1160

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 28, NO. 2, MAY 2013

Damping of Inter-Area Oscillations in Mixed

AC/DC Networks Using WAMS Based
Supplementary Controller
Robin Preece, Student Member, IEEE, Jovica V. Milanovi, Fellow, IEEE, Abddulaziz M. Almutairi, Member, IEEE,
and Ognjen Marjanovic, Member, IEEE

AbstractThe paper presents a supplementary VSC-HVDC

Power Oscillation Damping (POD) controller based on wide
area measurement signals (WAMS). The controller is designed
as Multi Input Single Output (MISO) using a Modal Linear
Quadratic Gaussian (MLQG) methodology in order to target
critical inter-area electromechanical modes. The approach has
been tested on a large (16 machine, 68 bus) test network incorporating parallel HVDC/AC transmission and has shown improved
damping compared to a traditional Power System Stabilizer
(PSS) based controller structure utilizing local signals. The design
process has incorporated the effects of wide area signal transmission delays. Variation in these signal delays and the complete loss
of signals has been also investigated to establish the robustness of
the WAMS based controller and its sensitivity to loss of signals.
Extension of the controller to incorporate reactive power modulation has been investigated, as has variation in available active
power modulation capacity. The proposed controller performance
has been assessed through small and large disturbance analysis.
Index TermsElectromechanical modes, Linear Quadratic
Gaussian (LQG) control, power oscillation damping, VSC-HVDC.

I. INTRODUCTION

LECTRICITY transmission networks of the future are

expected to incorporate large numbers of HVDC lines,
leading to many instances of HVDC operation in parallel with
AC lines. Worldwide interest in large renewable generation
sources, often either offshore or long distances from traditional
load centers is increasing. Moreover, in many cases planning
consent for new overhead lines is becoming increasingly
difficult to obtain, discouraging AC network reinforcements
as too lengthy and costly. HVDC systems with higher power
transfer capacities per line, feasible lengthy subsea installations
and no technical line length limitations provide an attractive
alternative [1]. In the U.K., this is especially true with two

Manuscript received January 09, 2012; revised February 28, 2012 and May
15, 2012; accepted June 30, 2012. Date of publication August 15, 2012; date of
current version April 18, 2013. This work was supported in part by the Engineering and Physical Sciences Research Council (EPSRC) and in part by National Grid Plc. Paper no. TPWRS-00030-2012.
R. Preece, J. V. Milanovi, and O. Marjoanovic are with the School of Electrical and Electronic Engineering, The University of Manchester, Manchester
M60 1QD, U.K. (e-mail: r.x.preece@gmail.com; milanovic@manchester.ac.
uk).
A. M. Almutairi is with the College of Technological Studies, PAAET,
Kuwait.
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TPWRS.2012.2207745

planned HVDC links to operate in parallel with the existing

grid. These links will help to facilitate the increased power
transfer from the north to the south of the country when large
renewable generation capacity is connected [2].
A further benefit of HVDC that is largely unused in practical
installations is that of power oscillation damping (POD). With
fast acting power electronics within the converter stations it is
possible to rapidly vary power flow through the HVDC line.
The potential for power injection control at non-generator buses
(where HVDC systems are typically installed) has attracted recent interest and it has been shown that HVDC systems can be
used with POD controllers to damp inter-area electromechanical oscillations [3][7].
This paper first presents a traditional Power System Stabilizer (PSS) based design for the supplementary HVDC POD
controller with local signal input. Following this a more robust
Wide Area Measurement Signals (WAMS) based controller
is designed using a targeted novel modal Linear Quadratic
Gaussian (MLQG) methodology.
LQG POD controllers have been applied to both centralized
generator control [3], [4] and fast acting FACTS devices such
as Thyristor Controlled Series Capacitors (TCSCs) [5][7]. The
paper shows that the novel MLQG controller design is effective with an HVDC application. This design approach utilizes
multiple global network signals subjected to transmission delays but controls only the HVDC line active power injection.
The MLQG approach enables targeted damping only of the critical inter-area system modes, leaving local modes unaffected.
Through this design, vastly improved post-disturbance system
stability is observed.
The simplicity and transparency of this design approach
distinguishes this controller from other existing multivariable
synthesis techniques. Controller tuning is completed simply
and effectively by applying nonzero weights only to those
targeted modes requiring supplementary damping. This is far
easier than the case of standard LQG, requiring participation
analysis and complex weightings, and
approaches, where
weighting functions must accurately model the uncertainties to
ensure the practical robustness of the controller.
II. MODAL LINEAR QUADRATIC GAUSSIAN CONTROL
The LQG control design is a cornerstone of modern optimal
control theory and its advantages led to widespread research
into it use in power system damping [3][7]. However, the design approach is rarely straightforward, especially within large
power systems where many generators participate in the critical

0885-8950/$31.00 2012 IEEE

PREECE et al.: DAMPING OF INTER-AREA OSCILLATIONS IN MIXED AC/DC NETWORKS

modes which require additional damping. In these situations,

the controller tuning process can become prohibitively complex.
Specifically, the issues arise surrounding the correct selection of
system states upon which to target controller action.
Participation factor analysis is required to identify the electromechanical states involved in targeted system modes. Weightings can then be assigned to these states. However, if these states
are involved in other targeted modes (or modes that do not require altering), the damping of these modes will also be affected,
sometimes adversely. This results in a complex and time consuming tuning process in which it is often not possible to obtain
exact target damping factors. These complexities and problems
can be overcome through the novel use of a modal representation of the control design problem discussed below.
Consider the following linearized state-space plant (power
system) model [8] described by (1) and (2):

