28 d jovcic - analytical modeling of tcsc dynamics
TRANSCRIPT
-
7/29/2019 28 D Jovcic - Analytical Modeling of TCSC Dynamics
1/8
1
AbstractThis paper presents an analytical, linear, state-
space model of a Thyristor Controlled Series Capacitor (TCSC).
Firstly, a simplified fundamental frequency model of TCSC is
proposed and the model results are verified. Using frequency
response of the non-linear TCSC segment, a simplified non-linear
state space model is derived, where the frequency of the
dominant TCSC complex poles shows linear dependence on the
firing angle. The non-linear element is linearised and linked with
the AC network model and the TCSC controller model that also
includes a Phase Locked Loop model. The model is implementedin MATLAB and verified against PSCAD/EMTDC in the time
and frequency domains for a range of operating conditions. The
model is sufficiently accurate for most control design applications
and practical stability issues in the subsynchronous range.
Index TermsModeling, Power system dynamic stability,
State space methods, Thyristor converters.
I. INTRODUCTION
Thyristor Controlled Series Capacitor (TCSC) is a series
FACTS device which allows rapid and continuous changes of
the transmission line impedance. It has great application
potential in accurately regulating the power flow on a
transmission line, damping inter-area power oscillations,
mitigating subsynchronous resonance (SSR) and improving
transient stability.
The characteristics of a TCSC at steady-state and very low
frequencies can be studied using fundamental frequency
analytical models [1],[2]. These particular models recognize
the importance of having different approach from SVC
modeling (assuming only line current as a constant) which,
although more demanding on the derivation, gives the most
accurate TCSC model.
The fundamental frequency models cannot be used in wider
frequency range since they only give the relationship among
fundamental components of variables when at steady-state.
Conventionally, the electromechanical transient programs likeEMTP or PSCAD/EMTDC are used for TCSC transient
stability analysis. These simulation tools are accurate but they
employ trial and error type studies only, implying use of a
large number of repetitive simulation runs for varying
parameters in the case of complex analysis or design tasks. On
the other hand, the application of dynamic systems analysis
This work is supported by The Engineering and Physical SciencesResearch Council (EPSRC) UK, grant no GR/R11377/01.
D. Jovcic and G.N. Pillai are with Electrical and Mechanical Engineering,University of Ulster, Newtownabbey, BT370QB, United Kingdom, (e-mail:[email protected], [email protected])
techniques or modern control design theories would bring
benefits like shorter design time, optimization of resources and
development of new improved designs. In particular, the
eigenvalue and frequency domain analysis are widely
recognized tools and they would prove invaluable for system
designers and operators. These techniques however always
necessitate a suitable and accurate dynamic system model.
There have been a number of attempts to derive an accurate
analytical model of a TCSC that can be employed in system
stability studies and controller design [3-6]. The model
presented in [3] uses a special form of discretisation, applying
Poincare mapping, for the particular Kayenta TCSC
installation. The model derivation for a different system will
be similarly tedious and the final model form is not convenientfor the application of standard stability studies and controller
design theories especially not for larger systems. A similar
final model form is derived in [4], and the model derivation is
improved since direct discretisation of the linear system model
is used, however it suffers other shortcomings as the model in
[3]. The modeling principle reported in [5] avoids
discretisation and stresses the need for assuming only line
current as an ideal sine, however it employs rotating vectors
that might be difficult to use with stability studies, and only
considers the open loop configuration. The model in [5] is alsooversimplified because of the use of equivalent reactance and
equivalent capacitance that might be deficient when used in
wider frequency range. Most of these reported models are
therefore concerned with a particular system or particular type
of study, use overly simplified approach and do not include
control elements or Phase Locked Loops (PLL).
In this study we attempt to derive a suitable linear
continuous TCSC model in state-space form. The model
should have reasonable accuracy for the dynamic studies in
the sub-synchronous range and it should incorporate common
control elements including PLL. To enable flexibility of the
model use with different AC systems, the model structure
adopts subsystem units interlinked in similar manner as withSVC modeling [6].
We also seek to offer complete closed loop model
verification in the time and frequency domains.
