Adjusted Solutions in Fast Decoupled Load Flow

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

726 IEEE Transactions on Power Systems, Vol. 3, No.

2, May 1988

AWUSTBD 90LUTIONS IN FAST D S C O Q P m IOAD Mw

Show-Kang Chang Vladimir Brandwajn


Member, IEEE Member, IEEE
Systems Control, Inc.
1801 Page Mill Road
Palo Alto, CA USA

ARSmAtx Various methods had been proposed for handling certain


adjustments in the FDLF [71-[91. Computer storage
The Fast Decoupled Load Flow method is commonly requirements, efficiency, as well as practicality for
accepted by the power industry, due to its simplicity large systems have been major problems. This paper
and computational efficiency. The inclusion of adjust- first explores the nature of the common adjustments
ments in the method, however, has not been given proper with the objective of overcoming the interaction prob-
attention even though it can have a major effect on the lem between the different adjustments. Empirical cri-
computational efficiency, if not implemented properly. teria as to when various adjustments are started and
This paper presents generalized criteria for including their relation to the load flow solution are
the common adjustments in the Fast Decoupled Load Flow described. Experience with many practical systems has
and the methods by which the different adjustments can indicated that these criteria can have a major effect
be implemented without significantly affecting the on the overall computational efficiency.
solution efficiency. Test results on several large
systems are also presented to demonstrate the effec- Relevant guidelines and methods for implementing the
tiveness of the methods. adjustments are also described. The methods are tied
to the use of the highly efficient sparsity techniques
-ION [lo], [ll] which substantially speed up the adjusted
solution. Test results on several large systems are
The Fast Decoupled Load Flow (FDLF) method [l] has presented to demonstrate the effectiveness of the
gained wide acceptance in the power industry, due to methods.
its simplicity and computational efficiency. The
inclusion of adjustments in the method to simulate the
control strategies of system operation has not been
given proper attention even though it can degrade the The Fast Decoupled Load Flow method involves the itera-
solution efficiency, e.g., the number of iterations for tive solution of the following systems of equations:
the adjusted solution can increase significantly, the
solution may become oscillatory or even diverge. AP/V = [B'l Ae (1)

Common approaches to incorporate adjustments into the AQ/V = [B"] AV (2)


load flow solution process are [21:
where
1. error-feedback adjustment;
2. automatic adjustment. AP, AQ : vectors of real and reactive power
mismatches
The error-feedback adjustment involves modifications of
a control variable to maintain another functionally 89. AV : vectors of corrections to voltage
dependent variable at a specified value, in a closed- angles and magnitudes.
loop feedback mechanism. The automatic adjustment, on
the other hand, involves modifications of the Jacobian B'ij = -l/xij and B g i i= z 1 / x i j
matrix to include the adjustment equality constraints
for direct solution of the control variables [31-[51.
Some success had been achieved in the application of
the automatic adjustment approach to the Newton's
method [6]. The practical implementation is, however,
complicated and the handling of the control variable xij, bij : reactance and susceptance of branch i-j.
limits is inefficient. In the FDLF, the error-feedback
approach is advantageous for most of the adjustments, SHi : total shunt admittance at bus i.
except for the handling of W-bus Q-limiting for which
the conventional bus-type switching approach is more The matrices [B]' and [B"] are both sparse and sym-
suitable. metrical. They are factorized at the start of the
solution so that (1) and (2) can be solved repeatedly
by forward and backward substitutions.

The matrices [B'] and [B"] are different dimensionally


since the reference bus is excluded from [B'] and
W-buses are excluded from [B"]. It may, however, be
disadvantageous to make such distinctions when con-
sidering adjustments. Both matrices can be tri-
angularized using the same ordering regardless of the
bus types. Whenever a row/column is to be excluded, a
large diagonal ( 0 if the inverse diagonal factor is
stored) is used to mask out the effects of the lower-
diagonal terms. This technique is especially useful in
This paper was sponsored by the IEEE Power Engineering Society the treatment of W-buses in the matrix [B"] [lll.
for presentation at the IEEE Power Industry Computer Applica-
tion Conference,Montreal, Canada, May 18-21, 1987. Manuscript
was published in the 1987 PICA Conference Record.

0885-8950/88/0500-0726$01 .WO 1988 IEEE


AwusrBD SOLWTI~S the bus mismatches provide an accuracy indicator by
which the activation of the different adjustments can
Common adjustments in the power flow solution include be determined. As the solution progresses, different
the control of phase shifter flows, area interchanges, adjustment types are thus switched on or off, depending
transformer taps, capacitor/ reactor switching, gen- on the most recent maximum bus mismatch.
erator remote voltage control, and the enforcement of
Q-limits at W-buses. To be consistent with the The adjustments are made either before or after the
decoupling principle of the FDLF, the adjustments are main solution of ( 1 ) or ( 2 ) , depending on their
categorized into real and reactive ones. The real nature. The computational efficielicy varies widely as
adjustments (phase shifter6 and area interchanges) are a function of the relative sequence of different
made in the =-cycle and the reactive ones (transformer adjustments. Normally, it is best to perform the
taps, capacitor/reactor switching, etc.) in the injection-based adjustments before the solution, and
Qv-cycle. the sensitivity-based adjustments after the solution.

