Alexandria Engineering Journal, Vol. 44 (2005), No.

1, 25-33 25
Faculty of Engineering Alexandria University, Egypt.
Dynamic modeling/simulation of complex power
systems using PSS/E


N.H. Abbasy and W.M. Al-Hasawi
College of Technological Studies, Electrical Technology Dept.,, Shuwaikh, P.O. Box 42325, 70654 Kuwait


Planning and operation studies of complex power systems require accurate and reliable
modeling of each system component, in addition to a powerful dynamic simulation tool. This
paper handles the dynamic modeling and simulation of complex power systems using
PSS/E, as being one of the most efficient dynamic simulation tools. The building blocks of
the dynamic model of a power system are discussed. Details of modeling Kuwait power
network are presented. Dynamic simulations are conducted, where a comparison is made
between simulation results of real disturbances and those recorded by actual disturbance
recorders (historical data). Based on this comparison, model parameters are tuned-up for
more realistic system simulation. Sample contingencies are simulated and system
performance is assessed.
-'=,', ,'=.\ .'.>= _= _>= >\ .=.== -,. -=.=. -,>=\ -,.',,>\ _\ ,=.\ ,==.\ =,==.\ .'=. ,\=..
-\'= '>'== .' .= _\; g='., ,.=.=', -,>=\ -,>,='.,.\ -,.',,>\ _\ ,=. '>'== -=.=. .=,\ .= g\'=,
PSS/E .= ',.=', =\. -,>,='.,.\ '>'==\ ..' '>' _= _>,='.,.\ _.=.\\ _.'.,\ ,,>.\ ..= .=,\ _=.=,
.,>\ -\., -,.',,>\ _\ ,'=.\ _>,='.,.\ _.=.\ .'., ,., '=> _ ,'=. _. -,>,='.,.\ '>'==\ .=,, .=,\ ,,
, -.'=\ .= ,., .,= -\.==\ ->,=\ _\= -\===\ ,'=.\\ -,=,'.\ .'.',,\ ,'=.\ _ -,,= .',==. '>'==\ g.'.. _,
.',==. .>=== -==, ==\ =' _= =\. -=.=.==\ -,>,='.,.\ _.'=.\ .,. ,, =,= _ -.'=\ =\. ,.=.=.
-,= .>' '>'== g.'.. _\= =\ ..'= _= .'.,= '>'== ,., '=> ..' ,,,. ,'=.\\ _ _=.=', .=,\ ,..=,
.'='...=. -=>=\

Keywords: Dynamic simulation, PSS/E, Model validation, Complex networks


1. Introduction

The requirements for improving the
stability of electric power systems and
reducing the effect of abnormal system
conditions on sensitive customers can be met
only by better understanding of the behavior
of the system and optimized configuration of
the different protection and control systems.
This has brought-up the need for improved
quality of simulation of normal and abnormal
system conditions, which is possible only
when using advanced simulation tools based
on accurate system models. Confidence in the
results from different steady state, dynamic or
transient studies is possible when the
recordings of power system events or
disturbances closely match the results from
the simulation of the same event [1]. The
accuracy of the different simulations and the
assessment of steady state and dynamic
security are determined to a great extent by
the accuracy of the models.
State-of-the-art power system simulation
programs [2-6] are designed to provide com-
prehensive and accurate simulation of differ-
ent power system conditions, including power
flow, dynamic stability, short circuit analysis,
motor start-up, protective relays coordination,
and electromagnetic transients. In addition,
research efforts have been conducted for
modeling specific power system components
and/or power systems of special design [7-10].
Using these simulations, planning, operations,
protection and control or power quality
engineers can analyze the behavior of the
system and optimize the performance of
different primary or secondary power system
equipment. However, the results of the
simulations will be accurate and reliable only
when the model of the system is correct. The
modeling of the system is very challenging
because it is continuously changing and
includes a huge number of elements.
In this paper, the building blocks of the
dynamic model of a power system are
N.H. Abbasy, W.M. Al-Hasawi / Complex power systems
26 Alexandria Engineering Journal, Vol. 44, No. 1, January 2005
discussed. Details of modeling Kuwait power
network are presented. Dynamic simulations
are conducted, where a comparison is made
between simulation results of real distur-
bances and those recorded by actual distur-
bance recorders (Historical data). Based on
this comparison, model parameters are tuned-
up for more realistic system simulation. Sam-
ple contingencies are simulated and system
performance is assessed.