II. TESTSYSTEM
The test system for the study is a long transmission system
compensated by a TCSC and connected to a firm voltage
source on each side. Figure 1 shows a single line diagram of
the test system where the transmission line is represented by a
lumped resistance and inductance in accordance with the
approach for sub-synchronous resonance studies [7]. Each
phase of the TCSC is composed of a fixed capacitor in parallel
with a Thyristor Controlled Reactor (TCR). The TCSC is
Analytical Modeling of TCSC Dynamics
D.Jovcic,Member, IEEEandG.N. Pillai
-
7/29/2019 28 D Jovcic - Analytical Modeling of TCSC Dynamics
2/8
2
controlled by varying the phase delay of the thyristor firing
pulses synchronized through a PLL to the line current
waveform. The controller is of a PI type with a feedback filter
and a series compensator.
Three different test systems are used in order to verify the
model accuracy with different system parameters and differentoperating points:
system 1 75% compensation, capacitive mode, system 2 40% compensation, capacitive mode, system 3 75% compensation, inductive mode,
The test system parameters are practically selected
according to the recommendations in [8] (regarding Xl/Xc and
natural resonance) and the test system data are given in the
Appendix.
- Calculated firing angle- PLL reference angle- Actual firing angle- Phase angle of- Mag. of voltage
PLL - Phase Locked loopPI - Voltage PI controllerTCR - Thyristor controlledreactor
TCR
TCSC
l
l
c
c
l
c
Figure 1. Test system configuration.
III. FUNDAMENTALFREQUENCYMODEL
The fundamental frequency model of a TCSC is derived
firstly to enable initialization of the steady-state parameters.
The voltage across the TCSC capacitor (vc) comprises an
uncontrolled and a controlled component [8], and it ispresumed that the line current is constant over one
fundamental cycle in accordance with [1-2],[5]. The
uncontrolled component v1, is a sine wave (unaffected by
thyristor switchings) and it is also constant over a fundamentalcycle since it is directly related to the amplitude of the
prevailing line current. The controlled component v2 is a non-
linear variable that depends on circuit variables and on the
TCR firing angle.
In this study, the controlled component is represented as a
non-linear function of the uncontrolled component and firing
angle, v2=N1(v1,,s), as shown in Figure 2. With this
approach, N1(v1,,s) captures the non-linear phenomena
caused by thyristor switching influence and all internal
interactions with capacitor voltage assuming only that the line
current and v1 are linear. We seek in our work to study
dynamics ofN1(v1,,s) in a wider frequency range and also to
offer a simplified representation for fundamental frequencystudies.
cs
1
-
+1v
1v
2v
2v
cv
cv
li
li - line current- linear component of TCSC voltage- non-linear component of TCSC voltage- TCSC voltage- firing angle
- Non-linear part dynamics
),,( 112 svNv =
),,( 11 svN
Figure 2. TCSC model structure.
The fundamental components of reactor current itcr and the
voltages v1 and v2 are selected as state variables and the non-
linear state-space model is presented as:
li
c
sv1
1 = (1)
tcric
gsv1
2 = (2)
21
11v
lgv
lgsi
tcrtcr
tcr = (3)
21 vvvc = (4)wheregrepresents the switching function:g=1 for thyristor in
conduction, andg=0 for thyristor in blocking state.
To derive the corresponding linear model at the nominal
frequency, it is postulated that the model has the following
structure:
lic
sv 11 = (5)
tcric
sv1
2 = (6)
2211 vkvksitcr = (7)
21 vvvc = (8)
where k1=k1() and k2=k2() are the unknown model
parameters dependent on the firing angle . The above model
structure is correct for zero firing angle (g=1,=0, i.e. full
conduction with measured from the voltage crest) and we
presume the same model structure but different parameter
values for(0,90).
Considering the two components of the thyristor current on
the right side of (7) we note that11vk produces the current
that is driven by a constant voltage v1 (over one fundamental
cycle) in accordance to the earlier assumption. As a result, the
configuration is similar to the shunt connection of TRC in a
SVC, and the constant1k can be calculated using the approach
of equivalent reactancealpl for the TCR current in a SVC
[6,8]:
)2sin(2/1 1
==
tcr
alp
l
lk(9)
-
7/29/2019 28 D Jovcic - Analytical Modeling of TCSC Dynamics
3/8
3
In order to determine the constant k2, we represent (6)-(7) as
a transfer function in the following form:
1/)(
)(
)(22
3
1
2
+==
d
fos
ksT
sv
sv
(10)
withc
kd
2= ,2
13
k
kk = (11)
Observing (10) it is seen that d is the resonant frequencyof an un-damped second order system. The proposed
expression for this frequency is given below in (12) and it is
determined using the experimental frequency response of the
model for the non-linear TCSC segment as presented later in
section IV.