1161

(3), and the final LQG feedback control law can be written as
(5):
(5)
where is an estimate of the states
filter, described by (6):

obtained using Kalman

(6)

where is a constant estimation error feedback matrix. The

optimal choice of is that which minimizes
. It is calculated by solving the ARE associated with the cost
function (7):
(7)

(1)
(2)
where w is process noise and v is measurement (sensor)
noise. They are usually assumed to be uncorrelated zero-mean
Gaussian stochastic processes with constant power spectral
density matrices
and
respectively. The LQR control
problem, in the modal formulation [9], is to devise a feedback
control law, which minimizes the cost function (3):

and
are tuned
In this paper, the weighting matrices
using the loop transfer recovery (LTR) procedure at plant input
[8]. The Kalman filter is synthesized such that the loop transfer
function
, where
is the plant transfer function, approaches the LQR loop transfer function
. The tuning parameters of Kalman filter are
calculated as in (8) and (9):

(3)

(8)
(9)

and are appropriately chosen weighting matrices

where
such that
and
, and
is
a real matrix which provides mapping between system modal
variables and state variables as in (4):
(4)
where modal variables are directly associated with system
modes (
where
). The real transformation matrix
is obtained using Real Schur Decomposition [10] and
relates to the matrix of right eigenvectors as
.
The weighting matrices
and are commonly constructed
as diagonal. Values of the diagonal elements of are set in order
to penalize the corresponding controllers outputs from high actions. Values of the diagonal elements of
are set in order
to penalize the corresponding modal variables when deviating
from their steady-state values. Each diagonal element in
is directly associated with a modal variable and hence with
the corresponding mode
. A higher value of modal weight
corresponds to a higher effort by the controller to stabilize the
corresponding mode. In order to focus on adding damping to the
modes of interest only, these modes will be given some weights
in
while the other modes weights are set to zero. In this
way, control effort of the designed LQR is directed towards the
modes of interest only, by shifting them to the left in the complex plane, while keeping locations of other modes unaltered.
The LQR controller gain is computed by solving the associated algebraic Riccati equation (ARE), based on cost function

where
and
refer to the nominal model, is any positive
definite matrix and is a constant.
Full recovery of robustness is achieved as
. Care is required as full recovery would lead to excessively high gains and
therefore deteriorates the nominal performance of the true noise
problem. For non-minimum phase systems, which is a common
case in power systems, only partial recovery can be achieved
[8].
The closed-loop dynamics of the LQG controller can be described by (10):
(10)
and the transfer function of the LQG controller, from
then given by (11):

to , is

(11)
The novel modal LQG (MLQG) controller provides many
benefits of the traditional LQG approach:
Participation factor analysis is no longer required as LQR
weightings directly correspond to system modes.
It is simple to address the damping of several modes by
providing non-zero weightings to all modes requiring additional damping.
Exact damping factors can be achieved if desired through
fine tuning of the weightings.

1162

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 28, NO. 2, MAY 2013

TABLE I
ACTIVE POWER IMPORTED INTO NETS FROM SURROUNDING AREAS
WITH NO HVDC LINK INSTALLED

Fig. 1. Injection model for an HVDC line in parallel with an AC line.

All untargeted modes are left completely unaffected by

the controller, allowing local controllers to maintain good
damping.
A full comparison between the traditional and modal LQG
controller designs is presented in [9].
III. HVDC MODELING
HVDC system modeling can range in complexity from full
detail models which include valve switching to simple models
which are only acceptable when the HVDC system is remote
and has no significant impact on the stability analysis [11]. This
research is concerned with electromechanical oscillations and
so the operation of the power electronics within the HVDC converter stations is neglected. As such, the converter stations are
represented simply as sinks and sources of active and reactive
power [12].
A. Injection Modeling

reactive power control reference setpoints are met instantly.

Converter station controllers are
or integral regulators
with clamped anti-windup. These controllers and the DC line
dynamics (modeled as a -section) are utilized to determine
the expected flow into and out of the VSC-HVDC system with
full description given in [14] and [16]. A common control
scheme is used with one converter maintaining the DC voltage
and the other regulating active power flow. Reactive power
control is independent at each converter station. Additional
signals,
and
, are available to vary the active and
reactive power reference setpoints and modulate power flow
for stabilizing purposes.
IV. TEST SYSTEM

The method employed to model the HVDC system is injection modeling [13]. The model complexity is determined by the
level of detail required for the studies being performed. Conceptually, the HVDC converter stations are replaced by equivalent
generator buses connected to the AC system across a reactance
, representing the converter transformer. Hence, the equivalent injection model for an HVDC line in parallel with an AC
transmission line is given by Fig. 1.
The voltage and angle at the equivalent generator buses are
varied to produce the desired power flow into and out of the
HVDC system as dictated by (12) and (13). This method can be
used to produce both voltage source conversion based HVDC
(VSC-HVDC) and line commutated conversion based HVDC
(LCC-HVDC) dynamic system models [14]:
(12)
(13)
B. VSC-HVDC Injection Model Development
The more recently operationally viable VSC-HVDC provides some benefits over its more established counterpart
LCC-HVDC, and is seeing an increase in popularity. Its
smaller footprint, decreased harmonic injection and reactive
power injection capabilities were previously compromised by
higher losses and lower operating capacities [15]. However
modular multi-level converters (MMC) have reduced these
losses to levels comparable with LCC systems and many
VSC-HVDC projects are currently under construction.
The internal current control loop of the VSC is modeled as
ideal for the system stability studies, meaning the active and

