Real-time implementation of a measurement-based adaptive wide-area control system considering communication delays

Real-time implementation of a measurement-based
adaptive wide-area control system considering
communication delays
S. Ray and G.K. Venayagamoorthy
Abstract: A wide-area control system (WACS) uses wide-area measurement signals to provide
auxiliary stabilising controls to power system devices. An adaptive WACS has been designed to
provide damping control signals to the excitations of generators. The delays in signal transmission
and the reliability of the communication network is a major concern with wide-area measurement-
based control. The adaptive WACS is designed to compensate for a wide range of communication
delays and provide robust damping to mitigate system oscillations. A single simultaneous recurrent
neural network is used in the realisation of the adaptive WACS for both identification and control
of the power system. The WACS has been implemented on a digital signal processor and its per-
formance is evaluated on a power system implemented on a real-time platform – the real-time
digital simulator. The additional damping provided by the WACS is enumerated using Prony
analysis.
1 Introduction
Large interconnected power systems are the basis of modern
life. The number of bulk power exchanges over long dis-
tances has increased as a consequence of deregulation of
the electric power industry. Several measures have been
taken in recent decades to provide better stability and
reliability to stressed systems. Slow-mode oscillations are
among the major concerns for stability of the entire power
system. Thus, local compensators such as power system
stabiliser (PSS) and flexible alternating current transmission
systems (FACTS) are installed on different sections of
power systems to provide damping.
Local controllers have good performance when local
measurements provide all information about the dynamics
of disturbances. But, if there are adverse interactions
between multiple adjacent areas of the power system, a
wide-area-based measurement provides better stabilising
control [1–5]. A wide-area control system (WACS) coordi-
nates the actions of distributed agents using supervisory
control and data acquisition (SCADA), phasor measurement
unit or other sources that provide wide-area dynamic infor-
mation [6–9]. The WACS receives information/data from
different areas in the power system and, based on some pre-
defined objective function, sends appropriate control signals
to distributed agents for enhancing system’s dynamic per-
formance [1–9]. The design and implementation aspects
of a SCADA-based wide-area monitoring and control
system is explained in [7, 9–10].
Multiple-linear model-based adaptive and hierarchical
wide-area control for damping post-disturbance oscillations
have been reported [11, 12]. The wide-area measurement-
based control suffers from signal transmission delays [13].
Linear matrix inequality (LMI), gain scheduling and
H1
-based WACS have been used to provide effective
auxiliary damping control accounting signal transmission
delays [14–17]. Feedback delays ranging from 50 ms to
1.0 s have been compensated using different classical tech-
niques for PSSs [14–16], FACTS devices such as static var
compensators and thyristor controlled series compensators
[17]. In [18], the implementation of a similar WACS is
reported on a real-time platform (RedHawk RT-Linux
operating system on a dual-Xeon processor) with a time
step of 1–10 ms and without considering any signal trans-
mission delays. All these designs require a mathematical
model of the system, which is difficult to implement in
practice. An alternative solution is a measurement-based
design.
In a recent study, an intelligent and adaptive modern
power grid has been envisioned with intelligence built-in
controllers, that will provide fast and accurate action
during contingencies [19]. Neural networks, known also
as intelligent agents, are universal approximators that are
able to identify and control multiple-input–multiple-output
(MIMO) time-varying systems like turbogenerators based
on input–output measurements [20, 21].
This paper presents the design and real-time implemen-
tation of an intelligent adaptive wide-area controller based
on measurements. The primary contributions of this paper
are:
† design of an adaptive WACS to provide auxiliary
damping to generators through their excitations;
† compensation of transmission delays ranging from 0 to
1.4 s in measured and control signals;
† the use of a single simultaneous recurrent neural network
(SRN) for both identification and control functions for the
WACS design and
† demonstration of the real-time implementation of the
proposed WACS on a digital signal processor (DSP) inter-
faced to the real-time digital simulator (RTDS).
# The Institution of Engineering and Technology 2008
doi:10.1049/iet-gtd:20070027
Paper first received 17th January and in revised form 7th June 2007
The authors are with Real-Time Power and Intelligent Systems (RTPIS)
Laboratory, University of Missouri-Rolla, Rolla, MO 65409, USA
E-mail: ganeshv@ece.umr.edu
IET Gener. Transm. Distrib., 2008, 2, (1), pp. 62–7062

2 Multi-machine power system with WACS
A power system is a complex network with thousands of
buses, several hundreds of generators and interactions
between multiple areas with several inter-area modes of
oscillations. The two-area power system [22, 23] of Fig. 1
is a test system, commonly used to show the effectiveness
of controllers in damping slow-mode oscillations. It is
important to demonstrate that the real-time implementation
of the adaptive WACS is practical; therefore this system is
chosen because of the limitation of the existing real-time
implementation platform (RTDS) in the Real-Time Power
and Intelligent Systems laboratory, but the concept is appli-
cable to any power system. The RTDS simulator platform is
explained in Section 4 [24].
