF
55
to the synchronous reference frame using the dq0 transformation.
Under the conditions that the three phase system is balanced, this transforma-
tions results in the three time varying quantities (abc) being reduced to two (αβ)
and one stationary zero (γ = 0). The two time varying quantities can then be
made stationary by rotating the system about γ with the fundamental frequency
(typically 60Hz for terrestrial domestic systems).
cos(θ) sin(θ) 0
Trotate = − sin(θ) cos(θ) 0
0 0 1
Z τ
θ= ωdt , where ω is the fundamental frequency
0
56
Figure 3.12: dq0 Transformation of Three Phase Voltage
(a) abc (b) αβ (c) direct quadrature
57
Figure 3.13: Injector Block
The AC impedance block follows the same procedure as the DC block, section
3.2.2, except with two key differences.
58
behind the A phase (by convention). In this work we determine the reference angle
θ using either a phase locked loop (PLL) or by assuming a proportional relationship
with the more easily measured rotor reference frame.
• PLL: A phase locked loop uses a low bandwidth PID controller to synchro-
nize the phase of an exogenous signal with an internal reference. This internal
reference is supplied to transform abc measurements into the dq frame and
dq injection commands into the abc frame. PLLs have been used in AC
impedance measurement systems [52], but they can negatively interact with
the disturbance, leading to errors. These errors can cause significant inaccu-
racy in the resulting impedance measurement [55]. The solution to this is to
ensure the bandwidth of the PLL is much lower than the lowest frequency
being measured. This however increases the time needed to synchronize and
reduces the PLL’s ability to follow changes.
• Rotor Reference: In this work, as in [20], the rotor reference frame and syn-
chronous reference frame of synchronous generators is assumed to be rigidly
coupled, at least in quasi-steady circumstances. This allows the rotor posi-
tion to be used as an alternative source for synchronous reference angle. In
practice this might be avoided due to the extra effort of mechanical mea-
surement, but in simulation this is not a problem. It obviates the need for a
PLL, removing the associated error and computational expense.
59
Figure 3.15: Simple AC LTI Circuit Diagram
plots for phase angle, see figures 3.16 and 3.17. There is excellent agreement be-
tween the plots.
The input impedance of the six pulse diode bridge in [51] was also found via
simulation, Figure 3.18. An analytical solution for the impedance using an aver-
aged model, although proven to be incorrect a higher frequencies, was available
for reference in [51]. The results of the AC impedance measurement block can be
found in figures 3.19 and 3.20.
60
Figure 3.16: Simple AC LTI Circuit Impedance Magnitudes
61
Figure 3.18: Six Pulse Diode Bridge Diagram
62
Figure 3.20: Six Pulse Diode Bridge Input Impedance Angles
63
Chapter 4 |
Notional Microgrid
A small scale notional microgrid containing a wind turbine, battery, and inverter
is developed and presented as a subject for stability investigation. It was modeled
in detail down to the switching level elements and inner current control loops
in Matlab and simulink. Alternate rectifiers and inverters are implemented for
comparison. The performance of the system is simulated along with the impedance
measurement blocks from chapter 3. The impedance characteristics of each device
are systematically determined for a range of conditions and the data stored in files
for easy assessment by stability routines later.
4.1 Architecture
In chapters 2 and 3 tools and methods to analyze the stability of a microgrid were
developed. In this chapter a DC power electronics-based microgrid is developed
for the demonstration in Chapter 5.
In this work microgrid architecture refers not just to physical layout but also
the selection of nominal interface variables and dispatch strategy. The architec-
ture of a microgrid can take many forms, and design is based on a number of
factors. The literature is rich with design strategies [57, 58] and control methods
to automate and regulate DC microgrids [11, 59–63] and AC microgrids [64, 65]. A
general hybrid microgrid architecture, Figure 4.1, is often assumed in higher level
optimization [24]. Adoption of true hybrid microgrids like this for non-vehicle ap-
plications has been slow [12, 13], due at least in some part to the ubiquity of AC
systems and lack of mature commercial DC protection devices [17, 66]. Microgrids
64
Figure 4.1: General Hybrid Microgrid
with high penetrations of renewables and power electronics will naturally contain
many DC buses, Figure 4.2. With proper control, the voltages of the DC buses
vary proportionally with each other, effectively operating like Figure 4.1 from a
high level perspective. The distinction is however important for assessing power
electronics stability. To ease stability analysis the grid can be broken into smaller
parts, in this chapter the analysis of a single DC leg will be presented.
65
Figure 4.2: Power Electronics-based Microgrid
ally or not, any DC bus will have at least some capacitance associated with it.
This naturally changes the voltage, a variable locally available to all connected
devices, in response to a net power imbalance, therefore making it a "soft grid
maker". The inverter can either inject power according to an exogenous signal,
"hard grid taker", or inject power in response to the bus voltage, "soft grid taker".
These designations are not absolute and depend on perspective, this is illustrated
in Figure 4.4 by applying to the conventional grid and droop control. The des-
ignation even applies to mechanical components such as the governor and prime
mover. The governor acts as a "soft grid taker" changing power in response to
local line variables (speed/system frequency). The inertia of the prime mover and
synchronous generator act as "soft grid maker" because it translates a mismatch of
power into a locally measurable bus variable, system frequency. When many units
are in parallel the energy stored in the rotating inertia becomes very large and the
frequency of the AC bus is regulated very tightly, making the collection of sources
a "hard grid maker". While these cases are simple they illustrate how the categoriz-
ing of microgrid components can be useful in creating a more complicated network.
66
Figure 4.3: DC "leg" Architecture Examples
• Every bus (common node) must connect at least one "maker" and "taker"
component.
• Every bus (common node) can only connect to one "hard grid maker" which
will negate the effects of any "soft grid makers".
67
Figure 4.4: Conventional Grid Architecture Example
Parameter Value
DC bus voltage 300 V
AC bus voltage 120 Vrms L-N
AC bus frequency 60 Hz
68
Figure 4.5: Notional Microgrid Layout
ponent (along with variants and alternatives) were designed and implemented in
Matlab simulink.
