You are on page 1of 10

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

1, February 1999

Voltage Stability Contingency Screening and Ranking


E. Vaahedi C. Fuchs W.Xu' Y. Mansour H. Hamadanizadeh G. K. Morison
Senior Member Member Senior Member Fellow Member

B.C. Hydro Powertech Labs. Inc


Burnaby, Canada Surrey, Canada
ABSTRACT - The objective of contingency screening and project was initiated to address the above needs with the
ranking function is to shortlist a specified number of critical following basic functions:
contingencies from a large list of credible contingencies and
rank them according to their severity. This paper summarizes * contingency screening and ranking
the work conducted as part of the EPRIA3.C. Hydro on-line secure region determination
voltage stability project in developing a Contingency remedial action determination
Screening and Ranking (CS&R) module. The two methods of
Reactive Support Index (RSI) and Iterative Filtering are The contingency screening and ranking function has the
derived in this paper and tested on the large scale systems of objective of short-Iisting an operator specified number of
critical contingencies from a large list of credible
B.C. Hydro and another Major Unnamed Utility. The results contingencies and rank them according to their severity.
obtained indicate that RSI on its own or in combination with
the Iterative Filtering method can be used for CS&R The majority of the relevant papers have attempted to obtain
depending on the acceptable level of misranking. The RSI a faster method for voltage stability margin calculation [ 1, 3-
method is a very fast and powerful CS&R method and is 81. Reference [l] uses two operating conditions to estimate
suggested for systems where some misclassification of the margin. While the method is fast and the results presented
contingencies can be tolerated. On the other hand, for in the paper give proper ranking for the contingencies, the
systems where an exact list of critical contingencies is actual margins calculated have relatively noticeable errors.
intended, then the Iterative Filtering method can be used for Reference [2] investigated the problem using the concept of
screening complemented with another method like RSI for equivalenced one-branch system. While the method may
ranking. The latter technique is being integrated in the work for radial systems, technically, it is not applicable to
EPRU6.C. Hydro On-line Voltage Stability tool. practical systems. Reference [ 5 ] computes approximate
Keywords: Voltage Stability, Contingency, Screening, margin to voltage collapse using voltage gradients
Ranking determined at a point on the stable equilibrium branch of the
voltage versus stress curve and at the corresponding point on
1- Introduction the unstable equilibrium branch. The margin is a first order
approximation of the voltage stability margin. The basis of
Currently many utilities perform off-line studies to determine this method is that in a heavily loaded power system, there
voltage stability margin and control actions necessary to exist two very close load flow solutions with the same stress
ensure stability for a restricted set of system conditions. value but one with higher voltage solutions than the other.
These results are then compiled into a look-up table for use However, test results obtained in reference [l] showed that
by the system operators. This approach has the following the accuracy of the method is not very satisfactory.
drawbacks: Reference [7] proposed a test function method for the
estimation of critical stress level. The test function is
0 significant amount of off-line analysis essentially derived from a linear approximation of the PV
e Iimited number of considered operating conditions curve. Accuracy of the results, as observed by reference [l]
0 limited number of considered contingencies are not acceptable. A second order approximation of the QV
limited or no information on remedial actions is proposed in reference [SI. This method is further extended
to the PV curve case as well [ 11 .
In light of the importance of voltage stability to system
afid the short-cominge of the
pe~fol-fnafice off-line methods. Another diiejorion of oonringeiioy swlsirtion rrurwrh is to ' ~ 1 s ~
there exists a strong need for a full-function on-line voltage indices for contingency ranking. There are two different
stability method. EPRUB.C. Hydro On-line Voltage Stability types of indices proposed for this purpose; a) structural
indices and b) physical indices. The 'structural indices are
A "

PE-002-PWRS-0-10-1997 A paper recommended and approved by those which use indices derived from system models
the IEEE Power System Operations Committee of the IEEE Power indicating singular value decomposition ll. The
EngineeringSociety for publication in the IEEE Transactions on Power
Systems. Manuscript submitted April 25, 1997; made available for physical indices are those which take advantage of Some
printing October 21, 1997. system quantities for measuring their closeness to voltage
instability. Reference [9] suggests to use the total reactive