2. Model preparation

The dynamic simulation of a power system
can be accomplished via two major modules;
Load Flow (LF) and Dynamic Simulation (DS)
modules. The earliest step in PSS/E simula-
tion of a power system is the preparation of
the Load Flow case. The output file of this
stage is referred to as the Load Flow Uncon-
verted Case (LFUC). In the DS module, a
dynamic raw data file is prepared. This file
includes data relevant to the dynamic models
of the system components. Thereafter, both
network LF data and dynamic data are
combined into one file, called the snapshot
file. Then, the simulation (run) stage follows
immediately. Different system components
are modeled using built-n dynamic models.
The selection among the available models is
justified according to the manufacturer data
provided with the plant installation. For exam-
ple, generators are modeled as solid rotor
generators at the sub-transient level. Exciters
are modeled as a potential source controlled
rectifier, where excitation power is supplied
through a transformer from the generator ter-
minals (or the units auxiliary bus) and is
regulated by a controlled rectifier.
As per load modeling, there are four
distinct methods for load modeling based
PSS/E; namely, constant MVA, modeling,
constant current modeling, constant admit-
tance modeling, and composite modeling.
In constant MVA modeling, the load
boundary condition is a specification of load
real and/or reactive power consumption, as
given by eqs. (1) and (2) (for bus # k):

. P ) i v ( al Re
k
*
k k
= (1)

. Q ) i v ( ag Im
k
*
k k
= (2)

In constant current modeling, loads may
be specified as a given active or reactive
component of current. These models make
load dependent on frequency in accordance
with:

, I
v
) i v ( al Re
pk
k
*
k k
= (3)

. I
v
*) i v ( ag Im
qk
k
k k
= (4)

In constant admittance models, load may
be specified by given real and reactive parts of
shunt admittance such that:

. B j G
v
i
k k
k
k
+ = (5)

As per the composite load modeling, all
PSS/E network solutions allow the load at
each bus to be a composite of arbitrary
amounts of load with each of the above three
characteristics. The composite characteristic
becomes the boundary condition used in it-
erative power flow solutions. A fraction of the
constant MVA load at each bus is transferred
to each of the other two load characteristics,
according to the following equation:

,
V
aS
S S
p
i I
+ =
,
V
bS
S S
2
p
y Y
+ =
, ) b - a - 1 ( S S
p P
= (6)

where Sp, Si, Sy are the original constant MVA,
constant current and constant shunt
admittance load, respectively, and SP, SI, SY
are the corresponding final nominal values.
Constants a, b are arbitrary chosen load
transfer fractions (where a + b <1), and V is
the magnitude of bus voltage when load
conversion is made. In case of Kuwait system,
loads are arbitrary composed according to the
following ratios: 80, 0 for active power and 40,
N.H. Abbasy, W.M. Al-Hasawi / Complex power systems
Alexandria Engineering Journal, Vol. 44, No. 1, January 2005 27
20 for the reactive power. This means that
active power load is modeled with 80%, 0%
and 20% as constant current, constant
conductance and constant MVA, respectively.
Similarly, reactive power load is modeled with
40%, 20% and 40% as constant current,
constant conductance (admittance) and
constant MVA, respectively.
One more relevant aspect is the frequency
dependence of loads. The LDFR family of mod-
els provided by PSS/E makes the constant
current and MVA components of all loads, to
which the model is applied, dependent on bus
frequency. These models make load dependent
on frequency in accordance with:

=
n
o
o
m
o
o
) ( Q Q , ) ( P P

=
s
o
qo q
r
o
po p
) ( I I , ) ( I I

, (7)

where constants m, n, r, and s are selected by
the user and stored in as CONs. These indices
are not necessarily integers, and may be zero
if the corresponding load components are
independent of frequency. In case of Kuwait
system, loads are categorized into 4
categories; industrial, auxiliary, distillation
and residential. Constants m, n, r, and s are
selected for each category. Typical values are:
m = 2, n = 0, r = 2, s = 0. The current setting
for these load models is as follows: for
residential/industrial load: m = 1.5, n = 0, r =
1.5, s = 0; for plant auxiliary /distiller load: m
=2, n = 0, r = 2, s = 0. Obviously, there is no
rule of thumb for assigning these constants.
However, these constants can be fine-tuned
through a comprehensive comparison between
simulation results and actual results obtained
by recorders.

3. Simulations and assessment

Kuwait network is used as a case study to
validate and assess the dynamic models
presented in this paper. In this system,
underfrequency relays are set such that they
automatically disconnect predetermined load
centers according to the level of the declined
frequency, in case of generation deficiency.
This underfrequency load-shedding scheme
adopts an overall number of 5 stages, where
relays automatically trip at 49, 48.7, 48.4,
48.2, and 48 Hz for stages 1 through 5
respectively
Table 1 presents the details of abnormal
incidents encountered the system during
2002-2003 period. Figs. 1 through 4 show the
simulated frequency response of the system
for cases 1 through 4 of table 1, whereas the
actual recorded tracings for case 1 is provided
in fig. 5. Actual frequency tracings of the
remaining cases were omitted for limited
space.

3.1. Model validation

In order to validate the models used in this
system, 4 different abnormal incidents were
recorded and simulated. It is noticed that
among these cases, case #1 is the only one
that required underfrequency load shedding.
In order to simulate these disturbances using
PSS/E, the dynamic simulation was run for
one half-cycle (0.01 s based-50 Hz) prior to the
disturbance. This brought up the simulation
up to t =0 (the instant of disturbance). At that
instant the disturbance was applied by
dropping specific units as listed in table 1.
Thereafter, simulation was advanced for two
seconds right after the disturbance in order to
determine the time instant at which system
frequency would hit the first stage load
shedding frequency threshold (49.0 Hz). Once
this instant was determined, the simulation
was repeated where the first stage load
shedding was activated at its predetermined
time instant (after accounting for an
additional 100 ms as a time delay for relay
operation). Then, the whole procedure was
repeated to check for the necessity of
activating the 2
nd
, 3
rd
,, stages, respectively.
In this simulation, it was found that 3 load-
shedding stages were necessary to restore the
system frequency, with a total reduction in
system loads of 714 MW. This result totally
agreed with the actual system behavior
described in table 1, case # 1. The close
inspection of figs. 1 to 5 show that both
responses yielded almost identical values for
the minimum frequency. (48.34 Hz). The time
elapsed for the system to reach this minimum
frequency was also identical (2.5 s). Moreover,
N.H. Abbasy, W.M. Al-Hasawi / Complex power systems
28 Alexandria Engineering Journal, Vol. 44, No. 1, January 2005
Table1
Details of the abnormal incidents during 2002-2003