1
PW = Cp (λ)ρAV 3
2
ωR
λ=
V
For modeling purposes it is convenient to fit the power coefficient and tip speed
ratio data to a third order polynomial. An example of this is presented in Figure
4.6. Each wind turbine will have unique aerodynamic characteristics and therefore
have a unique set of coefficients to describe the power coefficient function, Cp (λ).
The turbine parameters are presented in Table 4.3.
69
Figure 4.6: Example Cp vs TSR Curve
Parameter Value
Radius 1m
kg
Air Density ρ 1.2 m 3
a3 -0.0022
a2 0.0207
a1 0.0098
a0 0.0124
Cp (λ) = a3 λ3 + a2 λ2 + a1 λ + a0
70
Figure 4.7: Inside Wind Turbine Simulink Block
71
SimPowerSystems includes a library of motor models, each capable of receiving
rotor speed as an input and having electromagnetic torque as an output Te . A
wide range of machine preset models are available as well as the capability of
defining parameters, negating the need to create a custom model.
The two models are combined in an algebraic loop to form a combined wind
turbine and motor model that connects to the variable voltage and frequency three
phase bus, Figure 4.8.
4.4 Rectification
The rectifier fulfills two important functions for the wind turbine and motor.
• MPPT: Controls the amount of current to properly load the turbine for
given wind conditions
72
As shown previously in Figure 4.6, the aerodynamics of the blades result in
the turbine being very sensitive to the ratio of wind speed and rotation speed.
To produce the maximum power, the wind turbine would ideally be spinning at
the optimum speed for every given set of wind conditions. This process is called
maximum power point tracking (MPPT). It is accomplished by varying the motor
currents drawn to speed or slow the turbine to the optimal speed. A wealth of
MPPT algorithms are available in the literature [67–70]. For this work the torque
control MPPT from [70] is implemented.
T (ω) = Kopt ω 2
1 R5
Kopt = Cpmax ρπ 3
2 λopt
The speed of the MPPT algorithm is limited by the turbine inertia, which
is large. This is beneficial from a stability viewpoint as it won’t interact with
the relatively high speed power electronics controls that cause instability. Thus
allowing the MPPT control to be neglected in the small signal analysis. However
the means of implementing the MPPT torque control does effect the stability
characteristics. Three rectifiers are designed, each with a different power electronic
converter structure, but equally valid components for the microgrid, Figure 4.9.
73
the price of converter efficiency [71]. Other more sophisticated hardware exists,
but even then the diode bridge is still present with the additional active compo-
nents, Sections 4.4.2 and 4.4.3.
Since the passive diode bridge rectifier has no active controls, proper design is
critical to achieving acceptable overall efficiency of the system. This is primarily
accomplished through the selection of the DC bus voltage. Until the motor line-to-
line voltage exceeds that of the DC bus, the diodes will prevent any current (and
therefore electrical power) flowing from the motor. The produced motor voltage
amplitudes depend on the rotational speeds. A convenient expression describing
this uses a constant Cv (L-N volts per rad/s) that is specific to a given motor.
√
VLL = 3Cv ω
Until the turbine reaches the speed necessary to overcome the DC bus voltage,
the motor is essentially open circuited, the turbine free spinning. Once sufficient
speed is obtained, current begins to flow and the resulting electrical counter torque
resists further increases in speed. Under quasi steady conditions the wind speed
and turbine speed are therefore linked. TSR for a given speed and overall power
production of the system is therefore dependent on selection of the DC bus voltage.
ω = f (Vdc , Vwind )
ωR
λ= = f (Vdc , Vwind )
Vwind
1
PW = Cp (λ)ρAV 3 = f (Vdc , Vwind )
2
The selection of DC bus voltage will be subject to the availability of DC storage
options (battery string voltages) and the requirements of the post rectifier inverter.
A problem of competing specifications can be overcome with the addition of gear-
ing between the motor and turbine; making PW = f (GR , Vwind ).
74
Figure 4.10: Turbine Diode Rectifier Diagram
These operating points are varied over their expected operating ranges, Table 4.4.
The mechanical and electrical power are measured in simulation along with other
variables so that reduced order impedance functions can be made.
Once the impedance measurement runs are complete, a data file containing
impedances and operating points was analyzed for trends. Figure 4.12 displays
impedance plots for different operating point combinations and a pattern is ev-
ident. A parametric analysis, Figures 4.13 and 4.14, shows the dependence of
impedance on each of the input variables, ω and V . In Figure 4.13, each pathway
represents a set of impedances with constant voltage V and varying motor speed
ω. Red indicates high values, blue low. Figure 4.14 is the same with the roles of
75
Table 4.4: Diode Rectifier Operating Ranges
voltage and speed reversed. It is evident that impedance is not easily predicted
from speed or voltage set points alone.
76
Figure 4.13: Diode Bridge Impedances with Constant Voltage
high impedance for low power is due to the diode blocking, essentially creating an
open circuit when the bus voltage is higher than that produced by the motor at a
given speed.
77
Figure 4.15: Diode Mechanical Power Impedance Surface
78
It allows MPPT control and increased efficiency but without as many switching
components, sensors, or computational requirements as a full switching bridge rec-
tifier [72]. Generally speaking, the converter does not necessarily need to be of
boost type, but in this case the battery voltage is higher than that produced by
the motor under typical operation. Voltage boosting alleviates this issue without
the need for mechanical gearing, a preferred choice in small scale systems.
79
Figure 4.18: Boost Converter Circuit
Parameter Value
Filter C 200 µF
Boost L 100 µ H
Relay Tol ± 0.25 A
MPPT kopt 0.004757
Current Control kp 10
Simulation of the switching model over long time scales is impractical, requiring
about 5 minutes of calculation per second of simulated time. To test the MPPT,
the boost converter is replaced by a controllable ideal current source. Time do-
main simulations were then carried out, Figure 4.20 illustrates the control system
80
Figure 4.19: Motor Rectifier Boost Control Flow
effectively administering MPPT for a wind turbine start up scenario with constant
wind speed (7 m/s). Figure 4.21, illustrates the system following changing wind
speeds. The switching model was simulated over a short time period (about 1
second) to confirm the current control was working as intended, Figure 4.22.
81
Figure 4.21: Boost MPPT and Variable Wind
82
Impedance measurement at the boost converter output is conducted as in Fig-
ure 4.17. The wind turbine model is excluded because wind is a poor choice for
an input variable given the decoupling between turbine speed and wind speed.