* Presently with the University of Alberta


0885-8950/99/$10.00 0 1997 IEEE
257

power generation of a system as an index and develops This method was later substituted by another similar method
linearized methods to determine the sensitivity of the reactive named Reactive Support Index (RSI) which proved to have a
power with respect to system disturbances. Similarly, better performance (accuracy, speed and implementation
reference [12] introduces an index based on the change of the ease) than RCI. Due to space limitations, no results will be
generated reactive power as well as the sensitivity of reactive given here for the RCI method.
load to voltage. A similar concept was developed [ 131 where
an index based on generator reactive power output was The Reactive Support Index is also based on the premise of
proposed to measure the severity of contingencies. Also an defining the severity index, for a contingency, as the extra
index based on the derivative of the total reactive power to amount of reactive generation required to get from the
the active and reactive components of loads at different buses normal case nose to the contingency nose. Unlike the RCI
has been suggested [l 11. index which needed artificial reactive sources at load buses,
the Reactive Support Index is defined as the extra amount of
In the following sections, the new methods derived here for reactive generation from all the existing dynamic VAr
CS&R are discussed together with their test results on the devices (generation, SVC, etc.). To establish the extra
systems of B.C. Hydro and another major unnamed utility. reactive generation the reactive limits at the dynamic VAr
devices are ignored (Figure 2)

2.0 Reactive Support Index


Reactive Gen. after
Pre-contingency Contingency ;Open Limits
This method evolved from the severity indices based on
\ /
reactive generation [1,2] as well as an earlier method
developed in this prqiecc at B.C. Hydro called Reactive
Compensation Index (RCI) described here briefly. The RCI
method is based on the premise that the distance between the \ r=..
normal case (pre-contingency) nose and the contingency
case nose can be approximated by the total reactive injection Contingency
required at the load bwes to establish similar voltage levels
for both cases (Figure 1 ). In order to implement this method
\ :.
the following steps needed to be followed:
-e--.
_ _ _ - - ----
-
Reactive Gen
0 Obtain a basecase the nose of the normal case or a very ________------ Basecase Nose
stressed case close I O it
Add artificial synchronous condensers at the load buses
with voltage settings equal to the load bus voltage value
Apply the contingency
I Figure 2: Reactive Support Index
Q

I Solve the load flow, establish the reactive output of the


The RSI is computed as follows:
1 artificial condenser!;
Define RCI as:
0 Obtain a basecase reflecting the nose of the normal case
e or a very stressed case close to the nose.
RCI, = C m ,
J=1
QYl (1) Implement the contingency
0 Solve with the dynamic reactive devices limits open
where RCI, is the relative RCI index for contingency i,
Q,,” is the reactive power generation at the artificial Define RSI as:
dynamic Var resource added at the load bus j, I is the
total number of the load buses and m, is the weighting
factor for the Var resource at load bus J.
where RSI, 1s the relative RSI index for contingency
1, QY and QYlo are the reactive generation of the jth
RCI Index dynamic VX device in the pre-contingency case and
after contingency i with open reactive limits respectively
, g 1s the total number of the dynamic Var devices, m, is

Contingency) Basecase the weighting factor for the 1 th dynamic Var resource
and p is a factor reflecting the order of the index for
removing the masking effect.

In our implementation of RSI, the best results were obtained


I for p=l and normalization of the reactive generation at
different units based on their Qmax; if the Qmax was zero
Figure 1: BC Hydro RCI Index the unit’s MVA was used instead.
258
2.2 Studv Results reference. Similar results were also obtained for the other
system.
2.2.1 Studv Systems
2.2.3 RSI Results
The B.C. Hydro and another major power systems which
exhibit completely different characteristics were used as the The RSI method was used to screen and rank the
test systems for this project. The particulars of the system contingencies for the systems of B.C. Hydro and, another
basecases used are given in Table 1. The contingencies major utility. To obtain the RSI, the 80% stressed basecases
considered in these systems were all branches with voltages of the normal systems were used. Figures 4 and 5 compare
higher than 100 kV. the performance of normalized RSI against X-Stress results.
The scaling described in Section 2.2.2 was used here for
Basecase Details 1 BCHydro I AMajor Utility plotting the results. Also, a value of zero was assigned to
any contingency which could not be solved.

These figures indicate that the RSI method closely follows,


with some noise, the pattern of exact margin calculations.
Table 1: BCH and unnamed utility Power System Statistics This success is owed to the fact that RSI and RCI methods
both take into account the stress pattern by starting from the
nose (or very close to it) of the normal case.

To assess the perfbrmance of the screening and ranking


methods, the voltagd stability margin of all the contingencies 1.4 T
1'
1
were first obtained using a repetitive load flow method. p
Contingencies were then ranked based on their voltage
stability margins. This ranking is used as a reference for
measuring the accuracy of other techniques. Figures 3 shows
the ranking of all selected system contingencies for the B.C.
Hydro system based on S-Stress (increasing P and Q of a
load bus in proportion to the bus MVA) and X-Stress
(increasing only the Q of a load bus in proportion to the bus I- X-Stress I
MVA). The contingencies used were those over 100 kV (670 . . Normalized RSI
for B.C. Hydro and 202 for the other one). The X-axis in ,: 1
0 I I I 1 I I I I I I I I I I I I1 I I+ H I I I I I I I I H I I I I I I- I I- I I
these figures are contingency numbers sorted according to
the S-Stress values. The Y-axis corresponds to the
normalized value of voltage stability margin. The scaling of
Y-axis is such that zero value corresponds to the most critical
contingency (lowest stability margin) whereas "1 .O' Figure 4: BC Hydro Normalized RSI Index
corresponds to the base case margin without any
contingency.