A two-area system with the WACS shown in Fig. 1 con-
sists of two fully symmetrical areas linked together by two
transmission lines. Each area is equipped with two identical
synchronous generators rated 20 kV/900 MVA. All the
generators are equipped with identical speed governors
and turbines, AVRs and exciters (Fig. 2). The switch S1
(Fig. 2) can be used to provide auxiliary control signals to
the generators. The switch S2 (Fig. 2) is used to add the
PSS signal to the excitation system. Generators G1 and
G3 are only equipped with PSSs. Loads are represented as
constant impedances and split between the two areas in
such a way that Area 1 transfers 413 MW to Area
2. Three electromechanical modes of oscillation are
present in this system: two inter-plant/intra-area modes,
one in each area, and one inter-area low-frequency mode
[23]. The parameters of the two-area system are given in
[23].
3 Adaptive WACS
The WACS actions are based on remote signal measure-
ments from different areas. These remote signals are
exchanged over communication channels that have trans-
mission delays. This results in late arrival of the measured
signals by t1 s at the wide-area control station. Thus, the
actions taken by the WACS based on these signals are no
longer valid for the current state of the system. Similarly,
the control signals of the WACS communicated to the
respective areas/generators also arrive late by t2 s. Hence,
the stabilising control signal generated by the WACS is
out of phase and it adversely affects the stability of the
system. As mentioned in Section 1, the WACS design is
needed to provide stabilising control despite a substantial
delay in the communication of both measurement and
control signals. In this study, it is assumed that the WACS
can be located in one of the areas to eliminate delays in at
least one set of measurement and control signals. For the
two-area system in Fig. 1, the WACS is located in Area
1. Thus, the delays are in communicating generator G3’s
speed deviation signal to the WACS, and the WACS com-
municating the damping control signal to generator G3 in
Area 2. The objective of the WACS is to provide satisfac-
tory damping signals for a wide range of communication
delays. According to previous papers [14–17], the delay
in communication channels and in sensor measurements
are typically in the range of 50 ms to 1.0 s. For additional
delays, the signal is considered to be missing. Hence,
taking the maximum delay, the WACS has been designed
for a maximum delay of 1.0 s though, during testing,
satisfactory performance was achieved up to a delay of
1.4 s. The WACS design is based on a single SRN [25].
The SRN is used to provide an accurate representation of
the system dynamics at all time instances and sub-
sequently to generate appropriate control signals to damp
oscillations.
3.1 Simultaneous recurrent neural network
SRNs are robust and ideal for multi-step predictions. The
SRN is a dynamic or recurrent structure in which the
outputs depend on the inputs to the SRN and partially or
fully on the outputs of the different layers of the SRN.
The difference between a conventional recurrent neural
network (RNN) and SRN is that the feedbacks are time
delayed in RNNs whereas instantaneous in SRNs. In the
implementation of the SRN, much higher sampling rates
are used to emulate instantaneous feedback. To implement
an SRN in real-time, the simultaneous recurrences have to
be done multiple times within one sampling period of the
system measurements. Hence, for ten simultaneous recur-
rences, the outputs of the SRN hidden layer need to be cal-
culated 10 times and fed back as inputs to the hidden
neurons. Hence, to achieve this methodology, the SRN cal-
culated is 10 times per sampling period of 0.1 s; thereby, the
internal sampling rate used for the SRN is 100 Hz. The
output of a typical SRN is given by (1)
y ¼ W2 : d (1)
where
d ¼
1
1 þ eÀa
(2)
and a¯ ¼ W1X¯ is the input vector to the hidden layer, X¯ ¼ [x,
1, d]T
the input vector, d¯ the output of the hidden layer, y¯ the
output vector, W1 the input weight matrix and W2 is the
Fig. 1 Two-area power system with a WACS considering signal
transmission delays
Fig. 2 Block diagram of the AVR-exciter model
IET Gener. Transm. Distrib., Vol. 2, No. 1, January 2008 63

output weight matrix. More details on the SRNs can be
found in [25, 26].
3.2 Training of WACS
The design of the adaptive WACS is based on an adaptive
control method [27, 28]. Owing to the nonlinearity and
time-varying nature of the power system, this control
method requires a model/identifier to represent the power
system dynamics accurately from moment to moment thus
requiring online training. In this paper, a single SRN is
used for both identification and control in the adaptive
WACS design. The SRN operates in the forward mode as
an identifier and in the reverse mode as a controller. It
should be noted that the WACS parameters may not
always be updated, provided it is trained for steady state
as well as transients for different operating conditions of
the system.
To minimise the adverse effects of signal transmission
delays from the wide-area measurements, the WACS (in
the forward mode) is trained to predict the speed deviation
of the generators several steps ahead, ten steps in this paper.
As in the classical methods for power system identification
[29], the WACS is initially trained with small amplitude
pseudo-random binary signals (PRBS) (up to +10%) that
are added to the excitation inputs of generators G1 and
G3 (with switch S1 in Fig. 2 in position 1). The PRBS
signal excites all possible dynamics of the power system.