Motor speed ω and current command I ∗ are selected as operating point variables.
Sweeps are conducted over nominal operating point ranges, Table 4.6, using the
DC impedance measurement block. As in section 4.4.1, impedance is characterized
as a function of electrical power by aggregating results together and creating an
impedance surface using multidimensional interpolation. Results are presented in
Figure 4.23 and 4.24. Note high (infinite) impedance values are rounded down to
make the plots easier to read.
83
Figure 4.24: Boost MPPT Rectifier Impedance Angles
The popular technique of motor torque loading through q-axis current com-
mands is adopted to apply the MPPT [73–75].
3 3
Te = (Ld − Lq )id iq + λm iq
2 2
λm = λP where P = motor pole paris and λ = flux linkage
84
Figure 4.25: Switching Bridge Rectifier
motor. By eliminating the d axis current, the motor can produce a maximum
torque for a given total current magnitude purely through the q axis current.
To control the switches a pulse width modulated (PWM) PI current controller
with decoupling terms is designed and implemented in Matlab simulink. A flow
diagram of the control is presented in Figure 4.26. Rotor angle is used to establish
a synchronous reference frame to transform the abc variables into the dq frame. In
the dq frame a permanent magnet synchronous generator (PMSG) can be modeled
using the following equations [73, 75].
did
Ld = −Ris + Lq ωr iq + Vd
dt
diq
Lq = −Riq − Ld ωr id − λm ωr + Vq
dt
A full derivation of the PMSG equations along with PI controller design equa-
tions are presented in Appendix C.3. The parameter values and resulting controller
gains are summarized in Table 4.7.
Time domain simulations with the turbine connected are carried out to confirm
proper MPPT operation. Figure 4.27 shows system response to start up with con-
stant wind speed of 7 m/s and Figure 4.28 shows MPPT operation during variable
wind speeds of 5 m/s to 12 m/s.
85
Figure 4.26: Switching Bridge Rectifier Control Flow
Parameter Value
Motor R 0.9585 Ω
Motor L 0.00525 H
Motor Poles 6
Motor λ 0.1218 V s
PWM fs 10 kHz
Control gain, kp 9.25E1
Control gain, ki 958E1
86
Figure 4.27: Switching Bridge Rectifier: Startup
87
Figure 4.28: Switching Bridge Rectifier: Variable Wind
88
Figure 4.30: Switching Bridge Rectifier Impedance Magnitudes
89
Figure 4.31: Switching Bridge Rectifier Impedance Angles
4.5 Battery
As discussed, section 4.1, every bus needs a grid "making" component. For the
DC bus, this component will regulate voltage. For this notional system a battery
is chosen as the regulator. Individual cells are assumed to be placed in series
and parallel to be nominally 300 V across terminals and capable of handling the
expected power requirements without an unacceptable rise or drop in voltage.
Batteries themselves are dynamic, changing with age, temperature, and charge [76].
90
Battery modeling is a sophisticated area of study, being dependent on a number
of complicated electrical, chemical, and mechanical factors [77].
• aging effects
• mass transport
Over the timescales that power electronic converter instability occurs (fractions
of a second) many of those battery effects are irrelevant. A common modeling tech-
nique is to develop an equivalent circuit to represent the battery [78]. Given the
premise that impedance measurements will be in the 1 Hz or higher range, an
equivalent circuit model including electric double layer and high frequency electric
and magnetic effects is created, Figure 4.32. Battery parameters are chosen to
produce reasonable results, Table 4.9, they will later serve as centers for ranges
of battery values. Figure 4.33 shows the impedance plot of the battery nominal
values with real and reverse imaginary axes, as is convention in the literature,
for comparison. Note, the semicircular curve in the upper quadrant is due to the
charge resistance and double layer capacitance while the bottom is leg is due to
the series inductance (a high frequency effect).
Parameter Value
R 100 mΩ
L 0.1 µH
RCT 0.5 Ω
CDL 1 mF
91
Figure 4.32: Battery Equivalent Circuit
Figure 4.34, showing the impedance curves for different sets of battery conditions.
In this case the model is a linear circuit, making loading irrelevant. Instead the
charge transfer resistance RCT and double layer capacitance CDL are varied to
mimic the uncertainty associated with the exogenous factors affecting batteries;
age, cycle, etc.
92
Figure 4.34: Battery Equivalent Circuit Impedances
4.6 Inverter
The inverter transfers power from the DC bus to the three phase 120/208 V AC bus.
In this notional microgrid all the inverters act as a power port from the perspective
of the DC bus. Injecting real and reactive power according to an exogenous signal,
therefore classifying them as "hard grid takers" for the DC bus. Two scenarios will
be considered for the AC side of the system, a "stiff" grid connection and a "weak"
grid connection.
• "Stiff" Grid Scenario: Represents the case where the connected AC grid is
tightly controlled and regulated. It could be a direct connection to the macro-
grid or an autonomous microgrid with another "hard grid maker" present.
In this scenario the inverter injects real and reactive power according to a
communicated set point, accepting whatever voltage conditions are at the
interface. This is a common mode for small inverters to run in, especially
if they are connected to a renewable source, as they export the maximum
power available at all times. The downside to this type of inverter is that it
only functions in the presence of an AC bus regulated by other entities.
• "Weak" Grid Scenario: Represents the case where the connected AC grid
93
is not regulated. An example would be an isolated microgrid that has is-
landed from the main grid. In this scenario the inverter seeks to maintain
nominal voltage and frequency by injecting an appropriate amount of power.
Therefore the inverter is a "hard grid maker" from the AC bus’ perspective.
This allows the renewable battery microgrid "leg" to operate an autonomous
AC microgrid either alone or in conjunction with the grid and other genera-
tors (with the addition of droop controls, Appendix A.2.2).
Three inverter alternatives for the notional microgrid are designed and imple-
mented in Matlab. Each is a two-level three phase voltage sourced converter but
with different control strategy. All enact their switching using pulse width mod-
ulation (PWM). Figure 4.35 shows the basic hardware structure of the inverters.
It contains six power electronic switches, typically IGBTs or MOSFETS. The DC
bus is divided with a neutral point for analysis purposes, Appendix C.1 for more
details.