I I

2 0.41
. . - - Normalized RSI

0.2
El S-Stress

I I

Figure 5: Normalized RSI Index for unnamed utility


I These results also indicate that this method misclassifies
Figure 3: BC Hydro X-Stress Versus S-Stress some local contingencies which need small reactive support
as opposed to system-wide voltage problems requiring large
reactive support. Close examination of the results indicate
The results indicate that the S and X-Stress rankings are very that out of the 20 most critical contingencies RSI
similar and any of the two can be used interchangeably as a misclassifies 6 for B.C. Hydro and 3 for the other system.
259
While this may be acceptable for many systems, it was contingencies. To find that particular stressed loadflow, an
decided to devise other methods which can identify exactly initial pre-contingency stressed case is started with and the
an operator specified number (e.g. 20) of the most critical number of unsolved loadflows are noted. This is repeated for
contingencies. The method derived in the following section another stressed point and the iterative process continues
addresses this requirement. using interpolation (extrapolation) until the exact number is
arrived at within the acceptable tolerance. It should be noted
In these studies, the best results were obtained when p=l and that, in general, the majority of contingencies are in the Very
the weighting factors equal to the inverse of the Qmax. Mild category and hence, after the first iteration, the number
However, it should be noted that the variation of these of loadflows to be solved are significantly less than the
parameters should not be expected to eliminate the original number of contingencies. The implementation steps
misclassifications cotnpletely because of the algorithm for the method are given below:
intrinsic approximation in calculating the voltage stability
margin. 1. Use the normal basecase loadflow (0% stressed case) to
obtain the normal case nose (100% stressed case).
2. Obtain an m,% stressed case (start with a very high initial
3.0 Exact Screening Using Iterative Filtering
value, e.g. m,=90%).
3. Solve the case with all unsolved credible contingencies
3.1 Theorv
from previous iteration and establish d, where d, is the
number of contingencies for which the loadflows do not
Figure 6 shows a series of QV curves for post-contingency
systems relative to the QV curve representing the base case solve in iteration number i.
with no outage. The closer the post-contingency Q-curve to 4. If n-t<dl<n+t then stop otherwise go to 5 where n is the
the normal curve, the less severe the contingency. Vice versa, specified number of most critical contingencies needed
the farther the post-contingency Q-curve to the left of the and t is the tolerance.
normal curve, the mora severe the contingency. Based on this 5. Modify m, to ml+Ami where Ami is obtained by
argument, the area can be arbitrarily divided into four
segments of Very Mild, Mild, High and Very High interpolation (extrapolation) of the last two values as
indicating the degree of severity of contingencies falling in given below. In the first iteration a fixed value (e.g.-
these segments. 10%)will be used for Ami.
In Figure 6, to identify Very Mild contingencies, a stressed
loadflow corresponding to the normal case at point C should
-
Am,= (n-di)-(mi-mi.J4didi.J (3)
be solved for different contingencies. Only the loadflows 6. Repeat the process from step 2.
corresponding to Very Mild contingencies will solve because
those are the ones which their QV curves cross the vertical This algorithm is very general in that it finds the user defined
line representing the loading at point C. If the ones that do number (n) of the most critical contingencies for the stress
not solve are solved using the stressed loadflow method of choice. For example, depending on the stress
corresponding to normal case at point B, then only Mild method of choice (Q, X or MVA), step 1 and 2 should be
ones will solve and so on. So given a pre-contingency case adjusted correspondingly. The results obtained will be exact.
with a particular stress level, there will be a unique group of
contingencies that should not be able to solve. Moreover as Also from a theoretical viewpoint, there will not be any
misclassifications between the groups established by this
one raises the stress level, the number of unsolvable method including the final group of the critical contingencies.
contingency cases will rise. This fact can be used as the basis
for filtering the contingencies. 3.2 Tests of Iterative Filtering

V The Iterative Filtering method described above was used to


Basecase screen out the 20 most critical contingencies for the systems
of BC Hydro and the other system.

Table 2 shows the results for the B.C. Hydro system. This
table indicates that starting with a 90% stressed case in the
first iteration, out of 670 contingencies 87 do not solve. From
this 87, 50 contingencies solve in the second iteration until
iteration number 5 where the number of unsolved
contingencies is 24. The tolerance used here was 5. It is
interesting to note that the total number of loadflow solutions
required in this case is 853 which is 1.27 times the total
number of contingencies (670).