cltcr
d
2= (12)
Substituting thats= jo= j2fo (fo fundamental frequency)in (10), and using (11) in terms of input current and the output
voltage as given in (5)-(8) the TCSC gain at fundamental
frequency is derived:
( )( )[ ]1/
1/22
3
22
+
+=
doo
do
l
c
jcj
kj
i
v
(13)
and the TCSC fundamental frequency impedance is:
( )22
3
22
csc
/1
1/
doo
dot
c
kX
+= (14)
where
[ ]23 )2(
)2sin(2
=k (15)
and d is defined in (12).
Compared with the fundamental frequency TCSC model in
[1-2], the derived model (13) is considerably simpler, yet it
will be shown that the accuracy is very similar.
The above model is tested against PSCAD in the following
way. The system is operated in open loop, i.e. with a fixed
reference thyristor firing angle, in the configuration of Figure
1. The value of the reference thyristor angle is changed in
interval (0,90o) and for each angle the value of the TCSCvoltage is observed. Since all other parameters are constant,
the TCSC voltage is directly proportional to the TCSCimpedance and this is an effective way to obtain accurate
information on the fundamental TCSC impedance.
Figure 3 shows the steady-state TCSC voltage against the
range of firing angle values, where the above linear model
results are shown as model 2 and also the results fromresearchers [1,2] as model 1. It is seen that the proposed model
shows good accuracy across the entire firing angle range and
especially the results are very close to those from [1,2]. In fact
the above model differs negligibly from the model [1,2],
except in the less used low firing angles in range (20,30o)
where the difference is still below 5%. It is also evident thatthe two analytical models show small but consistent
discrepancy against PSCAD/EMTDC, and despite all
simplifications in PSCAD it was not possible to obtain better
matching.
Figure 3. Testing the fundamental frequency model. Steady-state TCSCvoltage for different values of firing angles.
IV. SMALL-SIGNALDYNAMICANALYTICALMODELA. TCSC model
The transfer function in (10) is accurate only at fundamental
frequency and it cannot be used for wider frequency studies.
The goal of the dynamic modeling is to derive a dynamic
expression forN1(v1,,s) of satisfactory accuracy in the sub-
synchronous frequency range and for small signals around the
steady-state operating point. The steady-state for an AC
system variable is a periodic waveform at fundamental
frequency with constant magnitude and phase.
The frequency response is of TCSC is studied assuming that
the input voltage is:
)2sin()2sin(01 tfAtfAv injinjo += (16)where the first sine signal denotes the steady-state operation
with all parameters constant. The second term is the input in
the experimental frequency response and it has small
magnitude Ainj
-
7/29/2019 28 D Jovcic - Analytical Modeling of TCSC Dynamics
4/8
4
Gain2
v1
v2
b1
a1
Switch2
Switch1
sine
cos
period
Source
PulseGenerator
1s
Int2
1s
Int1
1/(ct*ltcr)
Gain1
v2
sine
cos
period
a1
b1
Fourrieranalysis
0
Const.
1/(ct*ltcr)
Figure 4. Experimental frequency response set-up with the SIMULINK modelfor the non-linear TCSC segment.
magnitude and phase of the first harmonic of the output
voltage v2 is observed.
The pulse generator produces equally spaced conduction
intervals that are based on the fundamental frequency period
and which are dependent on the firing angle. Based on the
above assumption these firing instants are unaffected by theinjection signal.
Figure 5 shows the obtained time-domain responses for the
particular firing angle =72o and for two different frequencies
of the input signal. It is seen in Figure 5 b) that the output
trace for the 60Hz input frequency gives the familiar square
shape which is consistent with the results for the fundamental
frequency studies in [1,2,8]. The experimental frequency
response over the range of frequencies is repeated for all
values of the firing angle (0,90o) with a one-degree
increment.
