IEEE Transactions on Power Systems, Vol. 14, No.

3, August 1999 967

Model Validation for the August 10,1996 WSCC System Outage

Dmitry N. Kosterev, Member Carson W. Taylor, Fellow William A. Mittelstadt, Fellow
Transmission Operations and Planning
Bonneville Power Administration
Portland, Oregon USA

Abstract: A major power outage occurred in western North Table 1

America on August 10, 1996. In this paper we describe simulation Power flows on major PNWpaths prior to the disturbance.
of the disturbance using a transient stability program. Initially, the
sequence of events was simulated using the standard WSCC
dynamic data base, and the simulations did not agree with the
disturbance recordings. Several model additions and data
modifications were made before the simulations approximated the
disturbance recordings. These model changes are discussed in the
paper.
,,from lngledow (6C Hydro)
Keywords: power system modeling, power system stability, power
system disturbance, power system performance high voltage direct
current.

I. THE AUGUST 10,1996 WSCC DISTURBANCE

On August 10, 1996, a major failure occurred in the

Western Systems Coordinating Council (WSCC) system
resulting in break-up into four islands, with loss of 30,390
MW of load affecting 7.49 million customers in western
North America [l].

A, Pre-Event Conditions and Outages

The pre-event system conditions in the Pacific
Northwest (PNW) were characterized by high north-to-south
flows on the California-Oregon Intertie (COI) and the
Pacific HVDC Intertie (PDCl). Figure 1 shows a one-line
diagram of major 500-kV lines in the PNW with detail of the
Big Eddy area. Table 1 shows major path power flows and
the percentage loading of each path rating. The PNW
generation pattern was characterized by high imports from
Canada, high generation in the upper Columbia (Grand
Coulee, Chief Joseph) and low generation in the lower
10 Oh& to Round ML DC to Sylmar
Columbia area (John Day, The Dalles). Prior to the
disturbance, there were two forced outages of 500-kV lines: Fig. 1. One-line diagram of major 500-kVlines in !he Pacific Northwest.
John Day-Marion-Lane and Big Eddy-Ostrander. A
scheduled maintenance outage of the Keeler 500/230-kV B. The Initial Stage of the Disturbance
transformer reduced reactive power support and voltage
control at the 500-kV bus from the static var compensator at At 15:42:37, the AlIston-Keeler 500-kV line sagged
the Keeler 230-kV bus. close to a tree and flashed over. The line was tripped
following unsuccessful single-pole reclosure. Due to the
PE-226-PWRS-0-12-1997 A paper recommended and approved by Keeler breaker configuration, the Keeler-Pearl 500-kV line
the IEEE Power System Dynamic Performance Committee of the IEEE also opened. Prior to the outage, the Allston-Keeler line was
Power Engineering Society for publication in the IEEE Transactions on
Power Systems. Manuscript submitted July 29, 1997; made available carrying 1300 MW. This power shifted to 500-kV lines east
for printing December 12, 1997. of the Cascades Mountains (through the Hanford bus), and to
the underlving 115-kV and 230-kV network. The lower

U.S. Government Work Not Protected by US. Copyright

Authorized licensed use limited to: University of Waterloo. Downloaded on December 01,2020 at 19:45:07 UTC from IEEE Xplore. Restrictions apply.
 968
voltage lines parallel to the Allston-Keeler 500-kV line d) British Columbia-PNW tie power (both Ingledow-
became loaded up to 115%of their thermal ratings. Voltages Custer 500-kV lines).
in the lower Columbia area were depressed to 508 kV at For up to 40 seconds on the plot scale, the power oscillations
Hanford, 504 kV at Big Eddy, 510 kV at John Day, and 505 were sustained at near zero damping.
kV at McNary.
About five minutes later, the Merwin-St. Johns 115-kV C. The Final Stage before Cascading
line tripped due to a relay failure and the overloaded Ross-
Following initial tripping of McNary generators, the
Lexington 230-IcV sagged into a tree. These lines are parallel
system frequency dropped, the CO1 export decreased, and
to the Allston-Keeler 500-kV line.
significant BPA area control error developed. Figure 3 shows
About the same time, at 15:47:37, sequential tripping of
thirteen McNary units began due to exciter protection a) McNary generation (same as Figure 2a),
malfunctions at high field voltage. This started system power b) BPA Area Control Error, and
and voltage oscillations (Figure 2). c) Grand Coulee unit 20 frequency.
Subsequent AGC and governor actions tried to restore
(a) McNary Power House Output
the system frequency and the CO1 interchange flow. The
, ( 00 ) 11 I I AGC power pick-up occurred at Grand Coulee, John Day,
and Chief Joseph generators. Figure 4 shows the power
output increase at:
a) Grand Coulee,
b) Chief Joseph,
c) John Day.
3-0
0 10 20 30 40 50 60 70 60 90 100 Increase in Canadian exports due to governor action was also
Time [sec]
(b) Malin 500kV Bus Voltage recorded.
II ' 1
!i40 (a) McNary Power Output
IOW, I
$530

!j20
>
!<IO

l500
0 10 20 30 40 50 60 70 80 90 100
Time [sec]
. .
(c) California-Oregon Intertie Power 0 10 20 30 40 50 60 70 80 90 100
Time [sec]
4500 (b) BPA Area Control Error
--.woo
O,$
f 1200
a .$IO0
.MOO
:Boo
0 10 20 30 40 50 60 70 80 90 100
Time [sec]
(d) Ingledow-Custer Power 0 10 20 30 40 50 60 70 80 90 100
:!zoo- , Time [sec]

loo 60.02
., Grand Coulee unit #20 Freauencv

E
g :2000 - 60
259.98
n
$59.96
1900 $59.94
0 10 20 30 40 50 60 70 80 90 100 59.92
Time [sec]
59.9
Fig. 2. August 10th disturbance recordings (starting at 15:47:30).
0 10 20 30 40 50 60 70 80 90 100
Time [sec]
Fig. 3. Disturbance recordings of the McNary tripping, BPA Area Control Error
and the Coulee generator frequency (starting at 15:47:30)
Figure 2 shows:
a) power output from the McNary power plant, The power pick up occurred primarily in the upper
b) Malin 500-kV bus voltage, Columbia area, further stressing the east of Cascades 500-kV
c) CO1 power (sum of flows in the Captain Jack-Olinda transmission lines, and further depressing Northwest
and both Malin-Round Mountain 500-kV lines), voltages.