Figure 4.35: Two-Level Three Phase Voltage Sourced Converter Diagram
94
sists of the two level three phase converter plus an output filter, Figure 4.37. The
inductive output filter is critical to the control, since a current control scheme is
implemented. Transformed into the αβ frame, the inverter power output is de-
scribed by the following equations [75].
αβ AC power:
3
P = [Vα iα + Vβ iβ ]
2
3
Q = [−Vα iβ + Vβ iα ]
2
Rearranged, current commands can be derived from real and reactive voltage and
power set points:
2 Vα 2 Vβ
i∗α = 2
P∗ + Q∗
2
3 Vα + Vβ 3 Vα2 + Vβ2
2 Vβ 2 Vα
i∗β = 2 2
P∗ − 2 2
Q∗
3 Vα + Vβ 3 Vα + Vβ
di
The inductor equation, L dt = ∆V , in the αβ reference frame [75]
diα VDC
L = −Riα + mα − Vα
dt 2
diβ VDC
L = −Riβ + mβ − Vβ
dt 2
A filter and current controller is designed in [75] using the αβ frame equations,
a diagram of the control is presented in Figure 4.36. Most of the components in this
work are self developed but this design is taken as is and implemented in Matlab
simulink, allowing the simulated impedance measurement blocks to be run on a
more realistic, externally developed design. The parameter values are provided in
Table 4.10 for reference, and the control transfer function follows.
95
Table 4.10: P Resonant Parameters
Parameter Value
R 1.63 mΩ
L 100 µH
VDC 1450 V
fs 25 kHz
VAC 480 Vrms LL
The inverter is was successfully implemented and simulated with different power
commands. Figure 4.38 shows setting of P=1MW and Q = 0.5 MVAR. Impedance
sweeps on the DC an AC side are conducted while varying the real and reactive
power commands, Figures 4.39 ,4.38, and 4.41. Table 4.11 contains the input
ranges swept.
96
Figure 4.37: P Resonant Diagram
97
Figure 4.39: P Resonant DC Impedance Magnitudes
98
Table 4.11: P Resonant Operating Range
In this section, a kW-class inverter is designed in Matlab simulink and will later
serve as the base for the "weak" dq0 inverter in section 4.6.3. It is a "stiff" inverter
delivering power into a frequency and voltage regulated AC bus, Figure 4.42 shows
basic structure. The current commands are obtained from the power commands
in the dq0 frame [75].
99
3
P = [vd id + vq iq ]
2
3
Q = [−vd iq + vq id ]
2
Since the PLL aligns the d axis with the source voltage, vq = 0 in steady state,
simplify and rearrange
2P
i∗d =
3vd
2Q
i∗q = −
3vd
The inner control loop is designed to be fast by pole-zero canceling a stable (but
slow) transfer function pole. Details and derivation of the inner current loop are
in Appendix C.2. The PI gains are defined in terms of the component parameters
and an arbitrary time constant τ . Theoretically with the PI control, the actual
currents will have a first order response to current commands, described by time
constant τ . Selection of smaller τ ensures tighter control but power electronics
stability considerations from section 3.2 will be affected.
L
kp =
τ
R
ki =
τ
Circuit elements are designed to meet current ripple requirements given a se-
lected switching frequency and power loading. Procedure for this is presented in
Appendix C.4.1. This along with the preceding PI control gain equations allow
inverter customization for a given application. A set of parameters is designed for
3kVA rated inverter, allowing 5% ripple, Table 4.12.
100
Figure 4.42: "Stiff" dq0 Diagram
Parameter Value
Power 3 kVA
Allowed Current Ripple 5%
VDC 300 V
fs 200 kHz
R loss 1.5 mΩ
τ 1 × 10−5
Outputs
L 0.6 mH
kp 600
ki 150
101
4.6.3 "weak" grid dq0 Inverter
In this section an grid regulating kW class inverter is designed, implemented, and
impedance tested in Matlab. In contrast to the P resonant and "stiff" dq inverters,
sections 4.6.1 and 4.6.2, this inverter regulates the voltage and frequency of the
AC bus under variable loading. It can be classified as a "hard grid maker" and
can support an islanded AC microgrid. This is accomplished by adding an output
capacitor and feeding forward the load current, Figure 4.44.
In addition to the inner control loop, an outer voltage control loop is added.
Details of the derivation are in Appendix C.3.1. PI control is again used, the
method of symmetrical optimum determining the gains, Appendix C.3.2. The ca-
pacitor is sized using voltage ripple requirements. The details are presented in
Appendix C.4.2. The resulting element and design parameters are presented in
Table 4.13.
102
Table 4.13: "Weak" dq0 Inverter Parameters
Parameter Value
Power 3 kVA
Allowed Current Ripple 5%
VDC 300 V
fs 200 kHz
R loss 1.5 mΩ
τ 1 × 10−5
δm 53◦
Ouputs
L 0.6 mH
C 6.9 µF
kp 0.0462
ki 2239
simulated impedance measurement blocks are applied for varying power outputs,
the results are presented in Figures 4.46, 4.47 and 4.48.
103
Figure 4.46: "Weak" dq0 DC Impedance Magnitudes
104
Figure 4.47: "Weak" dq0 DC Impedance Angles
105
Chapter 5 |
Results and Conclusions
In this chapter the techniques and data developed thus far were used to demon-
strate stability assessment, stability margin, and operating point stability for non-
trivial circuit models. First a circuit containing the P resonant inverter and an
ideal source with variable line resistance was analyzed with emphasis placed on
measurement of stability margin. Second, a more complex notional wind power
microgrid was analyzed. This system contains component choices and a wider
range of operating conditions. The usefulness of the unstable operating point as-
sessment is illustrated. Finally conclusions are presented about both the notional
wind microgrid under study and the developed methods in general.
5.1 Results
The line resistance is varied, and Nyquist plots generated. Note separate com-
106
Figure 5.1: P Resonant Inverter System
The stability margin is tracked as a function of the line resistance for different
loadings, Figure 5.4. Stability margins rapidly deteriorate as the line resistance
increases beyond 0.1 Ω, although higher loadings are shown to be affected more.
Note that this resistance corresponds to less than 50 ft of 14 gauge power wire.