During this initial training period, the SRN is trained with
higher learning rate (0.06) for 40 min at the sampling rate
of 10 Hz. After the weights of the SRN stabilise, the SRN
training is continued with a small learning rate of 0.001
while in closed loop operations. During the online oper-
ation, the WACS outputs are applied to the excitation
systems of G1 and G3 (switch S1 in Fig. 2 in position 2)
and the online training of the SRN can be continued with
a small learning rate. Online training means that the SRN
is trained at each sampling instance using the signals
measured from the system on-the-fly. In practical appli-
cations, PRBS signals can be substituted with different
low signal-to-noise ratio probing signals [29]. The
ambient measurements exhibit small oscillations in the
system because of small changes in generation, loads and
operating conditions, which can also be utilised to identify
the modes of the system [30]. In most of the neural network-
based identification, a probing signal-based excitation is
preferred to other methods if possible [20]. To investigate
the ambient measurement-based training technique,
random load changes of +10% of total installed load is
also carried out to train the controller, and corresponding
results are given in case study 5. The input layer of the
SRN consists of 41 inputs [measured speed deviations
Dv1 and Dv3, and exciter input (see Appendix) voltages
u1 and u3 of generators G1 and G3, respectively, received
by WACS at time t, (t 2 DT ), . . . , (t 2 T ) and a bias
term] in which DT is the sampling period. The hidden and
output layers consist of ten sigmoidal and two linear
neurons, respectively. The outputs of the WACS in the
forward mode are estimated speed deviations
Dv1(t þ 10DT) and Dv3(t þ 10DT ) of the generators G1
and G3, respectively, ten sampling periods ahead. Pruning
techniques are used to obtain the minimum number of
hidden neurons for the SRN. The complete SRN is com-
posed of input weight W1 [10 Â 51], output weight W2
[2 Â 10], input vector X [51 Â 1], hidden layer output
vector d [10 Â 1] and output vector y [2 Â 1].
In practical systems, there may be several individual gen-
erators connected to one bus or distributed in one area, but
as the generators within one area are coherent, the measure-
ments from the dominant generator can be used for inputs to
the WACS. The WACS output can be fed either to the exci-
tation system of the dominant generator contributing to the
inter-area mode or to all the generators’ excitations within
one area depending on the feasibility and cost of implemen-
tation. The modal observability of measured states is critical
for damping inter-area oscillations. The schematic for the
WACS inputs, outputs and training is illustrated in Fig. 3.
The objective of the training is to minimise the square of
the error Em(t) in (3) by using (4) and (5) where a
(0 , a , 1) is the learning rate. The learning rate during
initial training used in this paper is 0.06 whereas during con-
tinuous online training under operation, it is reduced to
0.001.
Em(t) ¼ Dv(t) À D ^v(t) (3)
w1(t þ 1) ¼ W1(t) þ a
@Em(t)
@W1(t)
(4)
W2(t þ 1) ¼ W2(t) þ a
@Em(t)
@W2(t)
(5)
3.3 Operation of WACS
The control signals (WACS operating in the reverse mode)
are generated using the difference between the outputs of a
desired response predictor (DRP) and the outputs of the
WACS in the forward mode, as shown in Fig. 4. The two
control signals, u1 and u3, are the auxiliary signals (7)
added to generators G1 and G3’s excitation systems,
respectively, using the switch S1 in position 2 (Fig. 2).
The low-pass filter is used to eliminate high-frequency
Fig. 4 Operation mode of the WACS
Fig. 3 WACS training schematic (TDL, time-delayed lines)
IET Gener. Transm. Distrib., Vol. 2, No. 1, January 200864

components and smooth out the control signals generated
with a sampling frequency (10 Hz). The desired responses
for the speed deviations ten sampling periods ahead,
Dv1des(t þ 10DT ) and Dv3des(t-t1 þ 10DT ), are taken to
be a weighted summation of the past speed deviations.
