Professional Documents
Culture Documents
DISSERTATION
By
Approved by:
Dissertation Committee
Dr. Ali Keyhani, Adviser ____________________________________
Dr. Hooshang Hemami Adviser
Dr. Donald Kasten Graduate Program in Electrical Engineering
ABSTRACT
This PhD research discusses digital control strategies for three-phase Pulse Width
Modulated (PWM) voltage inverters used in Uninterruptible Power Supplies (UPS) for
single unit and parallel unit systems. For the single inverter system, this research
proposes a novel control strategy which utilizes the perfect Robust Servomechanism
harmonics from the output voltages under non-linear load conditions and to achieve fast
sliding mode current controller that provides fast current limiting capability needed for
overload or short circuit conditions. For the parallel inverter system, a combination of
two control methods is proposed: average power control method and droop control
method. The average power method is used in order to overcome the sensitivity of load
between units, while the droop method allows the control to still maintain proper load
sharing in the event that inter-unit communication is lost. In addition to this, a harmonic
droop scheme for sharing of harmonic content of the load currents is introduced based on
the proposed single unit control. Control analysis, experimental and simulation studies,
using two parallel three-phase PWM inverters, are presented to show the effectiveness of
the proposed control strategies, both for single and parallel inverter systems.
ii
Dedicated to Selvy, Arfan, and Farhan
iii
VITA
PUBLICATIONS
Research Publication
Voltages and Currents Control,” IEEE Transaction on Power Electronics, Vol. 19,
iv
2. M.N. Marwali, J.-W. Jung, and A. Keyhani, “Control of Distributed Generation
integrated virtual learning system for the development of motor drive systems,” IEEE
FIELDS OF STUDY
Systems
v
TABLE OF CONTENTS
Page
Abstract ...............................................................................................................................ii
Vita .....................................................................................................................................iv
List of Tables...................................................................................................................xvii
List of Abbreviations......................................................................................................xviii
Chapters
1. Introduction ................................................................................................................... 1
vi
2.1.1 State space model of the plant................................................................ 30
2.1.2 Perfect control of robust servomechanism problem............................... 39
2.1.3 Discrete-time sliding mode control ....................................................... 43
5.1 Combined droop method and average power control ...................................... 118
vii
5.2 Stability analysis of combined droop and average control method ................. 123
5.3 Generation of the reference angle θ ref ............................................................. 132
5.4 Harmonic sharing ............................................................................................. 134
Bibliography.................................................................................................................... 199
Appendices
B. Matlab script for robust stability analysis of single inverter control system.............. 219
viii
LIST OF FIGURES
Figures Page
1.4 Instantaneous Current Sharing with Master/Slave Control Type 1 [89, 90]............ 10
1.5 Instantaneous Current Sharing with Master/Slave Control Type 2 ([91]) ............... 12
ix
2.6 Output voltages controller using robust servomechanism controller....................... 54
3.6 Interconnection of the nominal open loop plant with ∆ to form uncertain open
loop model................................................................................................................ 73
3.9 Transient response for Case 1: w p = 0.5 , wS = 5 × 105 , and wSH = wS (Top:
Output voltage and its reference, Middle: Load current, Bottom: RMS
variations of output voltage). ................................................................................... 85
3.10 Transient response for Case 4: w p = 0.005 , wS = 5 × 105 , and wSH = wS (Top:
Output voltage and its reference, Middle: Load current, Bottom: RMS
variations of output voltage). ................................................................................... 85
3.11 RMS output voltage variations during 0 to 100% and 100% to 0 for
different w p ,and wSH = wS ...................................................................................... 86
3.12 Upper bound of the structured singular values for different w p ,and wSH = wS ....... 87
x
3.13 Transient response for Case 6: w p = 0.5 , wS = 5 × 105 , and wSH = 0.01 × wS
(Top: Output voltage and its reference, Middle: Load current, Bottom: RMS
variations of output voltage). ................................................................................... 89
3.14 Transient response for Case 9: w p = 0.005 , wS = 5 × 105 , and wSH = 0.01 × wS
(Top: Output voltage and its reference, Middle: Load current, Bottom: RMS
variations of output voltage). ................................................................................... 89
3.15 RMS output voltage variations during 0 to 100% and 100% to 0 for
different w p ,and wSH = 0.01 × wS ............................................................................ 90
3.16 Upper bound of the structured singular values for different w p ,and
wSH = 0.01 × wS ........................................................................................................ 91
3.18 Pole-zero map of closed loop uncertain system for different values of
individual-perturbation in inverter filter inductance parameter ............................... 94
3.25 Simulation result: Fundamental with 5th harmonic references tracking................. 104
xi
3.26 Simulation result: Fundamental with 7th harmonic references tracking................. 105
3.27 Simulation result: Fundamental with 5th and 7th harmonic references tracking..... 105
3.28 Simulation result: Short circuits at the output terminals ........................................ 106
4.3 Steady state linear load (a) 100% resistive balanced (b) 100% 0.8pf load (c)
100% resistive unbalanced (phase A unloaded) (d) 100% resistive unbalanced
(phase A&B). Top: load currents; middle: load voltages; bottom: inverter
voltages. ................................................................................................................. 113
4.4 Resistive load transient: 0% to 100%. Top: three-phase load currents, bottom:
three-phase load voltages resistive load transient .................................................. 113
4.7 Synchronous Reference Frame Voltage and Current Control with PI controllers. 116
4.8 Steady state performance comparisons under non-liner load (a) using proposed
control (b) PI controllers. Top: three phase load currents. Bottom: Three-phase
load voltages........................................................................................................... 117
5.1 Combined droop and average power methods for load sharing control ................ 119
xii
5.4 System responses (a) Case 1, (b) Case 2 (c) Case 3............................................... 131
5.5 Generation of reference angle for the case when bypass source is available......... 132
5.6 Generation of reference angle for the case when bypass source is available......... 133
5.7 Harmonic Drooping thru the harmonics compensators pole shifting .................... 135
xiii
6.13 Vq (Vd) Control Implementation........................................................................... 157
6.16 Output Currents of Unit 1 and Unit 2 under linear balanced resistive load ........... 164
6.17 Output Currents of Unit 1 and Unit 2 under 0.8 lagging balanced load ................ 165
6.18 Output Currents of Unit 1 and Unit 2 under 0.9 lagging balanced load ................ 166
6.19 Output Currents of Unit 1 and Unit 2 under single-phase resistive load ............... 167
6.20 Output Currents of Unit 1 and Unit 2 with harmonic drooping enabled................ 168
6.21 Output Currents of Unit 1 and Unit 2 with harmonic drooping disabled (gains
set to zero) .............................................................................................................. 169
6.22 DQ real (top) and reactive (bottom) powers of Unit 1 and Unit 2 ......................... 170
6.24 Phases and Amplitude Adjustments of Unit 1 and Unit 2 ..................................... 171
6.27 Load transient from no-load to balanced 0.8 pf lagging ........................................ 172
6.28 Load transient from balanced 0.8 pf lagging to single-phase resistive .................. 173
xiv
6.30 Units Output Frequency ......................................................................................... 174
6.33 Modifying the Reference Angle Generator Block for simulating mismatched
crystal clock frequencies ........................................................................................ 177
6.34 DQ real (top) and reactive (bottom) powers during free-run with mismatched
frequency and frequency locking disabled............................................................. 178
6.35 Units phases adjustments during free-run with mismatched frequency and
frequency locking disabled .................................................................................... 178
6.36 Frequency locking adjustments during free-run with mismatched frequency ....... 179
6.37 DQ real (top) and reactive (bottom) powers during free-run with mismatched
frequency and frequency locking enabled.............................................................. 180
6.38 Units phases adjustments during free-run with mismatched frequency and
frequency locking enabled ..................................................................................... 180
6.39 Setting up the control gains for unlike units sharing simulations .......................... 182
6.40 Steady state performance of load sharing with unequal proportion....................... 183
6.41 DQ real (top) and reactive (bottom) powers of Unit 1 and Unit 2 for unequal
load sharing proportion .......................................................................................... 184
6.42 Phases and Amplitude Adjustments of Unit 1 and Unit 2 for unequal load
sharing proportion .................................................................................................. 185
6.43 Units Output Currents during transient for unequal load sharing proportion ........ 185
6.44 COMM_OK flag was cleared at t=4 sec to simulate lost of communication ........ 186
xv
6.45 Units output DQ real (top) and reactive (bottom) powers when communication
was lost at t=4 sec .................................................................................................. 187
6.46 Units’ output currents when communication was lost at t= 4 sec.......................... 187
6.47 Units output currents transient from 8kW to 4 kW after communication was
lost .......................................................................................................................... 188
6.48 Units output currents transient from 4kW to 8 kW after communication was
lost .......................................................................................................................... 188
6.50 Steady state load sharing performance for balanced resistive load........................ 192
6.51 Steady state load sharing performance for unbalanced resistive load.................... 192
6.52 Steady state load sharing performance for balanced non-linear rectifier load....... 193
6.53 Dynamic performance: l00% to 0% and 0% to 100% load changes. Each trace
box shows the two units output currents for phase a, b, and c (from top to
bottom) ................................................................................................................... 193
6.54 Dynamic performance 0% to 100% load acquire (zoom in of Fig. 6.53). ............ 194
6.55 Unit 2 is reconnected with different load applied on each phase. Phase a and
phase b: linear resistive load. Phase c: single-phase non-linear load..................... 194
xvi
LIST OF TABLES
Tables Page
6.2 Steady state results: Resistive Balanced Load (2 x 0.8 x 5.0 kW= 8.0 kW)............. 164
6.3 Steady state results: 0.8 lagging Balanced Load (2 x 5kVA 0.8 pf lag) ................... 165
6.4 Steady state results: 0.9 leading Balanced Load (2 x 5kVA 0.9 pf lead).................. 166
6.5 Steady state results: Single-phase Loaded (2 x 4.0 kW/3=2.666 kW on phase A .... 167
xvii
LIST OF ABBREVIATIONS
A/D : Analog/Digital
AC : Alternating Current
DC : Direct Current
xviii
CHAPTER 1
INTRODUCTION
1.1 Background
With the growth of internet and information technology industry, the need for
reliable continuous power to protect data integrity and ensuring uninterrupted service is
continuous power delivery to a system with sensitive and critical loads have become
A typical UPS configuration is shown in Fig. 1.1. The UPS system provides two
paths of power flow to the load: i) rectifier-inverter path (through the rectifier (AC-DC
converter), battery, and the inverter (DC –AC converter)), ii) bypass static switch path.
When the input power from the utility is available, the rectifier provides a means for
charging the battery, while the inverter provides regulated output voltages to the load.
When the input utility is lost, the battery, which is in parallel with the dc bus of the
interruption of the power delivery to the load. The bypass static switch provides a means
for fast transfer of power to the load from the inverter path to the utility input or another
1
AC power source, and vice versa. Typically, a secondary backup power source such as an
AC generator is connected to the bypass static switch to provide a longer period of power
supply to the load than the battery. In the event of utility loss, the battery and inverter
system maintains uninterrupted power delivery to the load for a sufficient amount of time
until the secondary power can be made available, at which time the load may be
BYPASS STATIC
SWITCH
BYPASS SOURCE
BATTERY
LOAD
UTILITY INPUT
OUTPUT
RECTIFIER INVERTER
The power delivered by the inverter system must satisfy certain requirements in
order to qualify as being regulated power. First, the inverter output voltages at steady
state must be maintained at certain Root Means Square (RMS) value independent of the
load conditions [6]. For this requirement, a load regulation at certain load is defined as
the percentage deviation of the output voltage RMS at that load from its value at no load.
A high performance UPS must present low load regulation number for all types of load
up to its rated power level. Second, the inverter must be able to provide output voltages
2
with low total harmonic distortion (THD) under various load conditions. The latter is an
important requirement, since the high harmonic load currents produced by non-linear
loads such as rectifiers used in typical computer power supplies, can significantly distort
the output voltage of the inverter [7, 8]. This distorted output voltage is undesirable to
other types of loads such as electrical motors that may be connected to the UPS, since it
can result in unnecessary power loss in the form of heat and mechanical vibrations in the
motor. The third requirement relates to the ability of the inverter to produce an output
voltage with minimal undershoot and overshoot and provides fast recovery in the event of
load transient (load change). A recovery time of one line cycle or less is typically
required from a high performance UPS, with the RMS output voltage deviates less than
In addition to providing regulated power to the load, the inverter in a UPS system
must be able to provide system protection from excessive overload conditions such as
short circuit to ground faults. In these events, rather than just turning itself off, the
inverter must be able to maintain an adequate level of inverter currents for sufficient
amount of time necessary to trigger upstream circuit breakers to open and clear the faults.
For this requirement, the inverter must have a fast inverter current limiting control to
3
The most common inverter circuits currently used in UPS systems are the Pulse
Width Modulated (PWM) inverters using Insulated Gate Biolar Transistor (IGBT) power
switches, as shown in Fig. 1.2 for a three-phase electrical system. The inverter system
consists of a three-phase IGBT bridge converter with L-C filter at the output and
⎡Vinvab⎤
⎡Vloadan ⎤
⎡Vpwmab⎤ ⎢Vload ⎥
⎢Vpwm ⎥ ⎢Vinv ⎥
⎢ bc ⎥ ⎢ bc ⎥ ⎢ bn ⎥
⎢⎣Vpwmca ⎥⎦ ⎢⎣Vinvca ⎥⎦ Delta-Wye Transformer ⎢⎣Vloadcn ⎥⎦
Linv
x
Iinva
U
Iloada L
Cgrass
Cinv Cinv n
O
Vdc Iinvb A
y
V
Iloadb
D
Iinvc W
z
Iloadc
4
As the power requirement of a system supplied by a UPS grows, two options are
possible, either to use a large UPS unit or to use multiple UPS units in parallel. The
difficulties, and reliability. Paralleling UPS units, by contrast, provides for system
expansion and redundancy. Fig. 1.3 shows a typical parallel UPS units’ system setup.
BYPASS SOURCE
UTILITY INPUT
BATTERY
LOAD
OUTPUT
RECTIFIER INVERTER
BATTERY
RECTIFIER INVERTER
One of the technical challenges involved in paralleling UPS units is to ensure that
the load is shared properly by each inverter operating in parallel while still maintaining
the voltage regulation at the load. To achieve this, a proper load sharing technique must
5
be employed. The difficulties in the load sharing are due to the non-uniformity of the
units, component tolerances, and variations in the connecting line impedances. A good
control algorithms for a single IGBT PWM inverter shown in Fig. 1.2 and load sharing
control for the PWM inverters in the parallel configuration shown in Fig. 1.3. Several
prior works have been published in the area of control of PWM inverters for UPS
applications. This PhD research addresses some of the issues raised on the previous
works and presents alternative methods for achieving better performances and for
1.2.1 Low Total Harmonic Distortion Output Voltage Control and Inverter Current
Control
inverters (single-phase or three-phase) have discussed in several prior works. In the early
days, the carrier-modulated PWM techniques such as the triangular wave comparison
type PWM were very popular [9-11]. Microcomputer based techniques using
preprogrammed PWM pattern have also been utilized [12-21]. In these techniques, the
PWM patterns are generated by computing switching edges which satisfy certain
specified harmonics. Two main disadvantages of these techniques are: slow voltage
6
regulation response due to average voltage control, and phase displacement between the
reference sine wave and the filter output varies with the load [19]. More recent PWM
instantaneous real time control of voltage and current for PWM inverters has been the
focus of many research works. Among these are the real-time deadbeat-controlled PWM
[26-33] and other techniques based on instantaneous feedback controls [34-57]. These
techniques may achieve very fast response for load disturbances when properly designed,
but it is also known that these systems have a high THD for non-linear load (crest-load)
[19]. Techniques based on sliding mode control theory [58-65] and intelligent control
using fuzzy logic or neural networks [66-75] have also been attempted. Like the
previously mentioned techniques, these techniques also lack the ability to completely
reject the harmonic disturbances in the output voltage waveforms due to non-linear loads.
Recently, the use of the internal model principle to achieve very low THD output
voltage in single-phase PWM inverters has been reported [76-85]. The theory is based on
the internal model principle proposed by [86] which states that the asymptotic tracking of
(zero steady state tracking error) can be achieved if the models that generate these
references and disturbances are included in the stable closed loop systems. In other
words, if we include the frequency modes of the references and the disturbances to be
eliminated in the control loop, then the steady state error will not contain these frequency
modes. Applying the internal model principle into the output voltages control in a three-
7
phase PWM inverter means that the fundamental frequency mode (50 Hz or 60 Hz) has to
be included in the controller since the references vary at this frequency. Elimination of
the voltages errors due to the load currents at other harmonics frequencies can then be
achieved by including the frequency modes of these harmonics into the controller.
guarantees zero steady state error at all the harmonic frequencies less than half of the
sampling period. However, the repetitive control is not easy to stabilize for all unknown
load disturbances and cannot deliver very fast response for fluctuating load [77]. In [76],
the latter problem is solved by including a ‘one sampling ahead preview controller”, and
in [77-85], the stability result of [76] is enhanced by providing adaptive mechanisms for
The following sub sections provide reviews of existing techniques for paralleling
inverters available in the literature [89-114]. The basic methods have been classified into
4. Signal Injection Method for Reactive and Harmonic Current Sharing [113, 114]
A summary of each type is presented and its advantages and disadvantages are discussed
8
1.2.2.1 Instantaneous Current Sharing with Master/Slave Control [89-96]
The instantaneous current sharing technique is a load sharing technique that uses
the load current as a feedback signal to the paralleled inverters in order for them to share
the output currents equally. In this technique, one of the inverters acts as a master
controller that has the responsibility for establishing the required voltage at the load,
while the other inverters acting as slave controllers, will try to match their output currents
Fig. 1.4 shows a block diagram of the instantaneous load sharing technique as
described in [89] and [90]. Each inverter in Fig. 1.4 consists of a PWM inverter with
output LC filter and transformer. Each inverter is equipped with a voltage regulator,
which regulates the capacitor voltage, and an inner inverter current control loop. Unit no.
1 acts as the master controller and is responsible for establishing the load voltage based
on the reference voltage. In typical UPS applications, the reference voltage is either
master control contains an outer RMS voltage control loop for load voltage regulation.
The output current produced by the master unit is distributed to the other units that act as
slaves. Each of the slave inverters then tries to regulate its output current to match the
master’s load current. This is made possible by the inclusion of an outer load current
9
Synchronization SYNCH. Vref (t ) = Vm sin (ωt + φ )
signal LOGIC
&
SINEWAVE
GENERATOR
Vref (t )
Instantenous UNIT 1 (MASTER)
Current Controller
Cap Voltage Controller
*
V pwm PWM
Inverter
Vrms* + + V1 VL
+ PWM LOAD
PI VC CC
Method
- - -
I1
V1
Vrms
RMS
I L1
Vref (t )
UNIT 2 (SLAVE)
Instantenous
Current Controller
Cap Voltage Controller
*
I L1 V pwm PWM
Inverter
+ V1
+ + PWM
PI VC CC
Method
- - -
I2
I L2 V2
Vref (t )
Instantenous UNIT 3 (SLAVE)
Current Controller
Cap Voltage Controller
I L1 *
V pwm PWM
Inverter
+ V1
+ + PWM
PI VC CC
Method
- - -
I3
I L3 V3
Figure 1.4 Instantaneous Current Sharing with Master/Slave Control Type 1 [89, 90]
10
Note that in the control scheme shown in Fig. 1.4, the reference voltage will need
to be distributed to all the inverters in order for them to generate synchronized voltage at
the output. This means that each inverter must be equipped with a Phase Locked-Loop
for synchronization to this reference signal. This requirement is eliminated in [91] which
presented a variation of the instantaneous load current sharing technique. In [91], the
slave units are configured as current source PWM inverters rather than as voltage
sources. Fig. 1.5 shows a block diagram of the load sharing control in [91]. The master
unit is a voltage source PWM inverter (VSI) with a single loop voltage control. The total
Distribution Center which in turn sends proportion of this total load current to each slave
based on its power rating. Each slave unit controls its output current based on the current
command sent by the Parallel Distribution Center. Other variations of the instantaneous
The advantage of the instantaneous load current technique is that it provides fast
load sharing response since the control is performed instantaneously. However, the need
for having one unit to act as a master unit represents drawback to this method, since it
provides for the occurrence of a single point failure, i.e. if the master unit fails the overall
system will be down. Also the needs to distribute instantaneous analog current signal to
each of the inverters in parallel may present some practical problems, such as noise,
operations.
11
UNIT 1 (MASTER-VSI)
Voltage Controller
PWM
Inverter
SYNCH. VL
Synchronization
LOGIC + PWM
LOAD
& VC
signal Method
SINEWAVE
GENERATOR -
VL
UNIT 2 (SLAVE-CSI)
Current Controller
PWM
Inverter
IL I L* 2 + PWM
CC
Method
-
I2
POWER
DISTRIBUTION
CENTER
UNIT 3 (SLAVE-CSI)
Current Controller
PWM
Inverter
*
I L3 + PWM
CC
Method
-
I3
Figure 1.5 Instantaneous Current Sharing with Master/Slave Control Type 2 ([91])
12
1.2.2.2 Difference from Average Active/Reactive Power Method [97-101]
This technique of load sharing is based on the power flow theory in an ac system,
which states that the flow of the active (P) and reactive power (Q) between two sources
can be controlled by adjusting the power angle and the voltages magnitudes. It can be
shown that the active power flow P is predominantly controlled by the power angle,
Figure 1.6 shows two inverters represented by two voltage sources connected to a load
simplification purpose.
jωL1 jωL2
I1 I2
S1 = P1 + jQ1 = V ⋅ I1
*
(1.1)
*
where I 1 is the complex conjugate of the inverter 1 current and is given by:
*
⎡ E cos δ 1 + jE1 sin δ 1 − V ⎤
I =⎢ 1
*
⎥ (1.2)
jωL1
1
⎣ ⎦
13
*
⎡ E cos δ 1 + jE1 sin δ 1 − V ⎤ (1.3)
∴ S1 = V ⎢ 1 ⎥
⎣ jωL1 ⎦
This gives the active and reactive power flowing from inverter 1 as:
VE1
P1 = sin δ 1 (1.4)
ωL1
VE 1 cos δ 1 − V 2
Q1 = (1.5)
ωL1
Similarly for the second inverter
VE 2
P2 = sin δ 2 (1.6)
ωL2
VE 2 cos δ 2 − V 2
Q2 = (1.7)
ωL2
From equations (1.4) thru (1.7), it can be seen that, if δ 1 and δ 2 are small then the real
power flow is mostly influenced by the power angles δ 1 and δ 2 , while the reactive
power flow predominantly depends on the inverter voltages E1 and E 2 . This means that
to certain extent the real and reactive power flow can be controlled independently. Since
controlling the frequencies dynamically controls the power angles, the real power flow
Fig. 1.7 shows a block diagram of this load sharing technique for two inverters in
parallel. Each inverter controls its share of real and reactive power to match the
corresponding average power by modifying its internal reference voltage. In order for
each unit to know the average power, the real and reactive power of each unit must be
communicated by some means digitally or analog. In [97-99], the active and reactive
14
powers are obtained by decomposing the load current of each UPS into its active and
reactive components. In [100] and [101] the real and reactive powers deviations are
obtained from the differences of the output currents of the two inverters PI controllers are
used to force each power to equal its average. The real power affects the amplitude of the
reference voltage, while the reactive affects the phase or the frequency. Each inverter
then regulates its voltage to the modified reference voltage as shown in Fig. 1.7.