Time domain simulations confirm instability approximately at the indicated resis-
tances and loadings.
107
Figure 5.2: P Resonant Inverter Nyquist Plots
108
Figure 5.4: P Resonant Inverter System Margins
Throughout this work the term "source" and "load" have been used in quota-
tions deliberately. The definitions are given in section 2.2.1 and can lead to counter
intuitive descriptions, as is the case for this notional microgrid. The definitions
lead the wind turbine and inverter subsystems to be classified as loads while the
battery is considered a source. Although the wind turbine is obviously a power
source, with the connected power electronics it behaves like a constant current
source. It is ill designed to be connected with a constant current load, section 4.4,
109
Figure 5.5: DC Bus Stability Assessment Interfaces
Note that designations are not exclusive and there is some flexibility in how
a system is divided up. The battery is also stable when connected to a constant
voltage source, making it equally applicable as a "load". However batteries gener-
ally have very low impedance and therefore would make a poor choice for a "load"
subsystem. Designations and interface choices obviously don’t affect the actual
stability of the system, but they do change the shape of the Nyquist plot, causing
conservative criteria to fail for some selections. In general dual components with
high impedance should be considered "loads" and low impedance ones considered
"sources" to minimize conservatism.
110
Figure 5.6: Microgrid Component Impedances
construct the return ratio, component impedance data was loaded from data files
and processed. Collected impedance data will vary in resolution and range, no
attempt to standardize collection was desired, Figure 5.6. Data is interpolated for
the set of operating conditions considered, resized and re-sampled in the frequency
domain to make compatible, Appendix B.2
The selected operating parameters and ranges are presented in Table 5.1. Since
the system has energy storage the inverter’s rated power can safely be less than
the wind turbine subsystem’s rated power. The battery, being a modeled linear
circuit, does not vary with loading so instead circuit element parameters are var-
ied an order of magnitude around the nominal values developed in section 4.5 to
simulate aging and uncertainty.
A Matlab script was developed, Appendix B.3, to generate and stack Nyquist
plots for an interface. The results for each component combination and interface
are presented in Figures 5.7, 5.8, and 5.9. The number of variables leads to an
almost incomprehensible number of plots, although for this microgrid, the com-
ponents are very well behaved, making visual determination of stability possible.
Minimum stability margin for each system and interface are presented in Table 5.2.
111
Table 5.1: DC System Operating Variables
Note that interface return ratios can be calculated off-line from the individual
component impedance data for the interfaces by the following equations:
~ op )
Zbattery (ω, X
~ op ) = Zs =
L1 (ω, X
Zl ~ op )kZinverter (ω, X
Zturbine (ω, X ~ op )
~ ~ ~
~ op ) = Zs = Zbattery (ω, Xop )kZturbine (ω, Xop )(ω, Xop )
L2 (ω, X
Zl ~ op )
Zinverter (ω, X
Zs ~ op )kZinverter (ω, X
Zbattery (ω, X ~ op )(ω, X
~ op )
~ op ) =
L3 (ω, X =
Zl Zturbine (ω, X~ op )
Figure 5.7: Nyquist Plots: Notional Wind System with Diode Rectifier
(a) Interface 1 (b) Interface 2 (c) Interface 3
112
Figure 5.8: Nyquist Plots: Notional Wind System with Boost Rectifier
(a) Interface 1 (b) Interface 2 (c) Interface 3
Figure 5.9: Nyquist Plots: Notional Wind System with Switching Bridge Rectifier
(a) Interface 1 (b) Interface 2 (c) Interface 3
Comparison of the rectifier choices using the resulting stability margins and
Nyquist plots is difficult because they are all assessed to be very stable for the given
conditions. In the previous section it was shown that increasing the line resistance
destabilized the P resonant controlled inverter. A similar tactic was applied to
stress the wind system. A 5Ω line resistance added to the battery, corresponding
to about 200 ft of 24 gauge std hookup wire. The stability analysis is repeated,
producing more interesting results, Figures 5.10, 5.11, and 5.12. The switching
and boost converter are largely unaffected but the system with the passive diode
rectifier becomes unstable. Furthermore, analysis of the operating points show
that the diode rectifier system experiences this instability for only one of the 405
combinations of conditions tested, Table 5.3. Note that this instability was only
detected for inverter power output of 2.5 kW. All other variables held constant,
the system was assessed stable when the inverter was producing less that 1.25
kW or more than 3.75 kW. If the battery were used as part of an EMS, section
2.4.4, steady inverter operation in this range could be avoided, thus avoiding the
instability.
113
Table 5.3: Unstable Operating Point
Parameter Value
Wind Power 2.25 kW
Inverter Power 2.50 kW
Battery RCT 1Ω
Battery R 1Ω
Battery CDL 0.1 mF
5.2 Conclusions
The overall dissertation objective was largely achieved and unique contributions
delivered in the form of a Matlab impedance measurement block and high level op-
timization framework that includes power electronics stability assessments. Small
signal stability assessment techniques were adapted for microgrids and in this chap-
ter successfully applied to non-trivial models. The three requirements for a lower
level stability routine in the proposed approach; qualitative stability assessment,
quantitative stability assessment, and failure conditions assessment, were demon-
strated via simulation.
The Nyquist stability criterion (and GNC for AC systems) was selected and
shown to be effective in predicting the stability of both linear an nonlinear compo-
nents in AC and DC electrical systems. Although alternatives exist, this method
was particularly useful because it requires only information about individual com-
ponent impedances; data that can be captured empirically from any arbitrary
device.
114
Figure 5.11: Boost Rectifier and High Resistance Battery Line
(a) Interface 1 (b) Interface 2 (c) Interface 3
Figure 5.12: Switching Bridge Rectifier and High Resistance Battery Line
(a) Interface 1 (b) Interface 2 (c) Interface 3
Time domain simulations showed that while the Nyquist criterion alone could
be used for stability assessment, there was some error associated with the inex-
act impedance data measured in simulation. This necessitates the use of stability
margins. The criteria presented in the literature all intrinsically contain some
margin and conservatism. Maximum peak criteria was chosen to assess stability
115
margin, and was done separately from the stability analysis. Results in section
5.1.1, showed how it could be used to quantitatively compare the the stability of
different conditions. This verified the intuitive conclusion that higher source resis-
tance and higher inverter loading lead to lower stability.