For each firing angle, the gain and phase values across thefrequency range are plotted and Figure 6 shows an example
for the particular firing angle =76o
. Observing the frequencyresponse curves in Figure 6 and others for other firing angles,
it is concluded that the system shows higher order properties
and it becomes clear that it has dominant second order
dynamics. The following transfer function is proposed:
input v1
input v1
output v2
output v2
first harmonic v2
first harmonic v2
time [s]
a)
b)
time [s]
magnitude[kV]
m
g
itd
[
]
Figure 5. Experimental frequency response ofN1(v1,,s) using SIMULINK
model. Firing angle =72o, input frequency a)finj=12Hz. b)finj=60Hz.
gain
[dB]
phase[
deg]
frequency [rad/s]
fo
linear model
non-linear model
fund. frequency model
fo
Figure 6. Frequency response for the non-linear TCSC segment. Test system 1
with the firing angle =76o.
system 1 experimental
system 2 experimental
system 1 model
system 2 model
frqu
n
y[rd
/s]
firing angle [deg]
Figure 7. Linear relationship between the denominator characteristic
frequency (d) and the firing angle .
12
12
)(
)(
2
2
2
2
1
2
++
++=
d
d
d
n
n
n
w
s
w
s
w
s
w
s
sv
sv
(17)
The characteristic peak in the gain frequency response is
caused by the dominant complex poles and each of the troughs
by a pair of complex zeros. Observing the frequency response
across the range of firing angles, the location of the dominant
peak is recorded against the firing angle values and the circuit
parameters, and the final plot is shown in Figure 7. It soon
emerges that the denominator characteristic frequency d is a
linear function of the firing angle where the actualmathematical formula can be readily derived as given in (12).
-
7/29/2019 28 D Jovcic - Analytical Modeling of TCSC Dynamics
5/8
5
The linear formula shows very good accuracy and it reflects
well the variation in circuit parameters as seen in Figure 7.
Analysing (12) it is seen that the resonant frequency of a
TCSC starts at the natural resonance of L-C circuit (for=0)
and it linearly reduces with the increase in the firing angle.
The formula has importance for the practical design since itpoints to the TCSC firing angles that should be avoided if the
connected system is sensitive to resonance in a particular
frequency range (SSR problems for example).
There are four unknowns in (17) and they are determined in
a similar manner, by matching the frequency response and the
final step responses, and the proposed final formulae are given
below.
clw
tcr
d
2= (18)
)cos(38.0 =d (19)
+=
)(tan107.110
4
3
cl
fw
tcr
on (20)
)cos(2.0 =n (21)
The simulated frequency response with the above transfer
function is shown in Figure 6 where satisfactory matching is
observed, except at higher frequencies. It is emphasized here
that the model zeros fall in higher frequency region, implying
influence of sampling system phenomena, and the numerator
parameters in (17) are open for further research and
improvements.
It would be beneficial to use the simpler model as given in
(13) for the dynamic study in a wider frequency range. If wereplace the fundamental frequency in (13) with an independent
variable for frequency domain studies, jo=s, we can test themodel assuming the frequency variable represents deviations
around the nominal frequency. The resultant frequency
response curves are shown in Figure 6 labeled as the
fundamental frequency model. It is evident that such model
matches the frequency response only atfo; it has incorrect gain
for low frequencies, incorrect damping of the dominant mode
and it can not be used for wider frequency studies.
We point further that the complex poles of the transfer
function (17) have finite damping which is for capacitive
mode typically 0.07
-
7/29/2019 28 D Jovcic - Analytical Modeling of TCSC Dynamics
6/8
6
The filter time constant is of the order of2ms, which is in line
with studies in [6],[8] and [12]. It should be noted that the
voltage phase angle also affects the firing angle because the
actual firing angle is measured with respect to the voltage
curve. Here, we have direct correlation across each phase, and
the sampling frequency is 1/3 offs and therefore an additionallag is introduced represented by the delay filter with Td2.
Simulation results demonstrate improvement in the response
with the introduction of this delay element. The second order
filer with vc reduces the harmonics on the feedback signal.
PLL
Vref
Vref
cV
cV
+
+
+
+
+
PI controller Lead-lagFeedback filter Delay filter
Delay filter
/2
s
kk ip +22
2
2 fff
f
wsws
w
++
1
1
1 +sTd
1
1
2 +sTd
d
d
- TCSC voltage magnitude- reference voltage- voltage angle- delayed voltage angle
- line current angle- PLL reference angle- controller reference angle- actual firing angle
1
1
2
1
+
+
sT
sT
Figure 8. The controller model and firing angle calculation.