Day/date Case # 1
Monday 23-12-2002
Case # 2
Thursday 27-2-2003
Case # 3
Monday 10-2-2003
Case # 4
Monday 03-03-2003
Case Number Case # 1 Case # 2 Case # 3 Case # 4
Location DWPS DWPS ZSPS ZSPS
Time 13:30 10:57 10:00 17:37
System load
(MW)
2980 2680 2630 3020
Plants on bar
(MW)
3150 4550 4715 4640
ISR (MW) 490 675 725 675
Contingency All circuits, which are
connected to the 300
kV bus at DWPS,
tripped, causing an
outage of 900 MW.
Unit 8 tripped due to
tripping of it boiler on
drum level three low
indication, while it
was carrying 160 MW.
Unit 5 tripped due to
tripping of its boiler
on function group
control power supply
failure indication,
while it was carrying
120 MW.
Unit 2 tripped due to
tripping of its boiler
on boiler control
supply failure
indication, while it
was carrying 170 MW.
Number of
units in
service
SB (3); ZS (5); DE (4);
SS (4); DW (5)
SB (3); ZS (5); DW (4);
DE (4); SS (3)
SB (4); ZS (4); DW (5);
DE (4); SS (2)
SB(4); ZS(4); DW(5);
DE(4); SS(3)
Plant
generation
(MW)
SB (504); ZS (860);
DE (345); SS (368);
DW (900)
SB (480); ZS (800);
DW (800); DE (350);
SS (250)
SB (650); ZS (650);
DW (800); DE (350);
SS (180)
SB (600); ZS(850);
DW(850); DE(450);
SS(270)
Frequency
(Hz)
Dropped from 50.02
to 48.32
Dropped from 50.02
to 49.85
Dropped from 50.01
to 49.85
Dropped from 50.01
to 49.79
Plant pick-up
(MW)
SB (15); ZS (145); DE
(55); SS (55)
+3 stages of the
automated under
frequency scheme
operated causing a
loss of supply of total
714 MW.
SB (30); ZS (60); DW
(40); DE (20); SS (10).

No load shed.
SB (30); ZS (30); DW
(40); DE (15); SS (5).

No load shed.
SB (50); ZS (50); DW
(45); DE (15); SS (10).

No load shed.





Fig. 1. System frequency response for case # 1.



Fig. 2. System frequency response for case # 2.

N.H. Abbasy, W.M. Al-Hasawi / Complex power systems
Alexandria Engineering Journal, Vol. 44, No. 1, January 2005 29


Fig. 3. System frequency response for case # 3.



Fig. 4. System frequency response for case # 4.



Fig. 5. System tracing of disturbance recorder for case #1.

the new stabilized (steady state) frequency was
fairly close (about 49.65 Hz) in both
responses, giving the same steady state
frequency error (0.35 Hz). However, a
substantial deviation is observed in the
restoration time of both responses. The actual
system response required about 6 s to re-
stabilize, while the simulation response
required about 15 s to re-stabilize. From a
practical prospective, this discrepancy is of
less importance; as frequency restoration time
is not a planning parameter in any operational
scheme, e.g., load shedding. On the other
hand, the result is optimistic by itself, since
the actual system restoration is faster than
what the simulation gives. In cases 2, 3 and 4,
where load shedding was not required, the
simulation was advanced for adequate period
of time (e.g. 10 s) until frequency re-stabilized.
Table 2 shows a comparison between the
actual recorded results with those obtained
from simulations. This comparison is limited
to restoration time, value of minimum
frequency, the time required to reach the
minimum frequency, the initial rate of
frequency decline, and finally the steady state
frequency. The % deviation, defined as
[(simulated response - recorded response) /
recorded response]*100, is also computed and
shown in the same table.
From table 2, the following observations
can be made:
1. The actually recorded restoration times are
much longer than the simulated times.
2. The actually recorded minimum frequency
reached in each case is slightly higher than
the simulated values (except case #1).
3. The actually recorded times to reach
minimum frequency are sometimes higher
than the simulated times and sometimes
lower.
4. The actually recorded rate of frequency
decay is less than the corresponding
simulated values.
5. The actually recorded steady state
frequency is higher than the corresponding
simulated values.
6. In the actual recorded data, a slight
overshoot for the frequency is observed
before it settles at its steady state value. In
case of simulated data, the frequency rises
N.H. Abbasy, W.M. Al-Hasawi / Complex power systems
30 Alexandria Engineering Journal, Vol. 44, No. 1, January 2005
smoothly to the new equilibrium state without
overshooting.
In view of the above observations, it can be
generally concluded that the simulation
results are more accurate (closer to the actual
recorded results) in terms of frequency
(whether minimum or steady state) and less
accurate in terms of system restoration time,
initial rate of frequency decline and time to
reach minimum frequency. Nevertheless, the
deviation in simulation results is always in the
pessimistic side, in the sense that worst
system response is obtained via simulation.
This latter observation implies that the
simulation results are always in the safe side
in spite of not being highly accurate.