A 16-machine, 68-bus network is chosen as the test system

for the studies, shown in Fig. 2. This represents a reduced order
equivalent model of the New England Test System (NETS)
and the New York Power System (NYPS). Five separate areas
are present: NETS consisting of G1G9, NYPS consisting of
G10G13, and three further infeeds from neighboring areas
are represented separately by G14, G15 and G16. All generators are represented by full sixth order models. Generators
G1G8 are under slow DC excitation (IEEE-DC1A) while G9
is equipped with a fast acting static exciter (IEEE-ST1A) and
PSS. The remaining generators (G10G16) are under constant
manual excitation. Power system loads are modeled as constant
impedance. Full system details, generator and exciter parameters are given in [17] with PSS settings for G9 taken from [18].
All simulations are performed within the MATLAB/Simulink
environment making use of modified MATPOWER [19] to
perform initial load flows.
With loading as given in [17], the NYPS area is heavily
importing power from the surrounding areas due to an active
power demand of 8.57 GW but generation of just 5.86 GW.
Details of the active power import across inter-area ties are
given in Table I. The infeed from the G16 area (along two
lines) is largest, accounting for over half the total active power
import into NYPS.
A VSC-HVDC link is introduced between the NYPS and
G16 areas to support this power infeed. The link is installed
in parallel with the most heavily loaded tie line, from bus 18
to bus 50. Normal operating capacity for the HVDC link is selected as 400 MW. At this capacity, active power infeed from
the G16 area to NYPS through AC tielines is reduced compared with those given in Table I:
MW,
and
MW. Import from area G14 is also

PREECE et al.: DAMPING OF INTER-AREA OSCILLATIONS IN MIXED AC/DC NETWORKS

1163

Fig. 2. 16-machine, 68-bus test system. Separate areas (NETS, NYPS, G14, G15, G16) shown with inter-area ties highlighted. VSC-HVDC link shown.

TABLE II
INTER-AREA ELECTROMECHANICAL MODE DETAILS FOR TEST SYSTEM WITH STANDARD LOADING

slightly reduced with import from NETS largely unaffected. Details about the VSC-HVDC line parameters and controller gains
are included in the Appendix.
Small signal analysis of the linearized system including the
VSC-HVDC reveals that with standard loading, as given in
[17], four poorly damped oscillatory inter-area electro-mechanical modes are present which would benefit from improved
damping. These modes are detailed in Table II. All these
modes have damping factors, , below 5% which is considered
unsatisfactory in terms of control design objectives [18]. The
VSC-HVDC link is operating at 400 MW capacity with zero
reactive power injection
. All local electro-mechanical modes are adequately damped with
.
V. POD CONTROLLER DESIGN
The general control overview for the 16-machine, 68-bus
network is shown in Fig. 3. For all designed supplementary
HVDC POD controllers, output was limited to just one signal,
. This signal is sent to the converter station regulating
active power injection. In this study this is the inverter connected at bus 50. POD controller input signal transmission
delays are experienced only for wide area, or global, signals as
is discussed below.
A. SISO PSS Structure
1) Design Methodology: The initial design process for the
VSC-HVDC POD controller followed that of a conventional

Fig. 3. Test network control overview.

Fig. 4. PSS based SISO POD controller for HVDC.

PSS. This has been used previously with HVDC power modulation [20][23] and has been shown to be effective at modal
damping. The design consists of a washout filter, lead-lag blocks
and gain, as shown in Fig. 4. Controller parameters are dependent upon the modes of interest to be damped.

1164

TABLE III
RESIDUE ANGLES FOR MODES OF INTEREST FOR HVDC PSS POD DESIGN

The residue based tuning method is used [24]. First, the

angles of the residue values,
(of the open loop transfer
function between controller output and controller input), corresponding to the critical eigenvalues are assessed. In this case
the controller output will be the HVDC modulation reference,
, and the controller input is selected as a local signal,
the active power transfer into bus 50 from bus 18 through
the parallel AC transmission line,
. The phase compensation required by the PSS is then calculated simply as
(to shift the eigenvalue further into the
left half of the complex plane with no change in frequency). Parameters for the lead-lag blocks are then determined according
to the required phase compensation desired.
Shown in Table III are the residue angles and required PSS
compensation angles for the four poorly damped inter-area
modes. As the lead-lag block is tuned for a specific compensation angle at a specific frequency [24] it is clearly not possible
to optimally tune the PSS for multiple modes.
Tuning is carried out for the inter-area mode for which the
magnitude of the residue value suggests greatest modal controllability will occur given the controller input signal [11]. Residue
magnitude values are also shown in Table III, normalized for the
largest value. As Mode 1 is most controllable, tuning is carried
out for these modal characteristics at a standard VSC-HVDC
operating capacity of 400 MW. Final controller parameters are
given in the Appendix.
2) Small Signal Analysis: The improvement in modal
damping with the PSS based SISO POD controller installed
is shown in Fig. 6. As a local input signal is selected, it is assumed that there is no transport delay associated with receiving
the signal
. Similarly, it is assumed that the POD
controller is located at the VSC-HVDC converter station controlling active power regulation, and no output signal transport
delays are modeled.
Very slight improvement is seen in all inter-area modes,
though only Modes 1 and 2 cross the 5% damping threshold
with damping coefficient increasing to 7.08% and 5.44%,
respectively. All local modes remain damped with
.
The small signal analysis performed suggests that the PSSbased POD has limited influence upon the critical inter-area
electromechanical oscillations present within the test network.
This may be in part due to limited modal observability [11]
within the local input signal
. This is a problem that
can be overcome through use of global signals within a multiple
input control structure.
B. MISO Modal LQG Controller Design
To allow targeted damping of oscillatory electromechanical
modes, MLQG was selected as the controller design approach.