D. Model Connections
The final model consists of three independent state-space
models: AC system, TCSC model and the controller model,
which includes PLL. The linearised TCSC model is linked
with the firing angle which is the output of the controller
model, using the artificial rotating variables as presented in
[6].The three subsystem models are linked through the coupling
variables to create the final state space model in a similar
manner as with the SVC model in [6], and the final model has
the following structure:
outsssout
outssss
uDxCy
uBxAx
+=
+=o
(25)
where s labels the overall system, and the model matrices
are:
=
actcacactccoacacco
actctcactcctcco
accocoactccocco
s
ACBCB
CBACB
CBCBA
A
**
**
**
cot
cot
(26)
[ ] [ ]0,, ==
= sacouttcoutcoouts
acout
tcout
coout
s DCCCC
B
B
B
B
All the subsystems D matrices are assumed zero in (26)
since they are zero in the actual model and this noticeably
simplifies development. The Bs is the external input matrix,
i.e. it links with the reference step uout=Vref whereas Cs
specifies the monitored outputsyout
=vc.
The matrix As has the subsystem matrices on the main
diagonal, with the other sub-matrices representing interactions
between subsystems. The following indices are used ac AC
system, tc TCSC, co controller. As an example the
matrix Btcco is the TCSC model input matrix that takes input
signals from the controller. The model in this form hasadvantages in flexibility since, if the TCSC is connected to a
more complex AC system only the Aac matrix and the
corresponding input and output matrices need modifications.
Similarly, more advanced controllers can be developed using
modern control theory (H, MPC,..) and implemented directly
by replacing the Aco matrix. The above structure enables the
model to be readily interfaced with the MATLAB model of
other FACTS elements, HVDC or other power system units.
V. MODEL VERIFICATION
A. Time domain
The linear model is coded in MATLAB and it is verified in
time domain by comparing the reference step responses withPSCAD model for the three test systems. All the parameters in
the PSCAD model are identical to those in MATLAB model
(AC system, controller and PLL parameters) and model for
switches includes typical snubber circuits. Figures 9.10 and 11
show the three responses of systems 1,2 and 3 respectively.
We can conclude that the model accurately reflects the system
Figure 9. Test system 1 response following a 1kV voltage reference stepchange.
Figure 10. Test system 2 response following a 0.5kV voltage reference stepchange.
-
7/29/2019 28 D Jovcic - Analytical Modeling of TCSC Dynamics
7/8
7
Figure 11. Test system 3 response following a 0.05kVvoltage reference stepchange.
structure since with different circuit parameters and operating
points we consistently have good response matching. In
inductive mode, shown in Figure 11, it is most difficult toobtain high accuracy since in this mode v2 is larger than v1 and
the system is highly non-linear, but this mode is seldom used
in practice.
B. Frequency domain
The time domain testing clearly indicates the dominant
oscillatory mode and also it verifies model for different
operating parameters and inputs. Step input testing however
cannot indicate the frequency range for the model validity.The experimental frequency response is performed with the
PSCAD software in the frequency range 1Hz
-
7/29/2019 28 D Jovcic - Analytical Modeling of TCSC Dynamics
8/8
8
Using the root locus and frequency domain design
techniques with MATLAB, a lead-lag compensator is
designed with the parameters as shown in the Appendix,
system 1. The root locus for the compensated system is shown
in Figure 13 b). It is seen that the damping of the dominant
mode is much improved. Also, the eigenvalue location is notsignificantly changed with two times increased transmission
system impedance, as demonstrated by curves m=2, and m=3.
Our studies show that a fast capacitor voltage feedback control
with a suitable lead-lag compensator can give a very robust
stabilizing controller for a TCSC.
It is emphasized here that the above design is a
multidimensional problem involving simultaneous variation of
controller gains, system strength and the compensator
parameters. Such problem would require a very large number
of trial-and-error simulation runs with conventional time-
domain simulators. The design also involves higher frequency
dynamics requiring an analytical model with high accuracy.
VII. CONCLUSIONS
The fundamental frequency model of a TCSC is developed
in a fairly simple form and the model shows good accuracy.
The TCSC model is segmented into linear and non-linear part
and the non-linear model is analysed in frequency domain
using SIMULINK. A simplified linear-continuous model for
the non-linear dynamics is developed by matching the
frequency response. The analysis of the non-linear dynamics
concludes that the dominant oscillatory mode is a linear
function of the firing angle. The linear TCSC model is
implemented in MATLAB and tested against non-linear
digital simulation PSCAD/EMTDC and good accuracy is
observed across a range of system parameters and operating
conditions. The model is also tested in frequency domain and
it is concluded that accuracy is good at frequencies f