This desired trajectory provides a step-by-step transition
in getting the speed deviations to zero as reported in [21,
27, 28]. The DRP represents a desired linear model,
which satisfies required specifications of damping perform-
ance. The control actions are generated such that it cancels
the nonlinearity of the original system and tracks the
response of the desired linear model, thus providing
damping. The DRP is given in (6)
Dv1des(t þ 10DT) ¼ À
Xi¼N
i¼0
AiD ^v1(t À iDT)
(6)
Dv3des(t À t1 þ 10DT) ¼ À
Xi¼N
i¼0
AiD ^v3(t À t1 À iDT)
Dvˆ 1 (t) and Dvˆ 3 (t) are the outputs of the neuro-identifier or
model at time t. In this DRP design, the parameters are
chosen as N ¼ 3, A0 ¼ 3.04, A1 ¼ 23.883, A2 ¼ 2.422
and A3 ¼ 20.6416. The dominant poles of the desired
linear model in the discrete domain are located at
(0.8511 þ 0.3557i), (0.851120.3557i), (0.6691 þ 0.5536i)
and (0.669167 2 0.5536i). As the poles are within the unit
circle, the system is asymptotically stable. The discrete
desired pole locations are found by converting the desired
continuous system to discrete using 0.1 s as the sampling
frequency. The desired continuous linear model is chosen
to provide a damping of 0.2 for both inter-area 0.63 Hz
and intra-area 1.1 Hz modes. The error between the
desired and the actual speed deviations are backpropagated
through the WACS network and the differentiation of the
errors with respect to the control inputs (Vr) of the WACS
yields desired control signals, u1(t) and u3(t) as expressed
in (7)
u1(t) ¼
@Edr1(t þ 10)
@Vr(t)
u3(t) ¼
@Edr3(t À t1 þ 10)
@Vr(t)
(7)
3.4 Implementation of wide area controller
The implementation of the controller is done in the reverse
mode using the error backpropagation method used to train
neural networks. The error between the DRP output and the
WACS output, given by (8), is backpropagated from the
output layer to the hidden layer, and then from the hidden
layer to the input layer of the SRN, as given in (9) and
(10) to generate the control signals given in (11).
err1 ¼ Dv1des(t þ 10DT) À D ^v1(t þ 10DT)
err2 ¼ Dv3des(t þ 10DT) À D ^v3(t þ 10DT)
(8)
edi ¼
@err
@di
¼
X2
j¼1
W2jierri; i ¼ 1, . . . , N (9)
eai ¼
@err
@ai
¼ di(1 À di)edi; i ¼ 1, . . . ; N (10)
u1 ¼
XN
j¼1
W139j
eaj
u3 ¼
XN
j¼1
W140j
eaj
(11)
were N is the number of neurons in the hidden layer, ed the
derivatives of error in the output layer with respect to output
of the hidden layer and ea the derivative of error in the
output layer with respect to the input of the hidden layer.
The generated control signal is applied to the excitation of
generators G1 and G3, and the SRN is continuously
trained with past inputs and outputs with a small learning
rate (typically 0.001).
4 Real-time implementation platform
Owing to its complexity and expensive nature, it is very dif-
ficult to test new control methods and algorithms on a real-
world power system. The proposed WACS is implemented
on a DSP and its performance is tested on the two-area
power system simulated on a real-time power system simu-
lation platform, the RTDS. The RTDS is a fully digital
power system simulator capable of continuous real-time
operation. It performs electromagnetic transient power
Fig. 5 Laboratory hardware setup, DSP and the RTDS
Fig. 6 Speed deviation of generator G3 with the proposed con-
troller and WACS not designed for the delays for a three-phase
short circuit at bus 10 (Fig. 1) for ten cycles (166.67 ms) duration

system simulations with a typical time step of 50 ms utilis-
ing a combination of custom software and hardware. The
proprietary operating system used by the RTDS guarantees
‘hard real time’ during all simulations [24]. It is an ideal tool
for the design, development and testing of power system
protection and control devices.
With a large capacity for both digital and analogue signal
exchange (through numerous dedicated, high-speed I/O
ports), physical protection and control devices can be con-
nected to the simulator to interact with the simulated power
system. The WACS is implemented on the Innovative
Integration M67 DSP card (based on the TMS3206701 pro-
cessor), operating at 160 MHz, hosted on a Pentium III
433 MHz personal computer. The M67 DSP card is equipped
with two A4D4 modules [31]. Each A4D4 module is
equipped with four analogoue-to-digital converters and four
digital-to-analogoue converters. The DSP (implements
WACS) and RTDS (implements power system) interface
and laboratory hardware setup is shown in Fig. 5.
5 Evaluation of WACS damping performance
Performance of the WACS is evaluated and compared with
performance of the PSSs for small and large disturbances
applied to the power system. The case studies carried out
are given in Section 5.1. Prony analysis on the performance
of the WACS is given in Section 5.2. This section also quan-
tiﬁes the transient energy exhibited by the generators (G1
and G3) under different disturbances with only PSSs and
the combination of PSS and WACS.
5.1 Case studies
The WACS provides auxiliary damping signals to the exci-
tations systems of generators G1 and G3 when there is a
power swing in either one or both areas. The outputs of
the WACS are zero under steady-state condition. Five
case studies are carried out as explained below.
Case Study 1: Fig. 6 shows the typical responses of the
system for a ten-cycle three-phase line to ground short-
circuit fault applied at bus 10 considering signal trans-
mission delays with a WACS not designed to compensate
for time delays and a WACS designed to compensate for
time delays, respectively. With the WACS not designed to
compensate for delays, there are sustained oscillations.
The WACS signal causes adverse effects on the overall
system unlike the WACS design considering delays which
damps out the electromechanical slow mode of oscillations.