Much of the work encountered in the literature was aimed at electrical engi-
neers working on component circuit design and specification. The concerns of a
microgrid integrator are not so focused, needing to consider a range of factors
and trade-offs simultaneously. According to the proposed methods, fast off-line
determination of stability would use a database of component stability character-
istics. This was accomplished by measuring impedances as a function of operating
point and frequency via repeated simulations. The data was saved in files and
readily used by the developed Matlab assessment scripts, taking only seconds to
generate the results presented in this dissertation. This makes relatively complex
networks, such as microgrids, tractable problems once the stability characteristics
are obtained. Note however, that acquiring the necessary impedance data for some
of the developed components in this work required hours or even days of simulation.
The wind turbine subsystem studied was implemented with MPPT and three
alternative rectifier architectures. The advantages and disadvantages of each are
well documented in the literature for small wind applications. In this work a
comparison of their stability characteristics was sought. By themselves impedance
information was not particularly useful, as each had a similarly low impedance,
Figure 5.13. When incorporated into the notional system and assessed, the diode
bridge was shown to be particularly susceptible to instability if the battery was
connected by a long (200 ft) line. The other rectifiers, boost rectifier and switching
bridge, were unaffected according to the resulting Nyquist plots.
116
Figure 5.13: Comparison of Rectifier Impedances
(a) Diode Bridge (b) Boost Rectifier (c) Switch Bridge
affect allowable (stable) layouts. This is further evidence that the relationship
between microgrid design and power electronics induced instability is significant.
When a long resistive line was placed between the battery and the DC bus
of the notional microgrid, the system variant with the passive diode rectifier was
assessed unstable. Although this instability was only detected for a single set of
operating conditions. Conventionally this would disqualify the architecture entirely
but in the proposed method the instability could be mitigated. The conditions
of the battery and rectifier loading are considered exogenous and uncontrollable,
but the allowable loadings of the inverter could be restricted. If a diode rectifier
architecture was strongly preferred due to other (non-stability related) factors,
the instability could be overcome given the operating point stability assessment
information generated. The microgrid scenario presented demonstrates the benefit
of such analysis.
117
Chapter 6 |
Recommendations
During the course of this research, several topics were noted as being worthy of
additional investigation.
Some potential concerns about the interaction of devices when averaged models
118
and time invariance are no longer valid assumptions were discussed. In theory a
slowly switching device such as a passive diode rectifier violates these assumptions
and would therefore be difficult to incorporate into a stability analysis. How-
ever intuitively such a device with a large filter will behave mostly like the filter
in those frequencies, making stability assessment possible. A clear procedure for
determining when switching frequencies matter and how much would be of benefit.
The methods and techniques developed were for both AC and DC systems, the
primary difference being the non-scalar impedance characteristics of AC systems.
The assessment methods developed are modular and allow stability to be assessed
for a larger microgrid. The notional microgrid developed was conceived as part of
a large AC microgrid, Figure 6.1. AC impedance characteristics of the notional
wind subsystem could be combined with other the microgrid "leg" impedances.
Stability could than be analyzed using the same procedure as demonstrated in this
dissertation.
119
Figure 6.1: AC Microgrid Stability Diagrams
120
Bibliography
[1] Hammerstrom, D. J. (2007) “AC versus DC Distribution Systems Did We
Get it Right?” in Power Engineering Society General Meeting, Tampa, FL,
pp. 1–5.
[2] Sorrell, S. (2007) The Rebound Effect: an assessment of the eveidence for
economy-wide energy savings from improved energy efficiency, Tech. rep., UK
Energy Research Centre, iSBN 1-903144-0-35.
[3] Troester, E. (2009) “New German Grid Codes for Connecting PV sys-
tems to the Medium Voltage Power Grid,” in 2nd International Workshop on
Concentrating Photovoltaic Power Plants: Optical Design, Production, Grid
Connection, pp. 9–10.
[6] (1999) Regional Transmission Organizations, Tech. rep., Federal Energy Reg-
ulatory Commission, docket No. RM99-2-000; Order No. 2000.
121
[11] Zhang, L., T. Wu, Y. Xing, K. Sun, and J. M. Gurrero (2011) “Power
control of DC Microgrid using DC bus signaling,” in Applied Power Electronics
Conference and Exposition, Fort Worth, TX, pp. 1926–1932.
[20] Belkhayat, M. (1997) Stability Criteria for AC Power Systems with Regu-
lated Loads, Ph.D. thesis, Purdue University, West Lafayette, IN.
122
[23] Baring-Gould, E. I. (1996) Hybrid2 The Hybrid System Simulation Model,
National Renewable Energy Laboratory, 1617 Cole Boulevard, Golden, Col-
orado 80401-3393, 1.0 ed.
[24] National Renewable Energy Laboratory, 1617 Cole Boulevard, Golden, Col-
orado 80401-3393 (2005) Getting Started Guide for HOMER, 2.1 ed.
[25] Austin, P. (2011) Practical Use of HOMER In Preliminary Microgrid De-
sign, M.s. paper, The Pennsylvania State University.
[26] Rosado, S., R. Burgos, S. Ahmed, F. Wang, and D. Boroyevich
(2008) “Modeling of Power Electronics for Simulation Based Analysis of Power
Systems,” in Proceedings of the 2007 Summer Computer Simulation Confer-
ence, Society for Computer Simulation International, San Diego, CA, USA,
pp. 19–26.
[27] Dehghan, S. M., M. Mohamadian, and A. Y. Varjani (2009) “A New
Variable SPeed Wind Energy Conversion System Using Permanent Magnet
Synchronous Generator and Z Source Inverter,” in IEEE Transactions on
Energy Conversions, vol. 24, pp. 714–724.
[28] Haque, M. E., M. Negnevitsky, and K. M. Muttaqi (2010) “A Novel
Control Strategy for a Variable Speed Wind turbine With a Permanent Mag-
net Synchronous Generator,” in IEEE Transactions on Insustry Applications,
vol. 46, pp. 331–339.
[29] Baroudi, J. A., V. Dinavahi, and A. M. Knight (2007) “A review of
power converter topologies for wind generators,” Renewable Energy, 32, pp.
2369–2385, doi: 10.1016/j.renene.2006.12.002.