Authorized licensed use limited to: University of Waterloo. Downloaded on December 01,2020 at 19:45:07 UTC from IEEE Xplore. Restrictions apply.
 969
Existing converters and parallel Expansion converters, see
(a)
. . Grand Coulee Power Pick-Uo the Appendix.
200 -
1 PDCI operates in the constant power mode at the pole
level. In normal North-to-South operation, both Celilo
Expansion and Existing converters are in the current control
b
a
mode. The Sylmar Expansion converter operates in the
mixed (extinction angle - current) control mode, and the
Sylmar Existing converter is setting the dc voltage by having
0Oo
' I10O 20 30 40 50
50 60 70 80 90 i100
o0
Time [sec] the converter extinction angle at its minimum value. Sylmar
lool
(b) Chief Joseph Power Pick-up Existing converter LTC transformers are adjusted to control
the rectifier dc voltage to 500kV.
The PDCI was initially maintaining constant power
$ 60 control (Figure 6). As Celilo ac voltages continued to decay
40 on average, the Existing converters started limiting out
20 (reaching minimum firing angle and losing current control)
a during the lower portion of the ac voltage swings. During the
0 10 20 30 40 50 60 70 80 90 100
Time [sec] upper portion of ac voltage swings the converters regained
(c) John Day Power Pick-up current control. AC voltage oscillations and converter
150
control limits caused the PDCI power oscillations.
SI00 (a) California-Oregon Intertie Power
3
I ' 'I1 I I I
B 4500
g 50
~4400
24300

'0 10 20 30 40 50 60 70 80 90 100
t 4200
Time [sec] '4100
4000
Fig. 4. Disturbance recordings of the responses of major power generation 3900; I
10 20 30 40 50 60 70 80 90
plants in the Pacific Northwest to AGC action and frequency drop following Time [sec]
McNary tripping (starting at 15:47:30). (b) Pacific HVDC Intertie Power (total at Celilo)
29501

Due to increased line loading and limited reactive power

support from lower Columbia generators, the system voltages
continued to decay on average. Figure 5 shows Big Eddy and
John Day 500-kV voltages from BPA SCADA (2 second
data sampling). 2750'
0 10 20 30 40 50 60 70 80 90
Time [sec]
(c) Celilo Existing Converter Power
515.0, I I , I
19001 I

A./
510 0

F 5050
Y

;
->
0
5000
I 1 90
0 10 20 30 40 50 60 70 80
4950 Time [sec]
(d) Celilo Expansion Converter Power
490 0
1120
4050 . . .. ~1100
154720 154735 154750 1548.05 15:48:20 1548.35 15:48:50 +
..

Time t 1080
2 1060
IC40
Fig. 5. BPA SCADA voltages in the Lower Columbia area 1MO
0 10 20 30 40 50 €0 70 80 90
Time [sec]
The Pacific HVDC Intertie (PDCI) has the four-terminal
bi-polar configuration [2]. Each end each consists of Fig. 6. Disturbancerecordingof the PDCI response (starting at 15:42:30)

Authorized licensed use limited to: University of Waterloo. Downloaded on December 01,2020 at 19:45:07 UTC from IEEE Xplore. Restrictions apply.
970

Present PDCI controls have a function attempting to 11. AUGUST lOTH VALIDATION STUDIES
maintain dc power constant by transferring current control
We simulated the disturbance using the EPRI ETMSP
from a rectifier (Celilo) to an inverter (Sylmar) when a
program to verify the system dynamic models used in
rectifier becomes limited. The mode shift caused by this
planning and operational studies.
function contributed to dc power oscillations, resulting in dc
First, pre-event power flow conditions were accurately
power swings below and above the scheduled power.
represented using the SCADA data just prior to the Ross-
Figure 6 shows:
Lexington line outage.
a) CO1 power (same as Figure 2b),
b) total PDCI power at the Celilo converter,
A. Validation Criteriafor the Time-Domain Simulations
c) Celilo Existing converter total power,
d) Celilo Expansion converter total power. The main intent of the validation studies was to capture
As the PDCI started participating in the oscillations, the ac the mechanism and driving forces which led to the system
system oscillations started picking up in magnitude and the instability. Correct system reinforcements can then be
CO1 separation occurred at 15:48:52 (lines relayed due to identified and operating practices can be established to
low voltage high current conditions). The resulting prevent similar outages from reoccurring. The system
cascading [11 is outside the scope of this work. simulations are tested by the following criteria:
1. Frequency and damping of the inter-area oscillatory
D.Analysis of the Oscillations mode as well as their changes as the disturbance progressed.
2. Change in the PDCI behavior from constant power
Prony analysis [3] identified oscillatory components of
mode to power oscillations.
power quantities on major AC and DC paths. The oscillation
3. Average profile of the system voltages, line flows, and
“signature” changed as the transient progressed. At initial
system frequency.
stages, the oscillation frequency was 0.266 Hz with near zero
4. Distribution of the oscillatory mode in the system
damping (Prony analysis of the Maliri voltage oscillations on
(phase shift between power oscillations at the COI, PDCI,
the interval 20-35 seconds on the plot scale). With the AGC
Coulee, and British Columbia-PNW ties).
power pick-up, the oscillatory frequency decreased to 0.242
Hz and dampin:; changed to -2.66% (Prony on the interval
B. Time-Domain Simulations using the WSCC Dynatriic
35-50 seconds on the plot scale). As the PDCI started
Models and Data
participating in the oscillations, the oscillation frequency
continued to decrease and the system damping became more Initially, the August 10th sequence of events was
negative. At the final stage of the event, the oscillatory simulated using the standard WSCC dynamic data base. This
frequency was 0.217 Hi with damping of -7.62% (Prony on data base was assembled from dynamic data provided by
the interval 72-84 seconds). individual utilities.
We computed amplitude and phase of the power Figures 8-11 compare recorded and simulated data
oscillations for the COI, Celilo Expansion converter power, (Malin voltages, CO1 power, Grand Coulee frequency, and
Celilo Existing converter power, Celilo total power, Grand PDCI power respectively). The plots show that the response
Coulee power output, and the British Columbia-Northwest of the simulated system model does not agree with the
tie power. The complex plane in Figure 7 shows phase recorded system response. Specifically, the simulated system
relati onship between the specified power oscillations. This fails to capture:
phase relationship corresponds to the mode shape for the a) Damping of the inter-area oscillations. The simulated
approximately 0.23 Hz oscillations. Note that the Celilo system shows stable well-damped response.
Expansion and Existing converters had different behavior b) Averaged voltage and power profiles. The simulated
and impact on the system oscillations. system does not indicate voltage decay in the system.
Also, the simulated CO1 power does not increase
DC Expansion following as the transient progresses, because no AGC
action was modeled following initial tripping of McNary
generators.
PDCI Coulee c) PDCI mode shift and power oscillations. The simulated
PDCI maintains constant power control and does not
DC Existing
1 IBCHydro participate in oscillations.
d) System frequency excursion. The simulated frequency
dip following McNary tripping is only 60% of the
Fig. 7: phasor diagram of power oscillations at 0.23 Hz recorded.