As pointed out in [ l ] ,
conventional adjustment schemes All adjustments introduce perturbations into the load
such as error-feedback and bus-type switching are most flow solution. In general, large perturbations arise
suited to the FDLF. Although there are no definitive during the early stages of the sensitivity-based
guidelines for implementing these schemes, it is essen- adjustments and Q-limit enforcement. They invariably
tial that the effects of the adjustments on the solu- slow down the convergence, or more seriously, cause the
tion speed be minimized. The following two factors, solution to oscillate or even diverge. The
often overlooked in theoretical discussions, are oscillation/divergence problems can be effectively
crucial to the computational efficiency: alleviated by limiting the size of the correction in an
adjustment cycle. The slow convergence problem is
1. Starting criterion for an adjustment; mostly caused by excessive incremental power mismatches
due to the corrections. An efficient remedy to this
2. Relative sequence of the adjustments with respect problem is to perform an "auxiliary" solution after the
to each other and the solution of ( 1 ) or (2). adjustments. This, essentially, minimizes the incre-
mental power mismatches and, consequently, reduces the
fie starting criterion defines when the adjustment perturbations on the subsequent main solution cycle.
process should commence during the solution. In the
Newton's method [61, the iteration number is probably The auxiliary solution is always fast in comparison to
the most commonly used criterion, e.g., all adjustments the main solution. It is performed using the highly
are made at the end of the second iteration and there- efficient sparse-vector methods [ l o ] , and is par-
after. This criterion is heuristic and it does not ticularly well suited to the sensitivity-based adjust-
differentiate between the nature of the different ments. Details of its implementation and the tech-
adjustments. The weakness of this criterion lies in niques to handle bus type switching are described in
the fact that some adjustments are made at an inappro- the following sections.
priate moment or in an incorrect sequence and, con-
sequently, cause an unnecessary increase in the solu-
tion time.
Area Interchange Control
A simple but effective alternative is the use of bus
mismatches. They provide a measure of the solution The conventional area interchange control method, which
accuracy And allow for the coordinated control of the works reliably in the PDLF, is the adjustment of the
adjustments. To apply this criterion, it is helpful to scheduled real power generation of one or w r e area
characterize the adjustments as either sensitivity- "slack" generators between the power flow iterations.
based or injection-based. For the sensitivity-based The interchange error is shared among the area "slack"
adjustments, a set of computed or empirical sensitivi- generators according to prespecified proportions. The
ties ("feedback gains" in [ 2 ] ) is raquired to estimate interchange control is activated only when the solution
the control parameter variations. The control of phase is moderately converged because it directly affects the
shifter flow, transformer taps, and generator remote bus injections. It requires modifying the mismatch
voltage control belong to this group. The injection- vector AP immediately before the solution of (1). A
based adjustments involve the modification of bus limit checking process may also be included in the
injections to meet the controlled objective (area adjustment for detecting the possible "power imbalance'
interchange, voltage range for a switchable capacitor/ problem.
reactor). It should be pointed out that both types of
adjustments involve modification of the bus injec- An identical approach Can be used to perform the slack
tions. The principal difference between the two types distribution task.
lies in the way the modifications are calculated:
indirectly for the sensitivity-based adjustments and Phase Shifter Flow Control
directly for the injection-based ones.
The real power flow through a controllable phase
The enforcement of Q-limits at the W-buses is a shifter can be regulated by adjusting its shift
special case because it does not clearly belong to any angle. The effects of the angle 'shift are modelled by
of the two adjustment types. It involves the change of equivalent bus injections. In the FDLP method, this
injections as well as bus type switching. Experience adjustment is best handled via the "error-feedback"
has shown that this adjustment behaves similarly to the scheme in which the 'feedback gain" is approximated by
sensitivity-based adjustments. the sensitivity between the shift angle and the flow.