3.2. Tuning of model parameters

In view of the observation made at the end
of the previous section, an attempt was made
to narrowing the gap between the two
responses, during the restoration period. The
difference of restoration time may be
attributed to different factors. The complexity
and diversity of loads existing in the system,
the uncertainty of system model parameters,
and the inaccuracy of actual system
recordings may represent some of these
factors. However, there are actually 4 load
categories in the present system. These are:
residential, industrial, distillation and plant
auxiliary. Referring back to the LDFR family of
models given by eq. (6), three attempts of
frequency dependent load modeling were
made, with the following parameters:
1. m = 1.5, n = 0, r = 1.5 and s = 0
2. m = 2, n = 0, r = 2 and s = 0
3. m = 3, n = 0, r = 3 and s = 0
Fig. 6 shows the frequency behavior
during the present disturbance for these three
models. Compared with the actual recorded
frequency during this fault, it is obvious that
the third load model mimic the actual
response more accurately. This proved the fact
that better system performance can be
achieved through tuning some of model
parameters. This era requires extensive trials
and revaluation of the models adopted.

Table 2
Comparison between results of actual and simulated disturbances

Case # Case # 1
23-12-2003
Case # 2
27-2-2003
Case # 3
10-2-2003
Case # 4
3-3-2003
Recorded 8.28 3.76 s 4.10 s 3.77 s
Simulated 16.32 7 s 4.3 s 5.02 s
Restoration time
(s)

% Deviation 97% 86.17021 69.65551 33.1565
Recorded 48.238 49.85 Hz 49.85 Hz 49.79 Hz
Simulated 48.32 49.726 Hz 49.75 Hz 49.691 Hz
Min. freq. (Hz)

% Deviation -0.016 -0.24875 -0.2006 -0.19884
Recorded 1.7 1.654 s 1.91 s 1.34 s
Simulated 2.51 1.78 s 1.43 s 1.522 s
Time to reach
Min. freq. (s)

% Deviation 47.64 7.617896 -25.1309 13.58209
Recorded 1.728 0.0906 Hz/s 0.0785 Hz 0.162 Hz
Simulated 2.14 Hz 0.1539 Hz 0.175 Hz 0.203 Hz
Initial rate of
freq. decay
(Hz/s)
% Deviation -17.07 69.86755 122.9299 25.30864
Recorded 49.65 49.95 Hz 49.96 Hz 49.98 Hz
Simulated 49.61 49.899 Hz 49.914 Hz 49.892 Hz
Steady state
frequency (Hz)
% Deviation -0.08 -0.1021 -0.09207 -0.17607
N.H. Abbasy, W.M. Al-Hasawi / Complex power systems
Alexandria Engineering Journal, Vol. 44, No. 1, January 2005 31

Fig. 6. Frequency responses for three LDFR models.



Fig. 7. Response of ZSPS to outage of DWPS scale: Pe Pm:
150:220 MW, rotor angle:-10:15.


3.3. Performance assessment

The response of system generating plants
to the outage shown in case #1, table 2 was
monitored using activity CHAN. Fig. 7
summarizes the response of ZSPS to that
outage. This response includes the total
electrical power, total mechanical power,
machine rotor angle, and system frequency
measured at that plant. Similar plots for the
remaining plants were obtained but not
shown. Fig. 8 shows the sum of mechanical
power of all system units for that outage.
To check units response to generation
outages, a different simulation was performed.
Fig. 9 shows DWPS unit response at a trip of
one unit at ZSPS. The system frequency, the
mechanical and electrical power, as well as
the machine rotor angle of that particular unit
were monitored and plotted in that figure. The
mechanical power curve shows how the
instantaneous spinning reserve (ISR) builds
up to share in the alleviation of the deficit of
generation due to the outage. This ISR
response started fairly fast (within 5 s), then
its rate slowed down thereafter. Also both
electrical and mechanical power of that unit
coincided when the system frequency re-
stabilized at 49.6 Hz, leaving a steady state
frequency error of 0.4 Hz. Obviously, this trip
did not require any load shedding as the decay
of system frequency halted at 49.75 Hz before
the 1
st
stage frequency level of system
automatic load shedding scheme. Finally, a
third simulation was conducted to assess the
system response at line faults. The line
connecting JAHRW and SLYBW was faulted at
t =2 s. This fault sustained for about 5 s,
where the fault was cleared at t = 7 s. Fig. 10
shows the system frequency and voltage at
SRRD for this contingency. The simulation
shows that it took about 30 s for the
frequency to re-stabilize, where the system
frequency fluctuated between 53.75 and 48.85
Hz, whereas the voltage at SRRD fluctuated
between 85 and 408 kV. The voltage
monitored at SRRD stabilized at about 5.5%
above normal. This result would have impact
on insulation coordination of the system to
withstand such switching over voltages.