[30] de Freitas, T. R. S., P. J. M. Menegaz, and D. S. L. Si-
monetti (2011) “Converter Topologies for Permanent Magnetic Syn-
chronous Generator on Wind Energy Conversion System,” in 2011 Brazil-
ian Power Electronics Conference, IEEE, Praiamar, Brazil, pp. 936–942, doi:
10.1109/COBEP.2011.6085292.
[31] Tran, D.-H., B. Sareni, X. Roboam, and C. Espanet (2010) “Inte-
grated Optimal Design of a Passive Wind Turbine System: An Experimental
Validation,” in IEEE Transactions on Sustainable Energy, vol. 1, pp. 48–56.
[32] Sudhoff, S. D. (2000) “Admittance Space Stability Analysis of Power Elec-
tronic Systems,” in IEEE trans. Aerosp. Electron. Sys., vol. 36, IEEE, pp.
965–973.
[33] Bay, J. S. (1999) Fundamentals of Linear State Space Systems, Thomas
Casson.
123
[34] Chen, C.-T. (2013) Linear System Theory and Design, 4 ed., Oxford Uni-
versity Press.
[37] Feng, X., S. Ye, K. Xing, F. C. Lee, and D. Borojevic (1999) “Individ-
ual load impedance specification for a stable DC distributed power system,”
in 1999 Applied Power Electronics Conference and Exposition, IEEE, Dallas,
TX, pp. 923–929, doi: 10.1109/APEC.1999.750480.
[40] Li, A. and D. Zhang (2014) “Necessary and Sufficient Stability Criterion and
New Forbidden Region for Load Impedance Specification,” Chinese Journal
of Electronics, 23(3).
[44] Sudhoff, S., “DC Stability Toolbox 3.0,” Purdue University School of Elec-
trical and Computer Engineering website.
124
[46] Burgos, R., D. Boroyevich, F. Wang, K. Karimi, and G. Francis
(1998) “On the AC stability of high power factor three-phse rectifiers,” in
Energy Conversion Congress and Exposition, IEEE, Atlanta, GA, pp. 511–
521.
125
[56] MathWorks, “Matlab,” .
[61] Ito, Y., Y. Zhongqing, and H. Akagi (2004) “DC Microgrid Based Distri-
bution Power Generation System,” in Power Electronics and Motion Control
Conference, Xi’an, pp. 1740–1745.
126
[68] Broe, A. M. D., S. Drouilhet, and V. Gevorgian (1999) “A Peak
Power Tracker for Small Wind Turbine in Battery Charging Applications,”
IEEE Transactions on Energy Conversion, 14(4), pp. 1630–1635.
[69] Pan, C.-T. and Y.-L. Juan (2010) “A Novel Sensorless MPPT Controller for
a High-Efficiency Microscale Wind Power Generation System,” IEEE Trans-
actions on Energy Conversion, 25(1), pp. 207–216.
127
[78] Buller, S., M. Thele, R. W. A. A. D. Doncker, and E. Karden
(2005) “Impedance-Based Simulation Models of Supercapacitors and Li-Ion
Batteries for Power Electronic Applications,” IEEE Transactions on Industry
Applications, 41(3), pp. 742–747.
128
Appendix A|
Grid Control
129
• Power Quality: In the conventional grid the chief concern is power factor
angle, a mismatch of voltage and current waveforms resulting from inductive
effects. This can be easily corrected for in microgrids using power electronics
but in turn suffer from harmonic distortions from the switching components.
• Control: Due to the factors mentioned in this list, control methods for the
grid and microgrids are different. The grid uses the robust and unversal
Automatic Generator Control (AGC). This uses system frequency to bring
the electrical supply and demand on the grid into balance on a moment to
moment basis. Microgrids are not standardized and a plethora of control
schemes have been suggested in the literature.
A.2.1 AGC
Automatic generation control is the combination of several control schemes that
work together to maintain the balance, stability, and economics of the grid. These
controls are centered around the widespread use of synchronous generators and an
inductive distribution system of long lines and transformers [19].
130
• Load Frequency Control: While turbine governor control can be relied
upon to balance the system, it will inevitably result in steady state frequency
error if ∆pref is not also adjusted as load changes. LFC accomplishes this by
driving the steady state frequency back to the nominal value. It also ensures
that local generators pick up most of the local load changes by keeping the
net power flow between areas to the scheduled levels.
• Economic Dispatch: This level of control updates the least frequently and
enacts solutions to the economic problem of dispatch by alterning individual
generator setpoints. The economic problem considers the cost of generation
as well as losses in the transmission lines due to the geographic location of
generators.
• plug and play, peer to peer, automatic controls for (inertia-less) inverter
based microsources
131
the inverter controls to function properly. Although alternative control schemes
for resistive AC microgrids have been proposed [60].
132
Figure A.2: Simple Hybrid Wind System
rating, energy storage requirement (battery only), and response rate (mostly for
generator).
X
P (t) = 0 = Pbatt + Pdump + Pwind
Z τ
Ebatt = Pbatt (t)dt
0
" #
dPk (t)
P˙k = max
dt
133
Figure A.3: Simple Hybrid Wind System Time Domain Simulation Example
134
Appendix B|
Matlab Code
% Frequency Range
wmin = 1E-2; % begin sweep [Hz]
wmax = 1E5; % end sweep [Hz]
C1 = [zeros(1,(6-numel(B))), B];
135
C2 = [zeros(1,(6-numel(A))), A];
[B,A] = invfreqs(ZlV,freqs*2*pi,ONL,ODL);
C3 = [zeros(1,(6-numel(B))), B];
C4 = [zeros(1,(6-numel(A))), A];
C = vertcat(C1,C2,C3,C4);
% Empirical Calculation
s = tf(’s’);
SN = C(1,1)*s^5+C(1,2)*s^4+C(1,3)*s^3+C(1,4)*s^2+C(1,5)*s+C(1,6);
SD = C(2,1)*s^5+C(2,2)*s^4+C(2,3)*s^3+C(2,4)*s^2+C(2,5)*s+C(2,6);
LN = C(3,1)*s^5+C(3,2)*s^4+C(3,3)*s^3+C(3,4)*s^2+C(3,5)*s+C(3,6);
LD = C(4,1)*s^5+C(4,2)*s^4+C(4,3)*s^3+C(4,4)*s^2+C(4,5)*s+C(4,6);
Zs = SN/SD;
Zl = LN/LD;
136
freqs_r = logspace(lower_bound,upper_bound,n);
137
m=1;
for i=1:size(Zr_S,2)
for j=1:size(Z_L1_int,2)
for k=1:size(Z_L4_int,2)
L1(:,m) = Zr_S(:,i).*(Z_L1_int(:,j)+Z_L4_int(:,k))./ ...
