Application of FACTS Devices for Damping of

Power System Oscillations

Rusejla Sadikovic Petr Korba Göran Andersson
ETH Zürich, Switzerland ABB Switzerland Ltd. ETH Zürich, Switzerland
[email protected] [email protected] [email protected]

Abstract— This paper describes an adaptive tuning of FACTS devices can be represented by following equation:
parameters of a power oscillation damping (POD) controller
for FACTS devices. The FACTS devices considered here are ∆ẋ = A∆x + B∆u
the Thyristor Controlled Series Compensator (TCSC) and the (1)
Unified Power Flow Controller (UPFC). A residue method is ∆y = C∆x
applied to the linearized power system model to determine the
best siting for FACTS devices as well as for the selection of where B and C are the column-vector input matrix and the
measured signals. Information available from a higher control
level, e.g. from a wide-area monitoring and control platform, is row-vector output matrix, respectively. Let λi = σi ± jωi be
used for a fine tuning of the POD controller in case of changing the i-th eigenvalue of the state matrix A. The real part of
operating conditions. the eigenvalues gives the damping, and the imaginary part
gives the frequency of oscillation. The relative damping ratio
Keywords - TCSC, UPFC, power system oscillations, adaptive is given by:
control, damping controller design.
−σ
I. I NTRODUCTION ξ=√ (2)
σ2 + ω2
Satisfactory damping of power oscillations is an important
The critical oscillatory modes considered here are those having
issue addressed when dealing with the rotor angle stability of
damping ratio less than 3%.
power systems. This phenomenon is well-known and observ-
If the state space matrix A has n distinct eigenvalues, Λ, Φ and
able especially when a fault occurs. To improve the damping of
Ψ below are the diagonal matrix of eigenvalues and matrices
oscillations in power systems, supplementary control laws can
of right and left eigenvectors, respectively:
be applied to existing devices. These supplementary actions
are referred to as power oscillation damping (POD) control.
AΦ = ΦΛ
In this work, POD control has been applied to two FACTS
ΨA = ΛΨ (3)
devices, TCSC and UPFC. The design method utilizes the
Ψ = Φ−1
residue approach, see e.g. [1]. The presented approach solves
the optimal siting of the FACTS as well as selection of the In order to modify a mode of oscillation by feedback, the
proper feedback signals and the controller design problem. chosen input must excite the mode and it must also be visible
In case of contingencies, changed operating conditions can in the chosen output. The measures of those two properties are
cause poorly damped or even unstable oscillations since the the controllability and observability, respectively. The modal
set of controller parameters yielding satisfactory damping for controllability and modal observability matrices are defined as
one operating condition may no longer be valid for another follows:
one. In this case, an advantage can be taken from the wide- B 0 = Φ−1 B
area monitoring platform, [3], to re-tune the POD controller’s (4)
C 0 = CΦ
parameters. A lately developed algorithm for on-line detection
of electromechanical oscillations based on Kalman filtering The mode is uncontrollable if the corresponding row of the
techniques has been employed [2]. It gives the information matrix B 0 is zero. The mode is unobservable if the corre-
about the actual dominant oscillatory modes with respect to the sponding column of the matrix C 0 is zero. If a mode is either
frequency and damping as well as about the amplitude of the uncontrollable or unobservable, feedback between the output
oscillation obtained through on-line analysis of global signals and the input will have no effect on the mode. The open loop
measured at the appropriate place in the power system. This transfer function of a SISO (single input single output) system
has further been used as a basis for the fine adaptive tuning is:
of the POD parameters. ∆y(s)
G(s) = = C(sI − A)−1 B (5)
II. T HE RESIDUE METHOD ∆u(s)
In order to identify local and interarea modes of a multi- G(s) can be expanded in partial fractions of the Laplace
machine system, the total linearized system model including transform of y in terms of C, B, matrices and the right and
 G

left eigenvectors as: G

8

1
33 34 36 37
N N
X Cφ(:, i)ψ(i, :)B X Ri 11
G(s) = = (6) 35 32
9

i=1
(s − λi ) i=1
(s − λi ) 38 26 25 G

24 G
6
Each term in the denominator of the summation is a scalar 12

G 23
29 30
called residue. The residue for a particular mode gives the 10
22
13
sensitivity of that mode’s eigenvalue to feedback between the
output y and the input u for a SISO system. It is the product 14
20
27
31

15
of the mode’s observability and controllability. 16
19 21
28
7

2 18 G
17 G 5 4
39
3 G G
G
uref + u y(s)
G(s)
- Fig. 2. System configuration for the case study.