Fig. 7 Speed deviation of generators G1 and G3
a For different controller combination for a three-phase short circuit at
bus 8 (Fig. 1) for ten cycles (166.67 ms) duration
b With different controllers for a ten-cycle (166.67 ms) three-phase
short circuit at bus 8 with no delay in signal transmission
Fig. 8 Speed deviation of generators G1 and G3
a For different controller combination for a load decrease in Area 1
and an increase in Area 2 (Fig. 1)
b For different controllers and various delays in signal transmission
for a ten cycles (166.67 ms) three-phase short circuit at bus 8, which
is cleared by opening one of the parallel lines between buses 7 and
8 and buses 8 and 9

Case Study 2: A three-phase short circuit is applied at
bus 8 for the duration of ten cycles (166.67 ms). Fig. 7
shows the speed deviations of generators G1 and G3 with
no PSS (Uncompensated, dotted-line), with PSS on G1
and G3 (dashed-line), with only the WACS (solid line)
and, with the PSS and the WACS on generators G1 and
G3 (solid bold line) without any delay in signal transmission
(Fig. 7a) and with a delay of 1.0 s in wide-area measure-
ment and control (Fig. 7b). It is clear from this figure that
though the WACS itself does not show much better
response than PSSs, the combination of PSS and the
WACS (solid bold line) exhibit superior performance.
Prony analysis in Section 5.2 shows that though the
WACS itself can mitigate inter-area modes, there exists
some intra-area or local modes which are not damped as
well as the inter-area modes without the PSS. This is
because the WACS training is done with the PSSs online.
Hence, without any PSS, the WACS may not be able to
provide appropriate damping to the intra-area or local
modes for the controller design presented in this paper.
The WACS performs as a supplementary controller in this
study. But, from the results and Prony analysis, it can be
concluded that the WACS signal provide better perform-
ance even without PSSs.
Case Study 3: Fig. 8a shows the speed deviation of gen-
erators G1 and G3 for load changes in Areas 1 and 2. The
load in Area 1 is decreased from 967 2 j100 to
870 2 j90 MVA and the load in Area 2 is increased from
1767 2 j250 to 1863–j260 MVA. It can be observed that
PSSs, the WACS and combined PSS and the WACS, all
stabilise the power system. The WACS and PSSs combi-
nation damp out oscillations faster. There is a small
steady-state error in the speed response causing a 0.05 and
0.07% change in the frequency of generators G1 and G3,
respectively, due to a load change in the areas. This is as
a result of the speed governors having a 5% droop setting.
Fig. 8b presents a contingency where the post-fault top-
ology of the system changes. A three-phase line-to-ground
short circuit is applied at bus 8 for ten cycles and cleared
by opening one of the parallel lines between buses 7 and
8 and buses 8 and 9. Except for the extreme condition of
1.0 s communication delay, the combination of PSS and
WACS provides a good response for the change in top-
ology. Thus, the controller shows robustness to situations
with no prior knowledge.
Case Study 4: To verify the robustness of the proposed
design, a ten-cycle three-phase line-to-ground short circuit
is applied at bus 8. The transmission delays for the
WACS measurement (t1) and control signals (t2) are
varied from zero to a maximum of 1.5 s. Fig. 9a exhibits
acceptable controller performance of PSS and WACS for
a wide range of transmission delays. A graphical compari-
son of damping of one inter-area and two intra-area
modes with respect to different transmission delays in
WACS communication is also presented in Fig. 9b.
Along with these figures, the transient energies of the
generators (G1 and G3) in the first 15 s after the faults are
computed using (12)
TEGENi
¼
1
2
HGEN,
ðtfitþ15
tfit
Dv2
i dt (12)
where i is the generator number and tflt the time the fault is
triggered. Table 1 presents the transient energies of genera-
tors G1 and G3 because of a short-circuit fault at bus 10 for
167 ms. Calculations show observable improvement in the
transient energies of each generator with the WACS
signals under variable signal-transmission delays. The
transient energies of generators G1 and G3 because of a
Fig. 9 Speed deviation of generators and performance with PSS
and WACS for different delays
a Speed deviation of generators G3 with respect to G1 (v3 2 v1) with
different signal transmission delays for a three-phase ten cycles
(166.67 ms) duration short circuit at bus 8 (Fig. 1)
b Damping of different modes for PSS and PSS and WACS with
various communication delays
Table 1: Transient energy of generators 1 and 3 for different stability agents
Generators Transient Energy in MJ PSS PSS and WACS 1.0 s delay 1.4 s delay
No compensation 0.5 s delay
G 1 51.47 26.86 19.77 19.44 22.00
G 3 64.85 30.33 26.20 22.51 24.89

short-circuit fault (ten cycles) for different signal trans-
mission delays (t1 þ t2), ranging from 0 to 1.4 s, are
shown in Fig. 10. The performance of the WACS design
is seen to be robust for variable delays.