The possible "hunting" between different adjustments The sensitivity ia computed using the linearized net-
can be circumvented by activating them in a selective work model of (1 ). The change in the real povar flow
manner, depending on the solution accuracy and type of APij through the phase shifter branch i-j is first
adjustment. In general, the sensitivity-based adjust- expressed by the following relation:
ment could be started in the early stages of the solu-
tion and the injection-based adjustments only when the APij = (Aei - Aej + A+)/xij (3)
solution is moderately converged. At each iteration,
728

where AI) is the incremental shift angle, and xij the


phase shifter reactance. CaWnte
Next the effect of AI) on A0 and AO. is obtained from AP
(1) by the equivalent injection techAque:

APij = (-MIij tB'1-l Nij + 1) AI)/xij = S AI)/xij (4)

where Mij is a sparse vector with +1 and -1 in the ith


and jth positions and N.. is a sparse vector with
+l/xij and -l/xij in the i'th and jth positions. Control and Update
AP
The sparse-vector methods [ l o ] allow an efficient com-
putation of the sensitivity S via fast-forward/fast-
backward substitutions. The sensitivity is computed
1
only once during the solution process, due to the
invariant nature of the matrix [B' 1. A negative value
of S is used when Pji is controlled instead of Pij.

The shift limits are enforced by modifying the computed


increments, appropriately. At the same time, the pos-
sible limit back-off is also tested for. The correc-
tion of the phase shifter angle introduces incremental
real power mismatches at its terminal buses. The mis-
matches propagate to the next solution cycle and thus Phase Shifter
slow down the convergence. This problem can be &LSP
resolved by performing an auxiliary solution before Update Angle Shift
advancing to the QV-cycle. The auxiliary solution of
(1) is used to compute approximate angle changes caused
by the adjustment. Since the mismatch vector AP is
sparse (two nonzeros for each shift angle change),
sparse-vector methods are applicable for the solu-
tion. Experience has shown that the computational
overhead associated with the auxiliary solution is
usually very small and the benefits significantly out-
weigh the extra effort.
I Update 8
Auxy~-ion I
Inclusion of Real Adjustments in the =-Cycle

Figure 1 shows how the real adjustment described above To ( t V Cycle


are included in the PO-cycle of the FDLF method. Sa
and Sp represent the starting criteria for the area
interchange control and phase shifter flow control, Figure 1. P-8 Solution Cycle
respectively. The activation of the adjustments is
governed by the maximum bus mismatch which is identical problem of finding the binding constraints in the
to the normal convergence criterion (omitted from the optimal load flow 1121. An auxiliary solution which
figure for simplicity). Each iteration may, therefore, follows the main solution of (2) , remedies the problem
incur different adjustments, depending on the nmst by approximately resolving the inequalities before
recent maximum mismatch. advancing the solution.

The objective of the auxiliary solution is to identify


new Q-limit violations and/or limit back-offs, and to
W-Bus Q-Limit~ correct bus voltages should any new adjustments
occur. After the main solution of ( 2 ) and voltage
When the load flow solution is mderately converged, update, the reactive generation change at a W-bus i
any Q-limit violations at the W-buses should be cor- can be computed from the following formula:
rected. Any violated W-bus is converted to the PQ-
type with the reactive generation set at the limiting (5)
value. The voltage of the converted bus is sub-
sequently compared to the scheduled value and the bus The .linear approximation of ( 5 ) is generally ade-
is reconverted to the W-type, if any of the back-off quate. Each newly violated W-bus is converted to the
conditions are satisfied. FQ-type with its equation inserted into the matrix
[e"]. The incremental reactive power mismatch induced
The switching of bus types involves the insertion/ by this adjustment is (Qlh -
Q . ) , where Qlim is the
deletion of equalities into/from ( 2 ) and, hence, violated MVAR limit and Qgi isqthe updated reactive
requires. the retriangularization of the matrix [B"]. generation.
Two efficient methods are available for this task:
partial refactorization and factor updating [ill. Both Similarly, if a generator can back-off the limit, the
methods were found to be equally efficient. associated bus is converted to the W-type and the
corresponding equation deleted from the matrix (B-1.
The initial enforcement of the Q-limits normally Each back-off a150 causes a bus voltage correction:
perturbs the solution and can cause new Q-limit viola-
tions as well as limit back-offs. If not properly
resolved, it will slow down the convergence or even
AVi = (Vsi - Vi)

cause divergence. This problem resembles the classical where Vsi is the scheduled voltage.
729