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 28, NO. 2, MAY 2013

Selection of inputs to the controller is determined through a

full modal observability assessment for the network [11]. The
number of required signals depends upon the number of critical modes requiring additional damping and the observability
of these modes within the available system signals. Networks
containing a large number of PMUs will be able to exploit these
available signals to gain a highly accurate representation of the
networks oscillatory nature following disturbances. Within this
study, active power flow through lines from bus 45 to bus 51
, bus 68 to bus 18
, bus 65
to bus 17
, and from bus 67 to bus 42
were determined to display greatest observability of
the critical Modes 1 to 4, respectively.
A MISO MLQG power oscillation damping controller was
designed, as described in Section II, for the standard operating
conditions with a VSC-HVDC operational capacity of 400 MW.
Controller inputs are as described above and controller output
was limited to
.
Often when performing LQG control design it is necessary to
perform initial model order reduction to avoid ill conditioning
when solving high order matrix Riccati equations [6]. This
problem was not experienced and so controller design was
carried out using the full linearized system model.
1) Incorporating Signal Delays: As remote signals are
assumed to be sent through pre-existing communications links
(and not dedicated signal transmission hardware), delays between the instant of measurement and the signal reaching the
controller are experienced. Dependent upon various factors,
including distance and communication protocols, these delays
can be in the range of a few hundreds of milliseconds [25]. In
the simulations performed, global network signals are assumed
to have associated transport delays of 500 ms. These signal delays are modeled as 2nd order Pad approximations [26] during
model linearization and controller design. As with the PSS
based POD controller, it is assumed that the POD controller
is located at the VSC-HVDC converter station and no output
signal transport delays are modeled.
The introduction of delays to the controller input signals affects the LTR process during the design of the LQG controller.
The degree of recovery achieved can be assessed through
comparison of the singular value plots for both
and
. With parameters ,
, , and
set to
fixed values as given in the Appendix, the LTR procedure was
completed for
, shown in Fig. 5. Increasing
beyond 100 provides no further discernable improvement in
recovery in the frequency range of interest. Fig. 5 demonstrates
that with the signal transport delays included, full recovery of
the robustness properties of
is not possible, even
for large . For the rest of the design process, was fixed
at 1, providing a compromise between optimal recovery of
robustness properties and unacceptably high filter gains.
2) Small Signal Analysis and Controller Reduction: Small
signal analysis on the closed loop system (of Fig. 3) demonstrates the improvements seen in damping of all critical
inter-area modes. This is shown in Fig. 6 where it can be
seen that all critical modes now achieve damping greater than
5%. Modes 1 to 4 are now damped with values of 19.39%,
15.93%, 14.09%, and 11.63%, respectively. This is not only
greatly improved over the no POD case but also over the

PREECE et al.: DAMPING OF INTER-AREA OSCILLATIONS IN MIXED AC/DC NETWORKS

1165

Fig. 7. For base case, settling times for NYPS inter-area AC infeeds.

Fig. 5. LTR process at plant input with varying values with delayed controller
input signals.

Fig. 8. For outage case, settling times for NYPS inter-area AC infeeds.

Fig. 6. Modal placement with no POD, PSS-based POD, full order MLQG
POD, and reduced order MLQG POD (dashed line signifies 5% damping).

modal placements for the PSS-based POD controller. With the

targeted modal damping from the MLQG controller affecting
solely the inter-area Modes 1 to 4, all local modes remain
unaffected and adequately damped with
.
The designed MISO MLQG controller as defined by (11) is
of 182nd order, equal to the full linearized system model order.
It is desirable to reduce this in order to decrease the online computational burden of the controller whilst still maintaining the
improved critical mode damping.
The balanced Schur reduction method [27] was applied to the
controller, implemented within MATLAB. The limiting factor
upon the level of reduction permitted was chosen as the degradation seen in the damping of critical modes through the use of
the reduced order controller. Allowable degradation was set at
5% of the full order controller values for each critical mode.
With this methodology, the MLQG POD controller was reduced
to 28th order. Small signal analysis of the closed loop system
with the reduced order controller provides values of 18.90%,
15.43%, 14.34%, and 11.70% for Modes 1 to 4, respectively,
also shown in Fig. 6.
The small signal analysis performed is dependent upon the
linearization of the nonlinear power network and elements such
as the signal transmission delays. Furthermore, the improvement in inter-area mode damping is ideal and based on the
LQR state feedback control eigenvalue placement (which assumes full state knowledge). The true performance and robustness of the designed controller is dependent upon the Kalman
filter state estimator and is, therefore, most readily assessed
through transient simulations.
VI. CONTROLLER TRANSIENT PERFORMANCE
A. Variation in System Operating Conditions
Large disturbance transient studies have been performed for
two cases. These are:

1) The base case, the operating point for which the controllers
have been designed. System loading and generation is
as previously described. The VSC-HVDC is operating at
400-MW capacity with zero reactive power output. All
lines are in service. The system is subjected to a 100-ms
self-clearing fault at bus 38 at a time of 0.5 s.
2) The outage case. Still with standard loading, the line between bus 18 and bus 49 is removed from service. This line
provided a path for some of the power flow from the G16
area to NYPS. As a consequence, power flow through the
line from bus 18 to bus 50 is increased. VSC-HVDC operational capacity is increased slightly to 450 MW to aid this
transmission, still with zero reactive power injection. The
system is then subjected to a 100-ms fault near to bus 1 on
the line from bus 1 to bus 30, at a time of 0.5 s, cleared by
disconnecting the line.
These transient studies were performed for the case with no
POD controller installed, with PSS-based POD, and with the reduced order MLQG POD including signal delays. VSC-HVDC
modulation was limited to 100 MW.
1) Transient Results and Discussion: Settling times were
recorded from the point of fault clearance to the time at which
the power deviation was within 1% of the steady state value
(or 1 MW, whichever was greatest). These times are shown
for the NYPS inter-area AC infeeds (as detailed in Table I) in
Figs. 7 and 8.
Looking first at the base case (Fig. 7), the inter-area infeed
power flows demonstrate the considerably improved modal
damping of the MLQG controller. All infeeds settle in less than
12 s, compared with 23 s with the PSS-based POD controller
installed, and 31 s with no damping controller.
With the outage case (Fig. 8), the settling times of the AC
NYPS infeeds show the robustness of the MISO MLQG controller to varying operating point. The improved performance
over the PSS-based controller is still pronounced with all ties
settling within 19 s (compared to 33 s for the PSS-based POD
and 40 s when no POD controller is used).
Plots of the active power injected at bus 9 from bus 8 through
the AC infeed and the active power injected at bus 50 by the
VSC-HVDC are shown in Figs. 9(a) and 10(a). These show that
despite the separation between the point of control (the VSCHVDC link between buses 18 and 50) and a relatively distant
inter-area tie (bus 8 to bus 9), damping is still vastly improved.

1166

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 28, NO. 2, MAY 2013

Fig. 11. Deterioration in damping of critical modes by MLQG controller with

increasing signal delay.

line outages with both the PSS-based POD and MLQG POD
controller, available in [28]. It should be noted that the MLQG
controller significantly outperformed the PSS-based controller
across wide ranging operating scenarios.
Fig. 9. For base case: (a) Active power injected at bus 9 (NYPS) from bus
8 (NETS), and (b) Active power injected at bus 50 by VSC-HVDC inverter
: fault occurs,
: fault cleared, PSS POD power
station.
: MLQG POD power modulation evident.
modulation evident,

Fig. 10. For outage case: (a) Active power injected at bus 9 (NYPS) from bus
8 (NETS), and (b) Active power injected at bus 50 by VSC-HVDC inverter
: fault occurs,
: fault cleared, PSS POD power
station.
: MLQG POD power modulation evident.
modulation evident,

This is still evident for the outage case, demonstrating the ability
of the MLQG controller to perform well as operating conditions
change.
The plots of the active power injected by the inverter,
Figs. 9(b) and 10(b), present the control action of the various
POD controllers. The forced modulation of active power flow
through the VSC-HVDC link is used to stabilize the network.
With no POD it can be seen that the VSC-HVDC returns to
its steady state power injection setpoint rapidly. This can also
be seen to occur with the MLQG POD controller following
the disturbance at 0.5 s (labeled t1). It is not until 500 ms after
the fault instant (at
) that the controller input signals
display a disturbance and the HVDC active power modulation
begins. The active power modulation of the PSS POD using
signals with no delay is evident as soon as the fault is cleared
(at
).
Further investigation has been made into the effects of
varying operating conditions including generator and key tie

B. Variation in Wide Area Signal Delay and Loss of Signals

The use of wide area measurements has been demonstrated as
enabling a better, more robust controller design. These signals
are becoming more prevalent in modern power systems, providing reliable real-time data which can improve many aspects
of system performance. However, these signals will often be
sent through pre-existing satellite communication links (as dedicated hard-wired links may prove prohibitively expensive) and
as such are potentially subject to increased delay or even complete loss. If faster communication channels (e.g., fiber optic
links) are available then the shorter associated delays should be
included in the controller design stage. However, controller performance will only improve as signal delays shorten [9].
The effects of signal latency and mitigation techniques for
use with WAMS based controllers has been a topic of much interest. Readers are directed towards [29][34] for comprehensive analyses of these effects and novel mitigation techniques
to improve controller performance in the presence of signal latency. For the novel MLQG controller, the effects of increased
signal delays and the complete loss of signals have been investigated in order to demonstrate the controllers robustness to these
problems.
The controllers local signal to the VSC-HVDC converter station is assumed to be hard-wired and not subject to these issues.
In addition to the results investigating the robustness of the controller to varying system operating conditions, the MLQG controller has also been tested to determine the robustness to variation in signal delay and complete loss of input signals. Fig. 11
shows the deterioration seen in the damping of all critical modes
as the delays are increased to 1100 ms for the base case operating point. There is a gradual degradation in controller performance as the signals are delayed for longer than the 500 ms
assumed during the design process.
Table IV details the maximum permissible delay tolerances
beyond which the MLQG controller will exhibit worse performance than with the PSS-based POD controller, or no POD controller, installed. These delays can increase to 766 ms (53.2%
higher than the already pessimistic 500 ms assumed) before the
controller displays worse performance than the PSS-based POD
controller taking local signals. With respect to the complete loss
of signals, this has been tested for the loss of up to two of the
four input signals. As the signals selected all contain some information about all the critical modes, there is some redundancy

PREECE et al.: DAMPING OF INTER-AREA OSCILLATIONS IN MIXED AC/DC NETWORKS

1167

TABLE IV
MLQG CONTROLLER SIGNAL DELAY TOLERANCES

TABLE V
DAMPING FACTOR OF CRITICAL MODES WITH LOSS OF INPUT SIGNALS

Fig. 13. Active power injected at bus 9 (NYPS) from bus 8 (NETS) for base
case with and without reactive power modulation.