Case Study 5: To investigate the possibility of training
the controller with ambient measurements, random load
changes within +10% of normal loading is simulated and
the SRN is trained in a similar fashion as given in Section
3. The same number of input, output and hidden neurons
are used with a learning rate of 0.05 for 3 h. The controller
performance is tested using different delays in communi-
cation signal for a three-phase line-to-ground short-circuit
fault for ten cycles (166.67 ms). The speed of generators
G1 and G3 is presented in Fig. 11a. The damping ratios
for different delays with PSS and WACS combined are
also presented in Fig. 11b. Although the SRN initially
takes more time to train with ambient measurements,
results show that the trained SRN WACS performs
equally well, compared with the controller trained
with PRBS. For a practical real-world system, ambient
measurements may provide a better option for training the
WACS.
5.2 Stability analysis with the Prony method
Prony analysis [32, 33] is an extension of Fourier analysis
and it helps to ﬁnd the modal contents by estimating fre-
quency, damping and phase of a signal. This analysis tool
is useful to provide information on the stability of the
system at the operating point of concern without any need
of extensive techniques to linearise the complete system
model. The analysis is performed on the measured data;
hence, no prior information regarding the system is
required. The real-time implementation platform consisting
of RTDS and DSP does not provide any means to obtain the
linearised closed-loop model of the system. Therefore to
illustrate and quantify the improvement with the WACS
design in terms of damping ratios, the Prony method is
chosen. Although for different delays the linear models
obtained from the Prony analysis are different, Table 2 pre-
sents the results for a signal transmission delay of 1.0 s. This
method is applied on the speed data of generators G1. The
frequency and damping of each dominant mode is given
in Table 2. It can be seen that the designed adaptive
WACS improves the stability of the system even with the
signal transmission delays in both measurement and
control signals. The design is based on the measurement
data and no prior information is used in this design. For a
system with multiple modes of oscillations, during initial
training phase, the WACS identiﬁes slow modes of the
system. Thus, the WACS design with the SRN demon-
strated on a two-area power system is equally applicable
to larger systems with several modes of oscillations. The
SRN network will identify the complex modes and
Fig. 10 Transient energy of generators G1 and G3 for different
delays in the wide-area measurement and control signals
Fig. 11 Speed of generators and damping ratios for different
delays
a Speed of generators G1 and G3 for 10-cycle short-circuit fault at bus
8- the WACS trained with ambient measurements
b Damping ratios of inter-area and intra-area modes for the WACS
trained with ambient measurements
Table 2: Frequency and damping of the system for different power system stability
controllers
No compensation With PSS With WACS With PSS and
WACS
f, Hz z f, Hz z f, Hz z f, Hz z
0.61 0.01 0.62 0.09 0.63 0.22 0.57 0.23
1.22 0.07 1.23 0.1 1.19 0.08 1.20 0.18
1.17 0.07 1.30 0.09 1.20 0.09 1.15 0.09

provide damping to the power system when operating in
reverse mode.
6 Conclusions
This paper has presented the design of an adaptive WACS
for damping generator oscillations by providing auxiliary
signals to their excitation systems. This WACS is designed
to compensate for transmission delays ranging from 0 to
1.4 s with satisfactory performance. The WACS is realised
with a single SRN, which serves a dual purpose of continu-
ous identification of the power system dynamics and gener-
ation of appropriate damping control signals. This type of
design has merit in that it can provide damping for
several modes of oscillations. Real-time implementation
on the RTDS and DSP platform demonstrates the feasibility
of practical realisation of a wide-area measurement-based
SRN control system for a two-area four machine power
network. Multiple SRNs or ObjectNets implemented on a
number of high-speed processors will be needed to realise
this type of control system for large power networks.
The results with the WACS show better damping of
system oscillations caused by small and large disturbances
over a wide range of operating conditions. The combination
of PSS with WACS provides superior damping to the inter-
area as well as intra-area oscillating modes for the test
system in this paper. This superior WACS performance is
substantiated with eigenvalue analysis of the closed-loop
system. This adaptive WACS design shown on a small
power system is a promising and potential measurement-
based technique for large power systems.
7 Acknowledgment
The financial support from the National Science
Foundation, USA under the grant CAREER ECCS #
0348221 is gratefully acknowledged by the authors.