Authorized licensed use limited to: University of Waterloo. Downloaded on December 01,2020 at 19:45:07 UTC from IEEE Xplore. Restrictions apply.
 971
Actual Malin Voltage Actual PDCl Powei
550 2950 I I , , . I

540 2900

z530 Z2850
m b
g
>
520 g 26W
n
510 2750

500
0 10 20 30 40 50 60 70 80 90 2700'
0 10 20 30 40 50 60 70 80 90
Time [sec] Time [sec]
Simulated Malin Voltage (WSCC base case) Simulated PDCl Power (WSCC base case)

540

g 2800}

27
551 5000 10 20 30 40 50
Time [sec]
60 70 80 90 27000 10 20 30 40 50
Time [sec]
60 70 80 90

Fig. 8. Malin 500-kVvoltage -recorded and simulated Fig. 11. PDCI power -recorded and simulated using
using WSCC data base WSCC data base
Actual CO1 Power
4600 /I I C. Dynamic Data Modifications and Model Additions
g4400
L
We modified the modeling as follows:
4200
1) Detailed PDCl Model: The PDCI model in WSCC
a data base is two-terminal and has over-simplified control
4000 representation with current regulators. The model was in-
0 10 20 30 40 50 60 ,70 60 90 adequate to represent the PDCI behavior during the
Time [sec)
Simulated CO1 Power (WSCC base case) disturbance. A detailed stability model of the Pacific HVDC
Intertie and its controls was developed by BPA. The model is
four terminal and represents essential controls on the
bipolar, polar, station, and converter levels [2].The model
correctly captures the PDCI mode shift during the
disturbance and its impact on system damping.
2) Automatic Generation Control Action: AGC action is
0 10 20 30 40 50 60 70 80 90
represented at Grand Coulee, Chief Joseph and John Day
Time [sec]
power plants. These plants were on AGC control during the
Fig. 9. CO1 power - recorded and simulatedusing disturbance. Governors at those plants are represented by
WSCC data base user-defined models with an additional AGC input.
Actual Coulee Frequency
Typically, AGC is not considered in dynamic simulation.
3) Blocking turbine-speed controls on large steam-
turbine generators throughout WSCC: Depending on plant
design and operating practices, large turbo-generators may
not increase power output immediately following the system
frequency decay. If operated in turbine-follow mode, it may
I I )
0 10 20 30 40 50 60 70 80 90 take minutes before boilers respond to system frequency
Time [sec]
Simulated Coulee Freauencv (WSCC base case) excursion [4,5]. Therefore governors for large turbo-
generators were not represented. No changes were made for
hydro generation. This modification resulted in very close
agreement between the recorded and simulated system
frequency excursions following initial tripping of the
McNary generators.
4) Voltage Controls on Lower Columbia generators:
0 10 20 30 40 50
Time [sec]
60 70 BO 90 The US Army Corps of Engineers (Corps) operates hydro
Fig. 10. Grand Coulee unit 20 frequency -recorded and simulated plants in the Lower Columbia area. The Corps uses Data
using WSCC data base Acquisition and Control Systems (DACS) for high-side

Authorized licensed use limited to: University of Waterloo. Downloaded on December 01,2020 at 19:45:07 UTC from IEEE Xplore. Restrictions apply.
972
Actual CO1 Power
volutge control and balancing of reactive power of the many
hydro plant generators. As described above, voltages in the 4600

lower Columbia area were depressed for over five minutes z4400
follclwing the Allston-Keeler 500-kV line outage. The 2
B
DACS operate in this time frame to bring the generator 8 4200
a
reactive power output within unit capability curves. The
4000
DACS action is represented at John Day, The Dalles, and
Bonneville generators. Action of overexcitation limiters is Time [sec]
. .
modeled at McNary generators. Hermiston (near McNary) Simulated CO1 Power

and Coyote Springs combined-cycle plants were in power 4600 1

factor control, limiting their reactive power support. Reactive
power output at Boardman was reduced by operator
intervention, so that the plant excitation system is
represented by a constant field voltage in the simulations.
4000
Representing these voltage controls allowed matching of the
recorded and simulated voltages in the lower Columbia area. 0 10 20 30 40 50 60 70 80 90
Time [sec]
5 ) Load Modifications: After all the above system Fig. 13. CO1 power -recorded and simulated using
changes were made, the simulated system still showed more modified data base
Actual Coulee Freauencv
damping than the actual system. Load modifications were
made to improve the comparison. In WSCC data base, loads 60
in Northwest and Canada are represented by constant current w
E.
characteristics. On Saturday August loth, area loads were $59.95
characterized by air conditioning loads in the metropolitan i+
h
areas and irrigation loads in the rural areas. A combination 59.9
of induction motors and various static loads were used in the I
0 10 20 30 40 50 60 70 80 90
studies. Time [sec]
Simulated Coulee Freauencv
. .
D. Validation Studies with Modified Dynamic Data 60
w
We simulated the August 10th sequence of events using E.
the modified dynamic data base. The comparison between $59.95
U
3
recorded and simulated waveforms is shown on Figures 12- h
17 (Malin voltages, CO1 power, Grand Coulee frequency,
and PDCI power respectively). There is very good agreement
59.9