Very High High Mild Very Mild Unsolved


Severitv
50 37
Figure 6: Iterative Filtering 3 I 76.6 I 37 I32
4 I 68.5 I 32 15 I27
The key question in this algorithm is to come up with a
particular stressed basecase which can exactly filter out an
operator specified number n (e.g. n=20) of the most critical Table 2: Iterative Filtering Results for the B.C. Hydro System
260
Table 3 presents the screening results of the Iterative Filtering The results of the combined screening and ranking on the
method for the other system. In this case in total, 401 systems of B.C. Hydro and the unnamed utility are shown in
loadflows were required to get the final list of 18 critical Figures 7 and 8. As these figures show the results are very
contingencies which is 1.98 times the total number of close to the reference curves and there are no
contingencies (202) . misclassifications of critical contingencies.

I Iter. I % I No. of LF I No. of LF I No. of LF I 4.0 Computation Reauirements

The RSI method requires determining the nose point, which


is used for all contingencies, and one loadflow solution per
contingency. Therefore, the total computation requirement is
4 1 85.38 I 50 I 45 5 equal to (n+c) times the computation requirement of a single
5 I
86.5 1 45 27 18 I loadflow where n is the number of contingencies and c is the
Table 3: Iterative Filtering Results for the Other System number of loadflows to calculate the nose of the normal case.
For practical systems this ratio is very close to one. For
3.3 Test of Combined Screening and Ranking example, for the systems of B.C. Hydro and the unnamed
utility, this value was 1.01 and 1.02 respectively, Le. one
The iterative screening method described above was loadflow per contingency.
combined with the RSI method to screen and rank
contingencies. In each iteration of screening method only For Iterative Filtering method, the test results presented in
those contingencies which are solved are ranked based on Tables 2 and 3 indicate a total of 895 loadflows for the B.C.
RSI method by assessing the shift in reactive generation of Hydro system and 411 for the other system. These figures
the dynamic VAr devices. Also at the end of the process, the include 10 loadflows for the normal case nose point
remaining unsolved contingencies are ranked using RSI. calculation and intermediate stressed loadflow calculations.
These requirements translate into 1.33 load flows per
contingency for BC Hydro and 2.02 for the other system.

1.2 I 5.0 Conclusions

In this report two new techniques were derived for voltage


stability contingency screening and ranking; Reactive
Support Index (RSI) and Iterative Filtering method.

X-Stress I The RSI method uses the system normal nose point and ranks
0.2 -

0
I
I - - - - Iterative 1
I I I I I I II I I 1 I I I I 1 I I I 1 I+ I 1 I II I I1 I I I I I I 1 I I1 I I I I I I I1 j I I I I I I I1 I I t
contingencies based on the extra reactive injections required
at dynamic reactive devices to get a loadflow solution . The
'I/ ~ ~ ~

test results on the systems of B.C. Hydro and the unnamed


-b"W b " ? b c 3 o ) L o
w o ) c $ z b N a z o l o m ~ utility showed that this method performs a very good job of
N N c r ) m v b W " a
Contingencies screening and ranking contingencies with a computation
overhead of about one loadflow per contingency with a
Figure 7: BC Hydro Iterative Filtering capture ratio of about 70%.

To ensure that no critical contingencies are misclassified, the


T Iterative Filtering method was devised. In this method an
analytical algorithm is used to screen the coniingencies down
to the critical number required. The beauty of the Iterative
Filtering method is that it can be applied to any stress method
(Q, X or MVA) by applying proper stressing method to the
loadflows used within the algorithm. Theoretically, there
cannot be any misclassifications in the critical contingencies
obtained. This was confirmed by the tests conducted on the
systems of B.C. Hydro and another major utility with 100%
capture ratio with speed of between 1.3 to 2.02 loadflows per
contingency. Logically the Iterative Filtering method can be
combined with the RSI for screening and ranking.

The final recommendation of this paper is that RSI on its


Figure 8: Iterative Filtering Index for the Other System own or in combination with the Iterative Filtering method can
~

261
be used for CS&R dcpending on the acceptable level of 9. M. M. Begovic, Arun G. Phadke, “Control of Voltage
misranking. The RSI method is a very fast and powerful Stability Using Sensitivity Analysis”, Transactions on
Power Systems, Vol7. No.1, February 1992.
CS&R method and is suggested for systems where some
misclassification of local contingencies can be tolerated. On 10. J. Barquin, T. Gbmez, , F. Luis Pagola, “Estimating the
the other hand, for systems where an exact list (very high Loading Limit Margin Taking Into Account Voltage
capture ratio, e.g. 100%)of critical contingencies is intended, Collapse Areas”, IEEE Transactions on Power Systems,
the iterative Filtering method can be used for screening Vol.10. No. 4. November 1995.
complemented by another method like RSI for ranking. The 1 1 . S. Lefebvre, P. Jean Lagac6, R. Mailhot, “Voltage
latter technique is being integrated in the EPRID3.C. Hydro Stability as Part of On-Line Dynamic Security
On-line Voltage Stability tool. Assessment”, CEA Report. Project No. 347-T959.