C. Incorporating Reactive Power Modulation

Fig. 12. Active power injected at bus 9 (NYPS) from bus 8 (NETS) for base
and
signal failures.
case operating point with

within them. The damping of the four critical modes for the various cases is shown in Table V.
The shaded cells represent the cases when the modal damping
factor drops below that seen with the local PSS-based POD controller installed. It is evident, for example, that the loss of input
signal always results in the damping factor of Mode 3 being
heavily reduced to less than 5%. This signal was initially selected for the high observability of Mode 3, so this result is
not unexpected. Damping factors can be seen to still be higher
than with the PSS-based POD installed for the majority of cases
when signals fail.
Even with the (unlikely) loss of two input signals, the MLQG
controller can often maintain relatively high damping factors
on some modes. Looking at case 5 (highlighted in Table V),
even though damping of Mode 2 has dropped to lower than
PSS POD levels, the damping of the remaining low frequency
modes are still in the range of 8.87%9.60%. Due to this fact,
the transient performance of the controller is still highly competitive, with all infeeds settling within 16 s for the base case
scenario when signals and are lost. The oscillations present
on the tie line between buses 8 and 9 are shown in Fig. 12, the
robustness of the MLQG controller to the failure of wide area
signals is clearly visible with the oscillations quickly damped.

The studies presented have been concerned with the modulation of active power flow through the parallel VSC-HVDC
link in order to stabilize post-disturbance system oscillations.
One of the stated advantages of using VSC-HVDC over classic
LCC-HVDC is the availability of four quadrant operation of
the converters, allowing the generation or consumption of reactive power at each converter station.
The MLQG design approach can be readily extended to include multiple controller outputs into a MIMO structure. In addition to the single
signal,
signals at each converter station were incorporated and the MLQG controller design was completed once more.
Fig. 13 shows the oscillations present on the tie line between
buses 8 and 9 for the base case operating scenario both with, and
without, reactive power modulation included. It can be seen that
the additional reactive power modulation provides very limited
improvement in the system response (just 0.4 s improvement in
settling time).
With little benefit achieved, it is unlikely that VSC-HVDC
reactive power output would be modulated for power oscillation
damping purposes. Perhaps more probable would be the use of
fast reactive power modulation to ensure quickly stabilized bus
voltages at the points of interconnection with the VSC-HVDC
link during the post-disturbance oscillations.
D. Variation in Modulation Capacity
The simulations presented within this paper have assumed
a generous allowance of 100 MW for power oscillation
damping (equating to 25% of the VSC-HVDC link operating
capacity for the base case scenario). This modulation capacity
is the same for both designed POD controllers.
In a practical installation, the limit of available modulation
capacity will be determined by the system operator. The benefits of reserving this capacity for modulating purposes following
system disturbances must be compared with the costs of reducing the power transfer capability through the HVDC link.
Converter ratings will set the upper bound on possible power
transfer. As the required modulations in active power are relatively slow (typically about 1 Hz), DC voltage limit violations
should not be an issue provided the converter regulating DC
voltage has been designed to be suitably fast.
The effect of limiting the modulation capacity (from the previously considered 25%) to 10% and 5% has been investigated.
Fig. 14 shows the oscillations between buses 8 and 9 and the
controlled variations in the power injected by the inverter station
for the base case with varying limits on modulation capacity

1168

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 28, NO. 2, MAY 2013

Fig. 14. (a) Active power injected at bus 9 (NYPS) from bus 8 (NETS) and (b)
Active power injected at bus 50 by VSC-HVDC inverter station for base case
operating point with differing modulation capacity limits.

with the initially designed MLQG controller (varying only active power injection).
It can be seen that the settling time for this line increases
slightly as less capacity is reserved for POD action. The same
result is true of the PSS-based controller. With the exception
of the line 1850 (which always sees improved settling times
with the PSS POD due to the local signal selection), restricting
the capacity reserved for POD to just 5% with the MLQG controller still results in improved settling times over the PSS-based
POD controller operating with 25% modulation capacity. For
the MLQG controller 10% modulation capacity results in key
tie line settling times increasing by 1.92.7 s (to a maximum of
14.1 s); and 5% modulation capacity results in key tie line settling times increasing by 2.74.7 s (to a maximum of 16.1 s);
when compared with the initial 25% modulation capacity.
An idea of the likely availability of this modulation capacity
can be sourced from the publicly available data on the usage
of the 1 GW Britned HVDC link between July and December
2011 [35]. This data is taken for an HVDC link which does not
reserve capacity for modulation. During this six month period
(ignoring periods with no power transmission): 73.8% of the
time at least 5% link capacity was spare; 70.6% of the time at
least 10% link capacity was spare; and 63.1% of the time at least
25% link capacity was spare.
It is clear that there will be periods when large amounts of
HVDC link capacity may be available for active power modulation for system stabilizing purposes. At these times it would
be advisable to more fully exploit the damping capabilities of
the HVDC link as higher modulation capacities result in faster
system settling times. This may require flexible modulation
limits dependent upon system operating conditions.
VII. CONCLUSIONS
A MISO Modal LQG power oscillation damping controller
for VSC-HVDC lines has been shown to be effective at
damping inter-area electromechanical oscillations within a
large heavily meshed network. Furthermore, it has been shown
that even when accounting for transmission delays on wide
area controller input signals, the MISO controller is able to