0 10 20 30 40 50 60 70 80
1
90
between the recorded and simulated quantities. Time [sec]
Fig. 14. Grand Coulee unit 20 frequency -recorded and simulated using
modified data base
Actual Malin Voltage
3000 I 1
5450r-- A 1\11 2950

2900

2850

2800
2750

!io0
0 10 20 30 40 50 60 70
\I,II I I I
60 90
27000 10 20 30 40 50 60 70 80
Time [sec]
Simulated PDCI Power
Simulated Malin Voltage

5
0
0 0
10
1 ’
20 30 40 50
Time [sec]
Fig. 12. Malin 500-kVvoltage -recorded
modified data base
60
.
70
.
60

and simulated using

190
Time [sec]
Fig. 15. PDCI total power -recorded and simulated using
the modified data base

Authorized licensed use limited to: University of Waterloo. Downloaded on December 01,2020 at 19:45:07 UTC from IEEE Xplore. Restrictions apply.
 973

actual event. The behavior of individual converters agrees

Simulated PDCl Celilo Existing Power
well with the recordings.
3. Figures 12-14 show that average profiles of the CO1
power, system voltages and frequency agree with the
recordings.
4. Phase shift between simulated power oscillations on the
COI, British Columbia-PNW ties, Coulee, and PDCI is close
to the recorded.
. .
Actual PDCl Celilo Existing Power 111. CONCLUSIONS
1900

1850 1. The existing WSCC system dynamic data base was

$1800 inadequate to simulate the August 10, 1996 WSCC outage.
1750
2. The modified data base allowed a good reproduction
a of the August 10, 1996 event based on comparison of time-
17GO
domain simulations with disturbance recordings. The
16500 10 20 30 40 50 60 70 80 90 modifications include:
Time [sec]
a) a detailed Pacific HVDC Intertie model,
Fig. 16. PDCI Celilo old terminal power -recorded and
simulated using modified data base b) action of BPA Automatic Generation Controls,
c) blocking turbine-speed controls on large steam-turbine
It40
Actual PDCl Celilo Expansion Power generators,
1120
d) action of high-side voltage controls on the lower
~1100
Columbia hydro powcr plants,
z;1080 e) dynamic load representation.
3 1060
Model changes from the Validation studies are used in
1040
operating transfer capability studies by WSCC utilities on an
1020
interim basis.
0 10 20 30 40 50
Time [sec)
60 70 80 90 3. A continuing effort by WSCC utilities is needed to
Simulated PDCl Celilo Expansion Power improve dynamic models used in studies of inter-connected
power system. The WSCC has a large program underway to
1120
verify power plant dynamic data. Development and use of
dynamic load models is highly desirable [6].
4. Based on the validation study results, the BPA has
pursued various transmission system reinforcements to
1040 - improve system reliability. These reinforcements include
schemes for fast insertion of series and shunt capacitors,
Time [sec]
Fig. 17. PDCI Celilo new terminal power -recorded and improvement of PDCI dynamic control characteristics, and
simulated using BPA-modified data base modulation controls.

The modified model meets with the validation criteria: IV. ACKNOWLEDGMENTS
1. The frequency and damping of the inter-area oscillatory
mode are computed for the actual and simulated Malin We gratefully acknowledge contributions by many
voltages at the beginning and end of the oscillations, and engineers. Richard Bunch (BPA) was our expert on
presented in Table 11. modeling the PDCI controls. Dr. John Hauer (Battelle
Table I1 Pacific Northwest Laboratory) contributed to the Prony
Frequency and Damping of the Malin Voltage Oscillations analysis of power oscillations. Solomon Yirga (Powertech
Labs), and Mani Venkatasubramanian and his students
Time Frame Recorded Simulated
Beginning of Oscillations 0.266 Hz 8 0% 0.247 Hz Q 0%
(Washington State University) performed initial simulations
M e r AGC Pick-up 0.242 Hz Q -2.66% 0.237 Hz 8 -2.62% and eigenanalysis, as well as load sensitivity studies. Gary
Prior to Break-up 0.217 Hz Q -7.62% 0.208 Hz 8 -6.35% Keenan and Ravi Aggarwal (BPA) helped prepare the
dynamic data. Many WSCC engineers helped prepare the
2. Figures 15-17 indicate that the simulated PDCI behavior power flow base case.
and mode shift are identical to those observed during the

Authorized licensed use limited to: University of Waterloo. Downloaded on December 01,2020 at 19:45:07 UTC from IEEE Xplore. Restrictions apply.
974

V. REFERENCES related to system planning and reliability. He has participated in

various technical committees in WSCC, IEEE, and CIGRE. Mr.
Western systems Coordinating Council (Wscc), Mittelstadt is a registered Professional Engineer in the State of
Disturbance Report for the Power System Outage that Oregon.
Occurred on the Western Interconnection on August
loth, 1996 at 1548 PAST, October 1996. Available at
www.wscc.com. VII. APPENDIX
A. Hammad, R. Minghetti, J. Hasler, P. Eicher, R. The PDCI connects the Pacific Northwest (Celilo station)
Bunch, and D. Goldsworthy, “Controls Modeling and with the Los-Angeles area (Sylmar station) by a 1,361km
Verification for the Pacific Intertie HVDC 4-Terminal direct current line. The original bi-polar HYDC system was
Scheme,” IEEE Transactions on Power Delivery, Vol. commissioned in 1970 as a +/-400kV,1800A scheme. Each
8, No. 1, pp. 367-375, January 1993. converter consisted of three 133kV mercury arc groups. In
J. F. Hauer, “Application of Prony Analysis to the 1979, the converter current rating was increased to 2000A.
Determination of Modal Content and Equivalent Models In 1985, the terminals were upgraded to +/-5OOkV by adding
for Measured Power System Response,” IEEE a lOOkV thyristor group in series with the existing mercury-
Transactions on Power Systems, Vol. 6, No. 3, pp. arc groups. In 1989, the system was expanded by adding
1302-13 12. August 1995. 1100A, +/-5OOkV thyristor converter terminals in parallel
, U