8 References
1 Ni, H., and Heydt, G.T.: ‘Power system stability agents using robust
wide area control’, IEEE Trans. Power Syst., 2002, 17, (4),
pp. 1123–1131
2 Taylor, C.W., Erickson, D.C., Martin, K.E., Wilson, R.E., and
Venkatasubramanian, V.: ‘WACS-wide area stability and voltage
control system: R D and online demonstartion’, Proc. IEEE,
2005, 93, (5), pp. 892–906
3 Taylor, C.W., Venkatasubramanian, M.V., and Chen, Y.: ‘Wide-area
stability and voltage control’. Proc. VII Symp. Specialists in Electric
Operational and Expansion Planning, 2000, pp. 1–9
4 Taylor, C.W.: ‘The future in on-line security assessment and
wide-area stability control’. IEEE Power Engineering Society
Winter Meeting, 2000, vol. 1, pp. 78–83
5 Hauer, J.F., Mittelstadt, W.A., and Adapa, R.: ‘Chapter 11: power
system dynamics and stability. Section 8: direct analysis of wide
area dynamics’ (CRC Electric Power Engineering Handbook, CRC
Press, Cleveland, OH, 2000)
6 Kamwa, I., and Grondin, R.: ‘PMU configuration for system dynamic
performance measurement in large, multiarea power systems’, IEEE
Trans. Power Syst., 2002, 17, (2), pp. 385–394
7 Wilson, R.E., and Taylor, C.W.: ‘Using dynamic simulations to design
the wide-area stability and voltage control system (WACS)’. IEEE
Power Syst. Conference and Exposition, 2004, vol. 1, pp. 100–107
8 Aboul-Ela, M.E., Sallam, A.A., McCalley, J.D., and Fouad, A.A.:
‘Damping controller design for power system oscillations using
global signals’, IEEE Trans. Power Syst., 1996, 11, pp. 767–773
9 Zima, M., Larsson, M., Korba, P., Rehtanz, C., and Andersson, G.:
‘Design aspects for wide-area monitoring and control Systems’,
Proc. IEEE, 2005, 93, (5), pp. 980–996
10 Karlsson, D., Hemmingsson, M., and Lindahl, S.: ‘Wide area system
monitoring and control – terminology, phenomena, and solution
implementation strategies’, IEEE Power Energy Mag., 2004, 2, (5),
pp. 68–76
11 Choudhuri, B., Majumder, R., and Pal, B.C.: ‘Application of
multiple-model adaptive control strategy for robust damping of
interarea oscillations in power system’, IEEE Trans. Control Syst.
Technol., 2004, 12, (5), pp. 727–736
12 Okou, F., Dessaint, L.A., and Akhrif, O.: ‘Power systems stability
enhancement using a wide-area signals based hierarchical
controller’, IEEE Trans. Power Syst., 2005, 20, (3), pp. 1465–1477
13 Hongxia, W., Hui, N., and Heydt, G.T.: ‘The impact of time delay on
robust control design in power systems’. IEEE Power Engineering
Society Winter Meeting, 2002, vol. 2, pp. 1511–1516
14 Snyder, A.F., Ivanescu, D., HadjSaid, N., Georges, D., and
Margotin, T.: ‘Delayed-input wide area stability control with
synchronized phasor measurements and linear matrix inequalities’.
IEEE Power Engineering Society Summer Meeting, 2000, vol. 2,
pp. 1009–1014
15 Hongxia, W., Tsakalis, K.S., and Heydt, G.T.: ‘Evaluation of time
delay effects to wide-area power system stabilizer design’, IEEE
16 Hongxia, W., and Heydt, G.T.: ‘Design of delayed-input wide area
power system stabilizer using the gain scheduling method’. IEEE
power Engineering Society General Meeting, 2003, vol. 3,
pp. 1704–1709
17 Chaudhuri, B., Majumder, R., and Pal, B.C.: ‘Wide-area
measurement-based stabilizing control of power system considering
signal transmission delay’, IEEE Trans. Power Syst., 2004, 19, (4),
pp. 1971–1979
18 Majumder, R., Pal, B.C., Dufour, C., and Korba, P.: ‘Design and
real-time implementation of robust FACTS controller for damping
inter-area oscillation’, IEEE Trans. Power Syst., 2006, 21, (2),
pp. 809–816
19 Amin, S.M., and Wollenberg, B.F.: ‘Toward a smart grid’, IEEE
Power Energy Mag., 2005, 3, (5), pp. 34–41
20 Park,J.W.,Harley,R.G.,andVenayagamoorthy,G.K.:‘MLP/RBFneural
networks based on-line global model identification of synchronous
generator’, IEEE Trans., Ind. Electr. 2005, 52, (6), pp. 1685–1695
21 Venayagamoorthy, G.K., and Harley, R.G.: ‘Two separate
continually online-trained neurocontrollers for excitation and
turbine control of a turbogenerator’, IEEE Trans. Ind. Appl., 2002,
38, (3), pp. 887–893
22 Klein, M., Rogers, G.J., and Kundur, P.: ‘A fundamental study of
inter-area oscillations in power systems’, IEEE Trans. Power Syst.,
1991, 6, (3), pp. 914–921
23 Kundur, P.: ‘Power system stability and control’ (McGraw-Hill,
New York, 1994)
24 Forsyth, P., Maguire, T., and Kuffel, R.: ‘Real time digital simulation
for control and protection system testing’. IEEE 35th Annual Power
Electronics Specialists Conf., 2004, vol. 1, pp. 329–335
25 Serpen, G., Patwardhan, A., and Geib, J.: ‘The simultaneous recurrent
neural network addressing the scaling problem in static optimization’,
Int. J. Neural Syst., 2001, 11, (5), pp. 477–487
26 Werbos, P.J., and Pang, X.: ‘Generalization maze navigation: SRN
critic solve what feedforward or hebbian nets cannot’. Proc. World
Congress on Neural Networks, 1996, pp. 88–93
27 Narendra, K.S., and Parthasarathy, K.: ‘Identification and control of
dynamical systems using neural networks’, IEEE Trans. Neural
Netw., 1990, 1, (1), pp. 4–27
28 Shamsollahi, P., and Malik, O.P.: ‘An adaptive power system
stabilizer using on-line trained neural networks’, IEEE Trans.