The vpltage c o r r e c t i o n AV. introduces an incremental where Mk is a sparse vector with + 1 i n t h e kth p o s i t i o n
r e a c t i v e power mismatch (-b1 V . AV a t each connected and N a sparse vector with -bij/t and b . ./t i n t h e i t h
ij I i and j t h positions.
11
bp
After t h e i d e n t i f i c a t i o n of t h e new adjustments, t h e Each t a p increment c r e a t e s incremental r e a c t i v e mis-
bus voltages a r e corrected by solving ( 2 ) using t h e matches A t / t a t t h e transformer terminal buses,
,,/
/ incremental mismatches. Since t h e Q - l i m i t s are which shou?tih resolved by an a u x i l i a r y s o l u t i o n . For
enforced only when t h e solution is s u f f i c i e n t l y computational e f f i c i e n c y it i s b e s t t o combine t h i s
accurate, t h e number of back-offs i s normally small. s o l u t i o n with t h a t f o r t h e generator remote voltage
The incremental mismatch vector i s sparse and, there- control.
f o r e , t h e sparse-vector methods a r e again c a l l e d f o r .
If a bus voltage i s simultaneously c o n t r o l l e d by
The inclusion of t h e a u x i l i a r y s o l u t i o n s t e p has been s e v e r a l transformers, d i r e c t a p p l i c a t i o n of ( 8 ) w i l l
found to s i g n i f i c a n t l y increase t h e e f f e c t i v e n e s s of usually cause over-capensated adjustments which may
t h e Q - l i m i t adjustment. Independent of t h e number of r e s u l t i n an o s c i l l a t o r y solution. This adverse e f f e c t
W-buses, t y p i c a l l y , no more than two a d d i t i o n a l i t e r a - can be a l l e v i a t e d e i t h e r by l i m i t i n g t h e s i z e of AV t o
t i o n s a r e required f o r t h e adjustment. This i s espe- be corrected, o r , by coordinating t h e transformer
c i a l l y important t o a p p l i c a t i o n s involving a l a r g e adjustments. one p o s s i b l e coordination i s t o make t h e
number of load flow s o l u t i o n s , such a s contingency adjustments proportional to t h e i r s e n s i t i v i t i e s . The
analysis. s e n s i t i v i t i e s ccolputed from ( 9 ) a r e more appropriate
f o r t h i s purpose.
Capacitor/Reactor Switchinp
Generator R e m o t e Voltage Control
A bus voltage can be maintained within a s p e c i f i e d
range by switching a c o n t r o l l a b l e capacitor o r r e a c t o r A generator a t a PV-bus can c o n t r o l , within i t s MVAR
a t t h e bus. Although t h e c a p a c i t o r s / r e a c t o r s a r e dis- limits, t h e voltage of a remote bus. This type of
c r e t e i n nature, t h e common approach i s to ignore t h e c o n t r o l d i f f e r s from t h e l o c a l PV-bus r e g u l a t i o n i n
d i s c r e t e n e s s during t h e adjustment and to convert t h e t h a t the W-bus voltage i s t o be adjusted even though
appropriate parameter to i t s nearest d i s c r e t e s e t t i n g i t s Q - l i m i t s a r e not violated.
only a f t e r t h e i n i t i a l convergence. The load flow i s
then reconverged with t h e f i x e d s e t t i n g . I n i t i a l l y , t h e voltages a t t h e W-buses a r e used a s
c o n t r o l parameters. Whenever a W-bus i s converted t o
The error-feedback scheme i s used f o r t h i s adjust- t h e PQ-type due to Q-limiting, i t s c o n t r o l parameter i s
ment. In t h i s case, t h e feedback-gain r e p r e s e n t s t h e swktched t o t h e r e a c t i v e generation with t h e most
s e n s i t i v i t y between t h e voltage and shunt admittance of r e c e n t bus voltage serving as t h e scheduled voltage f o r
a bus. It i s found by computing t h e diagonal term of subsequent back-off t e s t i n g . The r e a c t i v e generation
t h e inverse of [B"]. For bus i, t h e s e n s i t i v i t y may change due t o t h e influence of o t h e r adjustments,
r e l a t i o n is thus expressed by: although t h i s r a r e l y occurs i n p r a c t i c a l applica-
tions. For s i m p l i c i t y , only t h e voltage control para-
tB"1-'Mi) Abi = S Abi (7) meter i s discussed i n t h i s paper.

where Mi i s a sparse vector with + 1 i n t h e i t h posi- The error-feedback scheme i s used, where the formula
t i o n , and Abi t h e shunt acbaittance change. f o r a W-bus i c o n t r o l l i n g t h e voltage of bus j
become$ :
The s e n s i t i v i t y S i s constant and it has t o be recal-
culated only following bus type switching. AV. = (Vjsp
3
- V 3. ) = S AVi (10)

The shunt admittance change Ab. causes an incremental An estimate of t h e s e n s i t i v i t y S i s c a l c u l a t e d using


mismatch Ab V which i s d i r a c t l y included i n t h e t h e l i n e a r i z e d model of ( 2):
mismatch ve%tkr A Q of ( 2 ) p r i o r to t h e main solution.
Experience has shown t h a t it i s not necessary t o modify S = -M ' [B"]-'N (11)
[B"] f o r the shunt acbaittance change, provided t h a t t h e j
mismatch c a l c u l a t i o n s c o r r e c t l y account f o r t h e change. where M . i s a sparse vector with + 1 i n t h e j t h p o s i t i o n
3
and N a sparse vector with +bik i n t h e kth p o s i t i o n f o r
Transformer Tap Voltage Control each bus k connected to i.