Both the RSI method and the Iterative Filtering method 12. N. Fllatabo, O.B. Fosso, R. Ogendal, T. Carlsen, K.R.
Heggland, “A Method for Calculation of Margins to
presented here are hased on loadflow solutions. The Voltage Instability Applied on the Norwegian System for
efficiency of the algorithms can be further enhanced using Maintaining Required Security Level”, IEEE Trans. on
many advancements of static CS&R methods [ 14,151. Power Systems, Vol. 8, No.3, August 1993.
13. B. Gao, “Powertech Labs Internal Report; Interim Report
6.0 Acknowledgment for On-line VSA Project”, Sept 13, 1995.

The work reported in this project was partially supported by 14. B. Stott, 0. Alsac, A. J. Monticelli, “Security Analysis
the Electric Power Research Institute. The discussions and and Optimization,“ Proceedings of the IEEE , Vol. 75,
no. 12, December 1987.
comments of the project manager Mr. D. Maratukulam is
gratefully acknowledged. 15. V. Brandwajn, M.G. Lauby, “Complete Bounding
Method for AC Contingency Screening”, IEEE Trans.
on Power Systems, Vo1.4. No. 2, May 1989.
7.0 References
1. G.C. Ejebe, G.D Irisarri, S. Mokhtari, O.Obadina, 8.0 BiograDhies
P.Ristanovic, J.Tong, “Methods for Contingency
Screening and Ranking for Voltage Stability Analysis of Ebrahim Vaahedi is presently the Manager of the Control
Power Systems”, IBEE Trans. on Power Systems Vol. 11 Centre Technologies Department of B.C. Hydro and serves as
No.1. Feb 1996
an Adjunct Professor at the Elect. Eng. Dept. of the
2. G.B. Jasmon, L.H.C.C. Lee, “New contingency ranking University of British Columbia.
technique incorporating a voltage stability criterion”,
IEEE Proceedings-C, Vol. 140. No.2. March 1993. Yakout Mansour is presently the Manager of the Grid
Operation and Inter-Utility Affairs Division of B.C. Hydro.
3. V. Ajjarapu and C. Christy, “The Continuation Power
Flow: A Tool for Steady State Voltage Stability He is the chairman of the System Dynamic Performance
Analysis,” IEEE T-PS, Vol. 7 No.1, Feb.1992 Committee of IEEE.

4. H.D. Chiang, A.J. Flueck, K.S. Shah and N. Balu, Chris Fuchs Obtained his M.Eng. degree from the Simon
“CPFLOW: A Practical Tool for Tracing Power System Fraser University of British Columbia and is currently an
Steady-State Stationary Behavior Due to Load and
Generation Varial.ions,” IEEEPES 94 WM 244-4 Engineer in the Control Centre Technology Department of
PWRD, New York, NY, Jan 30 - Feb 4.1994. B.C. Hydro.

5. A. Yokoyama and Y. Sekine, “A Static Voltage Stability Wilsun Xu obtained his Ph.D. degree from the University of
Index Based on Multiple Load Flow Solutions,” Proc. of
the Bulk Power System Voltage Phenomena-Voltage British Columbia and presently is an Associate Professor at
Stability and Security, Potosi, Missouri, Jan. 1989. the Electrical Engineering Dept. of the University of Alberta.

6. H.D. Chiang and R.Jean-Jumeau, “Toward a Practical Hamid Hamadanizadeh obtained his M.Sc degree from the
Performance Index. for Predicting Voltage Collapse In University of Toronto and presently works at the Powertech
Electric Power Systems,” IEEEPES 93 SM 512-4 Labs. as a Senior Engineer.
PWRS Vancouver, BC July 18-22. 1993.

7. T. Van Cutsem, Y Jacquemart, J.-N. Marquet, P.Pruvot. Kip Morison obtained his M.Sc from the University of
“A comprehensive analysis of mid-term voltage Toronto and is presently the Director of the Power System
stability”, Paper No. 94 SM 511-6 PWRS, presented at Studies Group at Powertech Labs Inc.
the IEEE Summer Meeting, San Fransisco, July 1994.