∆u

H(s) ϕcomp = 1800 − arg(Ri )

ϕcomp
1 − sin( )
Fig. 1. Closed-loop system with POD control. Tlead mc
αc = = ϕcomp
Tlag 1 + sin( ) (9)
Fig. 1 shows a system G(s) equipped with a feedback control mc
H(s). When applying the feedback control, eigenvalues of 1
the initial system G(s) are changed. It can be proved [1], Tlag = √ , Tlead = αc Tlag
wi α c
that when the feedback control is applied, movement of an
eigenvalue is calculated by: where
arg(Ri ) denotes phase angle of the residue Ri ,
∆λi = Ri H(λi ) (7) ωi is the frequency of the mode of oscillation in rad/sec,
mc is the number if compensation stages (usually mc = 2).
It can be observed from (7) that the shift of the eigenvalue The controller gain K is computed as a function of the desired
caused by a controller is proportional to the magnitude of eigenvalue location λi,des according to Equation 7:
the residue. The change of eigenvalue must be directed
towards the left half complex plane for optimal damping λi,des − λi
K=| | (10)
improvement. For a certain mode to be controlled, a same Ri H1 (λi )
type of feedback control H(s), regardless of its structure and IV. A PPLICATION TO FACTS DEVICES
parameters, is tried out at different locations. For the mode
of the interest, the residues at tried locations are calculated. The linearized power system dynamics can be represented by
The largest residue indicates the most effective location to an open-loop transfer function G(s). Variable y is used by
apply the feedback control. the POD controller as an input signal, variable u is where the
control is fed back, see Fig. 1. Since the FACTS devices are
located in transmission systems, local input signals like power
III. FACTS POD CONTROLLER DESIGN APPROACH deviation ∆P , bus voltages or bus currents, are always prefer-
able. As in case of choosing the feedback signal, the optimal
In order to shift the real component of ∆λi to the left, FACTS sitting of the FACTS device is also very important, since a
POD controller is employed. That movement can be achieved larger residue results in a larger change of the corresponding
with a transfer function consisting of an amplification block, oscillatory mode, (7).
a wash-out block and mc stages of lead-lag blocks. We adapt A one-line diagram of the New England test system is given
the structure of POD controller given in [1] and [6], i.e. the in the Fig. 2. The power flow data for this system can be
transfer function of the FACTS POD controller is: found in [4]; the corresponding dynamic data for generators
 m and exciters were chosen from [5]. TCSC and UPFC used
sTw 1 + sTlead c
H(s) = K = KH1 (s) (8) in the simulations are modeled using the current injection
1 + sTw 1 + sTlag
model, [7], [8]. To find the best siting for the TCSC and
where K is a positive constant gain, and H1 (s) is the transfer UPFC, different location in the test system are tested. Residues
function of the wash-out and lead-lag blocks. The washout associated with critical mode are calculated using the transfer
time constant, Tw , is usually equal to 5-10 s. The lead -lag function between the TCSC active power deviation ∆P and
parameters can be determined using the following equations: the TCSC input, that is control variable as well, characterized
by the compensation degree ∆kc , i.e. the compensation in p.u. is:
of the line reactance. For the UPFC the residues are calculated 
1

10s

0.0695s + 1
2
between active and reactive power deviations ∆P and ∆Q H(s) = 5.89
0.1s + 1 10s + 1 0.5042s + 1
individually, and the UPFC inputs (control variables), which (11)
are the changes of the UPFC injected series voltage magnitude In order to check controller ability to stabilize the system, the
and angle, ∆r and ∆γ. Tables I and II show the numerical fault is applied in the line 34-36. The fault is cleared after 100
results of siting TCSC and UPFC, respectively. ms by opening the faulted line.
The active power deviation ∆P is used as the input signal for The problem with a set of fixed controller parameters arises
the TCSC controller and active and reactive power deviations
∆P and ∆Q, as the input signals for the UPFC controller. 3
without POD control
with POD control
2