Note that this technique does not require any units to act as a master or slaves,
hence preventing single point failure in the system. Also good real and reactive power
sharing control can be achieved with control update in the order of once every line cycle.
Therefore, the communication link between each unit can be done digitally to achieve
One of the drawbacks of this technique is the fact that the average power control
only affects the fundamental component of the load current. It does not force the
harmonic contents of the load current to share. Hence, if the load is non-linear then
harmonic fluctuating currents will flow amongst the inverters. This is undesirable since it
increases losses in the system. Another disadvantage of this method is that it requires
communication link between inverters for the common synchronization signal and the
15
INVERTER 1
Vnom
+
P1 SINE WAVE + *
REFERENCE VC CC V pwm
GENERATION
- -
V1 I1
Qavg + ∆Q1 ∆θ1 θ1
PI
-
-
θ sync1 (k ) Pavg = (P1 + P2 ) 2
Q1
Qavg = (Q1 + Q2 ) 2
PLL
P1 Q1
Common reference
synchronization signal
P2 Q2
INVERTER 2 Pavg = (P1 + P2 ) 2
PLL Qavg = (Q1 + Q2 ) 2
θ sync (k )
Qavg + ∆Q2 ∆θ 2 θ2
PI
- Voltage Controller Current Controller
-
+
Q2 SINE WAVE + *
REFERENCE VC CC V pwm
GENERATION
- -
∆V2 V2 I2
Pavg + ∆P2 Vm 2
PI
-
-
Vref 2 (k ) = Vm2 sin θ 2 (k )
P2 Vnom
16
1.2.2.3 Frequency and Amplitude Droop Method [102-112]
This technique uses the same philosophy used in conventional power system with
multiple generators. The generators share the system load by drooping the frequency of
each generator with the real power P delivered by the generator. This allows each
generator to take up changes in total load in a manner determined by its frequency droop
between the generator control systems [102]. Similarly, a droop in the voltage magnitude
In the droop technique, the P-f droop and Q-V droop characteristic of each
f i = f 0 − mi ( P0i − Pi ) (1.8)
Vi = V0 − ni (Q0i − Qi ) (1.9)
where fi and Vi denote the frequency and voltage of the inverter i , f0 and V0 represent the
nominal frequency and voltage magnitude of the system, and mi and ni are the droop
coefficients. Figure 1.8 shows a block diagram of the frequency and amplitude droop
17
P0i f0i
Pi
- ∆f i +
+ mi + fi
mi < 0
Q0i V0i
- ∆V +
Qi
+ ni
+ Vi
ni < 0
The proportion of load shared by each inverter can be adjusted based on its
m1 ⋅ S1 = m2 ⋅ S2 = L = mn ⋅ Sn (1.10)
n1 ⋅ S1 = n2 ⋅ S2 = L = nn ⋅ Sn (1.11)
In Figure 1.9, two inverters can be seen to share the real and reactive loads proportionally
18
f0 V0
∆f ∆V
f V
m2 n2
m1 n1
P1 P2 Q1 Q2
The advantage of the droop method is that it does not require communication
signals amongst units in parallel, thereby increasing the reliability of the system.
However, because of the droop characteristics, the frequency and amplitude of the system
voltage will drop to such a value that all unit will be operating in a new lower frequency
and voltage different than the nominal values. Also as in the average power method, the
method only works well for linear loads, but it will not help to share the harmonics
components caused by non-linear loads. Another drawback of this method is the lack of
feedback signal used for the control can significantly affect the power sharing. This is
due to the lack of closed loop control of the power sharing method.
19
1.2.2 4 Signal Injection Method for Reactive and Harmonic Current Sharing [113], [114]
The signal injection method proposed in [113] and [114] provides a way to share
the reactive and harmonic current of paralleled inverters. In this method two signals at
frequencies different than the fundamental are injected to the fundamental reference
voltage. One injected signal is used to control the reactive power sharing and the other is
used for harmonic powers sharing. This is done in addition to the fundamental droop for
where ( V , ω ) are the fundamental voltage and frequency, ( Vh1 , ω q ) are the voltage and
frequency of signal injected to share the reactive power Q, and ( Vh 2 , ω d ) are the voltage
and frequency of signal injected to share the harmonic distortion power D. Vh1 and
Vh 2 are fixed, while the ω q and ω d are varied for reactive and harmonic power sharing.
To force reactive power sharing, the frequency of the ( Vh1 , ω q ) signal is drooped
ω q = ω q 0 − BQ (1.10)
Also, the small real power at frequency ω q due to the injected signal ( p q ) is measured
V = V0 − Dv p q (1.11)
Similarly, to force the distortion power sharing, the frequency of the second injected
20
ωd = ωd 0 − C ⋅ D (1.12)
where D is calculated using the apparent power S, the real power P, and the reactive
D = S 2 − P2 − Q2 (1.13)
In the case of the distortion power, the small real power at frequency ω d due to the
injected signal ( pd ) is measured and used to droop the bandwidth of the control signal as
in:
attractiveness of the signal injection method relies on the fact that it allows for sharing of
the harmonic distortion power in addition to real and reactive power sharing. Also it does
not require control connection to achieve the power sharing. One of the drawbacks of this
method is that the injected signals decrease the waveform fidelity of the output voltages.
This is undesirable in a very high performance UPS system where the output voltage
21
REAL POWER SHARING ω0
P +
∆ω ωq0
m - ω HARMOMIC POWER SHARING
Q ∆ω q
+ BW0
B - ωq
V0
+
pd
DBW ∆BW
- BW
+
pd
DV ∆V
- V
22
1.3 Research Outline
This PhD research addresses the development of control techniques for single
1.3.1 Development of Voltage and Current Control for Single Inverter System
The voltage control technique proposed in this research uses the perfect control of
robust servomechanism problem (Perfect RSP) theory developed in [87] to ensure perfect
tracking of the output voltages under unknown load by providing means for eliminating
errors at specified harmonic and at the same time ensuring good transient response. The
perfect RSP theory combines this internal model principle with optimal state feedback to
guarantee stability of the closed loop system and providing arbitrary good transient
response.
It is to be noted that, although the perfect RSP used in this research only
eliminates voltage harmonic at some specified frequencies, the perfect RSP control is still
three-phase system, most of the voltage harmonics, like the even harmonics or higher
order harmonics, are either non-existence or negligible in values. Therefore, not too
many significant harmonics are left for the control to handle. Moreover, the perfect RSP
In this research, the perfect RSP control of the voltages harmonics will be
combined with a fast current controller using a discrete time sliding mode controller [88]
for limiting the inverter currents under overload condition. This is one of the important
features necessary for a UPS, which is not addressed in the repetitive control work of
23
[76-85]. The discrete sliding mode controller has been chosen because of the fast and no-
overshoot response that it provides. The current controller acts as an inner loop to the
perfect RSP control of the output voltages in the outer loop. In this case, the perfect RSP
voltage control has been designed by accounting for the extra dynamic introduced by the
discrete time sliding mode controller. This way, the stability of the overall control system
is still guaranteed.
fundamental frequency reference and error regulation of the load disturbance at each of
the harmonic frequency included in the compensators. The RSP guarantees this property
independent of any perturbations in the plant as long as they do not destabilize system
[87]. Perturbations in a UPS system may come as results of load disturbances, noise and
analyze the stability property of the resulting controller under possible disturbances that
might destabilize the system in order to ensure proper operation of the UPS over its
framework has been used to study the stability robustness of the control system under
24
Finally, the effectiveness of the proposed controller has been verified thru
simulations and experimental studies. A 5 kVA experimental test bed has been designed
and built at the Mechatronics Laboratory of The Ohio State University for this purpose.
The results show that the proposed controller provides an output voltage with excellent
regulation and THD less than 2.5% under both linear and non-linear load while still
The load sharing technique developed in this research attempts to address and
improve the issues that exist in the previous research works as described earlier. To
ensure good load-sharing, the proposed scheme combines two control methods: droop
control method and average power control method. In this method, the sharing of real and
reactive powers between each UPS is implemented by two independent control variables:
power angle and inverter output voltage amplitude. Especially, the average power method
is used in order to significantly reduce the sensitivity of the load sharing to voltage/
current measurement error and wiring impedances mismatches. This scheme guarantees
good load-sharing of the fundamental components of the load currents. The droop control
method on the other hand provides a way for the paralleled inverter to still maintain load
25
To ensure sharing of harmonic contents of the load currents, a harmonic droop
sharing technique is proposed in this research. The technique is based on the voltages and
currents control for single PWM inverter developed in this research. The proposed
harmonic droop technique has the advantage that it only affects the harmonic
compensator gains and does not degrade the fundamental voltage regulation.
effectiveness of the proposed scheme. Specifically the following have been investigated
A second 5kVA test bed unit has been built at the laboratory for the purpose of testing the
begins with the development of a state space model of the single PWM inverter system,
followed by brief summaries of the perfect RSP control theory and Discrete Sliding
Mode control theory. It then proceeds with the development of the current controller and
26
voltage controller using the two control methods. Next, some practical considerations are
given in relation to the tuning and implementation of the controllers. The chapter ends
with explanations on how to use Matlab to compute and design the controller gains.
Chapter 2. A summary of robust stability test using the Structured Singular Value µ -
system necessary for the robust stability analysis. The chapter then provides robust
stability analysis and transient performance evaluations of the system for different
controllers resulting from different parameters choices in the cost function of the perfect
RSP control. Finally, this chapter presents simulation results of the complete three-phase
In Chapter 4, the experimental work involved for the single inverter system is
discussed. This chapter presents a description of the 5 kVA test bed used for the purpose
of verifying the effectiveness of the single inverter control. The experimental results
obtained from this test bed are then presented and discussed.
Chapter 5 and Chapter 6 discuss the results of this PhD research for control of the
parallel inverters system. Chapter 5 explains the parallel control strategy, while Chapter 6
Chapter 5 begins with the proposed load sharing control, followed by discussion of
stability analysis of the combined average power method and droop method. The chapter
then discusses the reference angle generation and provides explanation of the harmonic
27
implementation for the parallel inverter system is first described. The detailed
descriptions of the simulation system is necessary here in order to give the reader better
understanding of the practical implementation of the parallel control strategy and of the
extensive simulation results that follows. Chapter 6 ends with a description of the
experimental test bed used to verify the effectiveness of the parallel inverter control, and
This dissertation ends with Chapter 7 which gives conclusions to the overall
28
CHAPTER 2
The voltage control technique proposed here uses the perfect control of robust
tracking of the output voltages under unknown load by providing means for eliminating
errors at specified harmonic and at the same time ensuring good transient response. The
theory is based on the internal model principle proposed by [86] which states that
presence of disturbances (zero steady state tracking error) can be achieved if the models
that generate these references and disturbances are included in the stable closed loop
systems. In other words, if we include the frequency modes of the references and the
disturbances to be eliminated in the control loop, then the steady state error will not
contain these frequency modes. Applying the internal model principle into the output
voltages control in a three-phase PWM inverter means that the fundamental frequency
mode (50 Hz or 60 Hz) has to be included in the controller since the references vary at
this frequency. Elimination of the voltages errors due to the load currents at other
29
harmonics frequency can then be achieved by including the frequency modes of these
harmonics into the controller. The perfect RSP theory combines this internal model
principle with optimal state feedback to guarantee stability of the closed loop system and
For development of the control algorithm, a state space model of the system
shown in Fig. 2.1 needs to be developed. Each phase of the delta-wye transformer has
been modeled as an ideal transformer with leakage inductance Ltrans and series
resistance Rtrans on the secondary winding as shown in Fig. 2.2 The secondary
transformer currents are denoted as Isnd a , Isnd b , and Isnd c . The transformer’s turn
primary and secondary voltages respectively, Note that the transformer input-output
phase relationship shown in Fig. 2.1 and Fig. 2.2 corresponds to a particular type of
30
⎡Vinvab⎤
⎡Vloadan ⎤
⎡Vpwmab ⎤ ⎢Vload ⎥
⎢Vpwm ⎥ ⎢Vinv ⎥
⎢ bc ⎥ ⎢ bc ⎥ ⎢ bn ⎥
⎢⎣Vpwmca ⎥⎦ ⎢⎣Vinvca ⎥⎦ Delta-Wye Transformer ⎢⎣Vloadcn ⎥⎦
Linv
a
Iinva
A
Iloada L
Cload
Cinv Cinv
O
Vdc Iinvb A
c
B
Iloadb
D
b
Iinvc C
z
Iloadc
DELTA-WYE TRANSFORMER
Ltrans Rtrans
u
x
+ tr
tr − Vinv ab Isnd a
− ⋅ Isnd a −
tr
⋅ Isnd c 3
3 Ltrans Rtrans
v 3 -
y
tr n
− Vinvca
3 -
tr - Isnd b
− ⋅ Isnd b tr +
3 + − Vinvbc Ltrans Rtrans
w 3 z
Isnd c
31
Using the transformer model in Fig. 2.2, the dynamic equations of the output filter
3
r
Vinv abc = [Vinv ab
′ Vinv bc ′ ]T = 1
′ Vinv ca [Vinv ab Vinv bc Vinv ca ]
T
3
r
Vload abc = [Vload a Vload b Vload c ] ,
T
r
I load abc = [Iload a Iload c ]
T
Iload b
r
I snd abc = [Isnd a Isnd c ] ,
T
Isnd b
r
I invabc = [Iinvab Iinvca ]
T
Iinvbc
(2.2)
= 1 ⋅ [Iinv a − Iinvb Iinvb − Iinvc Iinvc − Iinv a ]
T
Note that in the dynamic equations of (2.1), the average values of PWM signals
commands [Vpwm ab Vpwmbc Vpwm ca ] have been used to model the IGBT PWM
T
32
switching voltages. Matrices Tri and Trv in equation (2.1.a) and (2.1.d) denote the currents
and voltages transformations of the delta-wye transformer. Based on Fig. 2.2 these
⎡− 2 1 1⎤ ⎡− 1 0 0⎤
Tri = ⋅ 1 − 2 1 ⎥ ,
tr ⎢
Trv = tr ⋅ ⎢ 0 − 1 0 ⎥ (2.3)
3 ⎢ ⎥ ⎢ ⎥
⎢⎣ 1 1 − 2⎥⎦ ⎢⎣ 0 0 − 1⎥⎦
rewrite the dynamic equations of (2.1) in some form of per-unit (PU) system. Define the
r r r r
v load abc = Vload abc V 2 , i load abc = I load abc I 1 (2.4)
where V1 and V2 are as defined earlier, and I1 and I2 denote the rated primary and
follows:
Z1 ≡ V 1 I1 Z2 ≡V 2 I2 (2.6)
Using the quantities defined above the dynamic equations (2.1) can be rewritten
as in (2.7):
r r r
dv invabc
dt
(
= ω f ⋅ xcinv ⋅ i invabc − tri ⋅ i snd abc ) (2.7.a)
33
r
di invabc
=
1
(− xrinv ⋅ i invabc − vrinvabc + vrpwmabc ) (2.7.b)
dt xlinv ω f
r
r r
dVload abc
dt
(
= ω f ⋅ xcload i snd abc − i load abc ) (2.7.c)
r
r
di snd abc
dt
=
1
xl tran ω f
( r r
− xrtran ⋅ i snd abc + trv ⋅ v invabc − v load abc ) (2.7.d)
where matrices tri and trv are now independent of the transformer turn ratio, and are
given by (2.8):
⎡− 2 1 1⎤ ⎡− 1 0 0⎤
1⎢
tri = 1 −2 1 ⎥, trv = 0 − 1 0 ⎥
⎢ (2.8)
3 ⎢ ⎥ ⎢ ⎥
⎢⎣ 1 1 − 2⎥⎦ ⎢⎣ 0 0 − 1⎥⎦
To obtain a state space model of the system, the dynamic equations in (2.7) are
with
⎡1 − 0.5 − 0.5 ⎤
2⎢
KS = 0 − 3 2 3 2⎥ ,
3⎢ ⎥
⎢⎣0.5 0.5 0.5 ⎥⎦
f qd 0 = [ f q , f d , f 0 ] , f abc = [ f a , f b , f c ]
T T
r r
where f abc denotes the abc voltages and currents defined in (2.4), and f qd 0 the
corresponding to the DQ0 stationary reference frame variables. The circuit dynamics can
34
r
r r
dv inv qd
dt
(
= ω f ⋅ xcinv ⋅ i inv qd − triqd ⋅ i snd qd ) (2.10.a)
r
di inv qd
=
1
(− xr ⋅ i inv qd − v inv qd + v pwmqd )
r r
(2.10.b)
xlinv ω f
inv
dt
r
r r
dv load qd
dt
(
= ω f ⋅ xcload i snd qd − i load qd ) (2.10.c)
r
r
di snd qd
dt
=
1
xl tran ω f
(− xr tran
r r
⋅ i snd qd + trv qd ⋅ v inv qd − v load qd ) (2.10.d)
dvload 0
= ω f ⋅ xcload (isnd 0 − iload 0 ) (2.10.e)
dt
disnd 0 1
= (− xrtran ⋅ isnd 0 − vload 0 ) (2.10.f)
dt xl tran ω f
r
where f qd = [ f q f d ], and matrices triqd and trvqd are defined as:
[
triqd = K s ⋅ Tri ⋅ K S
−1
]row 1, 2
⎡− 1 0 ⎤
=⎢ ⎥ (2.11.a)
⎣ 0 − 1⎦
col 1, 2
[
trv qd = K s ⋅ Trv ⋅ K S
−1
] row1, 2
⎡− 1 0 ⎤
=⎢ ⎥ (2.11.b)
⎣ 0 − 1⎦
col 1, 2
Notice that, due to the three-wire system of the inverter and filter, the zero
components of the inverter voltages ( vinv0 ), the inverter currents ( iinv0 ) and the input
PWM voltages ( vpwm0 ) are trivial and they do not appear in (2.10). Furthermore, define
35
r r
′ = triqd ⋅ i load qd
i load qd
r r
′ = trvqd −1 ⋅ v load qd
v load qd (2.12)
Using (2.12), the state space equations (2.10) can be rewritten in the form (2.13):
r
r r
dv inv qd
dt
(
= ω f ⋅ xcinv ⋅ i inv qd − ⋅i snd qd
′ ) (2.13.a)
r
di inv qd
=
1
(− xr ⋅ i inv qd − v inv qd + v pwmqd )
r r
(2.13.b)
xlinv ω f
inv
dt
r
′ r r
dv load qd
dt
(
= ω f ⋅ xcload i snd qd
′ − i load qd
′ ) (2.13.c)
r
′ r
di snd qd
dt
=
1
xl tran ω f
(− xr tran ⋅ i snd qd
r r
′ + v inv qd − v load qd
′ ) (2.13.d)
dvload 0
= ω f ⋅ xcload (isnd 0 − iload 0 ) (2.13.e)
dt
disnd 0 1
= (− xrtran ⋅ isnd 0 − vload 0 ) (2.13.f)
dt xl tran ω f
cinv ≡ 1 cload ≡ 1
ω f xcinv , ω f xcload ,
ωf ωf
linv ≡ , ltrans ≡ , (2.14)
xl inv xltrans
Substituting these definitions to the dynamic equations in (2.13), equivalent circuits can
be drawn in the DQ0 stationary reference frame as shown in Fig. 2.3. From Fig. 2.3, it
can be seen that the 0-axis equivalent circuit of the load-side of the transformer is
36
completely decoupled from the qd-axis equivalent circuits. This shows that the zero
components of the load voltages and the secondary transformer currents ( Vload 0 and Isnd 0 )
v
are not controllable by the input PWM voltages Vpwmqd . The 0-component of the load
currents will be non-zero under unbalanced load condition or under non-linear load in the
form of triplent harmonics. This 0-component of the load currents will create 0-
component of the load voltages, which is undesirable. Since the 0-component of the load
voltages is uncontrollable, no action can be done by the control to compensate for it.
However, as is apparent from Fig. 2.3, the small capacitor at the output along with the
leakage inductance of the transformer act as an LC filter which can be tuned to provide
attenuation to the effect of the 0-component of the load current to the load voltages’.
From Fig. 2.3, the amount of this attenuation at steady state can be calculated from:
jω ⋅ Ltrans + Rtrans
Vload 0 ( jω ) = ⋅ Iload 0 ( jω ) (2.14)
1 − ω Ltrans C grass + jω ⋅ R trans C grass
2
37
Iinvq Isndq′
- - -
Iinvd Isndd′
- - -
Isnd0
ltrans rtrans
+
cload Vload0 Iload0
38
2.1.2 Perfect Control of Robust Servomechanism Problem
involves finding a controller for the system such that: given a modeled class of unstable
perturbations which do not produce instability. A perfect controller for the RSP solves the
RSP and also provides arbitrarily good transient error, with no unbounded peaking in the
error response of the system. Below some of the main results of the perfect RSP
x& = Ax + Bu + Ed
y = Cx + Du + Fd
(2.15)
ym = Cm x + Dmu + Fm d
e = yref − y
the measurable outputs, d ∈ R δ are the disturbances, y ref ∈ R r are the reference input
signals, and e ∈ R r are the error in the system. Assume the disturbance d can be generated
with eig ( Ψ1 ) ⊂ C + and eig ( Ψ2 ) ⊂ C + , where eig (⋅) denotes the eigenvalues and C +
denotes the closed right complex half plane. This type of signals includes constant,
39
polynomial, sinusoidal, polynomial-sinusoidal, etc occurring in most applications [121].