vastly outperform a standard PSS based SISO POD controller

dependent upon local signals. Furthermore, the effects of
variation in the transmission delays of the wide area signals
the MLQG controller receives has been demonstrated. Not
only can the controller tolerate delays over 50% longer than
designed for, but it can even continue to outperform the PSS
based controller with the loss of half of its wide area inputs. It
has been shown that the addition of reactive power modulation
at each VSC-HVDC converter station offers little benefit with
respect to improved system settling times. The damping of
power oscillations is dominated by the active power modulation
through the HVDC link.
Furthermore, the effects of limiting the capacity available for
active power modulation have been demonstrated. The MLQG
controller performance was reduced as the capacity reserved for
POD controller action is also reduced; however the controller
maintained superior performance compared with the PSS-based
controller. Analysis of operational data for the Britned link
shows the availability of modulation capacity and suggests
that use of flexible controller limits may be advantageous in
fully exploiting the VSC-HVDC links ability to stabilize the
network. This paper has shown that active power modulation of
HVDC lines is effective at damping multiple inter-area modes
within a large, heavily meshed network. Also, the MLQG
control methodology has been shown to be implementable with
a VSC-HVDC application. The targeted damping of critical
electromechanical oscillatory modes afforded by the MLQG
POD controller design may be of particular interest in large
power systems where very selective additional damping is
desired.
Finally, It should be pointed out that although use of an LQG
controller synthesis approach cannot intrinsically guarantee robustness properties [36], robust controller performance can still
be achieved. Conversely, the use of controller synthesis techniques that guarantee the robustness of the controller can only
ensure this robustness within the bounds of the defined uncertainties. If these uncertainties are poorly formalized then the
controllers practical efficacy may not match its intended mathematical performance. In either case, the performance and stability of the final controller must be thoroughly assessed through
multiple means (including nonlinear simulations) in order to establish its true robustness.
APPENDIX
HVDC SYSTEM AND CONTROLLER DATA
VSC-HVDC Line Parameters (on 600-MVA HVDC base):

VSC-HVDC Controller Parameters:

SISO POD Control Parameters (on 100-MVA base):

Fixed Parameters during LTR Tuning:

PREECE et al.: DAMPING OF INTER-AREA OSCILLATIONS IN MIXED AC/DC NETWORKS

REFERENCES
[1] J. Arrillaga, Y. H. Liu, and N. R. Watson, Flexible Power Transmission: The HVDC Options. Chichester, U.K.: Wiley, 2007.
[2] National Grid Electricity Transmission plc, Offshore Development Information Statement, Sep. 2010. [Online]. Available: http://www.nationalgrid.com.
[3] D. Dotta, A. S. e Silva, and I. C. Decker, Wide-area measurementsbased two-level control design considering signal transmission delay,
IEEE Trans. Power Syst., vol. 24, pp. 208216, 2009.
[4] T. Michigami, M. Terasaki, N. Sasazima, K. Hayashi, and T. Okamoto,
Development of a new adaptive LQG system generator for high-speed
damping control techniques of power system oscillation, Elect. Eng.
Jpn., vol. 142, pp. 3040, 2003.
[5] A. C. Zolotas, B. Chaudhuri, I. M. Jaimoukha, and P. Korba, A study
on LQG/LTR control for damping inter-area oscillations in power systems, IEEE Trans. Control Syst. Technol., vol. 15, pp. 151160, 2007.
[6] K. M. Son and J. K. Park, On the robust LQG control of TCSC for
damping power system oscillations, IEEE Trans. Power Syst., vol. 15,
pp. 13061312, 2000.
[7] A. M. D. Ferreira, J. A. L. Barreiros, J. W. Barra, and J. R.
Brito-de-Souza, A robust adaptive LQG/LTR TCSC controller
applied to damp power system oscillations, Elect. Power Syst. Res.,
vol. 77, pp. 956964, 2007.
[8] S. Skogestad and I. Postlethwaite, Multivariable Feedback Control:
Analysis and Design. Chichester, U.K.: Wiley, 1996.
[9] A. Almutairi, Enhancement of power system stability using wide area
measurement system based damping controller, Ph.D. dissertation,
Univ. Manchester, Manchester, U.K., 2010.
[10] G. H. Golub and C. F. V. Loan, Matrix Computations. Baltimore,
MD: The John Hopkins Univ. Press, 1989.
[11] P. Kundur, Power System Stability and Control. London, U.K.: McGraw-Hill, 1994.
[12] P. E. Bjorklund, K. Srivastava, and W. Quaintance, Hvdc light modeling for dynamic performance analysis, in Proc. PSCE 06, 2006.
[13] H. F. Latorre, M. Ghandhari, and L. Sder, Active and reactive
power control of a VSC-HVdc, Elect. Power Syst. Res., vol. 78, pp.
17561763, 2008.
[14] R. Preece and J. V. Milanovic, Comparison of dynamic performance
of meshed networks with different types of HVDC lines, in Proc. IET
ACDC, London, U.K., 2010.
[15] N. Flourentzou, V. G. Agelidis, and G. D. Demetriades, VSC-based
HVDC power transmission systems: An overview, IEEE Trans.
Power Electron., vol. 24, pp. 592602, 2009.
[16] R. Preece, A. M. Almutairi, O. Marjanovic, and J. V. Milanovic,
Damping of electromechnical oscillations by VSC-HVDC active
power modulation with supplementary WAMS based modal LQG
controller, in Proc. IEEE PES General Meeting, Detroit, MI, 2011.
[17] B. Pal and B. Chaudhuri, Robust Control in Power Systems. New
York: Springer, 2005.
[18] G. Rogers, Power System Oscillations. Norwell, MA: Kluwer, 2000.
[19] R. D. Zimmerman, C. E. Murillo-Sanchez, and R. J. Thomas, MATPOWER: Steady-state operations, planning, and analysis tools for
power systems research and education, IEEE Trans. Power Syst., vol.
26, pp. 1219, 2011.
[20] C. Zheng, X. Zhou, and L. Ruomei, Dynamic modeling and transient
simulation for VSC based HVDC in multi-machine system, in Proc.
PowerCon 06, 2006.
[21] Y. Pipelzadeh, B. Chaudhuri, and T. C. Green, Wide-area power oscillation damping control through HVDC: A case study on Australian
equivalent system, in Proc. IEEE PES General Meeting, 2010.
[22] H. Jingbo, L. Chao, W. Xiaochen, W. Jingtao, and T. S. Bi, Design
and experiment of heuristic adaptive HVDC supplementary damping
controller based on online Prony analysis, in Proc. IEEE Power Engineering Society General Meeting, 2007.
[23] D. V. Hertem, R. Eriksson, L. Soder, and M. Ghandhari, Coordination
of multiple power flow controlling devices in transmission systems,
in Proc. IET ACDC, London, U.K., 2010.
[24] J. V. Milanovic and S. K. Yee, Roadmap for tuning power system controllers, in Proc. 3rd IASTED Int. Conf. Power and Energy Systems,
Marbella, Spain, Sep. 24, 2003.
[25] R. Majumder, B. Chaudhuri, and B. C. Pal, Implementation and test
results of a wide-area measurement-based controller for damping interarea oscillations considering signal-transmission delay, IET Gen.,
Transm., Distrib., vol. 1, pp. 17, 2007.