Mode residues, |Ri |, of the transfer 1

function ∆P /∆kc
0
line 34-37 0.4508

Active power flow [pu]

line 34-36 0.2331 −1

line 36-37 0.2043 −2

line 24-25 0.1462

−3
line 24-27 0.1007
line 25-35 0.0698 −4

line 11-12 0.0545 −5

line 33-34 0.0365
−6
line 31-32 0.0307
line 24-29 0.0245 −7
0 5 10 15 20 25 30
line 13-22 0.0061 time [s]
line 14-15 0.0004
line 23-24 0.0003 Fig. 3. Active power flow in the controlled line 34-37 with and without
TABLE I damping control after three phase fault is applied to line 34-36 cleared after
100ms.
S ITING INDICES OF TCSC
when the system topology is changed. A set of POD param-
eters that gives satisfactory damping for one operating point
does not have to work for another operating point at all. In
Mode residues, |Ri |, of the different transfer functions such cases, the re-tuning of POD parameters is required. One
∆P/∆r ∆P/∆γ ∆Q/∆r ∆Q/∆γ solution of this problem is to re-tune the controller parameters
(γ = γ0 ) (r = r0 ) (γ = γ0 ) (r = r0 )
line 34-36 5.1343 1.7122 1.2500 0.0720
for every new operating point based on a complete set of
line 34-37 4.8847 1.7724 1.2271 0.0794 the model parameters, see Fig. 4. The disadvantage of this
line 25-35 4.6879 0.2936 2.3433 0.1252 approach is the necessity of knowing all power system’s data
line 36-37 3.9875 4.6394 0.2512 1.0076 and performing on-line linearization for the new operating
line 24-25 2.7188 0.1016 1.5311 0.0930
line 33-34 2.3226 0.3797 1.2143 0.0986 point. In Figure 5, direct comparison between the active
line 24-27 1.6978 1.8707 0.5854 0.4276 power flow response of the system to the fault with old POD
line 24-29 0.5309 0.6053 0.2174 0.1519 parameters, and with newly calculated POD parameters is
line 14-15 0.4855 0.5554 0.1983 0.0782
line 11-12 0.2595 0.1274 0.2087 0.1380
shown.
line 23-24 0.1575 0.1447 0.0201 0.0118
line 13-22 0.1154 0.2474 0.0412 0.0411
line 31-32 0.0093 0.1793 0.0038 0.0354 Power Flow Control

TABLE II Pref + Power

_
FACTS
S ITING INDICES OF UPFC Network

u POD Control
Pline

A. TCSC POD parameter Linearization of information about

calculation power network operational point
The uncontrolled system, Fig.2, has one critical oscillatory
mode characterized by λ = −0.0784 ± j5.3677 with damping Fig. 4. POD controller tuning, method 1, general form
ratio ξ = 1.46%. According to Table I, the line 34-37 has
the largest residue and therefore the most effective location to Another possibility for re-tuning POD parameters is an adap-
apply the feedback control. Using the method presented above, tive on-line tuning, see Fig. 6 and 7, based on automatic
POD controller parameters are calculated in order to shift the detection of oscillations in power systems using dynamic data
real part of the oscillatory mode, to the left half complex plane. such as currents, voltages and angle differences measured
The obtained transfer function for the TCSC POD controller across transmission lines, [2]. They are provided on-line by
 1
DPline xref
+
G(s)
+
new POD
0 old POD DxPOD

1 Tws Tlead(fi)s + 1 Tlead(fi)s + 1

K(zi,fi) Ts+1 Tws+1 Tlag(fi)s + 1 Tlag(fi)s + 1
−1
Active power flow [pu]

Adaptive zi
−2 algorithm based fi
on Kalman filter

−3
Fig. 7. POD controller tuning, method 2, detailed form
−4
1

−5 adaptive tuned POD

0 old POD

0 10 20 30 40 50 60 70 80 90
time [s]
−1

Active power flow [pu]

Fig. 5. Active power flow in controlled line 34-37 after three phase fault −2
applied to line 24-25 with line 33-34 out of service.
−3

phasor measurement units. The on-line collected measured −4

data are subject to a further evaluation with the objective to

estimate dominant modes (frequencies and damping) during −5

any operation of the power system. This information is used 0 10 20 30 40

time [s]
50 60 70 80 90

then to up-date the POD parameters. The POD gain is a

function of the frequency and the damping, (10), whereas Fig. 8. Active power flow in the controlled line 34-37 after three phase fault
applied to line 24-25 and with line 33-34 out of service.
time constants, Tlead and Tlag are functions only of the
frequency, (9). Dominant relative damping [%]