4. Conclusions

This paper presented a procedure for
dynamic modeling and simulation of complex
power systems using PSS/E. Dynamic simula-
tions of real disturbances were conducted and
results were compared with those obtained by
system disturbance recorders. Based on this
comparison, the dynamic model parameters of
system loads were fine-tuned for more realistic
dynamic simulation. The system dynamic
behavior in case of unit outage, plant outage,
and line faults was assessed. Further studies
related to the switching impact on system over
N.H. Abbasy, W.M. Al-Hasawi / Complex power systems
32 Alexandria Engineering Journal, Vol. 44, No. 1, January 2005


Fig. 8. Unit's mechanical power for outage of DWPS.


Fig. 9. DWPS unit for trip at ZSPS scale: frequency: 49.6
50 Hz, Pe, Pm 175:195 MW, Rrotor angle 0:2.5.



Fig. 10. System frequency and bus voltage at SRRD for
line fault.
voltage and the proper allocation of dynamic
system reserve can be conducted by following
the simulation procedure shown in this paper.
It is believed that this paper would generally
assist students, researchers, and electrical
utility engineers in modeling and simulation of
complex power systems.

Acknowledgement

The authors are grateful to the staff of The
National Control Center, Kuwait, for providing
the data and relevant information of Kuwait
power network.

References

[1] Alexander Apostolov, Verification of
System Models for Steady-State and
Dynamic Security Assessment,
Proceedings of the 37
th
Hawaii
International Conference on System
Sciences (2004).
[2] GE Power Systems Analysis Software,
General Electric International Inc.
[3] Simsen: A Modular Software Package for
the Analysis of Power Networks and
Electrical Machines, ICEM 94.
[4] Matlab the Language of Technical
Computing, version 6.1.0.450, Release
12.1, 18 (2001).
[5] ETAP Power Station, 3.0, Operation
Technology, Inc. (2001).
[6] ATP-EMTP, Canam EMTP User Group.
[7] P. Sakis Meliopoulos, George J.
Cokkinides, and Robert Lasseter, "A
Multiphase Power Flow Model for Grid
Analysis", Proceedings of the 36
th
Hawaii
International Conference on System
Sciences (HICSS03), 0-7695-1874-5
(2003).
[8] W.J. Peet, Babcock and Wilcox,
Development and Application of a
Dynamic Simulation Model for a Drum
Type Boiler with Turbine Bypass
System, International Power
Engineering Conference, March,
Singapore (1995).
[9] Debabrata Chattopadhyay, and Ross
Baldick, Unit Commitment with
Probabilistic Reserve, Proc. of IEEE PES
Winter Meeting, New York, USA.
N.H. Abbasy, W.M. Al-Hasawi / Complex power systems
Alexandria Engineering Journal, Vol. 44, No. 1, January 2005 33
[10] F.W. Koch, I. Erlich, and F. Shewarega,
Dynamic Simulation of Large Wind
Farms Integrated In A Multi Machine
Network, IEEE PES 2003 General
Meeting, July 13-17, Toronto, Canada
(2003).

Received May 12, 2004
Accepted September 13, 2004