Let Λ := {λ1 , λ 2 ,K, λ p } be the zeros of the least common multiple of minimal polynomial
of Ψ1 and Ψ2 (multiplicities repeated), then a linear controller that solves the RSP for
u = ξ + K1η (2.18)
with
Ac = block diag (Ω, Ω,K, Ω )
r
Bc = block diag (γ , γ ,K , γ )
and
⎡0⎤
⎡ 0 1 0 L 0 ⎤ ⎢0⎥
⎢ 0 0 1 L 0 ⎥ ⎢ ⎥
Ω=⎢ ⎥ ,γ = ⎢M⎥
⎢ M M M M M ⎥ ⎢ ⎥
⎢ ⎥ ⎢0⎥
⎣− σ 1 −σ2 −σ3 L −σ p⎦
⎢⎣1⎥⎦
where the coefficients σ i , i = 1,2,K, p are given by the coefficients of the polynomial
∏ (λ − λ ) ; i.e.,
i =1
i
p
λ p + σ p λ p −1 + L + σ 2 λ + σ 1 := ∏ (λ − λi ) (2.20)
i =1
40
The servo compensator gain K1 in (2.18) and the stabilizing-compensator S
should be found to stabilize and provide intended behavior to the following stabilizable
⎡ x& ⎤ ⎡ A 0 ⎤⎡ x⎤ ⎡ B ⎤
⎢η& ⎥ = ⎢− B C +
Ac ⎥⎦ ⎢⎣η ⎥⎦ ⎢⎣− Bc D ⎥⎦
u
⎣ ⎦ ⎣ c
⎡ y m ⎤ ⎡C m 0 ⎤ ⎡ x ⎤ ⎡ Dm ⎤
⎢η ⎥=⎢ 0 +
I ⎥⎦ ⎢⎣η ⎥⎦ ⎢⎣ 0 ⎥⎦
u
⎣ ⎦ ⎣ (2.21)
input transformations [121]. The stabilizing compensators S are, however, not unique;
there are various classes of stabilizing compensators that can be used in a robust
S : ξ = K0 x (2.22)
where x can either be the measured states of the system, or the estimates found using an
observer (if not all states are measurable). In this case, the control u is given by state
feedback found to stabilize the augmented system of the plant and the servo-
compensator:
~ ~
x& = A ⋅ ~
~ x + B ⋅u (2.23)
~ ⎡ A 0⎤ ~ ⎡ B ⎤ ~ ⎡ x⎤
A=⎢ B=⎢ x =⎢ ⎥,
⎣ − Bc C Ac ⎥⎦ ⎥
⎣ − Bc D ⎦ ⎣η ⎦
u = K 0 x + K 1η = [K 0 K 1 ]~
x := K ⋅ ~
x
41
A perfect controller for the robust servomechanism problem can be stabilized
using Cheap Control method, where the gains K = [K 0 K1 ] are found by minimizing the
∞
J ε = ∫ (z ' z + ε ⋅ u ' u ) dτ , z = C ⋅ ~
~
x (2.24)
0
where ε > 0 is an arbitrarily small scalar. The optimal control gain K = K opt , which
~
minimizes this performance index, is given by: K opt = − 1 ε B ' Popt , where Popt is the unique
~ ~ ~ ~ 1 ~~
A' Popt + Popt A + C ' C − Popt B B ' Popt = 0 (2.25)
ε
42
2.1.3 Discrete-Time Sliding Mode Control
Discrete time sliding mode controller is an approach in the sliding mode control
theory, which is suitable for digital implementation since it does not exhibit the chattering
phenomena due to direct digital implementation of continuous time sliding mode control
[88].
x& (t ) = Ax (t ) + Bu (t ) + Ed (t ) (2.26)
y (t ) = Cx(t )
e(t ) = y (t ) − y ref (t )
input y ref (t ) , and disturbance d (t ) . System (2.26) can be transformed to a discrete time
x ( k + 1) = A* x ( k ) + B * u( k ) + E * d (k ) (2.27)
y ( k ) = Cx(k )
e(k ) = y ( k ) − y ref (k )
where :
TS A ⋅(TS −τ )
A* = exp( A ⋅ TS ) B* = ∫ e B dτ
0
TS A ⋅(TS −τ )
E* = ∫ e E dτ
0
It is desired to control the output y ( k ) to follow the reference y ref ( k ) . For this
purpose we can choose a sliding mode manifold in the form of: s( k ) = Cx(k ) − y ref ( k )
43
such that when discrete-time sliding mode exists we have y ( k ) → y ref ( k ) .Discrete-time
sliding mode exists if the control input u (k ) is designed as the solution of:
The control law that satisfies (2.28) is called ‘equivalent control’ and is given by:
If the control is limited to a value u0 i.e. u (k ) ≤ u0 then the following modified control
With control law (2.30), discrete-time sliding mode exists after a finite number of steps.
44
2.2 Control System Development
To achieve fast current limiting capability for the inverter, the control strategy
uses a two-loop control structure: an inner inverter currents loop and an outer load
v
voltages loop as shown in Fig. 2.4. The outer loop regulates the load voltages ( Vload qd ) to
v
follow 50/60 Hz balanced three-phase voltages references ( Vref qd ) and generates the
r
inverter currents commands ( I cmd qd ), which are limited. The inner loop in turn generates
the PWM voltage commands to regulate the inverter currents to follow the inverter
current command. A standard voltage space vector algorithm is then used to realize these
PWM command voltages. Notice that the 0-components of the load voltages are not
r
r *
r Vpwm qd (k )
I cmd qd I cmd qd
r r r
Vref qd (k ) eVqd eIqd Line-to-Line
Robust ServoMechanism + Discrete Sliding Mode
Voltage
Limiter Space
+ Controller Controller
Vector
- - PWM
r r
Vload qd (k ) I invqd (k ) PWM timing
states states
Fig. 2.5 shows the timing diagram of the PWM gating signals generation in
relation with the A/D sampling time of the DSP. It can be seen that there is a one-half
PWM period delay between the time the signals are sampled by the A/D and the time the
45
A/D sampling A/D sam pling
for time k for tim e k+1
D SP tim er for PW M
generation
new values of PW M timing new values of PW M timing
for tim e k take affect for time k+1 take affect
T pwm T pwm
T pwm=PW M period
The next two subsections summarizes the development of the two control loops,
the inner current loop using the discrete sliding mode controller and the voltage control
For designing the discrete time sliding mode current controller, consider the
r
di inv qd
=
1
(− xr ⋅ i inv qd − v inv qd + v pwmqd )
r r
(2.31.b)
xlinv ω f
inv
dt
r
Assuming the secondary transformer currents i snd qd as disturbances, this
46
r r r
r ⎡ 02 x2 ω f ⋅ xcinv ⋅ I 2 x 2 ⎤ r ⎡ 02 x 2 ⎤
A1 = ⎢ r r ⎥, B1 = ⎢ r ⎥,
⎢⎣ − (xl inv ω f ) ⋅ I 2 x 2 ⎢⎣(xlinv ω f ) ⋅ I 2 x 2 ⎥⎦
−1 −1
02 x 2 ⎥⎦
r ⎡− ω f ⋅ xcinv ⋅ triqd ⎤
E1 = ⎢ r ⎥,
⎣ 02 x 2 ⎦
r
r r
[
where the states are x1 = v inv qd , i inv qd ] r r
, the inputs u = v pwmqd and disturbances
r
d 1 = i snd qd .
where
TS
A1* = exp( A ⋅ TS ) B1* = ∫ e A1 ⋅TS B1 dτ
0
TS
E1* = ∫ e A1 ⋅TS E1 dτ
0
and TS is the A/D sampling time, which in this case is equal to the PWM period T pwm
To force the inverter currents to follow their commands, the sliding mode surface
r r r r r
is chosen as: s (k ) = C1 ⋅ x1 (k ) − i cmd (k ) where C1 ⋅ x1 (k ) = i inv qd (k ) , so that when
r r r
discrete sliding mode occurs, we have s (k ) = 0 or i inv (k ) = i cmd (k ) . The existence of
the discrete sliding mode can be guaranteed if the control is given by:
r r
⎧ u eq (k ) for u eq (k ) ≤ u0
⎪ r
u (k ) = ⎨ u eq (k ) r (2.33)
⎪u0 ur (k ) for u eq (k ) > u0
⎩ eq
r
where the equivalent control input ueq (k ) is calculated from:
47
r
r
(
ueq (k ) = C1 B1
*
) (ircmd
−1
qd
*r
)
− C1 A1 x1 (k ) − C1 E1 d1 (k )
*
(2.34)
and u0 denotes the maximum value of the PWM voltage command realizable by the space
vector algorithm.
Note that the secondary transformer currents are needed for the control, but these
currents are not measured in the system (see Fig.2.1). A linear Luenberger observer can
be easily designed to estimate these currents for control purposes. However, in most
practical cases we can approximate these currents with the load currents (i.e.
r r
I snd qd ≈ I load qd ) since the currents through the output capacitor filters are small. From
simulations and experiments, the effect of using this approximation is unnoticeable in the
control performance.
Due to the computation delay of the DSP (Fig. 2.5), the control action given by
(2.34) will result in undesirable overshoots during transients. This effect can be
r r
minimized, however, if the states x1 ( k ) and disturbances d 1 (k ) are replaced with their
r
r
(
u eq (k ) = C1 B1
*
) (Ircmd
−1
qd
*r
− C1 A1 x1p (k ) − C1 E1 d 1p (k )
*
) (2.36)
48
2.2.2 Voltage Controller Design using Discrete Perfect RSP
The voltage control loop designed in this research is based on the discrete form of
the technique developed in Davison [87] discussed in the previous section. To design the
load voltages controller, let’s first consider the entire plant system with the-0 components
of the voltages and currents omitted as given in (2.37). As explained in the previous
section, these 0-components are completely decoupled and uncontrollable from the
inputs, and therefore are not useful to be included in the controller design. In the system
(2.37), an input delay of one-half the PWM period ( 0.5T pwm ) has been explicitly included
r r r
⎡ 02 x 2 ω f ⋅ xcinv ⋅ I 2 x 2 02 x2 − ω f ⋅ xcinv ⋅ triqd ⎤
⎢ r r r r ⎥
r
Ap = ⎢
− ( xl inv ω
r f
)−1
⋅ I 2x2 02 x 2
r
02 x2
r
02 x 2
r ⎥
⎢ 02 x 2 02 x 2 02 x2 ω f ⋅ xcload ⋅ I 2 x 2 ⎥
⎢ r r r ⎥
⎢⎣(xl trans ω f ) ⋅ trv qd − (xl tran ω f ) ⋅ I 2 x 2 − xrtran (xl tran ω f ) ⋅ I 2 x 2 ⎥⎦
−1 −1 −1
02 x 2
,
r
⎡ 02 x 2 ⎤
⎢ r ⎥
r (xl
B p = ⎢ inv
ωf )
r
−1
⋅ I 2x2 ⎥
, (2.37)
⎢ 02 x 2 ⎥
⎢ r ⎥
⎣ 02 x 2 ⎦
49
r r r r
x p (k + 1) = Φ ⋅ x p (k ) + Γ1 ⋅ u (k − 1) + Γ2 ⋅ u (k ) (2.38)
where
TS 0.5TS
∫e ∫e
Apτ Apτ
Φ=e , Γ1 = B p dτ , Γ2 = B p dτ
ApTS
0.5TS 0
Discrete time system (2.38) can be written in a standard discrete time state space
r r v
equations by adding the extra states: x a (k ) = u (k − 1) = Vpwm qd (k − 1) to yield:
r r
⎡ x p (k + 1)⎤ ⎡ Φ Γ1 ⎤ ⎡ x p (k )⎤ ⎡Γ2 ⎤ r
⎢r ⎥ = ⎢r r ⋅ ⎢r ⎥ + ⎢ r ⎥ ⋅ u (k )
0 2 x 2 ⎥⎦ ⎣ x a (k )⎦ ⎣ I 2 x 2 ⎦
(2.39)
⎣ x a (k + 1)⎦ ⎣0 2 x 2
where:
r
r* ⎡ x p (k )⎤ ⎡ Φ Γ ⎤ ⎡Γ ⎤
x p (k ) = ⎢ r ⎥ , Ap = ⎢ r
*
r 1 ⎥ , B *p = ⎢ r 2 ⎥
⎣ x a (k )⎦ ⎣0 2 x 2 02 x2 ⎦ ⎣I 2x2 ⎦
To design the voltage controller, we need to consider the true plant (2.40) and the
discrete time sliding mode current controller as the equivalent ‘plant’ as seen by the outer
voltage loop. Using equation (2.36) and (2.40) the augmented true plant and discrete
r r
with u1 (k ) = I cmd qd
*
( k ) , and
(
Ad = A*p − B *p C1 B1 ) (B C
* −1 *
1 11 + E1*C12 )
(
Bd = B *p C1 B1
* −1
)
50
r r r r r
⎡ I 2x2 02 x2 02 x2 02x2 02 x2 ⎤
C11 = ⎢ r r r r r ⎥
⎣0 2 x 2 I 2x2 02 x2 02x2 02 x2 ⎦
r
[
C12 = 0 2 x 2
r
02x2
r
02 x2
r
I 2x2
r
02 x2 ]
Note that the augmented system given in (2.41) was found assuming the approximation
r r
I snd qd ≈ I load qd has been used.
Now, assume ωi = 2πf i i = 1,2,K n are frequencies of the reference voltages and
harmonics to be eliminated. For a 60-Hz UPS system with desire to eliminate 5th and 7th
r r r
eVqd = Vref qd − Vload qd (2.42)
where
r r r r r
η = [η1 ,η 2 ,Lη n ]T η i ∈ R 4 , i = 1,2,K n
with
r r
⎛ 02 x2 I 2x2 ⎞
Aci = ⎜⎜ r r ⎟ , i = 1,2, K n
− ω 0 2 x 2 ⎟⎠
2
⎝ i I 2x2
r r
(
Bci = 0 2 x 2 I 2x2 )
T
i = 1,2, K n
51
r r r
Note that each of the blocks η&i = Ac1η i + Bci eVqd represents a state space
yield:
r r r r r r
η (k + 1) = Ac*η ( k ) + Bc* eVqd ( k ) , eVqd ( k ) = Vref qd (k ) − Vload qd (k ) (2.43)
where:
Bc* = ∫ e A1⋅(TS −τ ) Bc dτ
TS
Ac* = exp( Ac ⋅ TS )
0
Now that we have determined the ‘plant’ and the servo compensator, the control
where the gains K = [K 0 K 1 ] are found by minimizing the discrete performance index:
∞
J ε = ∑ (z ( k ) ' z (k ) + ε ⋅ u( k ) ' u ( k ) ) ,
k =0
⎡ x *p ⎤
z=⎢ ⎥ (2.45)
⎣η ⎦
for the augmented ‘equivalent plant’ (2.41) and the servo compensator (2.43):
r r
⎡ x *p (k + 1)⎤ ⎡ Ad 0 ⎤ ⎡ x *p (k )⎤ ⎡ Bd ⎤
⎢ ⎥=⎢ + u (k )
Ac* ⎥⎦ ⎢⎣ η (k ) ⎥⎦ ⎢⎣ − Bc* D ⎥⎦ 1
(2.46)
⎣ η (k + 1) ⎦ ⎣− Bc C
*
52
Current Limit and Control Saturation Handling
r *
The current command I cmd qd (k ) generated by the robust servomechanism
r r
⎧ I cmd qd *
(k ) *
if I cmd qd (k ) ≤ I max
r ⎪⎪ r
I cmd qd (k ) = ⎨ I cmd qd (k ) r
*
(2.47)
⎪ Ircmd * (k ) I max
*
if I cmd qd (k ) > I max
⎪⎩ qd
Equation (2.47) limits the magnitude of the current commands but maintains their vector
that get excited by the harmonic contents of the error signals at frequency ωi . When the
break in the control loop. This problem is similar to the integrator windup problem that
occurs in an integral type controller. To prevent this, the servo-compensator in (2.43) can
be modified as follows:
r r r
η (k + 1) = Ac*η ( k ) + Bc*e1 ( k ) ,
r
r ⎧⎪erVqd ( k ) if I cmd qd
*
≤ I max
e1 ( k ) = ⎨ r (2.48)
⎪⎩ 0 if I cmd qd > I max
*
53
Using (2.48), during the current limit saturation, the servo compensator states will
continue to oscillate at the harmonic frequency with constant magnitude. The resulting
r *
I cmd qd (k ) > I max
r r
Vref qd (k ) eVqd Discrete implementation of
states η1
r
1 r
+ ⋅ I 2×2
s 2 + ω12
r
Vload qd (k ) Current limit
r Equation (4.17) r
K1 *
I cmd qd I cmd qd
Discrete implementation of r
r states η2
1
⋅ I 2×2 Servo
s 2 + ω2 2 Compensator
Gains
r
. ⎡ Vinv qd (k ) ⎤
.. K2 ⎢ r ⎥
⎢ rI invqd (k ) ⎥
Stabilizing compensator ⎢ Vload (k ) ⎥
gains ⎢ r qd ⎥
Discrete implementation of r
states η n
⎢ I load qd (k ) ⎥
⎢ r ⎥
⎢⎣Vpwm qd (k −1)⎥⎦
1 r
⋅ I 2×2
s 2 + ωn 2
directly applying discretization (2.43) to the continuous state space implementation of the
transfer function 1 s 2 + ωi ( 2
) as given in (2.42). The discrete time system resulting from
this process (2.43) generally results in coefficients of the matrices Ac* and Bc* that have
54
compensator in a fixed point Digital Signal Processor such as the one used in this
research. To alleviate this problem, a balanced state space realization may first be
range. Given the system (2.42), a balanced realization may be found by finding a
⎡ TAc T −1 TBc / α ⎤
⎢ ⎥
⎣αI 2 x 2T
−1
0 ⎦
has approximately equal row and column norms [120]. The balanced realization of (2.42)
with
The continuous state space (2.49) can then be discretized to obtain the discrete
balanced realization process outline above can easily accomplished in Matlab using the
2.3.2 Tuning the optimum stabilizing gains for desired dynamic performance
The discrete performance index given in (2.45) applies equal weighting to the
plant states x *p and the servo compensator states η . To obtain a desired dynamic
performance and achieve robust stability, the cost function may be modified to provide
different weighting for the plant states, fundamental and harmonic servo compensator
states as follows:
55
∞
⎛ ⎞
J ε = ∑ ⎜ w p ⋅ x *p (k ) ' ⋅ x *p ( k ) + wS1 ⋅η1 ( k )′ ⋅η1 (k ) + wSH ∑η h ( k )′ ⋅η h ( k ) + ε ⋅ u(k ) ' u( k ) ⎟ (2.50)
k =0 ⎝ h ⎠
where constants w p , wS , and wSH are weighting constants scalars for the plant states,
fundamental, and harmonic servo compensator states respectively. The effect of choices
of values for w p , wS , and wSH in the dynamic performance of the system will be further
Matlab can be used to find the optimum state feedback gains K = [K 0 K 1 ] that
From the system equations (2.14) and the equivalent circuits shown in Fig. 2.14, it
can be seen that the d-axis and q-axis components of the system are completely
decoupled. Therefore, a controller can be designed for one of the axis, and the same
controller can be used for the other axis. In this section, step-by-step instructions are
given to calculate the control gains for the controller using Matlab. The complete lsiting
56
1. Define the plant parameters, sampling frequency, and harmonic frequencies.
KVA=5e3;
inv_volt=240/sqrt(3); % L-N Inverter Cap Voltage
out_volt=120; % L-N output voltage
tr=out_volt/inv_volt; % transformer turn ratio
Cinv=55e-6; % Inverter filter capacitance
Linv=1.8e-3; % Inverter filter inductance
Rinv=0.1; % Inverter filter resistance
Cload=5e-6; % Output capacitor
2. Define power, voltages, and current bases, and compute the per-unit values of the filter
components.
57
3. Design the dicrete sliding mode current controller. First define the reduced order plant
A=[0 xCinv*wfun ;
-1/(xLinv/wfun) -xRinv/(xLinv/wfun)];
B=[0; 1/(xLinv/wfun)];
E=[-xCinv*wfun; 0];
F=[0; -1];
C=[0 1];
D=0;
Next, discretize the reduced plant using ZOH method, and compute the equivalent control
sysc=ss(A,B,C,zeros(size(C,1),size(B,2)),'inputdelay',0);
sysd=c2d(sysc,Tsamp,'zoh');
[Acurrd,Bcurrd,Ccurrd,Dcurrd]=ssdata(sysd);
CBinv=inv(Ccurrd*Bcurrd);
CA=Ccurrd*Acurrd;
CD=Ccurrd*F;
sysc=ss(A,E,C,zeros(size(C,1),size(B,2)),'inputdelay',0);
sysd=c2d(sysc,Tsamp,'zoh');
[Acurrd1,Ecurrd,Ccurrd1,Dcurrd1]=ssdata(sysd);
CE=Ccurrd1*Ecurrd;
The equivalent control with one-half step-ahead prediction 2.36 can be then implemented
by:
ueq = Ksm⋅ [Icmd(k ) Vinv(k ) Vinv(k − 1) Iinv(k ) Iinv(k − 1) Iload(k ) Iload(k − 1)]
T
>> see(Ksm_q)
Columns 1 through 7
58
5. Design the robust servomechanism controller. First define the true plant.
sysc=ss(Ao,Bo,Co,Do,'inputdelay',0.5*Tsamp);
sysd=c2d(sysc,Tsamp,'zoh');
[Aod,Bod,Cd,Dd]=ssdata(sysd);
C1=[eye(2) zeros(2,3)];
C2=[zeros(1,3) 1 0];
Ad=Aod-Bod*(CBinv*CA*C1+CBinv*CE*C2);
Bd=Bod*CBinv;
Cc_star=eye(size(Ch_star,1));
Compute the servo harmonic compensators. First define the analog implementation. The
following commands define the compensator with fundamental, 3rd, 5th, and 7th
harmonics included.
Ch1=[0 1;
-w1^2 0];
Ch2=[0 1;
-w2^2 0];
Ch3=[0 1;
-w3^2 0];
Ch4=[0 1;
-w4^2 0];
Ch5=[0 1;
-w5^2 0];
Ch7=[0 1;
-w7^2 0];
Ch_star=[[Ch1 zeros(2,2) zeros(2,2) zeros(2,2) ];
[zeros(2,2) Ch3 zeros(2,2) zeros(2,2) ];
[zeros(2,2) zeros(2,2) Ch5 zeros(2,2) ];
[zeros(2,2) zeros(2,2) zeros(2,2) Ch7 ]
];
59
Bh_star=[0; 1; 0; 1; 0; 1; 0; 1;];
Then compute the balanced realization of the compensators and obtain the discrete
implementation of controller.
csysc=ss(Ch_star,Bh_star,Cc_star,zeros(size(Ch_star,1),size(Bh_star,2)));
[csysbc,Tbal]=ssbal(csysc);
csysd=c2d(csysbc,Tsamp,'zoh');
[Acon_d,Bcon_d,Ccon_d,Dcon_d]=ssdata(csysd);
The numerical values of the servo compensators A matrices are given by:
>> Acon_d
Acon_d =
Columns 1 through 6
0.9976 0.0947 0 0 0 0
-0.0514 0.9976 0 0 0 0
0 0 0.9781 0.1883 0 0
0 0 -0.2296 0.9781 0 0
0 0 0 0 0.9397 0.3716
0 0 0 0 -0.3148 0.9397
0 0 0 0 0 0
0 0 0 0 0 0
Columns 7 through 8
0 0
0 0
0 0
0 0
0 0
0 0
0.8829 0.3644
-0.6050 0.8829
Ad_big=[Ad zeros(size(Ad,1),size(Acon_d,2));
-Bcon_d*Cd Acon_d];
Bd_big=[Bd ; -Bcon_d*Dd];
60
Next, we will compute the optimum gains that stabilize the system and minimize the cost
optimal gain matrix K such that the state-feedback law u(k)=-Kx(k) minimizes the
∞
J ε = ∑ (x (k ) ' ⋅ Q ⋅ x ( k ) + u ( k ) ' Ru ( k ) + 2 x ′(k ) Nu(k ) )
k =0
To use the dlqr command for cost function (2.50), define R = ε , N = 0 , and the states
⎡Q1 0 ⎤ ⎡w I 0 ⎤
Q=⎢ ⎥ Q1 = wP ⋅ I 5 x 5 Q 2 = ⎢ S1 2 x 2
⎣ 0 Q 2⎦ ⎣ 0 wSH I 6 xx 6 ⎥⎦
epsilon=1e-5;
WP=0.005; %WP
WS1=5e5; %WS1
WSH=1e-2*5e5; %WH
Q1=WP*eye(size(Ad,1));
Q2=eye(size(Acon_d,1));
Q2(1:2,:)=WS1_W*Q2(1:2,:);
Q2(3:8,:)=WSH*Q2(3:8,:);
Q=[ Q1 zeros(size(Ad,1),size(Acon_d,2));
zeros(size(Acon_d,1),size(Ad,2)) Q2];
R=epsilon;
[Kd,S,E]=dlqr(Ad_big,Bd_big,Q,R);
Kd=-Kd;
Scale the inputs by a constant to provide better dynamic range of matrix Bcon_d.
s_scale=600;
Bcon_d=s_scale*Bcon_d;
Kd(:,6:13)=Kd(:,6:13)/s_scale;
61
>> Kd
Kd =
Columns 1 through 6
Columns 7 through 12
Column 13
0.0006
>> Bcon_d
Bcon_d =
0.0053
0.1110
0.0105
0.1103
0.0209
0.1089
0.0207
0.1067
62
CHAPTER 3
The RSP controller developed in the previous chapter guarantees exact asymptotic
tracking of the fundamental frequency reference and error regulation of the load
disturbance at each of the harmonic frequency included in the servo compensators. The
RSP guarantees this property independent of any perturbations in the plant as long as they
do not destabilize system. Perturbations in a UPS system may come as results of load
important therefore to analyze the stability property of the resulting controller under
possible disturbances in order to ensure proper operation of the UPS over its intended
operating range.