An automatic on-load tap-changing transformer can keep The incremental r e a c t i v e mismatch induced by t h e W-bus
t h e voltage of a l o c a l or remote bus Within a s p e c i f i e d voltage adjustment i s s i m i l a r t o t h a t induced by a
range, by . a d j u s t i n g the in-phase tap. As i n t h e PQ-generator limit back-off. For computational
capacitor/reactor switching, t h i s adjustment i s per- e f f i c i e n c y it is combined with t h e incremental mismatch
formed using t h e continuous approach. For a trans- from t h e transformer adjustment and resolved i n a
former i - j c o n t r o l l i n g t h e voltage of bus k, t h e e r r o r - s i n g l e a u x i l i a r y solution.
feedback formula is:
The problem of voltage c o n t r o l by multiple W-buses can
AVk = S A t (8) be resolved s i m i l a r l y t o that of t h e multiple trans-
former control. I f a bus voltage is simultaneously
where AVk i s t h e c o n t r o l error, S the sensitivity, controlled by a transformer and a generator, t h e pre-
and A t t h e t a p increment. cedence is given to t h e generator u n t i l its c o n t r o l
c a p a b i l i t y i s exhausted, i.e., t h e r e a c t i v e generation
Normally, S = f l gives s a t i s f a c t o r y r e s u l t s . If hits a l i m i t .
desired, however, S can be e f f i c i e n t l y computed from:
Inclusion of Reactive Adjustments i n t h e Q-V Cycle
S = Mk' [ B"1-l N (9)
Figure 2 shows how t h e r e a c t i v e adjustments described
above a r e included i n t h e Q-V s o l u t i o n cycle of t h e
-..

730

e f f e c t i v e n e s s of t h e methods. The characteristics of


t h e s e systems a r e given i n Table I. For each system,
CalCJlte t h e number of buses, branches, W-buses. buses with
capacitor c o n t r o l , and LTCs i s l i s t e d . Also, shown iq
AQ t h e number of i t e r a t i o n s f o r t h e unadjusted s o l u t i o n .

TABLE I
C"SRIST1CS OP TSST SYSTR4S

Yes CapacitorlReactor ITERATION


-cNnga
VpdateAQ
-.
I 157 1 62 4

ITERATION

6
7
6
MAX
--
1240.98
107.84
-30.36
MW
MISMATCH MISMATCH

-3894.53

8.88
3.07
1.33

0.25
I
I

I
MAX MVAR

21907.52
-458.56
150.28
-36.73
10.75
-2.81

0.77
YQL N J p a
0
73
3
1
0
0

For the sake of c l a r i t y and b r e v i t y , only e f f e c t s of


t h e r e a c t i v e adjustments a r e described here. The real
adjustments behave s i m i l a r l y t o t h e r e a c t i v e ones. A l l
t e s t r e s u l t s w e r e obtained using t h e f l a t s t a r t condi-
10
0
0
0

* Auxiliary Solution
t i o n s and 1 MW/MVAR maximum bus mismatch tolerances.
S t a r t i n g c r i t e r i a f o r t h e sensitivity-based adjustments
and injection-based adjustments were 500 MW/MVAR and 5
A
MW/MVAR, respectively.

Effect of W-Bus Q-Limit Adjustment

The e f f e c t of t h e PV-bus Q - l i m i t adjustment on t h e


convergence of t h e test system B i s shown i n
Table 11. For each i t e r a t i o n , t h e maximum bus mismatch
p r i o r to t h e s o l u t i o n of ( I ) o r ( 2 ) i s shown, along
with t h e number of Q - l i m i t enforcements ( N Q l ) before
t h e s o l u t i o n of ( 2 ) , t h e number of new v i o l a t i o n s
(NQ21, and t h e back-offs (NQ3) a f t e r t h e s o l u t i o n of
(2). Most of t h e adjustments were performed a t t h e
t
To P-0 Cycle
U initial activation. Afterwards, only two more
a c t i v a t i o n s were needed t o i d e n t i f y t h e f i n a l 78
binding VAR l i m i t s . For t h e purpose of comparison, t h e
adjustment e f f e c t on t h e t e s t system C i s shown i n
Table 111.
Figure 2. Q-V Solution Cycle