8. T. Van Cutsem, R. Mailhot, “Validation of a fast voltage


stability analysis method on the Hydro-Qukbec system”,
Transactions on Power Systems, Vol 12. No.1, February
1997.
262

Discussion On the iterative filtering technique


This technique appears t o be similar t o the ((si-
T. Van Cutsem, C . Mokse, V. Sermanson (Uni- multaneous binary search” described in [A]. We be-
versity of LiBge, Belgium) : The authors are t o be lieve however that the latter might offer some compu-
commended for an interesting paper dealing with a tational advantages over the procedure used by the
very timely subject. We would like t o share some ex- authors.
perience on the contingency filtering problem, draw The binary search consists in building progres-
comparisons with the work reported in [A], and ob- sively an interval of stress [Sts,] such that St cor-
tain some clarifications from the authors. responds t o a stable scenario, S, t o an unstable one
and (S, - S t ) is smaller than a specified tolerance A.
On the use of load flow calculations The search starts with St = 0, corresponding t o the
Does B.C. Hydro plan t o have its Voltage Secu- base case, and S, = S ,,,, the maximum stress of
rity Assessment (VSA) based on load flow calcula- interest. At each step the interval is divided in two
tions only, or is the use of some dynamic simulation equal parts; if the mid-point is found stable (resp.
also envisaged ? Did the authors compare the two unstable) it is taken as the new lower (resp. upper)
approaches in terms of accuracy ? bound.
Load flow calculations may be appropriate for a When a secure operation limit has to be computed
first filtering, provided that they are not over-optimis- with respect to several contingencies, it is much more
tic with respect t o detailed time simulation (with the efficient to process the various contingencies “simul-
corresponding risk of discarding a dangerous contin- taneously” as outlined in Fig. A below. At a given
gency) nor too conservative (in which case, few con- stress level, the various contingencies retained from
tingencies will be filtered out). the previous step of the search are tested. If there
We have performed systematic comparisons of var- is at least one unstable contingency, the stable ones
ious security criteria and contingency ranking tech- (shown as parenthesized white dots in Fig. 1) can be
niques on the EDF (Electricit6 De France) and Hydro- discarded since the corresponding limits are higher
Qukbec [8] systems. Both utilities are using or im- than the current stress level. The unstable ones make
plementing the Quasi Steady-State (QSS) simulation up the reduced set of contingencies to be processed
approach described in [7,8,A] (incidentally, the test at the next step. The saving in computing time de-
function and QV technique quoted in the Introduc- pends obviously on how the individual limits are dis-
tion are not relevant t o [7,8]). tributed in the [0 S ,,,] interval. The stress level a t
Voltage stability in the EDF system is primar- which a contingency has been discarded provides a
ily influenced by load tap changers, overexcitation lower bound on the corresponding individual secu-
limiters and secondary voltage control. In this sys- rity limit.
tem, we found that load flow calculations were able t o
(pre-contingency)
provide fairly good estimates of secure operation lim-
its, except for some contingencies where the constant
power assumption of load flow calculations was inac-
?/ a
\
v w
f--)