A system is said to achieve robust stability if it remains stable for all considered
perturbations in the plant. In this chapter, the stability robustness of the system with the
controller developed in the previous chapter will be investigated using the Structured
63
Recall from the preceding chapter, that the chosen cost function
∞
⎛ ⎞
J ε = ∑ ⎜ w p ⋅ x *p (k ) ' ⋅ x *p ( k ) + wS1 ⋅η1 ( k )′ ⋅η1 (k ) + wSH ∑η h ( k )′ ⋅η h ( k ) + ε ⋅ u(k ) ' u( k ) ⎟ (3.1)
k =0 ⎝ h ⎠
provides a way of tuning the performance of the controller through the choice of the
weighting scalars w p , wS , and wSH . In this chapter, the effect of different choices of
these scalars to stability robustness of the system and its transient performance will be
investigated. It is important to remember that even though, each choice of the weighting
scalars results in stability of the nominal plant, each choice may not necessarily achieve
the robust stability of the system under plant uncertainties. The transient performance of
the system is evaluated by performing moving window RMS calculations of the three
phase output voltages under transient load change from 0 to 100% resistive load.
Structured Singular Value (µ) can be used to analyze the stability robustness of a
to use the µ-framework to analyze robust stability of a linear system under perturbation,
the problem needs to be recast into a feedback loop diagram of Fig. 3.1 where M
represents a known stable MIMO transfer function of the linear system with n inputs and
64
{
∆ = diag [δ 1 I r1 ,Kδ s I rs , ∆ 1 K ∆ F ] : δ i ∈ C , ∆ j ∈ C
m j xm j
} (3.2)
S F
∑ ri + ∑ m j = n
i =1 j =1
The structured singular value of M with respect to the uncertainty set ∆ is defined as
[116]:
1
µ ∆ (M ) = (3.3)
min{σ (∆ ) : ∆ ∈ ∆, det (I − M∆ ) = 0}
system using the structured singular value. It states that, if nominal M(s) is stable then the
65
Exact computation of µ ∆ (M ( jω )) is difficult. Commercial software, however, is
available (like Matlab µ-Tools toolbox), which computes upper and lower bound of µ by
As stated earlier, in order to use the µ-framework to analyze robust stability, the
problem needs to be recast to that of Fig. 3.1. A class of general feedback loops called
Linear Fractional Transformations (LFTs) can be used to achieve this. For complete
discussion of the LFT and its use in representing model uncertainties refer to [116].
66
3.1.2 Uncertain Open Loop Model
Figure 3.2 shows the single-phase equivalent circuit of the converter with RL
- - -
dvinv 1
= (iinv − ⋅isnd ′) (3.4.a)
dt cinv
diinv 1
= (− rinv ⋅ iinv − vinv + vpwm ) (3.4.b)
dt linv
dvload ′ 1
= (isnd ′ − yrload ⋅ vload ′ − il ) (3.4.c)
dt cload
disnd ′ 1
= (− rtran ⋅ isnd ′ + vinv − vload ′) (3.4.d)
dt l tran
dil
= yl load ⋅ vload ′ (3.4.e)
dt
67
Let us assume that the following parameter variations exist in the system due to
Furthermore, the specification of the unit requires that it operates stably without
degradation in performance for load from 0 to 200% with power factor of 0.8 lagging at
maximum load. Beyond this point, the UPS shall transfer the load to bypass static switch
68
yrload = yrload nom (1 + yrload tol ⋅ δ yrload ) , δ yrload < 1 (3.5.g)
Terms with subscript nom indicate the nominal values of the parameters as given in
(2.14), with yrload nom = 0.8 , and ylload nom = 0.6 . The tolerances are assigned as follow:
cinv tol = 0.06 , linvtol = 0.15 , cload tol = 0.06 , ltranstol = 0.15 , rinvtol = 0.5 ,
Parameters cinv , linv , cload , and ltrans appear as their inverses in the dynamic
equations (3.4), and can be represented using lower LFT as shown in Fig. 3.3. For rinv ,
69
1 1 1 1
cinv linv cload ltrans
⇔
⇔
⇔
⇔
⎡ 1 ⎤ ⎡ 1 ⎤ ⎡ 1 ⎤ ⎡ 1 ⎤
− cinvtol ⎥ ⎢ linv − linvtol ⎥ ⎢ cload − cload tol ⎥ − ltranstol ⎥
⎢ cinv ⎢ ltrans
⎢ nom
⎥ ⎢ nom
⎥ ⎢ nom
⎥ ⎢ nom
⎥
⎢ 1 ⎢ 1 ⎢ 1 1
− cinvtol ⎥ − linvtol ⎥ − cload tol ⎥ ⎢ − ltranstol ⎥
⎣⎢ cinv nom ⎦⎥ ⎣⎢ linvnom ⎦⎥ ⎣⎢ cload nom ⎦⎥ ⎣⎢ ltransnom ⎦⎥
⇔
⇔
⎡rinv rinvdel ⎤ ⎡rtrans rtrans del ⎤ ⎡ yrloadnom yrloaddel ⎤ ⎡ ylload nom ylload del ⎤
⎢ 1 ⎢ 1 0 ⎥⎦
0 ⎥⎦
⎢ 0 ⎥⎦ ⎢ ⎥
⎣ ⎣ 1 ⎣ 1 0 ⎦
⎣
70
For each uncertain term, the corresponding LFT in Fig 3.3 and Fig. 3.4 can be
substituted in to the dynamic model of equation (3.4), with the uncertain perturbations
separated out as shown in Fig. 3.5. The diagram shown in Fig. 3.5 contains all the known
model of the plant and will be referred to as the nominal open-loop plant model P. This
model can be combined with a structured perturbation model as shown in Fig. 3.6 which
71
iinv
∫
⎡ 1 ⎤ vinv
⎢ cinv − cinvtol ⎥
isnd
wcinv ⎢ nom
⎥ zcinv
⎢ 1 − cinvtol ⎥
⎢⎣ cinvnom ⎥⎦
iinv
⎡ rinv rinvdel ⎤
∫
iinv
vinv ⎡ 1 ⎤
⎢ 1 0 ⎥⎦
z rinv
⎢ linv − linvtol ⎥
wrinv
⎣ vpwm ⎢ nom
⎥
wlinv ⎢ 1 − linvtol ⎥ zlinv
⎢⎣ linv nom ⎥⎦
vload
∫
⎡ yrloadnom yrloaddel ⎤ il vload
⎢ ⎡ 1 ⎤
w yrload ⎣ 1 0 ⎥⎦ z yrload ⎢ cload − cload tol ⎥
isnd ⎢ nom
⎥
⎢ 1
wcload − cload tol ⎥ zcload
⎢⎣ cload nom ⎥⎦
isnd
∫
⎡rtrans rtrans del ⎤ isnd
⎢ 1 ⎡ ⎤
0 ⎥⎦
vload 1
wrtrans ⎣ z rtrans ⎢ ltrans − ltranstol ⎥
vinv ⎢ nom
⎥
⎢ 1
wltrans − ltranstol ⎥ zltrans
⎣⎢ ltransnom ⎦⎥
∫
vload il
⎡ ylload nom ylload del ⎤
⎢ ⎥
w ylload ⎣ 1 0 ⎦ z ylload
iload
il
72
vpwm vinv
iinv
vload
iload
wcinv zcinv
wlinv zlinv
wcload P (s ) zcload
wltrans zltrans
wrinv z rinv
wrtrans zrtrans
w yrload z yrload
w ylload z ylload
wcinv zcinv
⎛ δ cinv ⎞
wlinv ⎜ ⎟ zlinv
wcload ⎜ δ linv ⎟
⎜δ ⎟ zcload
wltrans ⎜ cload ⎟ zltrans
⎜ δ ltrans ⎟
wrinv ∆ (s ) = diag ⎜ ⎟ z rinv
δ
wrtrans ⎜ rinv ⎟ zrtrans
w yrload ⎜ δ rtrans ⎟
⎜δ ⎟ z yrload
w ylload ⎜ yrload ⎟ z ylload
⎜δ ⎟
⎝ ylload ⎠
Figure 3.6 Interconnection of the nominal open loop plant with ∆ to form uncertain open
loop model
73
The state space model of the nominal plant P can be derived from Fig. 3.5 with
Outputs:
where
⎡C ⎤ ⎡D ⎤
AP = [Anom ] BP = [Bnom Bdel ] C P = ⎢ nom ⎥ DP = ⎢ nom ⎥ (3.8)
⎣ C del ⎦ ⎣ Ddel ⎦
⎡ 1 −1 ⎤
⎢ 0 0 0 ⎥
cinvnom cinv nom
⎢ ⎥
⎢ −1 − rinvnom ⎥
0 0 0
⎢ linvnom linvnom ⎥
AP = ⎢ − yrload nom 1 −1 ⎥
⎢ 0 0 ⎥
⎢ cload nom cload nom cload nom ⎥
⎢ 1 −1 − rtransnom ⎥
⎢ ltrans 0 0 ⎥
ltransnom ltransnom
⎢ nom
⎥
⎣ 0 0 ylload nom 0 − rε ⋅ ylload nom ⎦
T
⎡ 1 ⎤
Bnom = ⎢0 0 0 0⎥
⎣ linvnom ⎦
74
⎡ − cinv tol 0 0 0 0 0 0 0 ⎤
⎢ − rinvdel ⎥
⎢ 0 − linvtol 0 0 0 0 0 ⎥
⎢ linvnom ⎥
− yrload del
Bdel = ⎢⎢ 0 0 − cload tol 0 0 0 0 ⎥
⎥
cload nom
⎢ − rtransdel ⎥
⎢ 0 0 0 − ltranstol 0 0 0 ⎥
⎢ ltransnom ⎥
⎢⎣ 0 0 0 0 0 0 0 ylload del ⎥⎦
⎡1 0 0 0 0 0 0 0⎤
⎢0 1 0 0 0 0 0 0⎥
C nom =⎢ ⎥
⎢0 0 1 0 0 0 0 0⎥
⎢ ⎥
⎣0 0 yrload nom 0 1 0 0 0⎦
⎡ 1 −1 ⎤
⎢ 0 0 0 ⎥
cinv nom cinv nom
⎢ ⎥
⎢ −1 − rinvnom ⎥
0 0 0
⎢ linvnom linvnom ⎥
⎢ − yrload nom 1 −1 ⎥
⎢ 0 0 ⎥
⎢ cload nom cload nom cload nom ⎥
C del =⎢ 1 −1 − rtransnom ⎥
⎢ ltrans 0 0 ⎥
ltransnom ltransnom
⎢ nom
⎥
⎢ 0 1 0 0 0 ⎥
⎢ 0 0 0 1 0 ⎥
⎢ ⎥
⎢ 0 0 1 0 0 ⎥
⎢⎣ 0 0 1 0 0 ⎥⎦
r
⎡ 03 x 9 ⎤
Dnom =⎢ ⎥
⎣0 0 0 0 0 0 0 yrload del 0⎦
75
⎡ 0 − cinv tol 0 0 0 0 0 0 0⎤
⎢ 1 − rinv del ⎥
⎢ 0 − linv tol 0 0 0 0 0⎥
⎢ linv nom linv nom ⎥
⎢ 0 − yrload del
0 0 − cload tol 0 0 0 0⎥
⎢ cload nom ⎥
⎢ − rtrans del ⎥
Ddel =⎢ 0 0 0 0 − ltranstol 0 0 0⎥
⎢ ltrans nom ⎥
⎢ 0 0 0 0 0 0 0 0 0⎥
⎢ ⎥
⎢ 0 0 0 0 0 0 0 0 0⎥
⎢ 0 0 0 0 0 0 0 0 0⎥
⎢ ⎥
⎢⎣ 0 0 0 0 0 0 0 0 0⎥⎦
76
3.1.3 Using Matlab to Derive Uncertain Closed Loop Model
As stated earlier, in order to use the µ -framework to analyze the robust stability
of the system, the system needs to be recast in to that of Fig. 3.1. In this case, the system
comprises of the nominal open loop P plant with the controller loop closed around it as
illustrated in Fig. 3.7.a. The system enclosed by the box in Fig. 3.7.a needs to be
combined to form the closed loop plant M as shown in Fig. 3.7.b. Matlab, specifically the
µ -Analysis and Synthesis Toolbox, has been utilized to help in achieving this purpose. A
error
vref
vref error
wcinv z cinv
wlinv zlinv
vpwm vinv wcload zcload
⇔
iinv wltrans zltrans
vload wrinv zrinv
wrtrans M (s ) zrtrans
iload
w yrload z yrload
wcinv zcinv w ylload z ylload
wlinv zlinv
wcload P (s ) zcload
wcinv z cinv
wltrans zltrans
⎛ δ cinv ⎞
wrinv z rinv wlinv ⎜ ⎟ zlinv
wcload ⎜ δ linv ⎟
wrtrans zrtrans ⎜δ ⎟ zcload
w yrload z yrload
wltrans ⎜ cload ⎟ zltrans
⎜ δ ltrans ⎟
w ylload
wrinv ∆ (s ) = diag ⎜ ⎟ zrinv
z ylload δ
wrtrans ⎜ rinv ⎟ zrtrans
⎜ δ rtrans ⎟
w yrload ⎜δ ⎟ z yrload
w ylload ⎜ yrload ⎟ z ylload
wcinv zcinv ⎜δ ⎟
⎛ δ cinv ⎞ ⎝ ylload ⎠
wlinv ⎜ ⎟ zlinv
wcload ⎜ δ linv ⎟
⎜δ ⎟ zcload
wltrans ⎜ cload ⎟ zltrans
⎜ δ ltrans ⎟
wrinv ∆ (s ) = diag ⎜ ⎟ zrinv
δ
wrtrans ⎜ rinv ⎟ zrtrans
w yrload
⎜ δ rtrans ⎟
⎜δ ⎟ z yrload
w ylload ⎜ yrload ⎟ z ylload
⎜δ ⎟
⎝ ylload ⎠
(a) (b)
77
Since the controller is implemented in the discrete time system, the following
steps are necessary to obtain the closed loop plant model in Fig. 3.7.b:
• Apply a Zero Order Hold transformation to the continuous plant P, to include the
effect of the sample and hold process of the digital sampling process. This step
should also include the half sampling time computation delay. Let Ap, Bp, Cp, Dp be
the Matlab variables that hold the matrices in (3.8), then this step can be
sysp=ss(Ap,Bp,Cp,Dp);
sysp.inputdelay=[0.5*Tsamp zeros(1,8)];
syspd=c2d(sysp,Tsamp,'zoh');
sysp_w=d2c(syspd,'tustin');
[Ap_c,Bp_c,Cp_c,Dp_c]=ssdata(sysp_w);
• Obtain state space representations of the RSP and Discete Sliding Mode controller
78
vref err
icmd
vinv
vload iinv vinv
vpwm
vload iinv
iload iload
vpwm
The discrete sliding mode controller with one-half step ahead predictor can be
r
written in state space form with the inputs u ( k ) = [icmd vinv iinv iload ] and
r r r
y ( k ) = C SM x + DSM u (3.9)
where
⎡0 1 0 0 ⎤
ASM = 0 3 x 3 BSM = ⎢0 0 1 0 ⎥
⎢ ⎥
⎣⎢0 0 0 1⎦⎥
79
r
u ( k ) = [err vinv iinv vload iload vpwm],
r r r
y ( k ) = C servo x + Dservo u (3.10)
where:
The combined controller state-space system can then be calculated using sysic
dsm=pck(Asm,Bsm,Csm,Dsm);
servo=pck(Aservo,Bservo,Cservo,Dservo);
systemnames='dsm servo';
inputvar='[err; plant{4}]';
outputvar='[dsm]';
input_to_dsm='[servo; plant(1,2,4)]';
input_to_servo='[err; plant(1:4) ; dsm]';
sysoutname='Hcontrol_d';
cleanupsysic = 'yes';
sysic;
continuous domain using the inverse Tustin transformation. In Matlab this can be
achieved using
[Acontrol,Bcontrol,Ccontrol,Dcontrol]=unpck(Hcontrol_d);
sys_con_d=ss(Acontrol,Bcontrol,Ccontrol,Dcontrol,Tsamp);
• Finally the closed loop plant model P is obtained by invoking the sysic command
80
plant_nom_w=pck(Ap_c,Bp_c,Cp_c,Dp_c);
controller_w=pck(Acontrol,Bcontrol,Ccontrol,Dcontrol);
systemnames='plant_nom_w controller_w';
inputvar='[vref ; w{8}]';
outputvar='[vref-plant_nom_w(3); plant_nom_w(5:12)]';
input_to_plant_nom_w='[controller_w; w(1:8)]';
input_to_controller_w='[vref-plant_nom_w(3);plant_nom_w(1:4)]';
sysoutname='M;
cleanupsysic = 'yes';
sysic;
The Matlab variable M contains the closed loop plant M as shown in Fig. 3.7.b.
81
3.1.4 Computing Individual Perturbation Frequency Response and the Structured
Now that the closed loop plant M has been obtained, we are ready to compute the
omega=logspace(-1,5,500);
clp_g=frsp(clp_w,omega);
In the robust stability study, we only need to consider the transfer functions seen by the
perturbations. These transfer functions can be obtained from clp_g above using:
clpstab_g=sel(clp_g,2:9,2:9);
insight as to how each parameter affects the system. The following commands are used to
plot these
for i=1:8
subplot(4,2,i)
vplot('liv,m',sel(clpstab_g,i,i));
grid
end
Finally the structured singular value of the system for robust stability study is obtained
82
3.2 Tuning the controller performance using w p , wS , and wSH
Recall from the previous chapter the w p , wS , and wSH in cost function (3.1) are
the weighting scalars for the plant states, fundamental, and harmonic servo compensator
states respectively. The selection of values of these scalars provides a way of tuning the
controller for desired transient performance and stability robustness. The structured
singular value discussed in the previous sections will be used to evaluate the robust
stability of each controller resulting from different choices of weighting scalars. A time
response simulation of the single phase equivalent circuit will be used to compare the
common in the industry to use the deviation of the RMS output voltage from its nominal
value during a 100% resistive load change as a performance measure. A less than 5%
deviation of output voltage under 100% resistive load transient is not an uncommon
In all the cases above, equal weighting scalars are applied to the fundamental
compensator states and the harmonic states ( wSH = wS ), while the plant states weighting
83
w p in each case is decreased from 0.5 all the way down to 0.0001. The time response
simulations for Case 1 and Case 4 are illustrated in Fig. 3.9 and Fig. 3.10 respectively,
each of which showing the output voltage, reference voltage, load current, and RMS
variation of the output voltage during both 0 to 100% and 100% to 0% resistive load
transients. From Fig. 3.9 and Fig. 3.10 it can be seen that the output voltage RMS
deviates as much as close to 20% for Case 1 and approximately 4% for Case 4.
The time response similar to Fig. 3.9 and Fig. 3.10 was obtained for each of the
cases 1 to 5 and the resulting RMS output voltage variations are plotted collectively in
Fig. 3.11. It can be seen that as the scalar weighting w p decreases, the transient
performance improves, with the RMS variation as little as 2% for the case w p = 0.0001 .
These results are not unexpected since the weighting scalars represent the penalty applied
to each state in the system. Intuitively, decreasing w p while keeping the compensator
states weighting the same, decreases the penalty cost applied to the plant states in the cost
function to be minimized. This results in allowing the plant states to move more freely
and hence faster response. Notice, however, that only Case 4 and Case 5 results in RMS
To analyze the robust stability, the upper bound of the structured singular value in
each case is plotted in 3.12. It can be seen that only Case 1, Case 2, and Case 3 achieve
robust stability under the considered structured perturbations with the peak value of in
µ ∆ (M ( jω )) each case being less than 1. Case 4 and case 5 --the only cases with
acceptable transient performances-- do not achieve robust stability with the peak values
84
Figure 3.9 Transient response for Case 1: w p = 0.5 , wS = 5 × 105 , and wSH = wS (Top:
Output voltage and its reference, Middle: Load current, Bottom: RMS variations of
output voltage).
Figure 3.10 Transient response for Case 4: w p = 0.005 , wS = 5 × 105 , and wSH = wS (Top:
Output voltage and its reference, Middle: Load current, Bottom: RMS variations of
output voltage).
85
Figure 3.11 RMS output voltage variations during 0 to 100% and 100% to 0 for
different w p ,and wSH = wS
86
Figure 3.12 Upper bound of the structured singular values for different w p ,and wSH = wS
87
Consider now the following cases of scalars weighting:
In the cases 6 to 10 the wP is assigned the same value as in Cases 1 to 5 respectively, but
smaller weighting is used for the harmonic compensator states as compared to the
fundamental’s. Fig 3.13 and Fig. 3.14 show the time responses of Case 6 and Case 9
respectively, while Fig. 3.15 shows the RMS variations for all cases. Comparing Fig.
3.13 thru 3.15 with Fig. 3.9 thru 3.11, it can be seen that the transient response improves
for each of the last 5 cases, with Case 9 and Case 10 now result in only 2% and 1% RMS
variations respectively.
Fig. 3.16 shows the upper bound of the structured singular value in Cases 6 to 10.
It can be seen that reducing wSH improves the stability robustness of each of the cases
previously considered. The counterpart of Case 4 (Case 9) now achieves robust stability
with µ ∆ (M ( jω )) peak value of 0.7 giving robust stability margin of 1 > 1 . Case 10 still
0.7
does not achieve robust stability with peak value of µ ∆ (M ( jω )) still greater than 1.
Comparing all the above cases, it can be seen that Case 9 gives the best transient
performance while still maintaining stability robustness of the system under the
88
Figure 3.13 Transient response for Case 6: w p = 0.5 , wS = 5 × 105 , and wSH = 0.01 × wS
(Top: Output voltage and its reference, Middle: Load current, Bottom: RMS variations
of output voltage).
Figure 3.14 Transient response for Case 9: w p = 0.005 , wS = 5 × 105 , and wSH = 0.01 × wS
(Top: Output voltage and its reference, Middle: Load current, Bottom: RMS variations
of output voltage).