[4] IEEE Committee Report, “MW Response of Fossil with the Existing converters.
Fueled Steam Units,” IEEE Transactions on Power Presently, the Pacific HVDC intertie is capable of
Apparatus and Systems, vol. PAS-92, no. 2, transmitting up to 3100MW in either North-to South or
March/April 1973. South-to-North directions. A one-line diagram of the present
[5] EPIC Engineering Company, Impacts of Governor Pacific HVDC intertie configuration is shown in Figure A1
Response Changes on the Security of North American (for one pole).
Interconnections, EPRI Report TR-101080, October
1992. CELILO SYLMAR
[6] IEEE Load Representation for Dynamic Performance .
EXPANSION

eEXPANSION

1
1

Task Force, “Standard Load Models for Power Flow and 5 0 0 k V m DC

at rectifier 220kV
Dynamic E’erformance Simulation,” IEEE Transactions
1100 A 1100 A
on Power Systems, Vol. 10, No. 3, pp. 1302-1313,
August 1995. 230W 2000 A 230kV

VI. BIOGRAPHIES
Dmitry N. Kosterev (M’ 97) received his Ph.D. degree in
EXISTING EXISTING
Electrical Engineering from Oregon State University in 1996. Dr.
Kosterev is currently with NSR at the Bonneville Power
Administration, Transmission Operations and Planning. He is Figure Al: PDCI block-diagram
involved in operating studies for the California-Oregon Intertie
projects, Pacific HVDC system modeling and controls, and wide- The PDCI controls have a hierarchical structure. PDCI
area monitoring systems. Dr. Kosterev was responsible for the normally operates in power control -at ‘the pole level. Pole
validation studies of the August 10, 1996 disturbance at BPA. controls compute current references for individual
Carson W. Taylor (F’88)joined Bonneville Power Administration converters, subjected to various static and dynamic
in 1969 after earning degrees from the University of Wisconsin and constrains. The converter controls are different for the
Rensselaer Polyiechnic Institute. His current position is Principal Expansion and Existing converters. While the Existing
Engineer in Transmission Operations and Planning. Mr. Taylor is a converter controls are proportional current regulators, the
Fellow of the E E E . He chairs the IEEE Power System Stability Expansion converter controls are of integral type with
Controls Subcommittee. He has also convened three CIGRE task several operating characteristics. Converter controls use
forces. Mr. Taylor is the author of the EPRI-sponsored book Power
different set of parameters depending on rectifier or inverter
Syslern Voltage Stability and has authored or co-authored many
IEEE and CIGRE papers. Mr. Taylor also operates Carson Taylor mode of operation.
Seminars, a training and consulting company established in 1986.

William A. Mittelstadt (F’92) received B.S. and M.S. degrees in

Electrical Engineering from Oregon State University in 1966 and
1968 respectively. Mr. Mittelstadt currently is Principal
Transmission System Planning Engineer at the Bonneville Power
Administration, where he responsible for a wide range of efforts

Authorized licensed use limited to: University of Waterloo. Downloaded on December 01,2020 at 19:45:07 UTC from IEEE Xplore. Restrictions apply.
 975

Discussion is indeed a small-signal stability problem. Also, it is

important to note that although system voltages were
indeed depressed during the incident, it can be
Kip Morison, Solomon Yirga, , and Prabha
demonstrated that the instability would have occurred
Kundur (Powertech Labs Inc.): The authors are to even if voltages were much healthier. The impact of
be complimented for a well written paper describing some capacitor additions on damping is shown in Table 1.
important modelling issues in the replication of the
August 10, 1996 WSCC system disturbance. However, Critical Modelling Issues
there are a number of critical issues which are not
presented in the paper and which merit discussion. The authors present some interesting work in
demonstrating the importance of accurate modelling of
Fundamental Nature of the Phenomenon
HVdc, AGC, turbine speed controls, and generator
voltage controls in matching certain simulated quantities
In determining critical modelling issues, it is essential to (such as frequency) to recorded responses. However, the
understand the fundamental phenomenon causing system authors have failed to show their relative importance in
instability. It is well known that the WSCC system is affecting the stability of the system.
prone to lightly damped low frequency interarea
oscillations in the frequency range of 0.2-0.3 hz; they Our analysis has shown that by far the two most critical
have been observed on the system many times. These modelling elements in reproducing this oscillatory
oscillations are a characteristic of groups of machines disturbance are load characteristics and generator
oscillating against other groups of machines through weak excitation controls. Both are relatively uncertain
ties. This is a phenomenon associated with small-signal (particularly load) and changes to either can profoundly
stability, and is a function primarily of networklgenerator change the system response. In fact we were able to
topology and excitation controls; our studies have shown obtain the undamped oscillations and replicate the actual
the August 10* event to be of this nature. system response without resorting to the use of dynamic
load models. With the "standard model" (the original
Following the Menvin-St. Johns 115kv line tripping and WSCC model as described in the paper), the system was
the initiation of the dropping of the McNary units, the well damped. As the percentage of constant MVA load
structure of the WSCC system became such that it is increased, the system damping decreases drastically.
exhibited small-signal instability and the 0.28 hz Table 1 shows how the system damping changes in
oscillations became very pronounced. Although the response to modelling modifications. Figure 2 shows the
actions of HVdc and AGC ultimately had some impact on dramatic impact on system stability of modifications
the system damping and frequency of the system made to load only.
oscillations, they were not the fimdamental cause. This is
a conclusion which we have demonstrated by time-
domain simulation and further illustrated by eigenvalue
analysis [ 13. Using eigenvalue analysis to determine
which machines are contributing to the oscillations and
which controls are most impactive, simple modifications
to the power system stabiliziers (PSS) on just three plants
in Southern California can be shown to eliminate the

0 10 20 30 40 50 60 70 80
Time in seconds

Figure 2 :Impact of Load Modelling

Simulation Model Damping of

Critical Mode
Standard WSCC Base Model 8.1%
Time in seconds
Load Model Modifications + AGC modeled I -3.2%
Figure I :Effutof PSS Modijications on Simulation Results Load Model Modifications + AGC modeled I -2.7%
1 -t John Day Capacitor added
undamped oscillations completely (see Figure 1) and
prevent system instability; clearly demonstrating that this Table 1: Impact on Damping of Various Modelling Changes