i
-i
sam<\
-
system
stress
contin& # 1 .....6 ....... .......4 ..................................
curate due t o LTCs hitting their limits in dynamic I
canting, # 2 .......................*..................................
t
simulation. contin& # 3 ..........................................................
(0 >
The Hydro-Qukbec system is equipped with au- e a n s t a b l e case
(0)stable case ( c o n t i n g e n c y discarded)
tomatic devices (mainly EHV shunt reactor tripping
and synchronous condenser voltage boosting) which Figure A : Simultaneous binary search
play a major r61e in maintaining voltage stability. In
such a system there is little hope t o obtain reliable In the example of Table 2, the Iterative Filter-
ing (IF) is stopped when the 20 most limiting con-
security limits with a conventional load flow, which
cannot reproduce the time sequence of the above con- tingencies are found. For system operation it may
be of interest t o find the security margin relative to
trols.
the single most constraining contingency. How many
Time (and in particular QSS) simulation also of-
additional load flows would be needed t o achieve this
fers the possibility of embedding more elaborate cri-
objective ?
teria, such as declaring a post-contingency situation
acceptable if the amount of unrestored load power is On contingencv rankine:
below some threshold. As explained in [A], we consider that contingency
263
ranking is necessary t o cut down the number of con- search is needed towards devising a reliable, while
tingencies to be processed by the simultaneous bi- computationally efficient, method.
nary search. This is particularly true when time
On computational requirements
simulation is used for deciding about system stabil-
Do the authors consider that 895 load flows for a
ity. Although QSS simulation is much faster than
single margin computation are compatible with the
full time simulation it is computationally more de- computational efficiency of a real-time VSA ? We
manding than a single load flow calculation.
guess that many contingencies out of the 670 elim-
We have also tested indices of the type RCI and
inated at the first step of the method (see Table 2)
RSI described in the paper. As a general rule, we could have been discarded based on the knowledge of
have been disappointed by their poor ranking capa-
the system. Considering relevant contingencies only
bilities. With RSI-type indices, we found that some
would raise the number of load flows per contingency
contingencies may require the non-limited generators
to a more realistic value.
to produce large arnounts of reactive power, while
these contingencies itre not dangerous from the volt- On security margins
age stability viewpcint. With RCI-type indices we Does B.C. Hydro envisage to determine security
found that the (fictitious) synchronous condensers lo- limits, taking into account the actions taken by oper-
cated at load buses closest to the contingency tend ators or controllers while the system is stressed in its
to provide the most part of the reactive power sup- pre-contingency configuration ? As discussed in [A],
port, thereby creating a distorsion with respect to this departs from a widely used practice of stress-
the system instability mode. Also, we found some ing the post-contingency configuration. Depending
cases where the system was unstable with no genera- on the system, the limits computed by the two ap-
tor reaching its reactive limit (hence the limit open- proaches may differ significantly. The above actions
ing would not help). can easily be taken into account in the IF (or the
The paper gives a misclassification rate for the simultaneous binary search), since the system is first
20 most constraining contingencies. Figures 4 and 5 stressed in its pre-contingency configuration.
seem to indicate that, for the whole set of contingen- The authors’ comments on the above points will
cies, the number of misclassifications is large. Could be highly appreciated.
the authors give some indication on the accuracy of
their indices, for the whole set of contingencies, and [A] T. Van Cutsem, C. Moisse, R. Mailhot, “De-
in particular how errors share between false alarms termination of secure operating limits with respect to
and non-identifications ? voltage collapse”, paper SM 97-942 PWRS presented
In Section 3.3, RSI ranking is combined with IF. at the IEEE 97 Summer Power Meeting, Berlin, July
Could the authors clarify the following points : 1997
- at each step of the IF, RSI ranking is performed on Manuscript received April 6, 1998.
solved cases. What is the purpose of ranking those
cases, since they are known to be less dangerous than E. Vaahedi, C. Fuchs, W. Xu, Y. Mansour, H.
the unsolved ones and will be discarded at the next Hamadanizadeh, and G. K. Morison: The authors would like
step of the IF ? Ranking would be useful to find the to thank the discussers who are regarded as experts in this area
most dangerous among the unsolved cases, and skip for enriching the paper by their valuable comments and
the less dangerous ones; contributions.
- the RSI computation requires to run a new load flow
We would like to point out a typo correction which the
(with generator limits opened). Hence some comput- discussers brought to our attention. In the Introduction
ing time could be saved by running this load flow on Section, the test function and QV technique quoted should
unsolved cases only. refer to References [5, 61 instead of [7, 81.
In our opinion, JtCI or RSI-type indices can be
used to screen contingencies, i.e. filter out the less On the use of loadflow calculations
dangerous ones, but they do not provide a severity
B.C. Hydro’s on-line VSA program consists of four modules
index that reliably reflects the value of the margin, of a) contingency screening and ranking; b) Secure region
i.e. an index such that the smaller the margin the Calculation; c) Remedial Action and d) Quasi-Steady State
larger the index. Although we found good results Simulation. The philosophy is that pre-specified number of
with the severity index described in [A], more re- (e.g. 20) most critical contingencies are short-listed from a
264

large list of the credible contingencies. The Secure Region based on the philosophy of our on-line VSA that we would
Calculation module then calculates the voltage stability have to identify preventive or corrective actions for the short-
margin for all the short-listed contingencies. This module is listed critical contingencies. If we did not have to find
based on load-flow calculations. The remedial action module corrective actions, then we would only need to find the most
then calculates the preventive action needed in the base case limiting single contingency [A]. Using Iterative Filtering the
or the corrective actions following each contingency. Finally total number of iterations to go to any pre-specified number is
the Quasi-Steady State Simulation module is used for almost the same (853) whether it is 20, 10 or one because it
verification of the results. uses the same projection equation [Equation 31. The tolerance
however, need to be reduced and that will tax the method a
Our experience with using the load flow based techniques on few more load flow solutions (5 to 10).
many large scale utility systems indicates that for the vast
majority of cases powerflow techniques give accurate results, On the Contingency Ranking
providing the modeling is done properly (i.e. generator
capability curve, load modeling, ULTC action, etc.). The Contingency ranking and Contingency screening terminology
approach to dynamic security assessment we have adopted is seems to create some confusion. The philosophy of our on-line
to provide a full range of capabilities to deal with a variety of VSA is to screen the contingencies to a short-listed number of
systems. These methods include full-time domain (in our critical contingencies which are then ranked for on-line
transient stability assessment or TSA package), quasi-dynamic processing. Therefore it is essential that screening is
or fast time domain simulation, and powerflow. While we conducted with zero misclassification and hence the
currently use powerflow methods as the primary VSA solution development of Iterative Filtering. Once the screening is
because of the high volume of cases which must be examined conducted the ranking can be done using RSI type method. If
on-line, if a system is found to require a more rigorous exact ranking among the short-listed critical contingencies is
method, they are available and would be used. For example, if also needed, the Iterative Filtering can be continued after the
voltage instability is a result of induction motor stalling, or the first short-list is obtained until smaller blocks of contingencies
action of fast HVDC or FACTS controls, neither powerflow are also found; the method can find exact ranking by
nor quasi-dynamic approaches may be sufficient and full time- individual contingencies if that is needed at some extra
domain may be required. However, a significant performance computing time. We suggest that the discussers use Iterative
compromise is required and so these methods need to be Filtering instead of RSI and RCI if accurate screening or
carefully justified. ranking is required. Another point is warranted here.