89
Figure 3.15 RMS output voltage variations during 0 to 100% and 100% to 0 for
different w p ,and wSH = 0.01 × wS
90
Figure 3.16 Upper bound of the structured singular values for different w p ,and
wSH = 0.01 × wS
91
Continuing with the analysis, it is instructive to see how each parameter
perturbation affects the stability robustness of the system. For this purpose, the frequency
response of each individual perturbation can be plotted as shown in Fig, 3.17 for the
92
It can be seen from Fig. 3.17 that for all the perturbations, the frequency response
values are less than 1.0, which confirms the stability robustness result of the structured
singular value presented earlier. The inverter filter inductor has the highest peak
frequency response value, with peak value of around 0.7, which means that an individual
perturbation in this parameter will be the closest to the margin of making the system to be
unstable. Closed loop pole-zero map of the uncertain closed loop system in Figure 3.6
with all other δ ’s set to zero except for the δ linv can be obtained for different values of
δ linv less than 1 and greater than 1. This is illustrated in Fig. 3.18. The system is stable at
nominal inductance value ( δ linv = 0 ), δ linv = −0.5 , and at the lowest value of the
component tolerance ( δ linv = −1.0 ). As the value δ linv is decreased further the system
93
Figure 3.18 Pole-zero map of closed loop uncertain system for different values of
individual-perturbation in inverter filter inductance parameter
94
3.3 Simulations of the Complete System
including the PWM inverter are presented. The simulations have been performed in
requiring only interconnection of library components models for building the circuits for
simulations.
The simulation results presented in this report have been obtained using the
proposed control strategy with the controller obtained in Case 9 in the previous section.
2. Transient performance.
The PWM switching frequency is chosen to be 5400 Hz and the control update
rate is equal to the PWM switching period giving an update rate of 90 times per 60 Hz
95
3.3.1 Steady state performance
Fig. 3.19 thru Fig. 3.21 show the steady state waveforms of load currents, load
voltages, and inverter voltages for different types of load both from simulations and
experimental results. The RMS values of the line-to-line load voltages for each case is
given in Table 3.1 It can be seen from Table 3 that all the numbers show that the
Load Simulation
Ph. A Ph. B Ph. C
No load 120.0 120.0 120.0
Full load 120.0 120.0 120.0
1-phase 119.7 121.1 119.2
2-phase 119 120.8 120.3
96
Figure 3.19 Simulation results for balanced full resistive load
97
Figure 3.20 Simulation results for unbalanced single-phase resistive load
98
Figure 3.21 Simulation results for unbalanced two-phase resistive load
99
3.3.2 Transient Performance
Fig. 3.22 and Fig. 3.23 show the simulation results respectively for 0-
100% load transient. It can be seen that the waveforms of the voltages and currents for
both cases look very similar. The voltage transients in both cases were characterized by
the existence of dips in the voltages and followed by slight overshoots exhibiting some
5th and 7th harmonic contents. The harmonics showed up as responses of the servo
harmonic compensator to load transients. Notice that the load transient lasts for only one
cycle of the fundamental, which shows the fast response of the controller. Also, in any
cases the RMS load voltages variation were less than the 5% meeting the specification
required.
100
Figure 3.22 Simulation results for 0-100% resistive load transient
101
Figure 3.23 Simulation results for 0-100% resistive load transient
102
3.3.3 Harmonic tracking and regulation under non-linear load
harmonics disturbances, the system can be tested under non-linear load. However, since a
non-linear load may contain other harmonics than those included in the servo
compensators, it is hard to judge whether or not each harmonic compensator has been
implemented correctly. As an alternative, test may be conducted to verify how well the
controller can track references containing the specified harmonics. This in turn will give a
good indication of how well the controller can reject those harmonic disturbances.
In Fig. 3.24, 3.25, 3.26, and 3.27, simulation results are presented for cases when
the DQ voltages references contain only the fundamental component, the fundamental
with 5th harmonic, the fundamental with 7th harmonic, and the fundamental with 5th and
7th harmonics, respectively. It can be seen that in all cases the actual DQ voltages follow
the harmonic references perfectly. This indicates that the controller in turn is able to
103
200
-100
-200
0.05 0.055 0.06 0.065 0.07 0.075 0.08 0.085 0.09 0.095 0.1
200
Vrefd & Vloadd (volts)
100
-100
-200
0.05 0.055 0.06 0.065 0.07 0.075 0.08 0.085 0.09 0.095 0.1
200
Vrefq & Vloadq (volts)
100
-100
-200
0.05 0.055 0.06 0.065 0.07 0.075 0.08 0.085 0.09 0.095 0.1
200
Vrefd & Vloadd (volts)
100
-100
-200
0.05 0.055 0.06 0.065 0.07 0.075 0.08 0.085 0.09 0.095 0.1
Figure 3.25 Simulation result: Fundamental with 5th harmonic references tracking
104
200
-100
-200
0.05 0.055 0.06 0.065 0.07 0.075 0.08 0.085 0.09 0.095 0.1
200
Vrefd & Vloadd (volts)
100
-100
-200
0.05 0.055 0.06 0.065 0.07 0.075 0.08 0.085 0.09 0.095 0.1
Figure 3.26 Simulation result: Fundamental with 7th harmonic references tracking
200
Vrefq & Vloadq (volts)
100
-100
-200
0.05 0.055 0.06 0.065 0.07 0.075 0.08 0.085 0.09 0.095 0.1
200
Vrefd & Vloadd (volts)
100
-100
-200
0.05 0.055 0.06 0.065 0.07 0.075 0.08 0.085 0.09 0.095 0.1
Figure 3.27 Simulation result: Fundamental with 5th and 7th harmonic references tracking
105
3.3.4 Short Circuit at the output terminals
The final verification to the proposed control algorithm is for its ability to limit
the inverter currents under short circuits at the output terminal. Fig. 3.28 shows the
simulation results for this condition. In this case, the current limit was set at 300% level..
From these figures it can be seen that the discrete sliding mode controller provides a fast
106
CHAPTER 4
The effectiveness the proposed control strategy has been verified on a 5 kVA
experimental test bed built at the Mechatronics Laboratory of the Ohio State University.
Fig. 4.1 shows a block diagram of the test bed. The converter system consists of the
following:
• Three-phase 240 V utility input that also acts as the bypass source.
driver module (SKHI-22), three-phase input filter inductors with inductance of 8.4
107
• A 6 kVA, 60 Hz, 240V Delta/208V Wye, Z=3% three-phase isolation
transformer
• Load for the PWM inverter consists of a linear load with resistive and inductive
108
⎡Vinv ab ⎤
⎡Vload an ⎤
⎡Vbyp ab ⎤ ⎡Vpwm ab ⎤ ⎢Vload ⎥
⎢Vpwm ⎥ ⎢Vinv ⎥
⎢Vbyp ⎥
⎢ bc ⎥ ⎢ bc ⎥ ⎢ bn ⎥
⎢ bc ⎥
⎣⎢Vpwm ca ⎦⎥ ⎣⎢Vinv ca ⎦⎥ ⎣⎢Vload cn ⎦⎥
⎣⎢Vbyp ca ⎦⎥
2.0mH
Iload b
Iin c Iinv c
Iload c
109
Figure 4.1 The 5 kVA Experimental Test Bed System
Figure 4.2 Three single-phase diode rectifier as non-linear load
have been designed to provide independent control of the three-phase PWM rectifier and
the three-phase PWM inverter as shown in Fig. 4.1. Signal conditioners circuits have
been designed and built to provide interfacing of the high voltages and currents
110
4.2 Experimental Results
Table 4.1 gives the steady state RMS output voltages regulation under different
types of loads. It can be seen that the control strategy provides good output voltages
regulation in all cases. Note that, the deviations in the output voltages for unbalanced
load are due to the uncontrollable 0-component of the load currents. However, as can be
seen the effect is minimal showing the effectiveness of the LC filter at the output side of
the transformer.
Fig. 4.3 shows the waveforms of the load currents, load voltages, and inverter
voltages under various linear loads: resistive, inductive, balanced, and unbalanced load.
Table 4.2 summarizes the output voltages THD under different types of loads showing
the superior THD performance of the proposed control strategy. Better results may be
controller.
Fig. 4.4 and 4.5 show the responses of the load voltages on resistive load
transients, 0-100% and 100%-0 respectively. It can be seen that the load voltages recover
within less than a cycle after the load is applied. Even though in both cases the output
voltages exhibit slight overshoot due to the under-damped nature of the control, the
output voltages magnitudes deviate less than 5% of the nominal. These results show that
the perfect RSP controller provides fast transient recovery under load transients with
111
Finally, the effectiveness of the current controller was verified by applying a
sudden three-phase short circuit on the output load terminals. The current limit was set at
300% level, and the inverter was shutdown deliberately after ten cycles of short circuit
condition. The results are shown in Fig. 4.6. It can be seen that the discrete sliding mode
Load Experiments
Ph. A Ph. B Ph. C
No load 119.9 120.1 120.1
Full load resistive 119.9 120.1 120.1
1-phase resistive 119.6 121.2 119.3
2-phase resistive 118.9 120.9 120.4
Full load 0.8pf lag 119.9 120.1 120.1
Full load Non-linear 119.9 120.1 120.1
112
(a) (b)
(c) (d)
Figure 4.3 Steady state linear load (a) 100% resistive balanced (b) 100% 0.8pf load (c)
100% resistive unbalanced (phase A unloaded) (d) 100% resistive unbalanced (phase
A&B). Top: load currents; middle: load voltages; bottom: inverter voltages.
Figure 4.4 Resistive load transient: 0% to 100%. Top: three-phase load currents, bottom:
three-phase load voltages resistive load transient
113
Figure 4.5 Resistive load transient:100% to 0% Top: three-phase load currents, bottom:
three-phase load voltages resistive load transient
Figure 4.6 Three-phase short-circuit on output terminals. Top: inverter currents, middle:
load voltages, and bottom: inverter voltages.
114
4.2.2 Comparison with Synchronous Reference Frame Control with PI controllers
quantities. Note that it is not necessary for the proposed control to be performed in the
synchronous frame, since the proposed control can guarantee zero steady state error for
signals varying at fundamental frequency. Fig. 4.7 shows the synchronous reference
frame control of the voltages and currents using the PI controllers. In this figure, the
voltages and currents are expressed in the synchronous reference frame using
transformation (4.1):
f qe = f qs ⋅ cos ωt − f ds ⋅ sin ωt
(4.1)
f ds = f qs ⋅ sin ωt + f ds ⋅ cos ωt
where the qs and ds subscripts denote variables in the stationary reference frame,
the qe and de denote the synchronous reference frame variables, and ωt represent the
phase angle of the voltages references. Given three-phase balanced voltages references, it
can be shown using (4.1) that the de and qe references voltages in the synchronous
reference frame are given by the maximum value of the reference voltage Vmax and 0
respectively as shown in Fig. 4.7. The control shown in Fig. 4.7 also includes decoupling
terms [26] that are used to compensate for the coupling of the d and q dynamics due to
transformation (4.1), and feed forward terms used to improve transient performance.
115
Iqeload Vqeinv
*
0
Iqeinv Vqe pwm
ω ⋅ Cinv ω ⋅ Linv
Vqeinv Iqeinv
Vdeinv Ideinv
ω ⋅ Cinv ω ⋅ Linv
Vde pwm *
V max
Ideinv *
Ideload Vdeinv
Figure 4.7 Synchronous Reference Frame Voltage and Current Control with PI
controllers
It is worth noting that, if properly tuned, the control given in Fig. 4.8 can easily
provide a good transient response for sudden load changes. Also, based on the internal
model principle [86], since the PI controller includes an integral term ( 1 s ), the controller
guarantees zero steady state error for references and disturbances at zero frequency in
references and disturbances in abc reference frame. Therefore, it is expected that the
control in Fig. 4.7 would have a good steady state voltage regulation and low total
harmonic distortion (THD) voltage when only linear balanced loads are applied.
However, when disturbances other than the zero frequency (in synchronous reference
frame) exist, then the PI controller can no longer guarantee a perfect regulation at steady
state. Note that, even though unbalanced linear loads in abc reference frame (or
116
frame they appear as signals varying at frequency other than 0 as apparent from
transformation (4.1). Therefore, the PI controllers shown in Fig. 4.7 would not yield
satisfactory steady performance for non-linear harmonics and unbalanced non-linear and
linear loads.
Figure 4.10 shows experimental results showing the steady state performances
comparison between the proposed control and the control shown in Fig 4.7 under 100%
non-linear load condition. Both controls are updated at the same control rate equal to the
3.2 kHz PWM switching frequency. The non-linear load applied consists of three single-
phase diode rectifiers with RC load as shown in Fig. 4.2 with small amount of linear load
added to obtain 100% rated power. It can be seen that the proposed control yields output
voltages with significantly lower THD than the PI controllers, which confirms the
(a) (b)
Figure 4.8 Steady state performance comparisons under non-liner load (a) using proposed
control (b) PI controllers. Top: three phase load currents. Bottom: Three-phase load
voltages
117
CHAPTER 5
In this PhD research, a combination of the droop method and the average power
control is proposed for load sharing of paralleled inverters as shown in Fig 5.1. The
average power method is used to overcome the sensitivity of the droop method towards
measurements errors and wiring impedance mismatches. Choosing phase angle (instead
of frequency) as the controlling variable for the real power control has the advantage that
the average power control can be deactivated by setting the input to the integrator to zero
if communication error is detected. This makes the load sharing control to be less
way of looking at this is that the average power control fine-tunes the droop method to
achieve proper load sharing when measurement errors or other non-idealities exist. The
simulations and experimental works show that proper load sharing at different load
conditions can still be achieved even after the average power is deactivated.
118
P0 θref
P + - m + ∆θ
+ θ
+
m<0
+
Pavg + T mi
z −1
- integrator mi < 0
Q0 Vnom
- + Vmax
Q + + ∆V
n
+
n<0
Qavg + T
z −1
ni
- integrator ni < 0
Figure 5.1 Combined droop and average power methods for load sharing control
Note that in Fig. 5.1, θ ref represents a reference phase angle which each inverter
needs to lock to. When the bypass source is available, the phase angle θ ref can be obtained
using a Phase Locked Loop algorithm that locks both phase frequency of each inverter to
a common source. When the bypass source is not available, no common reference exists
for the inverters to synchronize to. Each inverter in this mode will free-run at its nominal
frequency. In this case, the load sharing phase adjustment will attempt to lock the phase
angle; however some other means must exist to lock frequency of each inverter. A slight
119
different in frequency will cause each inverter to be out of synchronization which in turn
will break the load sharing control above. The solution to this issue will be further
Another issue to address is what power quantities should be used for the method
proposed. Most previous works in paralleled three-phase inverters use total three-phase
real and reactive power for this purpose. However, using total power does not guarantee
load sharing of individual phase in the case of unbalanced load or wire impedances. To
answer this question, in this research the load sharing control shall be applied to the
individual DQ axis circuit i.e. the method shall be applied to the real and reactive power
on each D and Q axis. The reasoning for this can be justified if one examines the DQ0
model of the paralleled inverters as shown in Fig. 5.2. The problem of paralleling three-
circuit. Since the 0-component is not controllable and the filter has suppressed its
quantities, the load sharing shall only be accomplished for the D and Q axis single-phase
circuitry.
120
Iload q1
Z out _ q1 Z wire _ q1
+ +
Vload q1
Iload q
-
Iload d 1
Z out _ d 1 Z wire _ d 1
+ +
Vload d 1 Iload d
-
Iload01 Z wire _ 01
Z out _ o1
+
Vload 01
Iload 0
-
Iloadq 2
Z out _ q 2 Z wire _ q 2
+ +
Vload q 2
-
Iload d 2
Z out _ d 2 Z wire _ d 2
+ +
Vload d 2
-
Iload 02
Z out _ o 2 Z wire _ 02
+
Vload 02
Applying the proposed load sharing control for the individual D and Q axis, we
obtain the following equations (5.1) to (5.6) for updating the phases and amplitudes of the
121
Q-Axis:
Phase Angle:
⎧P − Pq if activated
φq (k + 1) = φq (k ) + mi ⋅ ePq ePq = ⎨ q _ avg
⎩ 0 if deactivated
∆θ q (k ) = φq (k ) + m ⋅ (Pq − Pq 0 ) (5.1)
θ q (k ) = θ ref (k ) + ∆θ q (k )
Amplitude:
⎧Q − Qq if activated
υ q (k + 1) = υ q (k ) + ni ⋅ eQq eQq = ⎨ q _ avg
⎩ 0 if deactivated
Voltage Reference:
D-Axis:
Phase Angle:
⎧P − Pd if activated
φd (k + 1) = φd (k ) + mi ⋅ ePd ePd = ⎨ q _ avg
⎩ 0 if deactivated
∆θ d (k ) = φd (k ) + m ⋅ (Pd − Pd 0 ) (5.4)
θ d (k ) = θ ref (k ) + ∆θ d (k )
Amplitude:
⎧Q − Qd if activated
υ d (k + 1) = υ d (k ) + ni ⋅ eQd eQq = ⎨ d _ avg
⎩ 0 if deactivated
Voltage Reference:
122
5.2 Stability Analysis of Combined Droop and Average Control Method
In this section, stability of the paralleled UPS with the combined droop and
average control method described in the previous section will be investigated. Stability
analysis of paralleled UPS with only the frequency and voltage droop method has been
discussed in several prior works [106-108]. In these prior works, small signal analysis
models were developed to study the stability of the paralleled inverters network around
the state of equilibrium. The droop technique analyzed in [106] uses total three-phase
instantaneous active and reactive power, and therefore can not be directly extended to
analyze the proposed control in this research, since in this research the paralleling
problem is viewed as a three separate single-phase circuits (Fig. 5.2). Reference [107]
and [108], on the other hand, provide the analysis for two interconnected single phase
inverters using frequency and voltage droop method. The stability analysis provided in
this section, hence, follows [107] and [108], with the exception that the analysis will be
performed for the phase and voltage droop method combined with the average power
control method. Additionally, the analysis provided here is performed in discrete time
123
Fig. 5.3 shows an equivalent circuit of two paralleled inverters for one of the dq
phases in Fig. 5.2 with a resistive load and interconnection impedances modeled as
inductors. Due to the high bandwith of the voltage control loop as compared to the
bandwidth of the power dynamics, it is safe to assume the UPS inverters acting as ideal
voltage sources [106]. The variables E1 and E 2 represent the per-unit rms voltage, and
δ 1 and δ 2 power angles of the output voltages of UPS1 and UPS2 respectively. jX 1 and
jX 2 denote the per-unit impedances of the interconnection wires, and R represents the
E1∠δ 1 E2 ∠δ 2
jX 1 jX 2
I1 R I2
124
The active and reactive power delivered by each UPS can be calculated using
where
a1 = RX 2 D a 2 = RX 1 D
2 2
b = R 2 (X 1 + X 2 ) D c = RX 1 X 2 D
(
d1 = R 2 ( X 1 + X 2 ) + X 1 X 2
2
)D (
d 2 = R 2 (X 1 + X 2 ) + X 1 X 2 D
2
)
D = R 2 (X 1 + X 2 ) + (X 1 X 2 )
2 2
⎡ ∆P1 ⎤ ⎡ ∆δ 1 ⎤
⎢ ∆P ⎥ ⎢ ∆δ ⎥
⎢ 2 ⎥ = S ⋅⎢ 2⎥ , S = [sij ], i, j = 1K 4 (5.8)
⎢ ∆Q1 ⎥ ⎢ ∆E1 ⎥
⎢ ⎥ ⎢ ⎥
⎣ ∆Q2 ⎦ ⎣ ∆E 2 ⎦
125
s21 = −bE1e E 2 e cos(δ 2 e − δ 1e ) + cE1e E 2 e sin(δ 2 e − δ 1e )
s 43 = −b ⋅ E 2 e cos(δ 2 e − δ 1e ) + cE 2 e sin(δ 2 e − δ 1e )
In this research, each of the active and reactive powers is computed once every
one line cycle and digital low pass filtering is applied to each power computation for the
purpose of removing measurement noise. The digital filters used are given in (5.9)
p2 m (k + 1) = (1 − γ ) p2 m (k ) + γP2 (k ) (5.9.b)
126
Using the filtered power quantities in (5.9), the combined average power control
and droop method from the previous section can be written in per unit notation as
follows:
δ 1 (k ) = φ1 (k ) + m ⋅ p1m (5.10.a)
δ 2 (k ) = φ2 (k ) + m ⋅ p2 m (5.10.b)
E1 (k ) = 1 + υ1 (k ) + n ⋅q1m (5.10.c)
E 2 (k ) = 1 + υ 2 (k ) + n ⋅q 2 m (5.10.d)
mi
φ1 (k + 1) = φ1 (k ) + ⋅ ( p1m − p 2 m ) (5.11.a)
2
mi
φ2 (k + 1) = φ2 (k ) + ⋅ ( p 2 m − p1m ) (5.11.b)
2
ni
υ1 (k + 1) = υ1 (k ) + (q1m − q2 m ) (5.11.c)
2
ni
υ 2 (k + 1) = υ 2 (k ) + (q2 m − q1m ) (5.11.d)
2
Using small signal analysis (5.9), (5.10), and (5.11) can be linearized as shown in
127
Linearized Combined Average and Droop Control
∆E 2 (k ) = ∆υ 2 (k ) + n ⋅ ∆q 2 m (5.13.d)
mi
∆φ1 (k + 1) = ∆φ1 (k ) + ⋅ (∆p1m − ∆p 2 m ) (5.14.a)
2
mi
∆φ2 (k + 1) = ∆φ2 (k ) + ⋅ (∆p 2 m − ∆p1m ) (5.14.b)
2
ni
∆υ1 (k + 1) = ∆υ1 (k ) + (∆q1m − ∆q2 m ) (5.14.c)
2
ni
∆υ 2 (k + 1) = ∆υ 2 (k ) + (∆q2 m − ∆q1m ) (5.14.d)
2
Substituting (5.13) into the linearized power equations (5.8), we obtain linearized
equations in terms of the integrator states and power filter states perturbations as:
∆x (k + 1) = A∆ ⋅ ∆x (k ) (5.16)
128
where:
A∆ =
⎡(1 − γ ) + mγs11 mγs12 nγs13 nγs14 γs11 γs12 γs13 γs14 ⎤
⎢ mγs21 (1 − γ ) + mγs22 nγs 23 nγs24 γs21 γs22 γs23 γs24 ⎥
⎢ ⎥
⎢ mγs31 mγs32 (1 − γ ) + nγs33 nγs34 γs31 γs32 γs33 γs34 ⎥
⎢
⎢ mγs41 mγs 42 nγs 43 (1 − γ ) + nγs44 γs41 γs42 γs43 γs44 ⎥⎥
⎢ mi
− i
m
0 0 1 0 0 0 ⎥
⎢ 2 2 ⎥
⎢ m mi ⎥
⎢ − i 0 0 0 1 0 0 ⎥
⎢ 2 2 ⎥
⎢ ni ni
0 0 − 0 0 1 0 ⎥
⎢ 2 2 ⎥
⎢ n ni ⎥
⎢⎣ 0 0 − i 0 0 0 1 ⎥
2 2 ⎦
Equation (5.16) describes the system dynamics around the equilibrium point. The
eigenvalues of matrix A∆ can be used to determine the stability of the system around the
state of equilibrium and the type of dynamic response expected. For example, assume the
following system parameters: X 1 = 0.001 , X 2 = 0.002 , and R = 1.0 . The filter time
constant is chosen to be γ = 0.5 . Let us consider two cases of droop coeffiecients and
The state of equilibrium for each case needs to be calculated using power flow analysis.