Authorized licensed use limited to: University of Waterloo. Downloaded on December 01,2020 at 19:45:07 UTC from IEEE Xplore. Restrictions apply.
 976

Preventing Future Disturbances \

4 Loh
Model-Based Analysis
While w e appreciate the authors intentions in replicating
the system response through simulation, we feel a j I - I
I
Eigenvalue
Analysis
1 I

different perspective on modelling is needed when ! I Model b/ - I.’I I I I

studying the WSCC system. To prevent exposure to
similar disturbances in the future requires analysis and
mitigation of the fundamental stability phenomena
occurring in the system. The lesson from August lo’, is
that the WSCC system is prone to small-signal instability.
To :study this type of system we must,
(a) Use appropriate models in simulations. For small-
signal stability assessment, the most critical and 1 DISTURBANCE L--- System Tests
uncertain models are load characteristics and I
Measurements
generator excitation controls. ‘ Measurement-Based Analysis
I__----_______________________________J
‘I

(b) Use appropriate study methods. In addition to time-

Fig. HI. Systems analysis for model validation.
domain simulations and Prony analysis presented by
the authors, eigenvalue analysis is invaluable in SOWG was, in effect, a template for a “virtual” systems engineering
assessing and tuning controls. Also, the system must staff distributed across multiple institutions. Its work in model
be studied as a whole, and controls adjusted with due validation, represented in Fig. HI, required a range of staff skills
and tools that would rarely be found within a single utility. Very
regard to low frequency inter-area oscillations. similar resources and workflow are needed for wide area control
(e.g., FACTS), for which SOWG had essential support
Reference responsibilities to the WSCC [2,4]. Is some new organization
evolving to resume such functions?
[l] EPRI Final Report TR-108256 “System Disturbance
Stability Studies for Western System Coordinating
Council (WSCC),” September 1997, prepared by
Powertech Labs Inc.
Mariuscript received February I S, 1998.
JOHN F. HAUER (Pacific Northwest National Laboratory):
The authors have produced an excellent piece of technical work.
Speaking unofficially for the WSCC‘s System Oscillations Work
Groups (SOWG), I would like to address the decline in engineering
resources that has made that work necessary.
Both of the WSCC breakups in the summer of 1996 were the result
f2 610 .
242

of known problems that had persisted for too long. To some extent ,
this reflects a progressive fragmenting, and effective loss, of 1.1 B
collective knowledge under the erosive pressures of utility E
e
restructuring. It also reflects an emerging attitude toward system
reliability that is very different from that of the former regulated ReferenceT i m - OIUlOIoB 15H35mMs PDT
, -1-
power industry.
uo‘
i {oo io0 im ao iw im 700 Bw
210

T i m In Second,
From 1988 to 1994 the WSCC, through SOWG, conducted a
massive effort to improve its planning models and its planning Fig. H2. Voltage signatures following Keeler-Allstonline
practices [l-41. Key findings, based upon many tests and trip on August 10, 1996.
disturbances, were that
The risk of 0.7 Hz oscillations is overestimated. 008MalinRoundMountaint l MWMW
1550
t
The risk of 0.35 Hz (PACI) oscillations is underestimated.
“This could become important as imports from Canada
increase and Canadian generation resources expand
1500 4I II
northward.” [3]
Modeling for prime movers is quite optimistic, and affects
damping estimates.
4 Greater emphasis and resources should be given to
- direct (measurement based) analysis.
- advanced analysis tools.
With the arguable exception of HVDC effects, it appears that all of
the modeling problems and repairs found by the authors are among
those reported some years earlier by SOWG. The authors comments
on this would be appreciated.
SOWG advocated and, just a few years ago, drew upon a much
richer technical infrastructure than most planners enjoy today.
Prony analysis and the operating records that the authors use so IlSO{
t
extensively are both products of the SOWG effort. At BPA, support P.L-Tku=l&aaSO PDT ifi
of SOWG model validation work was an important formal 1100
300 350 400 450 5C#ei~$~Pe,m6$0 650 700 750 800
requirement in development of the monitor facilities that recorded
the cascading outages of July 2 and August 10. Recommendations
in 121 underlie most and perhaps all of the monitor projects in the Fig. H3. Oscillation buildup for the WSCC breakup of
WSCC. August 10,1996.

Authorized licensed use limited to: University of Waterloo. Downloaded on December 01,2020 at 19:45:07 UTC from IEEE Xplore. Restrictions apply.
 977