Energy Conver., 1997, 12, (4), pp. 382–387
29 Zhou, N., Pierre, J.W., and Hauer, J.F.: ‘Initial results in power system
identification from injected probing signals using a subspace method’,
IEEE Trans. Power Syst., 2006, 21, (3), pp. 1296–1302
30 Pierre, J.W., Trudnowski, D.J., and Donnelly, M.K.: ‘Initial results in
electromechanical mode identification from ambient data’, IEEE
31 OMNIBUS User’s Manual, Innovative Integration, CA USA, 2001
32 Hauer, J.F.: ‘Application of Prony analysis to the determination of
modal content and equivalent models for measured power system
response’, IEEE Trans. Power Syst., 1991, 6, (3), pp. 1062–1068
33 Hong, J.H., and Park, J.K.: ‘A time-domain approach to transmission
network equivalents via Prony analysis for electromagnetic
transients analysis’, IEEE Trans. Power Syst., 1995, 10, (4),
pp. 1789–1797
9 Appendix
Generators G1–G4:
† Rated MVA of the machine ¼ 900.0 MVA.
† Rated RMS line-to-line voltage ¼ 20 kV.
† Base angular frequency ¼ 60.0 Hz.

Exciter parameters for G1–G4
Vrated, KV Tr, s Tc, s Tb, s Ka Ta, s Vrmax, pu Vrmin, pu Kc Kf Tf, s
20.0 0.0 1.0 1.0 200.0 0.01 5.1 -4.0 0.0 0.0 1.0
PSS for G1–G4
K1 Tw, s T1, s T2, s T3, s T4, s Vcu, pu Vcl, pu
20 10 0.05 0.02 3.0 5.4 1.2 0.2
Turbine/governor for G1–G4
R T1, s T2, s T3, s Upper
limit, pu
Lower
limit, pu
0.05 0.36 2.16 8.0 1.07 0.0
Loadﬂow
Bus number Rated
voltage, kV
Bus
type
Voltage, pu Phase,
degree
1 20.0 2 1.030 14.46
2 20.0 2 1.010 4.67
3 20.0 2 1.030 -23.32
4 20.0 2 1.010 -30.88
5 230.0 3 1.005 37.99
6 230.0 3 0.974 27.85
7 230.0 3 0.953 19.36
8 230.0 3 0.930 2.01
9 230.0 3 0.970 -14.91
10 230.0 3 0.984 -7.62
11 230.0 1 1.000 0.00
Generator operating data
Generator number Real power, MW Reactive power, MVAR
1 700.00 196.18
2 700.00 261.46
3 719.00 227.34
4 700.00 198.14
Branch or transmission line data
From
bus
To bus Resistance, pu Reactance, pu Succeptance, pu
10 11 0.0025 0.0250 0.0438
1 4 0.0000 0.0167 0.0000
3 11 0.0000 0.0167 0.0000
5 6 0.0025 0.0250 0.0438
1 5 0.0000 0.0167 0.0000
2 6 0.0000 0.0167 0.0000
6 7 0.0010 0.0100 0.0175
7 8 0.0110 0.1100 0.1925
7 8 0.0110 0.1100 0.1925
8 9 0.0110 0.1100 0.1925
8 9 0.0110 0.1100 0.1925
9 10 0.0010 0.0100 0.0175
Load data
Bus number Real power, MW Reactive
power, MVAR
7 967.09 0.00
7 0.00 99.91
9 0.00 250.00
9 1767.00 0.00
Generator parameters
Generator H Ra, pu Xa, pu Xd, pu Xd0
, pu Xd00
, pu Xq, pu Xq0
, pu Xq00
, pu Td00
, pu Td000
, pu Tq00
, pu Tq000
, pu
G1 6.5 0.0025 0.2 1.8 0.3 0.25 1.7 0.55 0.25 8.0 0.03 0.4 0.05
G2 6.5 0.0025 0.2 1.8 0.3 0.25 1.7 0.55 0.25 8.0 0.03 0.4 0.05
G3 6.175 0.0025 0.2 1.8 0.3 0.25 1.7 0.55 0.25 8.0 0.03 0.4 0.05
G4 6.175 0.0025 0.2 1.8 0.3 0.25 1.7 0.55 0.25 8.0 0.03 0.4 0.05

Real-time implementation of a measurement-based adaptive wide-area control system considering communication delays

More Related Content

What's hot (14)

Similar to Real-time implementation of a measurement-based adaptive wide-area control system considering communication delays (20)

Real-time implementation of a measurement-based adaptive wide-area control system considering communication delays