FDLF. Sc, Sq, S t , Sr a r e t h e s t a r t i n g c r i t e r i a f o r


capacitor/reactor switching, Q-limit enforcement,
transformer t a p c o n t r o l , and generator remote c o n t r o l ,
respectively. Normally, Sq=St=Sr. The W-bus Q - l i m i t
adjustment is performed i n two steps. The a u x i l i a r y
s o l u t i o n s f o r the transformer t a p and generator remote
c o n t r o l a r e combined i n t o a s i n g l e s o l u t i o n a s
described above.
3
4 4 51.75 30.68 6 0 1
5 5 -8.43 -2.29 0 0 0
The adjustment methods described above have been t e s t e d
on systems of s e v e r a l u t i l i t i e s .
three systems a r e presented
Typical r e s u l t s on
t o demonstrate t h e 1 5 5 0.47 I -0.25
731

ITERATION W MW MAX W A R
Q MISMATCH YISMATCH MISMATCH

1 1 -2574.49 530 13.92 0 5841.97 14150.46 0 0 0 0


2 2 -344.85 -250.07 68 6437.75 -2737.09 0 0 0 0
3 3 41.97 16.27 10 198.48 262.27 69 11 10 7
4 4 -6.0 1 -103.20 56 51.75 30.70 6 0 1 3
5 5 22.80 9.61 0 -8.44 -10.99 0 0 0 0
6 6 -2.97 -0.83 ' 0 2.10 -1.60 0 0 0 0
7 7 8.95 3.13 4.05 25.94 0 0 0 0
1 0
I I I I -4.08 0 0 0 0
7 7 -0.84 -0.25
8 7 -0.08 0.74
Effect of Capacitor Switching and Transformer Tag
Control Effects of the Starting Criteria

Table IV shows the effect of capacitor switching and The effects of the starting criteria for the Q-limit
transformer tap control on the convergence of the test enforcement and tap adjustments in systems B and C are
system A. The number of capacitor adjustments (NCP) shown in Tables VI1 and VIII, respectively.
and transformer adjustments (NTT) is shown for each
iteration. In the initial test (Table VII) the starting criterion,
for the Q-limit enforcement, was varied while the taps
The idtial convergence without accounting for the were kept fixed. The starting criterion for the
discrete nature of capacitor/transformer tap adjust- Q-limit adjustment was then fixed at 500 W A R and the
ments was achieved in 6 iterations. An additional criterion for tap adjustments was varied. These
iteration was required to reconverge the case following results are shown in Table VIII.
the discretization of capacitors and transformer
taps. It is noteworthy that the auxiliary solution TABLE V I 1
prevented large perturbations even in the presence of KwBlcT CW Q-LmIT Sl'ARTIUG CRITElUoR
extensive adjustments.
STARTING SYSTEM B SYSTEM C
Combined Effect of Reactive Adjustments CRITERION

The combined effect, of different reactive adjustments


on the copvergence of test systems B and C, is shown in
Tables V and VI, respectively. In this test, the com-
bined effects of Q-limiting and tap control were
checked. The results indicate that the convergence is 10
largely dictated by the Q-limit adjustment. The inter-
action between the real and reactive adjustments is
insignificant, due to the use of the auxiliary
solution. An additional iteration was required in both TABLE VI11
cases to reconverge the solutions following tap dis- m?FPcT op TAP S
-!I CRITRRIW
cretization.
STARTING SYSTEM B SYSTEM C
CRITERION ITERATION ITERATION

1
ITERATION
-
P 8 U
! %z 15 16

1 I -3894.53 21907.52 0
2 2 1240.98 -458.56 73
3 3 105.53 156.42 4
4 4 -30.45 -24.70 1 Effect of Reactive Adjustments Without Auxiliary
-51.75
Solutions
5 5 1.82 0
6 6 3.08 -8.16 0 0 0 Most of the existing load flow programs perform the
7 7 1.32 13.13 0 0 0 adjustments without the use of auxilliary solutions.
Tables IX and X show the test results obtained on
systems B and C without the benefits of auxiliary
7 7 solutions. The Q-limit adjustment (limit
enforcement/back-off) was pe forned in a sinxlf step
prior to the solution of 7 2 ) and the tra S Orme
adjustment updated the taps without Computing the
resulting voltage increments. Note that in some cases
the solution failed to converge as indicated by I t ' .
732

B. Stott, "Review of Load-Flow Calculh


Methods", IEEE Proceedings, vol. 62, pp. 916-9a
July 1974.
I STARTING
CRITERION
I SYSTEM B
ITERATION
I SYSTEM C 1 J.P. Britton, "Improved Area Interchange Control
ITERATION for Newton's Method Load Flows", IEEE Trans. on
MVAR P B P B Power App. and Syst., vol. PAS-88, pp. 1577-1581,
October 1969.
500 12 11
5 13 12 11 11 N.M. Peterson and W.S. Meyer, "Automatic Adjust-
ment of Transformer and Phase-Shifter Taps in the
Newton Power Flow', IEEE Trans. on Power App. and
Syst., vol. PAS-90, pp. 103-108, Jan/Feb 1971.