Table H I . Observed behavior of the PAC1 mode P. Donalek (Ham Engineering CO., Chicago, IL) :
DatelEvent Frequency Damping
12/08/92 (Palo Verde trip) 0.28 Hz 7.5 % Would the authors comment on the usefulness of the
03/14/93 (Palo Verde trip) 0.33 Hz 4.5 %
07/11/95 (brake insertion) 0.28 Hz 10.6 % simulation models as a basis for a study of system
07/02/96 (system breakup) 0.22 Hz 1.2 % stability aids? Specifically, could the model be used
PAC1 mode on August 10,1996 to make a study of the effectiveness of 8 proposed
ThimelEvent Frequency Damping pumped storage hydro plant connected at the Captain
10:52:19 (brake insertion) 0.285 Hz 8.4 % Jack substation? It has been suggested that if the
14:52:37 (John Day-Marion) 0.264 Hz 3.7 %
1518 (ringing) 0.276 Hz proposed pumped storage plant had been in operation
15:42:03 (Keeler-Allston) 0.264 Hz 3.5 % and if the plant had adjustable speed machines, then
1545 (ringing) 0.252 Hz
15:47:40 (osc.startup) 0.238 HZ -3.1 % it could have prevented the spread of the disturbance
15:48:50 (0%. finish) 0.216 HZ -6.3 %
into the California system. For a description of this
As to technical details, records from the August 10 breakup should
suggestion see "System Reliability Improvement with
provide a very fruitful base for research. Figures H2-H3 show Adjustable Speed Machines." The letter was
indications of impending trouble for several minutes before actual published on page 45-46 of the February 1998 issue
breakup on August 10. While Table H1 indicates unusually low
frequency and damping for the PACI mode following both line of IEEE Power Engineering Review; vol. 18, no. 2.
trips, voltage recovery was much more robust for the earlier trip.
On-line use of such signature information, as a form of Manuscript received June 1, 1998.
measurement based dynamic security analysis, would be a useful
precaution against errors in estimating transfer capacities [5].
Other salient issues are whether WSCC practices concerning PSS Dmitry Kosterev, Carson Taylor, and Bill
tuning and operation are appropriate, and the role of HVDC controls Mittelstadt: We thank Messrs. Morison, Yirga, Kundur,
in the August 10 breakup. Disturbance records for Palo Verde
generator trips place that plant at the end of the PACI mode Donalek, and Hauer for insightful discussions.
eigenvector, and suggest that very small injections of real power
there would have quenched the PACI oscillations [4-51. Do the Response to Kip Morison, Solomon Yirga, and Dr.
enhanced planning models also indicate this - and do they replicate
the Palo Verde trips? Prabha Kundur (Powertech Labs): We see the August
HVDC effects may be a case of a disruptive limit cycle that was not 10, 1996 WSCC system outage as a broader system
visible to linear analysis. Concerns that mode switching might phenomenon than a small signal stability problem.
produce this are as old as the PDCI itself but, to my knowledge, the Although the instability was manifested in growing
right combination of model and "killer case" has not been found oscillations, the outage resulted from operating the
before. Studies of voltage-referenced inverter control at Sylmar,
reported in [I], changed the ratio of ac and dc swings but had little power system under very stressed conditions with
effect upon their sum or upon swing damping. This insensitivity several transmission facilities out-of-service com-
has been observed many times and is a primary reason for the low pounded by unforeseen factors.
priority previously assigned to full PDCI modeling. Using their Pacific Northwest generation patterns have changed
enhanced model, could the authors provide the participation factors
for HVDC involvement in the PACI mode, and those for the Palo due to anadromous fish recovery programs limiting
Verde and/or San Onofre machines? Also, is there any information number of hydro generating units on the lower
concerning performance of the SVC dampers near Sylmar? Columbia River. As the paper describes, power flows
characterized by high generation on the upper Columbia
J. F. Hauer, "Reactive Power Control as a Means for Enhanced River, and high imports from Canada lengthen the
Interarea Damping in the Western U. S. Power System-A Northwest-California transmission path. Low
Frequency-Domain Perspective Considering Robustness generation in the lower Columbia reduces dynamic
Needs," Application of Static Var Systems for System
Dynamic Performance, IEEE Publication 87THO187-5-PWR, reactive power and inertial support for the ac and dc
pp. 79-92. interties. The electrical center of the north-south
Evaluation of Low Frequency System Response: Study oscillations also moves further north, closer t o the
Results and Recommendations. Report of the WSCC 0.7 Hz Celilo HVDC terminal. With such generation pattern,
Oscillation Ad Hoc Work Group to the WSCC Technical the system is more susceptible to power oscillations.
Studies Subcommittee, September 1990.
On August loth, outages of several 500-kV lines in the
J. F. Hauer, "Update on 0.7 Hz Dynamics in the Western Power
System," BPA 1992 Engineering Symposium, March 3 I-April lower Columbia area and Keeler-Allston 500-kV line
I , 1992, Portland, OR, pp. 187-198. further increased the north-south angular separation and
J. F. Hauer and J. R. Hunt, in association with the WSCC caused reactive power deficiency in the Northwest. The
System Oscillations Work Groups, "Extending the Realism of BPA-developed phase angle monitor indicated that the
Planning Models for the Western North America Power relative phase angle between the Grand Coulee and
System," V Symposium of Specialists in Electric
Operational and Expansion Planning (SEPOPE), Recife (PE) Malin 500kV busses increased from 59 degrees to 69
Brazil, May 19-24, 1996. degrees following the Keeler-Alston 500kV line outage.
J. F. Hauer, D. J. Trudnowski, G. J. Rogers, W. A. Mittlestadt, The AGC action following McNary tripping spread the
W. H. Litzenberger, and J. M. Johnson, "Keeping an Eye on inter-area angle further. At that time, if not growing
Power System Dynamics," IEEE Computer Applications in oscillations, insufficient synchronizing power or voltage
Power, pp. 50-54, October 1997.
collapse would likely have caused the system breakup;
Manuscript received February 18, 1998. see SCADA recordings in Figure 5. Other contributing

Authorized licensed use limited to: University of Waterloo. Downloaded on December 01,2020 at 19:45:07 UTC from IEEE Xplore. Restrictions apply.
 978