However, since the amplitude and phase adjustments made by each UPS are very small
compared to the nominal value of the voltages and the phases are always initially in
129
synchronization with respect to a reference voltage, we can choose ( δ 1e , δ 2 e , E1e ,
E 2 e )=(0, 0, 1,1) for calculation of each terms in matrix S in (5.8). The eigenvalues of
Case 1:
Case 2:
Case 3:
In Case 1 and Case 2, all the eigenvalues are located within the unit circle, so both
systems are stable. In case 3, two of the eigenvalues lie outside of the unit circle so the
system is unstable. In Case 1, all eigenvalues are real, and therefore the system has a
damped response. Case 2 presents eigenvalues with imaginary components so the system
circuit of Fig. 5.2 with the proposed control have been performed using Simulink Power
System Blockset. Figure 5.4 (a), (b), and (c) show the system responses ( P1 , P2 , Q1 , and
Q2 ) for Case 1, Case 2, and Case 3, respectively. It can be seen that the small signal
stability analysis accurately predicts the nature of stability for each case.
130
(a) (b)
(c)
Figure 5.4 System responses (a) Case 1, (b) Case 2 (c) Case 3
131
5.3 Generation of the Reference Angle θ ref
As explained in the previous section, when the bypass source is available the
reference angle θ ref can be obtained using a PLL that locks the phase and frequency of
each inverter to those of the bypass source. Fig. 5.5 shows a single-phase digital PLL that
can be used to achieve this purpose. At each sampling time, the PLL computes the
average over one reference cycle ( Tref = 2π ωref ) of the product of the bypass voltage
Vbyp with the cosine of the reference angle θ ref . A PI controller is then used to force this
average to zero. Using the relation sin α ⋅ cos β = 0.5 sin(α + β ) + 0.5 sin(α − β ) , it can be
easily seen that when the one cycle average becomes zero, the reference angle θ ref is
t +Tref
∫ [V ⋅ cos(θ ref )] dt
1
⋅
ωnom
byp
Vbyp θ ref
T
ωref z −1
cos( )
Figure 5.5 Generation of reference angle for the case when bypass source is available
132
When the bypass source is not available, the reference θ ref needs to be updated
differently than that of Fig. 5.5. In this research, a method is proposed to force the
frequency of each inverter to stay lock together during free-running operation. This
method is shown in Fig. 5.6. The idea of this method originates from the fact that if the
frequencies of the inverters are not the same, then the load sharing angle adjustment ∆θ
in Fig. 5.1 will continue to ramp up or down at a rate proportional to frequency errors
between the two sources. To prevent this, the inverter frequency needs to be adjusted to
force the rate of change of the angle adjustment to zero. In the method shown in Fig. 5.6,
the rate of change of the angle adjustment is forced to zero through an integrator that
outputs the frequency adjustment from the nominal value. Note that since there is only
one θ ref for both DQ axis, the rate of change of ∆θ is calculated from the average rate of
ωnom
∆θ (k )
δω ωref θ ref
T T
kf
z −1 z −1
∆θ (k − 1)
Figure 5.6 Generation of reference angle for the case when bypass source is available
133
5.4 Harmonic Sharing
The proposed load sharing method above does not guarantee the harmonic
components of the load current to share, since it affects only the phase and magnitude of
the fundamental output voltage. A means must exist to affect the harmonic components
of the load currents based on its harmonic contents. The individual control of the
harmonics proposed for the single PWM control in this PhD research, presents a clear
advantage for achieving this purpose. The control gains affecting the harmonic in the
voltage control can be individually adjusted based on the harmonic content of the load
current at that harmonic frequency. For example, the pole frequencies of the harmonic
compensator can be shifted based on the harmonic content of the load current at those
frequencies. This is illustrated in Fig. 5.7, where I3, I5, and I7 denote the harmonic
contents of the load currents at third, fifth, and seventh harmonics respectively, and ω 3′ ,
ω 5′ , and ω 7′ are the third, fifth, and seventh harmonic frequencies of the harmonic
compensator poles. Unlike previous works, the harmonic drooping method proposed has
the advantage that it does not degrade the fundamental component properties; it only
134
ω3
I3 +
∆ω 3
m3 + ω 3′
ω5
I5 +
∆ω 5
m5 + ω5′
ω7
I7 +
∆ω 5
m7 + ω7′
Figure 5.7 Harmonic Drooping thru the harmonics compensators pole shifting
r r r
If η3 , η5 , and η7 denotes the states of the harmonic servo compensators for 3rd, 5th
and 7th, then the harmonics servo-compensator for the single unit PWM control are
modified as follows:
r r
η&3 = Ac 3 (ω3′ )η3 + Bc eVqd
r r
η&5 = Ac 5 (ω5′ )η3 + Bc eVqd (5.17)
r r
η&7 = Ac 7 (ω7′ )η 7 + Bc eVqd
135
where
r r r
eVqd = Vref qd − Vload qd
r r
⎛ 02 x 2 I 2x2 ⎞
Aci (ωi′ ) = ⎜⎜ r r ⎟ , i = 3,5,7
⎝ − ωi′ I 2 x 2 0 2 x 2 ⎟⎠
2
r r
(
Bc = 0 2 x 2 I 2x2 )
T
i = 1,2, K n
The ω3′ , ω5′ , and ω7′ are computed from the harmonic droop in Fig. 5.7 and are given by:
ω3′ = ω3 + m3 I 3 , ω3 = 2π ⋅ 3 ⋅ 60
ω5′ = ω5 + m5 I 5 , ω5 = 2π ⋅ 5 ⋅ 60 (5.18)
ω7′ = ω7 + m7 I 7 , ω7 = 2π ⋅ 7 ⋅ 60
where I3, I5, and I7 denote the harmonic contents of the load currents at third, fifth, and
system to yield:
r r r
ηi (k + 1) = Aci* (ωi′ )ηi ( k ) + Bc*eVqd ( k ) i = 3,5,7 , (5.19)
r r r
eVqd (k ) = Vref qd ( k ) − Vload qd ( k )
where:
Bc* = ∫0 S e A1⋅(TS −τ ) Bc dτ
T
Ac* = exp( Ac ⋅ TS )
136
The modified robust servo mechanism output voltage control with harmonic
r *
I cmdqd (k ) > I max
r r
Vref qd (k ) eVqd Discrete implementation of
states η1
r
1 r
+ ⋅ I 2×2
s 2 + ω12
r
Vloadqd (k ) Current limit
Discrete implementation of r r r
1 r states η3 K1 *
I cmdqd
Equation (4.17)
I cmdqd
⋅ I 2×2
I3 H ω3′ s 2 + ω3′
2
A Servo
R Compensator
M Gains
O Discrete implementation of r
r η5 r
⎡ Vinvqd (k ) ⎤
N 1 states
I5 ⋅ I 2×2
I
ω5′ s 2 + ω5′ K2 ⎢ r ⎥
⎢ rI invqd (k ) ⎥
2
C
D
Stabilizing compensator ⎢ Vload (k ) ⎥
R
gains ⎢ r qd ⎥
I7 Discrete implementation of r ⎢ I loadqd (k ) ⎥
O states η7
r ⎢r ⎥
⎣⎢Vpwmqd (k − 1)⎦⎥
O 1
P ω7′ 2 ⋅ I 2×2
s 2 + ω7′
Figure 5.8 Output voltages controller using robust servomechanism controller with
harmonic drooping
137
CHAPTER 6
CONTROL
The proposed load sharing control described in the previous chapter will be tested
in a simulation environment that represents the experimental test-bed. For testing the
effectiveness of the proposed load sharing control, it is not necessary to simulate the
switching devices in the three-phase IGBT PWM inverter since the voltages and currents
at the point where the two UPS are being paralleled are not affected by the high
frequency switching of the PWM inverter. Therefore, in order to reduce the computing
time required to simulate the entire system, the three-phase IGBT PWM inverter will be
modeled as a voltage controlled source with delay of half the PWM cycle time ( Ts / 2 ).
This is a valid assumption assuming that the DC bus voltage can be maintained at
sufficient regulation and a proper space vector PWM switching technique is used for
138
simulation of power electronic circuit devices and components in Simulink, requiring
only interconnection of library components models for building the circuits for
simulations.
The main Simulink block diagram for the system is shown in Fig. 6.1. The two
PWM inverter systems are implemented by the blocks PWM Inverter 1 and PWM
Inverter 2. A 3-phase Load Bank block implements different load conditions which are
1: 2 x 4 kW balanced Resistive Load (can be modified for arbitrary parallel RLC load)
2: 2 x 5 kVA 0.8 pf lagging balanced load (can be modified for arbitrary parallel RLC
load)
4: Unbalanced Load (can be modified for arbitrary unbalanced parallel RLC load)
loading.
139
Figure 6.1 Simulink main diagram of the system
The wiring between the two PWM inverters are modeled by the impedances
ZwA1, ZwB1, ZwC1, ZwA2, ZwB2, and ZwC2 as shown in Fig. 6.1. The values of these
A global flag COMM_OK in the main diagram (Fig. 6.1) is used for indication of the
condition of the communication link of between the two units. COMM_OK=1 indicates
that the communication link is ok and therefore the two units can use the average power
140
for the load sharing control, COMM_OK=0 indicates otherwise. The powers of each unit
are communicated using the labels PQqd1 and PQqd2 each of which consists of
The control flag free_run=1 tells both inverter to run in free-running mode, while
free_run=0 tells them to synchronize to the bypass voltage modeled as a sine wave
The labels PQ1 and PQ2 consist of the per-phase real and reactive powers of each
inverter in ABC frame (Pa, Qa, Pb, Qb, Pc, Qc). The abc load currents of each inverter
are available as Iload1 and Iload2, and the abc load voltages and currents are given by
The PWM Inverter 1 and PWM Inverter 2 in the main diagram both contain the
block diagrams shown in Fig. 6.2 The power circuitry consisting of the Inverter, LC
Filter, Transformer, and Output Capacitor subsystems are implemented using the
141
Figure 6.2 PWM Inverter block
SimPowerSystem library blocks and their block diagrams are given in Fig. 6.3(a), 6.3(b),
and 6.3(c), respectively. The values of the inductors and capacitors are given by Matlab
workspace variables Linv, Cinv, and Cout initialized by the control5kva.m file. The
inverter subsystem consists of three ideal controlled voltage sources with half PWM
142
a) LC Filter b) Transformer
143
The Transformer is built from a SimpowerSystem standard Delta/Y linear
The inverter inductor filter currents (Iinv), inverter capacitor filter voltages (Vinv),
load voltages (Vload), and load currents (Iload) are measured and multiplexed along with
the bypass voltage into the signal label ACT_SIGNALS. These signals are multiplied by a
constant array Measurement Error Constants to model measurement errors that exist in
the system. For example the Measurement Error Constants can be initialized as: [0.99
1.01 0.99, 1.01 0.99 0.99, 0.999 1.001 0.999, 1.01 0.99 0.99, 1] which represents -1%,
+1%, -1% errors in the inverter phases currents measurements, +1%, -1%, -1% in the
capacitor voltages, -0.1%, +0.1%, -0.1% in the load voltages, and no error in the bypass
voltage measurements.
The measured voltages and current affected by the measurement errors are then
fed through a one-sampling-time (Tsamp=1/60/90) Zero-Order Hold block into the DSP
Output block that implements the control by a digital signal processor. The
communicated power quantities from the other unit are also fed to this block through the
144
6.1.3 DSP Output
Fig. 6.4 shows the implementation of the DSP Output block. The block performs
the load voltages and inverter current control as well as the load sharing control for
parallel operation. The block functionalities are broken down in to four subsystems as
The DQ Power Calculation subsystem calculates the output real and reactive
powers of each individual D and Q axis from the DQ load voltages and currents. The first
output of this block is a multiplexed signal labeled PQqd which consists of the power
quantities Pq, Qq, Pd, and Qd. The block also generates the 3rd, 5th, and 7th harmonic
currents for each DQ axis which are multiplexed into signals labeled Iq_harm and
Id_harm. Fig. 6.6 shows the implementation of this subsystem. The power quantities are
calculated using PSB library block Discrete Active & Reactive Power with sampling time
equals to Tsamp=1/60/90 and fundamental frequency of 60Hz. The harmonic currents are
computed using the PSB library block Discrete Fourier which gives a discrete Fourier
60Hz.
145
Figure 6.4 DSP Output block diagrams
146
Figure 6.5 DQ Transformations & States Calculations
147
Figure 6.6 DQ Power Calculation
148
6.1.4 Load Sharing Control
Fig. 6.7 shows the implementation of the Load Sharing Control block. The real
power sharing of the D and Q axis (equations 5.1 and 5.4) are implemented by the P
Sharing block shown in Fig. 6.8 (a), while the reactive power sharing control (Equations
5.2 and 5.5) are implemented in the Q sharing block Fig. 6.8 (b). The Reference Angle
Generation block generates the reference angle θ ref as discussed in Chapter 5. The two
blocks at the bottom of the diagrams computes the sine-waves references for the Q and D
axis based on the reference angle θ ref , phases and amplitude adjustments for the load
sharing control.
149
(a) P Sharing Block (a) Q Sharing Block
150
Fig. 6.9 shows the implementations of P (Fig. 6.9 a) and Q (Fig. 6.9 b) sharing
controls for each D and Q axis in Fig. 6.8. The diagrams implements the equations (5.1)
or (5.4) for P sharing control and (5.2) or (5.5) for Q sharing controls. The signal
COMM_OK defined in the Main Diagram is used to switch off the input to the
integrators. When COMM_OK=1 the power errors are used as the inputs to the
integrators, when COMM_OK=0 constants 0 are used instead. The sampling time of the
discrete time integrators are set at one line cycle period (1/60 sec). The control gains for
these blocks are initialized by Mask Subsystems of the P Sharing and Q Sharing blocks in
151
(a)Pq or Pd Sharing Control
152
Figure 6.10 Masked-subsystems of P-Sharing and Q-Sharing for initializing load sharing
control gains
The Reference Angle Generation block is implemented as shown in Fig. 6.11. The
top part implements the frequency locking mechanism of Fig. 5.5 for free running
operation, while the bottom part implements the phase-lock loop in Fig. 5.6. The signal
label free_run defined in the Main Diagram selects the input to the Discrete Time
Integrator4 which generates the reference angle. When free_run=1, the frequency
locking frequency output is selected, when free_run=0 the frequency is obtained from the
PLL output. The implementation of reference angle generation in Fig. 18 also ensures
that the continuity of the integrators states in the two mechanisms. During free_run=1,
the integrator in the PI controller of the PLL is held reset, while during free_run=0 the
Discrete Time Integrator5 is held reset. An initial value equal to the output frequency of
the other mechanism is given for each integrator when the reset signal changes from 1 to
0. These values are denoted by the labels df1 and df2 in Fig. 6.11
153
Figure 6.11 Reference Angle Generation implementation
The harmonic robust servo mechanism voltage control and sliding mode control
developed in this PhD research will be used for the control of each inverter.
the M-File for calculating the servo and sliding mode control gains and its listing is
provided in Appendix A. This m-file needs to be executed prior to running the Simulink
simulation.
is shown in Fig. 6.12. Control for each axis Q and D are implemented as Vq control and
154
Vd control blocks respectively. Each block has df3, df5, and df7 for harmonic drooping
control which are obtained from multiplying the Iq_harm and Id_harm signals calculated
Fig. 6.13 implements the Vq (Vd) control block. The Var Freq Harmonic Control
block implement equation (5.19) for harmonic servo compensator with drooping
included. In order to implement the harmonic drooping, the coefficients for matrices
Aci* (ωi′ ) and Bc* in equation (9) have to be computed for all possible values of ωi′ . It can
155
be shown that for fundamental frequency in the small range around 60 Hz, each
coeffiecient in the matrices Aci* (ωi′ ) and Bc* can be approximated by a linear function of
The following Matlab code (included in the Matlab script in Appendix A) can be used to
calculate the coefficients of matrices for Aci* (ωi′) and Bc* based on the above method. Fig.
6.14 shows the implementation of the Var Freq Harmonic Control block using this
method.
harmonic=[1 3 5 7];
Bh_star=[0;1];
for h=1:length(harmonic) % iterate through harmonics 1, 3, 5, 7
i=1;
for f=59.0:0.02:61 % for each harmonic compute the A and B matrix in the range
w=2*pi*f*harmonic(h);
Ch_star=[0 1; % define the analog compensator matrix
-w^2 0];
csysc=ss(Ch_star,Bh_star,eye(2),0);
[csysbc,T]=ssbal(csysc); % perform balanced realization
csysd=c2d(csysbc,Tsamp,'zoh'); % discretize the compensator
[Acon_dx,Bcon_dx,Ccon_dx,Dcon_dx]=ssdata(csysd);
A11(i)=Acon_dx(1,1); % store each coefficient in an array for polyfit
A12(i)=Acon_dx(1,2);
A21(i)=Acon_dx(2,1);
A22(i)=Acon_dx(2,2);
B1(i)=Bcon_dx(1)*300;
B2(i)=Bcon_dx(2)*300;
i=i+1;
end
f=59.0:0.02:61;
pA11{h} = polyfit(f,A11,1); % find the linear fit of each coefficient
pA12{h} = polyfit(f,A12,1);
pA21{h} = polyfit(f,A21,1);
pA22{h} = polyfit(f,A22,1);
pB1{h}=polyfit(f,B1,1);
pB2{h}=polyfit(f,B2,1);
pA11{h}(2)=pA11{h}(2)+pA11{h}(1)*60; % shift the input at 60 Hz
pA12{h}(2)=pA12{h}(2)+pA12{h}(1)*60;
pA21{h}(2)=pA21{h}(2)+pA21{h}(1)*60;
pA22{h}(2)=pA22{h}(2)+pA22{h}(1)*60;
pB1{h}(2)=pB1{h}(2)+pB1{h}(1)*60;
pB2{h}(2)=pB2{h}(2)+pB2{h}(1)*60;
end
156
Figure 6.13 Vq (Vd) Control Implementation
157
(a) Var Freq Harmonic Control
158
The current controller implementation is shown in Fig. 6.15. The control gains are
159
6.2 Simulations Results
hypothetical components mismatches among the units will be assumed based on the
worst case tolerances of the components used. It is assumed that the magnetic
components all have ± 10% tolerance while capacitors have ± 6% tolerance. The
mismatches are introduced among the units to create worst case mismatches, including
unbalanced effect among phases within one unit. Table 6.1 shows the assumed
Unit 1 Unit2
LinvA +10% -10%
LinvB +10% -10%
LinvC -10% +10%
CinvAB -6% +6%
CinvBC -6% +6%
CinvCA +6% -6%
CoutAB -6% +6%
CoutBC -6% +6%
CoutCA +6% -6%
Xfmr Leakage Primary -10% +10%
Xfmr Leakage Secondary -10% +10%
ZwA 5.0+j5.0 mOhm 2.0+j2.0 mOhm
ZwB 4.8+j4.8 mOhm 2.2+j2.2 mOhm
ZwC 5.2+j5.2 mOhm 1.8+j1.8 mOhm
160
Assuming a ± 0.1% measurement error in each of the signals used by the control
implementation, the following measurement errors constants have been used for the
simulations (see section 6.1.2 for definition of the Measurement Error Constants)
[0.999 1.001 0.999 , 1.001 0.999 0.999 , 1.001 1.001 0.999 , 1.001 0.999 0.999 , 1]
[1.001 0.999 1.001 , 0.999 1.001 1.001 , 0.999 0.999 1.001 , 0.999 1.001 1.001 , 1]
The control gains for the load sharing have been selected as follows. Note that the load
P Sharing Control:
Q Sharing Control:
These gains selections give approximately 1 degree span of phase adjustment from no-
load to full load (4 kW= 5 x 0.8) and less than 1V span of amplitude voltage adjustment
for zero reactive power to full (5 x 0.6 kVAR = 3 KVAR). The offsets P0 and Q0 shift
the droop curve halfway in between no-load and full load. This cuts the total phase and
amplitude into half of the overall span for the range of operation from no-load to full
load. Clearly, the selection of the phase control gains above still ensures proper phase
synchronization for bypass operation. The amplitude adjustment is also considerably less
161
than 1% of the nominal voltages amplitude ( 120 2V = 169.7V ) , still ensuring good
Proportional and Integral Gains for the PI controller in the Phase Lock Loop:
proposed control are investigated for different linear and non-linear loads. To obtain the
results, the corresponding load was selected and configured in the Load Bank block and
the simulations were run for 5 seconds, at the end of which the results were collected.
The global flags free_run and COMM_OK are both set to 1, and the gains of the free
running mode frequency locking are also set to 0, so it was assumed that both units were
Table 6.2, 6.3, and 6.4 summarizes the active and reactive power of each inverter
for balanced linear load with resistive, balanced inductive 0.8 pf, and capacitive 0.9 pf,
with the corresponding steady state load currents shown in Fig. 6.16, 6.17, and 6.18
respectively. It can be seen that for each case the load is shared by the inverters within
1% accuracy.
162
Table 6.5 sunmarizes the result for linear single-phase load, with the load current
waveform in Fig. 6.19. For this case the load power is shared by each inverter within
approximately 3%, which is slightly worst than in the balanced load cases. This can be
expected since the 0-component of the load current flowing due to the unbalanced load
Table 6.6 and Fig. 6.20 show the simulation results for balanced harmonic load. It
can be seen that the load power is shared by the inverters to within less than 1%. Also
from Fig. 6.20 it can be seen that the harmonic load current is shared very well by each
inverter, verifying the effectiveness of the proposed harmonic drooping technique. For a
comparison, Fig. 6.21 shows a result of the simulation when the harmonic drooping
mechanism is disabled. It can be seen in this case the load current is not shared by each
163
Unit 1 Unit 2 % Difference
from Average
Pa (watt) 1338.12 1339.5 0.11%
Pb (watt) 1340.08 1338.30 0.13%
Pc (watt) 1338.76 1339.53 0.06%
Ptot (watt) 4017.9 4017.3 0.01%
Qa (var) -5.76 -4.25 N/A
Qb (var) -5.60 -2.00 N/A
Qc (var) -1.26 -6.35 N/A
Qtot (var) -12.62 -12.6 N/A
Load Volt Van=120.32 Vbn=120.31 Vcn=120.31
(V LN rms)
Table 6.2 Steady state results: Resistive Balanced Load (2 x 0.8 x 5.0 kW= 8.0 kW)
Figure 6.16 Output Currents of Unit 1 and Unit 2 under linear balanced resistive load
164
Unit 1 Unit 2 % Difference
from Average
Pa (watt) 1322.54 1323.50 0.07%
Pb (watt) 1325.08 1321.72 0.25%
Pc (watt) 1322.11 1324.61 0.19%
Ptot (watt) 3969.7 3969.8 0.01%
Qa (var) 984.86 988.31 0.35%
Qb (var) 987.31 988.18 0.09%
Qc (var) 989.87 985.57 0.44%
Qtot (var) 2962.0 2962.1 0.01%
Load Volt Van=119.59 Vbn=119.59 Vcn=119.58
(V LN rms)
Table 6.3 Steady state results: 0.8 lagging Balanced Load (2 x 5kVA 0.8 pf lag)
Figure 6.17 Output Currents of Unit 1 and Unit 2 under 0.8 lagging balanced load
165
Unit 1 Unit 2 % Difference
from Average
Pa (watt) 1518.81 1520.79 0.13%
Pb (watt) 1520.69 1519.68 0.07%
Pc (watt) 1520.38 1519.89 0.03%
Ptot (watt) 4559.9 4560.4 0.01%
Qa (var) -740.25 -740.17 0.01%
Qb (var) -741.95 -736.00 0.81%
Qc (var) -735.95 -741.96 0.44%
Qtot (var) 2218.2 2218.1 0.01%
Load Volt Van=119.59 Vbn=119.59 Vcn=119.58
(V LN rms)
Table 6.4 Steady state results: 0.9 leading Balanced Load (2 x 5kVA 0.9 pf lead)
Figure 6.18 Output Currents of Unit 1 and Unit 2 under 0.9 lagging balanced load
166
Unit 1 Unit 2 % Difference
from Average
Pa (watt) 1343.65 1293.78 3.78%
Pb (watt) -12.74 14.21 0.07%
Pc (watt) -13.55 15.03 0.03%
Ptot (watt) 1317.40 1322.2 0.367%
Qa (var) -5.83 -4.00 N/A
Qb (var) -29.69 18.73 N/A
Qc (var) 18.27 -29.26 N/A
Qtot (var) -17.25 -14.53 N/A
Load Volt Van=119.32 Vbn=120.76 Vcn=120.95
(V LN rms)
Table 6.5 Steady state results: Single-phase Loaded (2 x 4.0 kW/3=2.666 kW on phase A
Figure 6.19 Output Currents of Unit 1 and Unit 2 under single-phase resistive load
167
Unit 1 Unit 2 % Difference
from Average
Pa (watt) 1313.00 1314.03 0.08%
Pb (watt) 1315.53 1312.56 0.23%
Pc (watt) 1312.90 1315.02 0.16%
Ptot (watt) 3941.4 3941.6 0.01%
Qa (var) 772.35 775.40 0.39%
Qb (var) 774.05 775.50 0.19%
Qc (var) 777.12 772.63 0.58%
Qtot (var) 2323.5 2.3235 0.00%
Load Volt Van=119.93 Vbn=119.93 Vcn=119.93
(V LN rms) %THD = 2.77, 2.84, 2.84
Table 6.6 Steady state results: Balanced non-linear load (2x5 KVA, three single-phase
rectifier)
Figure 6.20 Output Currents of Unit 1 and Unit 2 with harmonic drooping enabled
168
Figure 6.21 Output Currents of Unit 1 and Unit 2 with harmonic drooping disabled (gains
set to zero)
In this section, simulation results are given for transient performance of the proposed
load sharing algorithm. The load was initially set to 8 kW balanced resistive load, and
resistive load.