In this work (where adaptive control is derived from a mon- 50

itoring algorithm where only the dominant frequency and 0

damping are known, see [2]), one assumes that the mode
−50
residue remains unchanged (since no information about the 0 10 20 30 40
Dominant frequency [Hz]
50 60 70 80 90

2
angle of compensation is available for an up-date) and one only 1.5

up-dates Tlead and Tlag according to (9). A relatively simple 1

and powerful adaptive controller tuning has been achieved in 0.5

0
considered test systems using this approach. 0 10 20 30 40
Predictive error
50 60 70 80 90

1
Note that assuming an unchanged dominant residue for two
0.5
(or more) different operating points means in words that the
0
controller designer believes in fixed proper selection of the
measured feedback signal and fixed proper location of the −0.5
0 10 20 30 40
time [s]
50 60 70 80 90

FACTS controller for all considered operating points.

Fig. 8 and 10 show the active power flow responds for two Fig. 9. Results of detection of oscillations for the case in Fig.8
different cases. It can be seen that adaptive on-line tuning
0
satisfies even the N-2 criterion, see the case on Fig. 10.
−0.5 old POD
adaptive tuned POD

Power Flow Control −1

Pref Power
−1.5
Active power flow [pu]

FACTS
_ Network −2

u POD Control −2.5

Pline −3

Adaptive zi Adaptive −3.5

Controller fi algorithm based

Design on Kalman filter
−4

−4.5
0 10 20 30 40 50 60 70 80 90
time [s]
Fig. 6. POD controller tuning, method 2, general form

Fig. 10. Active power flow in the controlled line 34-37 after three phase
Fig. 9 and 11 show the results of detection of oscillations; fault applied to line 33-34 and with lines 31-32 and 12-26 out of service.
relative damping of the dominant oscillatory mode, frequency
of the dominant oscillatory mode and predictive error, which
is the error between the filtered measured signal and its
 −0.6

prediction.
−0.8
without POD
with POD
−1
Dominant relative damping [%]
60
−1.2
40

Active power flow [pu]

20
−1.4
0

−20 −1.6

0 10 20 30 40 50 60 70 80 90
−1.8
Dominant frequency [Hz]
1.5
−2

1 −2.2

−2.4
0.5
0 10 20 30 40 50 60 70 80 90
Predictive error −2.6
0.5 0 10 20 30 40 50 60 70
time [s]

0
Fig. 12. Active power flow in the controlled line 34-36 after three phase
−0.5
fault applied to line 33-34 and with line 31-32 out of service.
0 10 20 30 40 50 60 70 80 90
time [s]
−0.6

Fig. 11. Results of detection of oscillations for the case in Fig.10

−0.8

−1

B. UPFC
Active power flow [pu]
without POD
with POD
−1.2
The UPFC is located subsequently in the same line as TCSC.
It has two control parameters, r and γ, the magnitude and −1.4

the angle of the series injected voltage, respectively. The third

variable, shunt reactive power, Qconv1 is inactive, so the UPFC −1.6

performs the function of the series compensation. −1.8

It is theoretically possible to consider four possible POD

−2
control loops. However, from Table II, where the critical mode 0 10 20 30 40 50 60 70 80 90

time [s]
residues of the resulting four transfer functions are calculated,
one can see that ∆Q is not a good choice for the POD Fig. 13. Active power flow in the controlled line 34-36 after three phase
controller as an input signal, since the residues of ∆P/∆r fault applied to line 12-26 and with line 33-34 out of service.
and ∆P/∆γ have almost always larger values than ∆Q/∆r
and ∆Q/∆γ. Based on this fact, ∆P is considered to be a
better input signal than ∆Q. Hence, there are two suitable conditions. In this case, a re-tuning is necessary. An algorithm
loops remaining: the first one based on the feedback signal for detection of oscillation has been utilized to automate this
∆r and the second one based on the signal ∆γ. Since the procedure.
residue’s value for ∆r as feedback signal is bigger compare In case of the more expensive UPFC, the residue approach for
to ∆γ, only one transfer function is employed with ∆r as tuning of its POD controller gives in presented cases directly
the feedback signal. From Table II, the line 34-36 has the one set of parameters which works for a variety of conditions
largest residue for the transfer function ∆P/∆r and therefore and no re-tuning is necessary.
it would be the most effective location to apply the feedback
VI. ACKNOWLEDGEMENTS
control on ∆r variable. The corresponding transfer function
is: The first author, R. Sadikovic, gratefully acknowledges the