1169

[26] J. Lam, Model reduction of delay systems using Pade approximants,

Int. J. Control, vol. 57, pp. 377391, 1993.
[27] M. G. Safonov and R. Y. Chiang, A Schur method for balanced-truncation model reduction, IEEE Trans. Autom. Control, vol. 34, pp.
729733, 1989.
[28] R. Preece, A. M. Almutairi, O. Marjanovic, and J. V. Milanovic, Effectiveness of a supplementary MLQG power oscillation damping controller installed at an HVDC line within a meshed network, in Proc.
Cigre Int. Symp.: The Electric Power System of the Future, Bologna,
Italy, 2011.
[29] J. W. Stahlhut, T. J. Browne, G. T. Heydt, and V. Vittal, Latency
viewed as a stochastic process and its impact on wide area power
system control signals, IEEE Trans. Power Syst., vol. 23, pp. 8491,
2008.
[30] W. Hongxia, K. S. Tsakalis, and G. T. Heydt, Evaluation of time
delay effects to wide-area power system stabilizer design, IEEE Trans.
Power Syst., vol. 19, pp. 19351941, 2004.
[31] B. Mirkin and P. O. Gutman, Robust output-feedback model reference
adaptive control of SISO plants with multiple uncertain, time-varying
state delays, IEEE Trans. Autom. Control, vol. 53, pp. 24142419,
2008.
[32] B. Chaudhuri, R. Majumder, and B. C. Pal, Wide-area measurementbased stabilizing control of power system considering signal transmission delay, IEEE Trans. Power Syst., vol. 19, pp. 19711979, 2004.
[33] R. Majumder, B. Chaudhuri, B. C. Pal, and Z. Qing-Chang, A unified
Smith predictor approach for power system damping control design
using remote signals, IEEE Trans. Control Syst. Technol., vol. 13, pp.
10631068, 2005.
[34] N. R. Chaudhuri, S. Ray, R. Majumder, and B. Chaudhuri, A new approach to continuous latency compensation with adaptive phasor power
oscillation damping controller (POD), IEEE Trans. Power Syst., vol.
25, pp. 939946, 2010.
[35] National Grid Transmission, Metered Half-Hourly Electricty
Demands: JulyDecember 2011, 2011. [Online]. Available:
http://www.nationalgrid.com.
[36] J. Doyle, Guaranteed margins for LQG regulators, IEEE Trans.
Autom. Control, vol. 23, pp. 756757, 1978.

Robin Preece (S10) received the B.Eng degree in electrical and electronic engineering in 2009 from the University of Manchester, Manchester, U.K., where
he is currently pursuing the Ph.D. degree.

Jovica V. Milanovi (M95SM98F10) received the Dipl.Ing. and M.Sc.

degrees from the University of Belgrade, Yugoslavia, the Ph.D. degree from
the University of Newcastle, Australia, and his Higher Doctorate (D.Sc. degree) from The University of Manchester, Manchester, U.K., all in electrical
engineering.
Currently, he is a Professor of electrical power engineering and Director of
External Affairs in the School of Electrical and Electronic Engineering at The
University of Manchester (formerly UMIST), Visiting Professor at the University of Novi Sad, Novi Sad, Serbia, and Conjoint Professor at University of
Newcastle, Newcastle, Australia.

Abddulaziz M. Almutairi (S06M11) received the B.Sc. degree from Kuwait

University, Kuwait, the M.Sc. degree from the University of North Carolina at
Charlotte, and the Ph.D. degree from the University of Manchester, Manchester,
U.K., all in electrical engineering.
Currently he is an Assistant Professor of electrical power engineering in the
College of Technological Studies, PAAET, Kuwait.

Ognjen Marjanovic (M08) received the First Class honors degree from the
Department of Electrical and Electronic Engineering, Victoria University of
Manchester, U.K., and the Ph.D. degree from the School of Engineering, Victoria University of Manchester, U.K.
Currently he is a Lecturer in the School of Electrical and Electronic Engineering at The University of Manchester, U.K.