From the results (Fig. 6.22 to Fig. 6.28) it can be seen that proper load sharing
was maintained during all load transients, and good load voltage regulations were also
169
Figure 6.22 DQ real (top) and reactive (bottom) powers of Unit 1 and Unit 2
170
(a)Phase Adjustment Unit 1 (a) Phase Adjustment Unit 2
171
Figure 6.26 Load transient from Balanced resistive to no-load
172
Figure 6.28 Load transient from balanced 0.8 pf lagging to single-phase resistive
To study the bypass synchronized mode and free-running mode operation the following
The bypass voltage (Main Diagram) frequency was set to 60.5 Hz, and the units were
free_run flag was set to 1 simulating the units lost its bypass voltage and entered free-run
mode. Prior to t=3.5 sec, the bypass voltage recovered and the units had to try to
resynchronize to bypass again. During all this time, a constant balanced resistive load of
8 kW was used.
173
Fig. 6.29 shows the bypass and units phase error obtained from the input to PI
controller in the PLL. Fig. 6.30 shows the output frequency of the UPS. It can be seen
that initially the units are phase and frequency locked to the bypass voltage at 60.5 Hz.
The frequency was then automatically set to 60 Hz when the units free-ran. When the
free_run flag was cleared at 3.5 sec, the units tried to resynchronize to bypass again, and
174
Fig. 6.31 and 6.32 show the units DQ powers and output currents during the
modes transitions above. It can be seen that load sharing was properly maintained during
the mode transitions. Only slight transient occurred during the time the units tried to
175
Mismatched Clock Frequency during Free-Running Mode
During free-run mode, the output voltage frequency of each UPS is determined by
the sampling time of each DSP in the unit which in turn is determined by the clock
frequency of the crystal used. The crystal frequency error is usually in the range of
hundreds of ppm (parts per million). This small mismatch in the crystals used will be
reflected as output voltage frequency mismatches in the units. To simulate this effect, the
sampling time of each unit must be changed to reflect this clock frequency mismatch.
However, the same effect can be observed by simply changing the constant used in the
Reference Angle Generation block to different values as shown in Fig. 6.33. This
simulated Unit 1 unknowingly generated output voltage at frequency 60.001 and unit 2
To study the effect of this mismatch, the frequency locking mechanism was first
disabled by setting the gains to zero as shown in Fig. 6.33. Fig. 6.34 and 6.35 show the
results of the simulations when the frequency locking control was disabled. It can be seen
that the units failed to maintain proper load sharing and the phase angle adjustment
Fig. 6.36, 6.37, and 6.38 show the results of simulations when the frequency
locking control was enabled (gains were set to 100). From Fig. 6.36, it can be seen that
the frequency locking control adjusted the frequency of each unit such that their
difference was compensated. This resulted in proper power load sharing as shown in Fig.
6.37 and prevented the phase adjustments to change continuously in Fig. 6.38.
176
Unit 1
Unit 2
Figure 6.33 Modifying the Reference Angle Generator Block for simulating mismatched
crystal clock frequencies
177
Figure 6.34 DQ real (top) and reactive (bottom) powers during free-run with mismatched
frequency and frequency locking disabled
Figure 6.35 Units phases adjustments during free-run with mismatched frequency and
frequency locking disabled
178
(a) Unit 1
(b) Unit 2
Figure 6.36 Frequency locking adjustments during free-run with mismatched frequency
179
Figure 6.37 DQ real (top) and reactive (bottom) powers during free-run with mismatched
frequency and frequency locking enabled
Figure 6.38 Units phases adjustments during free-run with mismatched frequency and
frequency locking enabled
180
6.2.5 Load Sharing with Unequal Proportion
In the following simulations, the effectiveness of the proposed control for load
sharing of unlike units’ sizes is investigated. The following simulations assume that it
was desired to have unit 1 handle only 1/3 of the total load while unit 2 handle 2/3 of the
total load. In other words, this simulated conditions where unit 1 is half the size of unit 2.
Fig 6.39 shows the changes that need to be made in the Load Sharing Control Block in
Figure 6.40 shows the results of the unequal proportion load sharing at steady
state for two different types of loads. It can be seen that the proposed load sharing control
effectively made the units share the load based on the desired proportion. The transient
performance was studied by changing one load to the other during simulation, and the
results are shown in Fig 6.41, 6.42 and 6.43. From these results, it can be seen that proper
181
Unit 1 Unit 2
Figure 6.39 Setting up the control gains for unlike units sharing simulations
182
0.8 pf lagging Balanced Load (2 x 3 KVA 0.8 Resistive Balanced Load (2 x 3KW x 0.8 = 4.8 kW)
lag)
Figure 6.40 Steady state performance of load sharing with unequal proportion
183
Figure 6.41 DQ real (top) and reactive (bottom) powers of Unit 1 and Unit 2 for unequal
load sharing proportion
184
(a)Phase Adjustment Unit 1 (a) Phase Adjustment Unit 2
Figure 6.42 Phases and Amplitude Adjustments of Unit 1 and Unit 2 for unequal load
sharing proportion
Figure 6.43 Units Output Currents during transient for unequal load sharing proportion
185
6.2.6 Lost of Inter-Unit Communication
communication when the two units were running in parallel. The system load was
condition, the global flag COMM_OK was changed from 1 to 0 at t=4 sec as shown in
Fig 6.44. After loosing the communication the units ran at the same load for a while, until
7.515, the load was changed again back to the balanced 8 kW resistive.
Figure 6.45, 6.46, 6.47, and 6.48 show the results of the simulation. It can be seen
that even after the inter-unit communication was lost, the two units maintain the proper
load sharing. This verifies that the dependency of the proposed load sharing technique on
the communication is not critical. The communication is used only to improve the
performance of the load sharing under components mismatches and the present of
measurements errors.
Figure 6.44 COMM_OK flag was cleared at t=4 sec to simulate lost of communication
186
Figure 6.45 Units output DQ real (top) and reactive (bottom) powers when
communication was lost at t=4 sec
Figure 6.46 Units’ output currents when communication was lost at t= 4 sec
187
Figure 6.47 Units output currents transient from 8kW to 4 kW after communication was
lost
Figure 6.48 Units output currents transient from 4kW to 8 kW after communication was
lost
188
6.3 Experimental Test Bed
To verify the effectiveness of the parallel inverters control, a second 5 kVA test
bed unit was built at the Mechatronics Laboratory at The Ohio State University. The two
units are interconnected as shown in Figure 6.49. Communication between the two units
is accomplished through an RS232 serial communication link between the two Inverter
189
⎡Vinv ab ⎤ ⎡Vload an ⎤
⎡Vbyp ab ⎤ ⎡Vpwm ab ⎤ ⎢Vload ⎥
⎢Vpwm ⎥ ⎢Vinv ⎥
⎢Vbyp ⎥
⎢ bc ⎥ ⎢ bc ⎥ ⎢ bn ⎥
⎢ bc ⎥
⎣⎢Vpwm ca ⎦⎥ ⎣⎢Vinv ca ⎦⎥ ⎣⎢Vload cn ⎦⎥
⎣⎢Vbyp ca ⎦⎥
2.0mH
Iina Iinv a Iload a
Iload b
Iinc Iinv c
Iload c
⎡Vinv ab ⎤
⎡Vload an ⎤
⎡Vbyp ab ⎤ ⎡Vpwm ab ⎤
190
⎢Vpwm ⎥ ⎢Vinv ⎥ ⎢Vload ⎥
⎢Vbyp ⎥
⎢ bc ⎥ ⎢ bc ⎥ ⎢ bn ⎥
⎢ bc ⎥
⎢⎣Vbyp ca ⎥⎦ ⎣⎢Vpwm ca ⎦⎥ ⎣⎢Vinv ca ⎦⎥ ⎣⎢Vload cn ⎦⎥
2.0mH
Iload b
Iinc Iinv c
Iload c
Fig.6.50, 6.51, and 6.52 show the steady state performance of the load sharing
control under various loads conditions. In Fig. 6.50 and 6.51, each trace box shows the
per-phase load voltage and the two units’ output currents. For Fig. 6.52 only the two
units’ output currents are shown. From top to bottom the quantities apply for phase a, b,
and c respectively. In Fig. 6.50, balanced three-phase resistive load is applied, while in
Fig. 6.51 resistive loads are applied with unbalanced loading (phase a being unloaded).
Fig. 6.52 shows the load sharing performance under non-linear load in the forms of three
single-phase diode rectifiers with RC loads. It can be seen that the load currents are well
shared between the two units, which verifies the effectiveness of the harmonic sharing
droop control.
Fig. 6.53 and Fig. 6.54 show the performance of the load sharing control under
changing load conditions. In Fig. 6.53, a 100% to 0% load change and 0% to 100% load
change are shown on the same plot. Fig. 6.54 shows more detailed view of the 0 to 100%
load change. It can be seen the load sharing was well maintained under both load
transients.
Finally, in Fig. 6.55 unit 2 is brought on-line while unit 1 was supplying different
types of loads on each of its phases. Phase a and b were loaded with resistive loads while
phase c was loaded with a single-phase diode rectifier load. It can be seen that within a
few cycles the different loads are shared by the two units, verifying the effectiveness of
the control under reconnection transients and its robustness towards different per-phase
loading conditions.
191
Figure 6.50 Steady state load sharing performance for balanced resistive load
Figure 6.51 Steady state load sharing performance for unbalanced resistive load
192
Figure 6.52 Steady state load sharing performance for balanced non-linear rectifier load.
Figure 6.53 Dynamic performance: l00% to 0% and 0% to 100% load changes. Each
trace box shows the two units output currents for phase a, b, and c (from top to bottom)
193
Figure 6.54 Dynamic performance 0% to 100% load acquire (zoom in of Fig. 6.53).
Figure 6.55 Unit 2 is reconnected with different load applied on each phase. Phase a and
phase b: linear resistive load. Phase c: single-phase non-linear load.
194
CHAPTER 7
7.1 Conclusions
The first part of this PhD research outlines the development of a digital control
strategy for a single three-phase PWM inverter used in UPS applications. The control
strategy combines the perfect RSP controller for low THD output voltages regulation and
the discrete sliding mode current controller for fast over-current protection. The voltage
controller was developed by including the dynamic of the current controller into the plant
with the computation delay of the DSP accounted for. It was shown that by including the
harmonic frequency mode to be eliminated into the perfect RSP controller, superior low
The stability robustness of the proposed control system was verified using
parameters errors and linear load variations. It was shown that the scalar weighting in the
optimal control cost function provides a way of tuning the transient performance of the
195
perturbations. The presented simulation and experimental results verified the
effectiveness of the proposed control strategy both in providing low THD output voltages
The second part of this research describes a novel load sharing method that can
properly control the load-sharing such as the real, reactive, and harmonic powers for
of droop method, average power control method, and harmonic droop sharing control
using only low bandwidth power information exchanged between each UPS unit. In
particular, the average power control method significantly reduced the sensitivity about
voltage and current measurement errors, the droop method allowed proper load sharing
when the communication between units was lost, and the harmonic control loop
guaranteed harmonic power sharing under nonlinear loads. Small signal stability analysis
was used to analyze the stability of the combined average power and droop method.
From the simulation and experimental results presented, it was shown that the
proposed parallel control method is effective in maintaining proper load sharing under
different types of loads including linear and non-linear loads in steady state and transient
conditions. It was also shown that when communication between units is lost, proper load
sharing can still be maintained. In this case, a control strategy is introduced for
maintaining the output voltage phase angle and frequency synchronization between
inverters in the event that common synchronization reference voltage is lost (inverters are
in free-running mode).
196
7.2 Future Work
Within the scope of the digital control for PWM inverters in UPS applications,
there are other areas that may be further investigated in order to improve the results
presented in this research. In this research, for example, standard space vector PWM
technique [126] has been used to generate the PWM switching patterns for the IGBTs at
5400 Hz. This switching frequency is common for low to medium power inverters. For
very high power applications, lower PWM switching frequency is desirable for
minimizing heat due to the losses generated in the devices and increasing overall unit
efficiency. Many PWM switching strategies optimized for low switching frequency
and/or IGBT losses have been investigated by the researchers over the years [127-132].
This topic, however, still presents vast research opportunities for improving the
performance of the PWM inverters in general. Also of interest is the effect of the lower
control sampling time to the control bandwidth and hence transient and steady state
Another area closely related to this research is the distributed generation system.
This area has become of high interest recently due to the tendency of smaller and more
distributed power generations using alternative energy like fuel cell, solar cell, etc.[132-
136]. The results from this PhD research are directly applicable to this application, since
one of the building blocks of a distributed generation system is a PWM inverter as the
one described in this research. In particular, the requirements of the PWM inverters
developed in this research are the same as those used in a distributed generation system
under a stand alone AC supply mode of operation. The results of this PhD research,
197
however, need to be extended for other modes of operation in a distributed generation
system such as parallel operation with utility (grid interconnected mode) and transitional
mode operations.
Finally, the idea developed in this research in achieving output voltage with low
THD using the robust servo mechanism maybe used for other power converters
applications where low THD of AC signals are required. For example, the idea may be
applied to PWM rectifiers or active filters systems used in UPS applications where low
198
BIBLIOGRAPHY
[2] J. Perkinson, “UPS systems: a review,” 1988 Third Annual IEEE, Conference
Proceedings Applied Power Electronics Conference and Exposition, pp. 151 – 154, 1-5
Feb. 1988
[4] Krishnan, R.; Srinivasan, S., “Topologies for uninterruptible power supplies,”
Conference Proceedings IEEE International Symposium on Industrial Electronics,
Budapest, pp. 122 – 127, 1-3 June 1993
199
[10] P.D. Ziogas, “Optimum voltage and harmonic control of PWM techniques for
three phase static UPS systems,” IEEE Trans. Ind. Appl., vol IA-16, no. 4, pp. 542-546,
July/Aug. 1980.
[11] B.K. Bose and H.A. Sutherland, “A high performance pulse-width modulator for
an inverter-fed drive system using a microcomputer,” in Conf. Rec. 1982 Ann. Meet. Ind.
Appl. Soc., pp 847-853.
[12] S.R. Bowes, “Advanced regular-sampled PWM control techniques for drives and
static power converters,” IEEE Transactions on Industrial Electronics, Vol. 42, Issue: 4,
pp. :367 – 373, Aug. 1995
[13] S.R. Bowes, “Advanced regular-sampled PWM control techniques for drives and
static power converters,” Proceedings of the IECON '93, pp. 662 - 669 vol.2, 15-19 Nov.
1993
[14] S.R. Bowes, “Efficient microprocessor real-time PWM drive control using
regular-sampled harmonic minimisation techniques,” Conference Proceedings, ISIE'93 –
Budapest, pp. 211 – 218, 1-3 June 1993
[18] H.S. Patel and R.G. Hoft, “Generalized technique of harmonic elimination and
voltage control in thyristor inverter, Part I,” IEEE Trans. Ind. Appl., vol IA-9, no.3, pp
310-317, Mar./Apr.1973; and “Part II,” IEEE Trans. Ind. Appl., vol. IA-10, no. 5, pp 666-
673, Sept./Oct. 1973.
[19] F.G. Turnbull, “Selected harmonic reduction in static dc-ac inverters, “ IEEE
Trans. Commun. Elec., vol. 83, pp 374-378, July 1964.
200
[21] G.S. Buja, “Optimum output waveforms in PWM inverters,” IEEE Trans. Ind.
Appl., vol. IA-16, no. 6, pp. 830-836, Nov./Dec. 1980.
[22] A. Kernik, D.L. Stechshulte, and D.W. Shireman, “Static Inverter with
synchronous output waveform synthesized by time-optimal response feedback,” IEEE
Trans. IECI, vol. IECI-24, no. 4, pp 297-305, 1977
[23] A. Kawamura and R.G. Hoft, “Instantaneous feedback controlled PWM inverter
with adaptive hysteresis,” IEEE Trans. Ind. Appl., vol IA-20, no. 4, pp. 769-775,
July/Aug. 1984
[24] P. Zoigas, “Delta modulation technique in static PWM inverters,” IEEE Trans.
Ind. Appl., vol. IA-17, pp. 289-295, 1981
[26] K.P Gokhale, A. Kawamura, and R.G. Hoft, “Dead beat microprocessor control of
PWM inverter for sinusoidal output waveform synthesis,” in Conf. Rec. IEEE Power
Elect. Spec. Conf., 1985, pp. 28-36.
[27] A. Kawamura, T. Haneyoshi, and R.G. Hoft, “Deadbeat controlled PWM inverter
with parameter estimation using only voltage sensor,” Conf. Rec. IEEE Power Elec. Spec.
Conf., 1986
[29] J.-S. Cho, S.-Y. Lee, H.-S. Mok, G.-H. Choe, “Analysis and design of modified
deadbeat controller for 3-phase uninterruptible power supply,” Proceedings of the 1999
IEEE Power Electronics and Drive Systems, Vol. 2, pp. 1003 – 1009, 27-29 July 1999
201
[32] P. Mattavelli, “A modified dead-beat control for UPS using disturbance
observers,” Power Electronics Specialists Conference, 2002, Vol. 4, pp. 1618 – 1623, 23-
27 June 2002
[33] J.-S. Cho, S.-Y. Lee, H.-S. Mok, G.-H. Choe, “Modified deadbeat digital
controller for UPS with 3-phase PWM inverter,” Conference Record Industry
Applications Conference, 1999, Vol. 4, pp. 2208 – 2215, 3-7 Oct. 1999
[35] E.G. Carati, , C.M. Richter, H.A. Grundling, “A robust high performance three-
phase uninterruptible power supply”, Proceedings of the 2000 IEEE International
Conference on Control Applications, pp. 896 – 901, 25-27 Sept. 2000
[36] C.M. Richter, E.G. Carati, H.A. Grundling, “Multivariable model reference
adaptive control for a three-phase uninterruptible power supply,” Proceedings of the 2000
IEEE International Conference on Control Applications, pp. 914 – 917, 25-27 Sept. 2000
[37] M.J. Ryan, R.D. Lorenz , “A high performance sine wave inverter controller with
capacitor current feedback and “back-EMF” decoupling,” Power Electronics Specialists
Conference 1995 Record, Vol. 1 , pp. 507 – 513, 18-22 June 1995
[38] S. Sivakumar, K. Natarajan, “Kalman filter based high speed measurement and
control of AC voltages for UPS applications,” Power Electronics Specialists Conference,
1993 Record, pp. 907 – 912, 20-24 June 1993
[40] T.J. Liang, F.J. Chang, J.W. Chen, “Design of PWM inverter via servo control
technique,” IECON 97, Vol. 2, pp. 704 – 709, 9-14 Nov. 1997
[41] L.-Q. Yi; Y.-X. Dai, “Digital servo control techniques based sine inverter design,”
International Conference on Machine Learning and Cybernetics 2003, Vol. 2 , pp. 888 –
892, 2-5 Nov. 2003
[42] H.S. Khalil, S.A. Shehata, M.A. El-Bakry, A.M. Metwaly, “Modeling and
simulating an UPS system with feedforward and feedback control,” Proceedings of the
36th SICE Annual Conference, pp. 1173 – 1178, 29-31 July 1997
202
[43] H.-J. Cha, S.-S. Kim, M.-G. Kang, Y.-H. Chung, “Real-time digital control of
PWM inverter with PI compensator for uninterruptible power supply,” IECON '90, vol.2,
pp. 1124 – 1128, 27-30 Nov. 1990
[44] Poh Chiang Loh, M.J. Newman, D.N. Zmood, D.G. Holmes, “A comparative
analysis of multiloop voltage regulation strategies for single and three-phase UPS
systems,” IEEE Transactions on Power Electronics, Vol. 18, Issue 5, pp. 1176 – 1185,
Sept. 2003
[45] N.M. Abdel-Rahim, J.E. Quaicoe, “Analysis and design of a multiple feedback
loop control strategy for single-phase voltage-source UPS inverters,” IEEE Transactions
on Power Electronics, Vol. 11, No. 4, pp. 532 – 541, July 1996
[46] A. von Jouanne, P.N. Enjeti, D.J. Lucas, “DSP control of high-power UPS
systems feeding nonlinear loads,” IEEE Transactions on Industrial Electronics, Vol. 43,
No. 1, pp. 121 – 125, Feb. 1996
[47] Y. Ito, S. Kawauchi, “Microprocessor based robust digital control for UPS with
three-phase PWM inverter”, IEEE Transactions on Power Electronics, Vol. 10 , No. 2,
pp. 196 – 204, March 1995
[48] V.F. Montagner, P.L.D. Peres, , “Robust state feedback control applied to a UPS
system,” IECON '03, Vol. 3, pp. 2245 – 2250, 2-6 Nov. 2003
[50] S. Buso, ”Robust control of single phase UPS,” APEC '97 Conference
Proceedings 1997., Vol. 2 , pp. 825 – 831, 23-27 Feb. 1997
[51] H.A. Grundling, E.G. Carati, J.R. Pinheiro, “A robust model reference adaptive
controller for UPS applications,” IECON 97, Volume: 2 , 9-14 Nov. 1997 Pages:901 -
905 vol.2
[52] Yong Kang, Qin Zhang, Jian Chen, “Analyses and implementation of a UPS
inverter with instantaneous feedback control and harmonic injection,” PESC '94 Record,
pp. 1462 - 1465, 20-25 June 1994
203
[54] A. Kawamura, K. Ishihara, “Real time digital feedback control of three phase
PWM inverter with quick transient response suitable for uninterruptible power supply,”
Conference Record of the 1988 IEEE Industry Applications Society Annual Meeting,
vol.1, pp. 728 - 734, 2-7 Oct. 1988
[58] S.-L. Jung, Y.-Y. Tzou, “Discrete feedforward sliding mode control of a PWM
inverter for sinusoidal output waveform synthesis,” PESC '94 Record, vol.1, pp. 552 -
559, 20-25 June 1994
[59] S.-L Jung, Y.-Y. Tzou, “Sliding mode control of a closed-loop regulated PWM
inverter under large load variations,” PESC '93 Record, pp. 616 – 622, 20-24 June 1993
[60] T.-L. Tai, J.-S. Chen, “UPS inverter design using discrete-time sliding-mode
control scheme,” IEEE Transactions on Industrial Electronics, Vol. 49, No. 1, pp. 67 –
75, Feb. 2002
[61] S.-L. Jung and Y.-Y. Tzou, “Discrete sliding-mode control of a PWM inverter for
sinusoidal output waveform synthesis with optimal sliding curve,” IEEE Transactions on
Power Electronics, Vol. 11, No. 4, pp. 567 – 577, July 1996
[62] J.M Ruiz, S. Lorenzo, I. Lobo, J. Amigo, “Minimal UPS structure with sliding
mode control and adaptive hysteresis band,” IECON '90., vol.2, pp. 1063 - 1067, 27-30
Nov. 1990
204
[65] S.J. Chiang, T.L. Tai, T.S. Lee, “Variable structure control of UPS inverters,” IEE
Proceedings-Electric Power Applications, Vol. 145, No. 6, pp. 559 – 567, Nov. 1998
[66] B.-R. Lin and C. Hua, “Uninterruptible power supply with fuzzy logic approach,”
Proceedings of the IECON '93, vol.2, pp. 1123 – 1128, 15-19 Nov. 1993
[67] F.-Y. Hsu and L.-C. Fu, “Adaptive fuzzy control for uninterruptible power supply
with three-phase PWM inverter,” Proceedings of the 1996 Asian Fuzzy Systems
Symposium, pp. 188 – 193, 11-14 Dec. 1996
[68] Yu Qin and Shanshan Du, “Control of single phase power factor pre-regulator for
an on-line uninterruptible power supply using fuzzy logic inference,” APEC '96
Conference Proceedings, Vol. 2 , pp. 699 – 702, 3-7 March 1996
[69] B.R. Lin, “Analysis of neural and fuzzy-power electronic control,” IEE
Proceedings-Science, Measurement and Technology, Vol. 144, No. 1, pp. 25 – 33, Jan.