... (Z_L1_int(:,j).*Z_L4_int(:,k));
sys = frd(L1(:,m),freqs_r);
[re,im] = nyquist(sys,freqs_r);
Nyq1(:,m) = complex(re,im);
m = m + 1;
end;
end;
end;
138
if A3<-1
Stab1(1,i) = 1;
end;
end;
end;
Stab1(2,i) = min(temp_dist1);
end;
139
Appendix C|
Voltage Source Converters
140
The derivations and equations provided as reference in this section are from
and were used to design the microgrid components used in this work [75].
d~i
L = −R~i + V~t − V~s
dt
d~i
= −R~i + V~t − V̂s ej(ω0 t+θ0 )
L
dt
From reference dq reference frame theory
Divide by eθ(t)
Let dθ(t)
dt
= ω(t), separate real and imaginary parts, and assume the fundamental
ω(t) is approximately constant
did
L − Lω0 iq = −Rid + Vtd − V̂s cos(ωt + θ0 − θ(t))
dt
" #
diq
j L + Lω0 id = −Riq + Vtq − V̂s sin(ωt + θ0 − θ(t))
dt
141
Control inputs are Vtd and Vtq which in turn are functions of the ideal VSC dq
duty cycles md and mq
Vdc
Vtd = md
2
Vdc
Vtq = mq
2
Rearrange and solve for duty cycles
" #
2 did
md = L − Lω0 iq + Rid + Vd
Vdc dt
" #
2 diq
mq = L + Lω0 id + Riq + Vq
Vdc dt
Decouple the d and q axis by creating new control variables ud and uq
did
ud = L + Rid
dt
diq
uq = L + Riq
dt
These allow d and q currents to be independently regulated by simple PI control
k(s) = kp s+k
s
i
where the plant P (s) relates control input u and current i.
!
u(s) 1 1 1
i(s) = (Ls + R)u(s) → P (s) = = = R
i(s) Ls + R L s+ L
ki
kp s + kp kp ki R
k(s)P (s) = R → k(s)P (s) = when =
Ls s + L Ls kp L
142
i(s)
Closed loop with of the transfer function i∗ (s)
= G(s) is therefore
kp
Ls 1
G(s) = kp → G(s) = L
1+ Ls kp
s+1
L
τ=
kp
After some algebra, proportional and integral constants can be defined from circuit
elements and a selected time constant
L
kp =
τ
R
ki =
τ
The d and q axis equations are identical so these PI constants can be used for both
the i and d axis current controllers. The control output of these controllers ud
and uq can then be substituted in the previously developed equations to calculate
duty cycles md and mq in feedfoward control. See Figure C.3 for the control flow
diagram.
2
md = [ud − Lω0 iq + Vd ]
Vdc
2
mq = [uq + Lω0 id + Vq ]
Vdc
did
Ld = −Rid + Lq ωr iq + V d
dt
diq
Lq = −Riq − Ld ωr id − λm ωr + Vq
dt
143
Figure C.2: Post VSC Filter Diagram
144
Figure C.4: PSMG Control Flow
They are very similar to the equations derived in section C.2 for a VSC with RL
filter. Using the same methods and PI controlled current control the PMSG only
requires modification of the feedforward loop to account for the magnetic flux,
variable rotor speed, and lack of voltage measurement, Figure C.4.
2
md = [ud − Lωr iq + Vd ]
Vdc
2
mq = [uq + Lωr id − λm ωr + Vq ]
Vdc
145
of the capacitor on the AC bus is as follows.
dV~s ~ ~
C = i − iL
dt
d[vd + jvq ]
Cejθ(t) + C[vd + jvq ]ejθ(t) jω = [id + jiq ]ejθ(t) − [iL d + jiL q]ejθ(t)
dt
Vd
C − CωVq = id − iLd
dt
Vq
+ CωVd + iq − iLq
C
dt
Set i = i∗ and solve for inner control loop reference
Vd
i∗d = C − CωVq + iLd
dt
Vq
i∗q = C
+ CωVd + iLq
dt
Let new control variables ud and uq be functions of voltages Vd and Vq
If the current control loop is fast, τ is small, then the overall voltage transfer
function including the inner current control loop can be described approximately
146
as
V (s) 1
= P (s)
u(s) Cs
where then inner control loop if designed according to section C.2 will simplify to
a first order filter, thus producing
V (s) 1 1
=
u(s) τ s + 1 Cs
A PI feedback, k(s), can be used, providing ud and uq from the error of Vd and Vd q
respectively. This results in the closed loop transfer function
1
!
k(s)P (s) Cs
V (s) = 1 V ∗ (s)
1 + k(s)P (s) Cs
s+z
k(s) = k
s
1
P (s) =
τs + 1
s+z 1 1 k s+z 1
l(s) = k → l(s) = −1
s τ s + 1 Cs τC s + τ s2
A method to select constants k and z to allow adequate performance is detailed in
section C.3.2.
147
z can be solved for analytically. Once found, calculation of the constant k is
straight forward
√
k = C zτ −1
di
L = vt − vs = ∆V
dt
Vdc
Lmin =
4∆Ifs
dv
C = iC
dt
148
Figure C.5: Current Ripple Diagram
Z v2 Z τ
C dv = iC dt
v1 0
ic τ
V2 − V1 =
C
Assume a steady maximum current from time 0 to τ
iCmax τ
∆Vmax =
Cmin
iLmax
Cmin =
∆Vmax fs
149
Vita
Peter M. Austin
Education
Publications
"Study of Diminutive and Subsurface Cracks Using SOnic IR Inspection". 34th
Annual Review of Progress in Quantitative Nondestructive Evaluation. American
Institute of Physics. Vol. 975 (2008). pp. 504-511