1

10s

0.094s + 1
2 support of ABB Switzerland Ltd for this project.
H(s) = 1.2 (12)
0.1s + 1 10s + 1 0.3764s + 1 R EFERENCES
where the lead-lag parameters were obtained according to (9). [1] M. E. Aboul-Ela, A. A. Salam, J. D. McCalley and A. A. Fouad,
Since the UPFC is more powerful than the TCSC, a set of ”Damping controller design for power system oscillations using global
POD parameters gives very satisfactory damping for variety signals”, IEEE Transactions on Power System, Vol 11, No. 2, May
1996, pp 767-773
of operating condition so that no re-tuning is necessary when [2] P. Korba, M. Larsson, C. Rehtanz, ”Detection of oscillations in power
N-1 criterion is considered. Fig. 12 and Fig. 13 show two such systems using Kalman fitlering techniques”, Proceedings of 2003 IEEE
cases. Conference on Control Applications, June 2003 Pages: 183 - 188 vol.1
[3] C. Rehtanz, M. Larsson, M. Zima, M. Kaba, J. Bertsch, System for
V. C ONCLUSION Wide Area Protection, Control and Optimisation based on Phasor Mea-
surements, Power Systems and Communication Systems Infrastructures
This paper presented a simple adaptive tuning method based for the Future, Beijing, 23. - 27. September, 2002.
on residue approach, applied to TCSC and UPFC. It is shown [4] P. M. Anderson and A. A. Fouad, Power System Control and Stability,
IEEE Press, 1994.
that in some cases the set of TCSC POD parameters can [5] P. Kundur, Power System Stability and control, McGraw Hill, New York,
not stabilize the power system under all admissible operating 1994.
[6] N. Yang, Q. Liu, J.D. McCalley, TCSC Controller Design for Damping
Interarea Oscillations, IEEE Transactions on Power System, Vol 13, No.
4, November 1998, pp 1304-1310
[7] R. Sadikovic, G. Andersson, P. Korba, ”Power Flow Control with FACTS
Devices”, World Automation Congres (WAC), Seville, 2004.
[8] R. Sadikovic, ”Damping controller design for power system oscillations”,
Internal report, ETH Zrich, December 2004.

VII. B IOGRAPHIES
Rusejla Sadikovic obtained Dipl. Ing. Degree and M.S. from the University
of Tuzla, Bosnia and Herzegovina in 1995 and 2001, respectively. Since 2002
she is PhD Student and research assistant at the Power System Laboratory
of Swiss Federal Institute of Technology (ETH) in Zurich.

Petr Korba received his M.Sc. degree in electrical engineering from

the Czech Technical University, Prague, Czech Republic, in 1995 and his
Ph.D. degree (with honours) from the University of Duisburg, Germany, in
2000. He was an invited scientist at the Delft University of Technology, the
Netherlands, and at the University of Manchester Institute of Science and
Technology (UMIST) in 1998 and 1999, respectively. He became a member
of staff at UMIST, Control Systems Centre, where he stayed until 2001.
He then joined ABB Switzerland Ltd. He is currently with ABB Corporate
Research. His research interests include model identification techniques,
robust and adaptive control theory and their industrial applications in power
systems.
Dr. Korba received the 2000 American Control Conference Best Paper Award.

Goran Andersson (M’86, SM’91. F’97) was born in Malm, Sweden. He

obtained his M.S. and Ph.D. degree from the University of Lund in 1975 and
1980, respectively. In 1980 he joined ASEA:s, now ABB, HVDC division in
Ludvika, Sweden, and in 1986 he was appointed full professor in electric
power systems at the Royal Institute of Technology (KTH), Stockholm,
Sweden. Since 2000 he is full professor in electric power systems at the Swiss
Federal Institute of Technology (ETH), Zrich, where he heads the powers
systems laboratory. His research interests are in power system analysis and
control, in particular power systems dynamics and issues involving HVDC
and other power electronics based equipment. He is a member of the Royal
Swedish Academy of Engineering Sciences and Royal Swedish Academy of
Sciences and a Fellow of IEEE.