1997
[71] Yu Qin and Shanshan Du, “Comparison of fuzzy logic and digital PI control of
single phase power factor pre-regulator for an on-line UPS,” Proceedings of the 1996
IEEE IECON, Vol. 3, pp. 1796 – 180, 15-10 Aug. 1996
[72] Yu Qin and Shanshan Du, “Fuzzy logic and digital PI control of single phase
power factor preregulator for an online UPS-a comparative study,” Proceedings of
Industrial Technology Conference 1996 (ICIT '96), pp. 103 – 107, 2-6 Dec. 1996
[73] H. Osterholtz, “Simple fuzzy control of a PWM inverter for a UPS system,”
INTELEC '95, pp. 565 – 570, 29 Oct.-1 Nov. 1995
[74] Y.-Y. Tzou, L.-H. Ho; R.-S. Ou, “Fuzzy control of a closed-loop regulated PWM
inverter under large load variations,” Proceedings of the IECON '93, vol.1, pp. 267 - 272,
15-19 Nov. 1993
[75] B.-R Lin and R.G. Hoft, ”Real-time digital control of PWM inverter with fuzzy
logic compensator for nonlinear loads,” Industry Applications Society Annual Meeting,
1993 Conference Record, vol.2, pp. 862 – 869, 2-8 Oct. 1993
Repetitive control techniques
205
[76] T. Haneyoshi, A Kawamura, and R.G. Hoft, “Waveform compensation of PWM
inverter with cyclic fluctuating loads, “ IEEE Trans. Ind. Appl., vol. 24, pp. 582-589,
1988.
[77] Y.-Y. Tzou, S.-L. Jung, and H.-C. Yeh, “Adaptive repetitive control of PWM
inverters for very low THD AC-voltage regulation with unknown loads,” IEEE
Transactions on Power Electronics , Vol. 14 , No. 5, pp. 973 – 981, Sept. 1999.
[78] Y.-Y. Tzou, R.-S. Ou, S.-L. Jung, M.-Y. Chang, “High-performance
programmable AC power source with low harmonic distortion using DSP-based
repetitive control technique,” IEEE Transactions on Power Electronics, Vol. 12 , No. 4 ,
July 1997 Pages:715 - 725
[79] Y.-Y. Tzou and H.-C. Yeh, “DSP-based adaptive repetitive control of a PWM
inverter for UPS with very low harmonic distortion,” Proceedings of the 1996 IEEE
IECON, Vol. 2, pp. 1122 – 1127, 5-10 Aug. 1996
[80] S.-L. Jung, H.-S. Huang, Y.-Y. Tzou, “A three-phase PWM AC-DC converter
with low switching frequency and high power factor using DSP-based repetitive control
technique,” PESC 98 Record, Vol. 1 , pp. 517 – 523, 17-22 May 1998
[81] S.-C. Yeh and Y.-Y. Tzou, “Adaptive repetitive control of a PWM inverter for
AC voltage regulation with low harmonic distortion,” PESC '95 Record, Vol. 1 , pp. 157
– 163, 18-22 June 1995
[82] V.F. Montagner, E.G. Carati, H.A. Grundling, “An adaptive linear quadratic
regulator with repetitive controller applied to uninterruptible power supplies,” Industry
Applications Conference 2000, Conference Record, Vol. 4, pp. 2231 – 2236, 8-12 Oct.
2000
[83] V.F. Montagner, E.G. Carati, H.A.Grundling , “Design and analysis of a linear
quadratic regulator with repetitive controller for AC power supplies,” 2003 IEEE
International Symposium on Industrial Electronics , Vol. 1, pp. 544 – 549, 9-11 June
2003
[85] H.A. Grundling, E.G. Carati, J.R. Pinheiro, , “Analysis and implementation of a
modified robust model reference adaptive control with repetitive controller for UPS
applications,” Proceedings of IECON '98, Vol. 1, pp. 391 – 395, 31 Aug.-4 Sept. 1998
206
[86] B.A. Francis, and W.M. Wonham, , “The internal mode principle for linear
multivariable regulators”, Appl. Math. Opt., vol. 2, pp. 170-194.
[89] J. Holtz, W. Lotzkat, and K.H. Werner, “ A high power multi transistor-inverter
uninterruptible power supply system,” IEEE Trans. Power Electronics, 1988, vol. 3, pp
278-285.
[90] J. Holtz, K.H. Werner, “ Multi-Inverter UPS System with Redundant Load
Sharing Control,” IEEE Trans. Industrial Electronics, vol. 37, no. 6, December 1990, pp
506-513
[91] J-F Chen, and C-L Chu, “ Combination Voltage-Controlled and Current-
Controlled PWM Inverters for UPS Parallel Operation,” IEEE Trans. Power Elect., vol.
10, no.5, September 1995, pp. 547-558.
[93] Yadong Liu, Yan Xing, Lipei Huang, K. Hirachi, , “A novel distributed control
scheme for the parallel operation of digital controlled UPS,” IECON '03, Vol. 1, pp. 668
– 672, 2-6 Nov. 2003
[94] H. Van Der Broeck and U. Boeke, “A simple method for parallel operation of
inverters,” Telecommunications Energy Conference 1998, pp. 143 – 150, 4-8 Oct. 1998
207
[98] Duan Shanxu, Meng Yu, Xiong Jian, Kang Yong, Chen Jian, “Parallel operation
control technique of voltage source inverters in UPS,” Proceedings of the IEEE 1999
International Conference on Power Electronics and Drive Systems, Vol. 2 , pp. 883 –
887, 27-29 July 1999
[102] M.C. Chandorkar, D.M. Divan, and R. Adapa, “ Control of Parallel Connected
inverters in standalone ac supply systems,” IEEE Trans. Industry Appl., vol. 29, no. 1,
January/February 1993, pp. 136-143
[106] E.A.A. Coelho, P. Cortizo, and P.F.D. Garcia, “Small-Signal stability for parallel-
connected inverters in stand-alone AC supply systems”, IEEE Trans. Ind. App, vol. 38,
No.2, Mar/Apr, 2002, pp.533-542
[107] L. Xinchun, F. Feng, D. Shanxu, K. Yong, and C. Jian, “Modeling and Stability
Analysis for two Paralleled UPS with no control interconnection”, IEMDC Conf. Proc.,
Madison, Wisconsin, June 2003, pp. 1772-1776
208
[109] J.M. Guerrero, L. Garcia de Vicuna, J. Matas, J. Miret, , “Steady-state invariant-
frequency control of parallel redundant uninterruptible power supplies,” IECON 02, Vol.
1, pp. 274 – 277, 5-8 Nov. 2002
[111] Y.B. Byun, T.G. Koo, K.Y. Joe, E.S. Kim, J.I. Seo, D.H. Kim, “Parallel operation
of three-phase UPS inverters by wireless load sharing control,” INTELEC 2000, pp.526 –
532, 10-14 Sept. 2000
[113] A. Tuladhar, H. Jin, T. Unger, and K Mauch, “Parallel operation of single phase
inverters with no control interconnections,” in Proc. IEEE APEC’97, vol. 1, 1997, pp. 94-
100
[115] T.S. Lee; K.S. Tzeng, M.S. Chong, “Robust controller design for a single-phase
UPS inverter using mu-synthesis,” IEE Proceedings-Electric Power Applications, Vol.
151, No. 3 , pp. 334 – 340, 8 May 2004
[116] G.J Balas, J.C. Doyle, K. Glover, A. Packard, and R. Smith, The µ Analysis and
Synthesis Toolbox, Mathworks, Natick, MA , 1991
[117] R. Chiang, and M. Safonov, Robust Control Toolbox, Mathworks, Natick, MA,
1998
[118] J.C. Doyle, Analysis of feedback systems with structured uncertainties, IEEE
Proc., 129 Part D(6), 242-250, 1982.
[119] G.J. Balas, and A. Packard, “The structured singular value ( µ ) framework”, The
Control Handbook, editor W. S Levine, CRC Press, 671-687, 1996.
[120] TheMathworks, Control System Toolbox User’s Guide Version 6, The Mathworks
Inc., Natick, MA, June 2004
209
[121] E J Davison, “ Robust Servo Mechanism Problem,” The Control Handbook,
editor W. S Levine, CRC Press, 731-747, 1996.
[122] G.F Franklin, J.D. Powell, and M.L. Workman, Digital Control of Dynamic
Systems, Addison-Wesley Publishing Company, Second Edition, New York, 1990
[123] J.C. Doyle, B.A. Francis, and A.R. Tannenbaum, Feedback Control Theory,
Macmillan Publishing Company, New York, 1992
[124] G.F. Franklin, J.D. Powell, A. E. Naeini, Feedback control of Dynamic Systems,
Addison-Wesley Publishing Company, Third Edition, New York, 1994.
[125] F.L Lewis, V.L. Syrmos, Optimal Control, John Wiley & Sons Inc., Second
Edition, New York, 1995
[126] H.W. Van Der Broeck, H.C. Skudelny, “Analysis and Realization of a Pulsewidth
Modulator Based on Voltage Space Vectors”, IEEE Trans.on Industry Applications, vol.
24, no. 1, January/February 1988. pp. 142-150
[128] H. Pinheiro, R. Blume, P. Jain, “Space vector modulation method for single
phase on-line three-leg UPS,” INTELEC 2000, pp. 679 – 686, 10-14 Sept. 2000
[129] Comparison of SV modulation methods for single phase on-line three-leg UPS
Pinheiro, H., Blume, R., Jain, P., IECON 2000, Vol. 2, pp. 1328 – 1333, 22-28 Oct. 2000
210
[134] C.V. Nayar, , M. Ashari, W.W. L. Keerthipala, “A grid-interactive photovoltaic
uninterruptible power supply system using battery storage and a back up diesel
generator,” IEEE Transactions on Energy Conversion, Vol. 15, No. 3 , pp. 348 – 353,
Sept. 2000
[137] J.-Y.Lee, W.I. Tsai, Y.M. Chang, “A simple method for improving input power
factor of single-phase UPS,” IECON '91, Vol. 1, pp. 830 - 835 28 Oct.-1 Nov. 1991
[138] W.-J. Ho, J.-B. Lio, W.-S. Feng, “Economic UPS structure with phase-controlled
battery charger and input-power-factor improvement,” IEE Proceedings-Electric Power
Applications, Vol. 144, No. 4, pp. 221 – 226, July 1997
[139] J.-A. Hyo, C. Aguilar, J. Arau, “Simplified AC UPS with power factor
correction”, Power Electronics Congress 2000, pp. 145 – 149, 15-19 Oct. 2000
[140] J. Hoyo, J. Arau, C. Aguilar, “Novel approach to design PFC converters for use in
AC UPS applications,” Power Electronics Specialists Conference, 2001, Vol. 1, pp. 271
– 276, 17-21 June 2001
[141] M.N. Marwali, Design and Implementation of 5kVA PWM Inverter Test Bed for
Uninteruptible Power Supply, Mechatronics Laboratory, The Ohio State University, 2004
211
APPENDIX A
212
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% Program to calculate Control Gains using RSP Voltage Controller and %%
%% discrete Sliding Mode Current Controller for Single Unit Inverter %%
%% Author: Mohammad N Marwali %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
clear all
213
%% Per Unit Current Limit
Ilimit=3; % 300% inverter current limit
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Start design of discrete SM current controller
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
A=[0 xCinv*wfun ;
-1/(xLinv/wfun) -xRinv/(xLinv/wfun)];
B=[0;
1/(xLinv/wfun)];
E=[-xCinv*wfun;
0];
F=[0;
-1];
C=[0 1];
D=0;
sysc=ss(A,B,C,zeros(size(C,1),size(B,2)),'inputdelay',0);
sysd=c2d(sysc,Tsamp,'zoh');
[Acurrd,Bcurrd,Ccurrd,Dcurrd]=ssdata(sysd);
CBinv=inv(Ccurrd*Bcurrd);
CA=Ccurrd*Acurrd;
CD=Ccurrd*F;
sysc=ss(A,E,C,zeros(size(C,1),size(B,2)),'inputdelay',0);
sysd=c2d(sysc,Tsamp,'zoh');
[Acurrd1,Ecurrd,Ccurrd1,Dcurrd1]=ssdata(sysd);
CE=Ccurrd1*Ecurrd;
Asm=zeros(3,3);
Bsm=[0 1 0 0;
0 0 1 0;
0 0 0 1];
214
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Start design of Perfect RSP voltages controllers
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Bo=[0;
1/(xLinv/wfun);
0;
0];
Co=[0 0 1 0];
Do=zeros(size(C,1),size(B,2));
Ch2=[0 1;
-w2^2 0];
Ch3=[0 1;
-w3^2 0];
Ch4=[0 1;
-w4^2 0];
Ch5=[0 1;
-w5^2 0];
Ch7=[0 1;
-w7^2 0];
Bh_star=[0;
1;
0;
1;
0;
1;
0;
1;];
215
% Discretize true plant
sysc=ss(Ao,Bo,Co,Do,'inputdelay',0.5*Tsamp);
sysd=c2d(sysc,Tsamp,'zoh');
[Aod,Bod,Cd,Dd]=ssdata(sysd);
Q1=state_W*eye(size(Ad,1));
Q2=eye(size(Acon_d,1));
Q2(1:2,:)=fund_servo_W*Q2(1:2,:);
Q2(3:8,:)=harm_servo_W*Q2(3:8,:);
Q=[ Q1 zeros(size(Ad,1),size(Acon_d,2));
zeros(size(Acon_d,1),size(Ad,2)) Q2];
R=epsilon;
216
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% Compute Variable Frequency Harmonic Servo Compensator
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
harmonic=[1 3 5 7];
Bh_star=[0;1];
for h=1:length(harmonic)
i=1;
for f=59.0:0.02:61
w=2*pi*f*harmonic(h);
Ch_star=[0 1;
-w^2 0];
csysc=ss(Ch_star,Bh_star,eye(2),0);
[csysbc,T]=ssbal(csysc);
csysd=c2d(csysbc,Tsamp,'zoh');
[Acon_dx,Bcon_dx,Ccon_dx,Dcon_dx]=ssdata(csysd);
A11(i)=Acon_dx(1,1);
A12(i)=Acon_dx(1,2);
A21(i)=Acon_dx(2,1);
A22(i)=Acon_dx(2,2);
B1(i)=Bcon_dx(1)*s_scale;
B2(i)=Bcon_dx(2)*s_scale;
i=i+1;
end
f=59.0:0.02:61;
pA11{h} = polyfit(f,A11,1);
pA12{h} = polyfit(f,A12,1);
pA21{h} = polyfit(f,A21,1);
pA22{h} = polyfit(f,A22,1);
pB1{h}=polyfit(f,B1,1);
pB2{h}=polyfit(f,B2,1);
pA11{h}(2)=pA11{h}(2)+pA11{h}(1)*60;
pA12{h}(2)=pA12{h}(2)+pA12{h}(1)*60;
pA21{h}(2)=pA21{h}(2)+pA21{h}(1)*60;
pA22{h}(2)=pA22{h}(2)+pA22{h}(1)*60;
pB1{h}(2)=pB1{h}(2)+pB1{h}(1)*60;
pB2{h}(2)=pB2{h}(2)+pB2{h}(1)*60;
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% Some useful matrices used in simulations etc.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
217
Kln=[1 0 -1;
-1 1 0;
0 -1 1];
Trv=[-1 0 0;
0 -1 0;
0 0 -1];
Tri=1/3*[-2 1 1;
1 -2 1;
1 1 -2];
Tri_qd0=Ks*Tri*Ksinv;
Tri_qd=Tri_qd0(1:2,1:2);
Trv_qd0=Ks*Trv*Ksinv;
Trv_qd=Trv_qd0(1:2,1:2);
218
APPENDIX B
219
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% ROBUST STABILITY ANALYSIS OF SINGLE INVERTER CONTROL SYSTEM
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
cinv_tol=0.06;
linv_tol=0.15;
cload_tol=0.06;
ltrans_tol=0.15;
rinv_tol=0.5;
rtrans_tol=0.5;
yrload_tol=1;
ylload_tol=1;
linv_nom=xLinv/wfun;
cinv_nom=1/(xCinv*wfun);
ltrans_nom=(xLtrans/wfun);
cload_nom=1/(xCload*wfun);
rinv_nom=xRinv;
rtrans_nom=xRtrans;
yrload_nom=0.8;
ylload_nom=0.6*wfun;
rinv_del=rinv_nom*rinv_tol;
rtrans_del=rtrans_nom*rtrans_tol;
yrload_del=yrload_nom*yrload_tol;
ylload_del=ylload_nom*ylload_tol;
reps=1e-1;
Bnom=[ 0;
1/linv_nom;
0;
0;
0];
Bdel=[ -cinv_tol 0 0 0 0 0 0 0 ;
0 -linv_tol 0 0 -rinv_del/linv_nom 0 0 0 ;
0 0 -cload_tol 0 0 0 -yrload_del/cload_nom 0 ;
0 0 0 -ltrans_tol 0 -rtrans_del/ltrans_nom 0 0 ;
0 0 0 0 0 0 0 ylload_del
];
Bp=[Bnom Bdel];
Cnom=[1 0 0 0 0;
0 1 0 0 0;
220
0 0 1 0 0;
0 0 yrload_nom 0 1];
Cdel=[ 0 1/cinv_nom 0 -1/cinv_nom 0 ;
-1/linv_nom -rinv_nom/linv_nom 0 0 0 ;
0 0 -yrload_nom/cload_nom 1/cload_nom -1/cload_nom;;
1/ltrans_nom 0 -1/ltrans_nom -rtrans_nom/ltrans_nom 0;
0 1 0 0 0;
0 0 0 1 0;
0 0 1 0 0;
0 0 1 0 0;];
Cp=[Cnom ; Cdel];
Dnom=[ zeros(3,9);
0 0 0 0 0 0 0 yrload_del 0];
Ddel=[ 0 -cinv_tol 0 0 0 0 0 0 0 ;
1/linv_nom 0 -linv_tol 0 0 -rinv_del/linv_nom 0 0 0 ;
0 0 0 -cload_tol 0 0 0 -yrload_del/cload_nom 0 ;
0 0 0 0 -ltrans_tol 0 -rtrans_del/ltrans_nom 0 0 ;
0 0 0 0 0 0 0 0 0 ;
0 0 0 0 0 0 0 0 0 ;
0 0 0 0 0 0 0 0 0 ;
0 0 0 0 0 0 0 0 0
];
Dp=[Dnom ; Ddel];
plant_nom_c=pck(Ap,Bp,Cp,Dp);
plant_nom_w=pck(Ap_c,Bp_c,Cp_c,Dp_c);
221
del=[del_cinv 0 0 0 0 0 0 0 ;
0 del_linv 0 0 0 0 0 0 ;
0 0 del_cload 0 0 0 0 0 ;
0 0 0 del_ltrans 0 0 0 0 ;
0 0 0 0 del_rinv 0 0 0 ;
0 0 0 0 0 del_rtrans 0 0 ;
0 0 0 0 0 0 del_yrload 0 ;
0 0 0 0 0 0 0 del_ylload
];
plant_del_c=starp(plant_nom_c,del);
rifd(spoles(plant_del_c));
plant_del_w=starp(plant_nom_w,del);
Asm=zeros(3,3);
Bsm=[0 1 0 0;
0 0 1 0;
0 0 0 1];
Bservo=[Bcon_d zeros(8,5) ;
zeros(1,5) 1];
Cservo=[Kd(6:13) Kd(5)];
dsm=pck(Asm,Bsm,Csm,Dsm);
servo=pck(Aservo,Bservo,Cservo,Dservo);
222
outputvar='[dsm]';
input_to_dsm='[servo; plant(1,2,4)]';
input_to_servo='[err; plant(1:4) ; dsm]';
sysoutname='Hcontrol_d';
cleanupsysic = 'yes';
sysic;
[Acontrol,Bcontrol,Ccontrol,Dcontrol]=unpck(Hcontrol_d);
sys_con_d=ss(Acontrol,Bcontrol,Ccontrol,Dcontrol,Tsamp);
clpstab_g=sel(clp_g,2:9,2:9);
del_cinv=0;
del_cload=0;
del_ltrans=0;
del_rinv=0;
del_rtrans=0;
223
del_yrload=0;
del_ylload=0;
figure
hold
del=[del_cinv 0 0 0 0 0 0 0 ;
0 del_linv 0 0 0 0 0 0 ;
0 0 del_cload 0 0 0 0 0 ;
0 0 0 del_ltrans 0 0 0 0 ;
0 0 0 0 del_rinv 0 0 0 ;
0 0 0 0 0 del_rtrans 0 0 ;
0 0 0 0 0 0 del_yrload 0 ;
0 0 0 0 0 0 0 del_ylload
];
clp_del=starp(clp_w,del);
[Adel,Bdel,Cdel,Ddel]=unpck(clp_del);
sys_del=ss(Adel,Bdel,Cdel,Ddel);
pzmap(sys_del);
end
224