J.P. Britton, "Improved Load Flow Performance


Through a More General Equation Form", IEEE Trans.
~~
on Power App. and Syst., vol. PAS-90, pp. 109-116,
STARTING Jan/Feb 197 1.
C RITEFSON SYSTEM B SYSTEM C
MVAR ITERATION ITERATION W.F. Tinney and C.E. Hart, "Power Flow Solution by
( Q-LIMIT ) P B P 8 Newton's Method", IEEE Trans. on Power App. and
Syst., vol. PAS-86, pp. 1449-1456, NOV. 1967.
500 14 12
P.S. Nagendra Rao, K . S . Prakasa Rao, and J. Nanda,
5 16 16 16 15 "An Exact Fast Load Flow Method Including Second
Order Terms in Rectangular Coordinates", IEEE
Trans. on Power App. and Syst., vol. PAS-101,
pp. 3261-3268, Sap. 1982.
It can ben se& fron Tables IX and X that the number of
iterations for both P and Q increases quite R.N. Allan and C. Arruda, "LTC Transformers and
significantly without the benefits of auxiliary MVAR Violations in the Fast Decoupled Load Flow",
solutions. Also, the adjustment may fail to correctly IEEE Trans. on Power App. and Syst., vol. PAS-101,
identify the binding set of Q-limits when a high pp. 3328-3332, Sep. 1982.
starting criterion is used.
Z.X. Han, "Phase Shifter and Power Flow Control",
IEEE Trans. on Power App. and Syst., vol. PAS-101,
coIIcLuG1ons pp. 3790-3795, Oct. 1982.
%npirical criteria and efficient methods for the imple- 1101 W.F. Tinney, V. Brandwajn, and S.M. Chan, "Sparse
mentation of adjustments in the Fast Decoupled Load Vector Methods", IEEE Trans. on Power App. and
Flov have been presented. By the use of auxiliary Syst., vol. PAS-104, pp. 295-301, Feb. 1985.
solutions and sparsity techniques, the adjusted solu-
tions can be obtained without substantial increase in [ll] S.M. Chan and V. Brandwajn, "Partial Matrix Refac-
the number of iterations and computational time. Since torization", Proc. of the Fourteenth PICA
only conventional schemes, i.e., error-feedback and Conference, pp. 9-15, May, 1985.
bus-type switching, are w e d , the basic E'DLF formula-
tion remains unchanged and the adjustments are easy to [121 H.W. Dommel and W.T. Tinney, "Optimal Power Flow
include in the program. The methods are also applic- Solutions", IEBE Trans. on Power App. and Syst.,
able to other adjustments such as transformer tap flow vol. PAS-87, pp. 1866-1876, Oct. 1968.
control, distributed system slack, etc. Test results
on several realistic systems indicate that the number
of extra iterations required for W-bus Q-limiting is
virtually independent of the size of the problems. The
Discussion
perturbations caused by the adjustments are minimized
via the auxiliary solution. The described techniques Fernando L. Alvprndo (The University of Wisconsin-Madison, WI):
can be implemented efficiently in other load flow This is a usetid and practical paper for all those interested in the fast
methods. The paper's major contributions consist of: decoupled method. This discussion concerns only the fourth of the stated
paper contributions. Sparse vector methods are indeed one of the most
a) analysis of the nature of different adjustments; important new ideas in sparse matrix technology. With all due credit to
Tinnex, Brandwajn, and Chan for reference [lo], sparse vector methods
b) development of methodology for efficient were also described in full detail in [All, which was presented concurrently
implementation of different adjustments; with [lo], and was published three months after [lo] only due to the quirks
of the publishing process. An unfortunate circumstance of [A101 was that
c) providing practical guidelines for the selection of sparse vector methods were not mentioned explicitly in the paper title, as
starting criteria; they were considered by the authors (in retrospect perhaps mistakenly) the
lesser of the three stated contributions of this paper.
d) pointing out the applications of sparse vector The main reason for bringing this issue up at this time is not so much to
methods in the implementation of different criticize the authors of this paper in any way, but to point out that in the
adjustments. experience of this discusser over the last two years, sparse vector methods
are indeed best explained using reference [lo]. However, many of those to
whom this discusser has exposed sparse vector methods have found the
implementation of the method to be most straightforward from the
description in reference [All. Thus, it is this discusser's opinion that much
[l] B. Stott and 0. Alsac, "Fast Decoupled Load Flow", could be gained by all in the future by insuring reference to [All alongside
IEEE Trans. on Power App. and Syst., vol. PAS-93, [lo] whenever implementation issues regarding sparse vector methods are
pp. 859-869, May/June 1974. to be addressed.
,/
733