On tlze Iterative Filtering technique The paper as the discussers point out gives a misclassification
rate for the 20 most critical contingencies. Figure 4 and 5
There are two assertions by discussers which are responded to show the actual ranking of the contingencies based on their
separately. Firstly, the discussers have compared the true margin versus the RSI method. Misclassification is
calculation solution using a binary search [A] to a Newton meaningless for all the contingencies because if a contingency
Correction method used in Iterative Filtering. Binary search is really mild (e.g. normalized margin=0.93) then whether RSI
uses the mid-point between a stable case and an unstable case indicates its value as 0.99 or 0.85 is irrelevant. Please note that
and continues the process until the tolerance is achieved. The the base case with no contingency has a normalized margin of
Newton Correction technique, however, uses the slope 1.00 and from these figures good 80% to 90% of the
between the last two solutions to project the next solution. The contingencies have normalized margin of above 80%. So
Newton Correction technique is more efficient than the Binary misclassification really becomes meaningful only at the end of
search because it uses the slope information whereas the the screening process since it is important to know how many
Binary search blindly selects the mid-point. The second of the most critical contingencies have been missed.
assertion is the efficiency resulted as a result of
simultaneously considering contingencies [A]. In the In section 3.3, RSI is combined with Iterative Filtering at each
technique given in [A] simulation are conducted for different step, as the discussers pointed out correctly, to find out how
contingencies at the same stress level and if they found to be closely this combination performs against the real margin. As
stable they are abandoned. This resembles to Iterative Filtering the discussers pointed out, in the actual implementation, the
Method's of solving contingencies at the same stress level and ranking only needs to be conducted on the critical short list In
carrying forward the unsolved ones to the next level of stress. the short list of critical contingencies, there are contingencies
However, we also see significant differences between the two that do not solve with zero stress (normal) base case. In our
methods in the approach and application. While IF is a very search for a screening method we first discovered RSI and
fast and accurate screening technique designed to move to any RCI type of methods which try to approximately calculate the
pre-defined number of most critical contingencies, the method distance to nose and moved eventually to IF because of the
given in [A] is an evaluation technique designed to find the restriction on zero misclassification of critical contingencies
single most critical contingency out of a short list of that RSI and RCI could not achieve. In our experience we
contingencies. found the methods that base their index based on the
unstressed base case condition such as [A] are less accurate
In the example of Table 2, the Iterative Filtering is stopped than those which attempt to identify the condition at the nose
when the 20 most limiting contingencies are found. This is simply because they do not incorporate the stress pattern and
265

they base their indices on the steady state condition of the conditions and series of events which can lead to instability.
unstressed (normal) base case before and after the Therefore, "experience" based screening, which relies on an
contingency. While better methods for ranking are always understanding of system performance from past incidents or
welcome, there seems to be no incentive for us to further studies, could be applied under normal system conditions, but
improve RSI since the misranking within the short-list is not under alert or emergency conditions, (which are often the
as critical and all the short-listed contingencies will be studied critical times for security assessment), more rigorous
individually in detail. assessment is required. With the use of distributed computing,
a rigorously screening of a large number of contingencies is
On Computation Requirements possible, and probably prudent. Ultimately, we envisage the
use of an event database (populated by measurements and
The functional specification for B.C. Hydro's on-line VSA studies) connected to automatic learning systems to provide
stated a requirement of handling 1000 contingencies. To be guidance on pre-screening.
able to handle this number of contingencies in real-time, it
also is recognized that rule based methods would be an asset On Security Margins
in performing the first level of filtering. However, we believe
that great care must be taken in relying too heavily on In our on-line VSA, the operator's pre-contingency actions are
"experience" to pre-screen contingencies. Contingencies properly simulated while the system is stressed to avoid
which are innocuous under general conditions may become inaccuracies resulting from control actions. As the discussers
critical under abnormal conditions. Events such as the July 2, pointed out, this control actions can be easily simulated by IF
1996 WSCC disturbance, clearly illustrated the complex and the method given in [A].

You might also like