factors were units operating on power factor control and studies. Inadequate modeling of Pacific HVDC Intertie
operator intervention at a key unit to disable voltage in WSCC stability program resulted in 150MW
control. reduction of the operating nomogram for 1997 Summer
Our sensitivity studies clearly demonstrate that by season. WSCC Modeling and Validation Working
reducing the north-south electrical distance (inserting Group set detailed HVDC representation as one of the
series capacitors at Fort Rock) and providing additional top modeling priorities in WSCC production programs.
reactive power support (shunt capacitors at Hanford and We agree that eigenanalysis complements time-
John Day) the August 10, 1996 outage could have been domain simulations. Eigenvalues and mode
prevented. Other sensitivity studies demonstrate that participation factors can identify controls that need
had the McNary units not tripped, the system would retuning. Controllability and observability indexes
have remained stable. indicate the most effective location for damping
We agree that low-frequency oscillations in WSCC, controllers (such as SVCs) and the input signals
0 . 2 4 3 3 Hz Pacific intertie mode, should be addressed providing robust performance. Such studies require
by proper tuning of power system stabilizers. This is complete system representation. For special wide-area
further discussed in response to Dr. Hauer. We have controls, however, we are finding that non-linear
been investig(3ting feasibility of reactive power controllers, developed mainly by time simulation, are
modulation and switching to provide additional positive producing more robust performance.
damping to the system. We designed and implemented Response to Dr. John Hauer: Dr. Hauer offers both
AC -Voltage Dependent Current Order Limiter at Pacific a broad perspective on root causes of the summer 1996
HVDC Intertie to correct the PDCI dynamic response disturbances and many specific comments. We agree
for low ac voltage conditions at Celilo. The new that power company engineering resources have
function is an emergency controller which releases declined over the last decade with downsizing because
reactive power from the converter when ac voltages are of ever increasing cost pressures. As in other industries.
depressed and also can provide positive damping. we must now do more with less, hoping that technology
Negatively-damped oscillations can be created in a advances can compensate for smaller staffs.
coniputer program in different ways. According to the Although reliability tends to be gradually neglected
discussors’ reference, their load model in BC Hydro and during extended periods without major outages, the
the Northwest was changed to constant P and Q models stated policy is that reliability must not be compromised
to create negative damping without regard to actual with restructuring (just as airline deregulation did not
estimated load composition and characteristics. Our compromise safety). This will be difficult, but possible,
purpose, however, was to realistically simulate the in future power systems. Transmission companies,
instability mechanisms. While we agree that load working together and with owners of other power
models have the most significant impact on the system system components, will bear primary responsibility.
damping, we do not think use of static load models New organizations such as security coordination
(particularly constant power models) is the best centers, and new mandatory standards may help.
approach. A good fit for a particular condition may not Information technology to improve on-line security
be a good fit for other conditions. The effect of dynamic assessment is required
loads on damping is known; see reference 7. Both the WSCC System Oscillations Work Group
We investigated various load, characteristics and and the WSCC Modeling Work Group (groups now
concluded that the most realistic load representation merged) provided recommendations that were approved,
included dynamic representation of motor load. but not implemented. Modeling Work Group load
Saturday August loth was a very hot day through the modeling recommendations are similar to IEEEIPES
western states and assumption of air-conditioning load, recommendations [ 6 ] .
along with irrigation and other pumping load, is very We think the WSCC PSS tuning and operation
reasonable. While more investigation is desirable, we practices are generally adequate. The main problem for
see no reason not to model estimated motor load as Pacific Northwest units has been hardware-related -
aggregated dynamic devices. noisy frequency transducers and old analog stabilizers.
We are convinced that the detailed HVDC model is Newer accelerating power stabilizers are more effective.
essential for accurate reproduction and understanding of Upgrading of over 100 hydro units is costly and time
the event. Especially important is the multi-terminal consuming. Several large plants have dc commutator
representation of PDCI, and the control differences of exciters that make PSS less effective.
the mercury-arc and thyristor converters. According to Regarding further use of the detailed multi-terminal
the discussors’ reference, it was necessary, using their dc model, we will report results in future papers.
simplified single terminal HVDC model and controls, to As far as on-line monitoring, the BPA has taken
raise the minimum firing angle from 5 to 9 degrees to several significant steps in this direction. A reactive
somewhat approximate the PDCI behavior. Again, this reserve monitor is implemented to alarm operators when
is i i change to fit the system response that doesn’t have a reactive reserves at lower Columbia hydro plants are
physical basis. becoming too low. The BPA also designed and
PDCI is a major transmission element in the WSCC implemented a Phase Angle Monitor, which alarms a
system, transferring up to 3100MW of power, and its dispatcher when the phase angles between Coulee, John
detailed representation is essential for the inter-area Day, and Malin are exceeding safe operating limits.

Authorized licensed use limited to: University of Waterloo. Downloaded on December 01,2020 at 19:45:07 UTC from IEEE Xplore. Restrictions apply.
 979

The BPA also initiated a probing signal project, when the first observed since 1976 [9] - which was the last
an operator can inject a step input in the PDCI current previous year with excellent hydroelectric generation
order to stimulate the ac system response. On-line conditions in the Pacific Northwest and British
analyzers will compute damping of power oscillations Columbia. The 1976 oscillations occurred during light
to ensure that the system has sufficient positive damping load hours with high power transfers from Canada to
margin. California, and with low generation at lower Columbia
Response to Peter Donalek: The developed River plants.
simulation model is being used for testing various Following the instability, the cost of the breakup
stability aids such as supplementary controls applied to was greatly increased because a controlled separation
existing static var compensators, the Slatt thyristor- scheme was out of service, and because of the
controlled series capacitor, and switched shunt undesirable tripping of generators in the southern
capacitors at Malin. Any device improving voltage islands despite adequate underfiequency load shedding.
support to California-Oregon Intertie, as suggested The WSCC has taken steps to restore use of the NE/SE
pump storage plant, should be beneficial for the system Separation Scheme which islands the northern and
stability. Implementation and operational costs of any southern parts of the WSCC grid for loss of the
solution should be carefully evaluated. California Oregon Intertie, and enhance it with selective
In summary: The discussion and closure process load tripping.
adds to the value of the paper, and to the documentation We conclude that defense-in-depth is essential to
of the power failure. Because of the importance of this limit cascading and prevent rare combinations of factors
outage to the power industry, we offer the following from causing very costly power failures. The modeling
concluding comments. issues discussed are critical elements in preventing
The WSCC has made simultaneous transfer rating future failures.
determination process much more rigorous. Season by
season studies are made, with the challenge being References
development of more accurate and less manpower [7] J. M. Undrill and T. .F. Laskowski, “Model
intensive on-line security assessment. Generator model- Selection and Data Assembly for Power System
ing is being improved by a WSCC-wide power plant Simulation,” IEEE Transactions on Power
testing program. Controls such as overexcitation Apparatus and Systems, Vol. PAS-101, NO. 9, pp.
limiters will be routinely modeled. Multi-terminal 3333-3341, September 1982.
HVDC models will be used in future studies. [SI C. W. Taylor and D. C. Erickson, “Recording and
Challenges remain for the modeling of loads and prime Analyzing the July 2 Cascading Outage,” IEEE
mover controls. Computer App[ications in Power, Vol. 10, No. 1,
Beyond the modeling, simulation and control pp. 26-30, January 1997.
desigdtuning issues, it’s evident that many factors [9] R. L. Cresap, D. N. Scott, W. A. Mittelstadt, and C.
contributed to the severity of the August 10” failure. W. Taylor, “Damping of Pacific AC Intertie
The many recommendations of reference 1 of the paper Oscillations via Modulation of the Parallel Pacific
are being implemented. HVDC Intertie,” CIGRE 14-05, 1978.
Except for the July 2, 1996 event [SI, the
undampednegatively damped intertie oscillations were Manuscript received June 29, 1998.

Authorized licensed use limited to: University of Waterloo. Downloaded on December 01,2020 at 19:45:07 UTC from IEEE Xplore. Restrictions apply.