References S. K. Chang and V. Brandwajn: We wish to thank the discussers for their
interest in our paper and for their comments.
/L.. Alvarado, M. K. Enns, and S . K. Mong, “A Fault Program Regarding the comments of Prof. Alvarado, reference [lo] was listed in
, with Macros, Monitors and Direct Compensation of Mutual the bibliography primarily because it provides fundamental insight into the

-/ Groups,” ZEEE Trans. Power App. Syst., May 1985, pp. 1109-
1120.
sparse vector methods. The general concept has been described in a number
of earlier papers, although not as extensively (see especially [A]). The cited
paper co-authored by the discusser addresses the algorithmic aspects of fast
forwardlfast backward substitution in considerabledetail and may serve as a
convenient reference document for those who need guidance in implement-
B. Stott, 0. Alsac, and M. Prais (Power Computer Applications, Mesa, ing these techniques.
AZ): For many years, control adjustments have been the main algorithmic The authors are grateful to Messrs. Stott, Alsac, and Prais for their
challenge in conventional power flow, a challenge that, judging by the valuable comments on the problem areas associated with the power flow
literature, has partially gone ignored. This paper is therefore particularly adjustments. The choice between precomputing alI local sensitivities ushg
welcome, and is a valuable partner to the original fast decoupled reference. the sparse inverse algorithm and computing only the needed ones during the
We are in agreement with the general discussion of adjustment principles solution with the sparse vector methods is problem dependent. With good
and with the specific techniques described, having evolved our fast implementation, however, both approaches should cause only a small
decoupled versions in much the same ways. V. Brandwajn is acknowledged overhead on the overall solution time.
for suggesting the Q-limit scheme some years ago. The approach of using the self-adapive sensitivities to determine the
In order to realize overall convergence speed-up, the kinds of refinement control parameter variations is commonly employed in the load flow
described in the paper must be applied rigorously to all types of control. calculations in which the sensitivities are costly to compute. It works well
Any one slowly converging device will largely nullify all of the others. for the tap/phase shifter flow control. Tuning is g e n e d y required for other
We can confirm that the use of a control’s local sensitivity, calculated by types of adjustments. Our experience of applying it to solve the external
sparse vector methods, generally works well, except in the case of generator model in the state estimator has indicated that the approach can be quite
var limits where bus-type switching is needed. In a power system with large efficient even though we have not implemented it in our fast decoupld load
numbers of voltage controls, particularly LTC transformers, the computa- flow program.
tion of the sparse inverse matrix might be a more economical way of There is no unique solution for control variables that regulate the same
obtaining most of the required local sensitivities. quantity without specified var-sharing between them. If the var-sharhg is
The main remaining problems are where controls have high interactions, defined, the control adjustments can be made proportional to their
and where the sensitivities change during the solution process, due to composite sensitivities which are functions of var-sharing factors and the
limiting or backoff of neighboring controls. We note that the authors have relevant sensitivities. It is found beneficial to limit the size of the control
not included self-adaptive sensitivity updating (at each iteration as a correction at each step to avoid the possible solution oscillation due to
function of the changes during the last iteration). Do they have any control interactions.
experience with this? Should it be used in conjunction with the precomputed We concur with the comment on the starting criterion. The criterion is
local sensitivities? effective if it can affect the activation of different adjustments in a
The paper is nonspecific where control of a quantity is shared between determined manner. Our experience has shown that the use of bus
multiple devices. The most common example is a PQ bus voltage controlled mismatches indeed fulfills this requirement. The starting criteria are system
from a number of remote PV buses, with a defined var allocation between dependent and can be developed based upon the magnitudes of bus net
them. Another case is where several transformers control the same bus injections.
voltage, with specified var-sharing or tap gauging, etc. Are they catered for The simulation of control precedences, based on speeds of response, in
by the techniques in the paper? the adjustment processes deserves careful consideration. The effects on
The paper brings out the importance of the starting criterion for a control alleviating the multiple solution problem as well as the impact on the
such as var limiting and transformer tapping. This appears to be a system- solution speed are not known to us, although the ways to incorporate them
dependent aspect of tuning that even the best algorithms cannot overcome. in the power flow can be envisioned.
One of the greatest unsolved problems strongly associated with reactive
controls is the existence in some power systems of multiple solutions. These References
can be attributed to the different ways in which a system state is reached.
The power flow solution makes no attempt to simulate the different [AI K. W. Edwin and G. G. Wagner, “On-Line Security Constrained
precedences, based on speeds of response, of the various controlled Economic Dispatch Using Sensitivities and Convex Linear Program-
devices. How critical are such factors, and would it make sense to mine.” oresented at WAC Symposium on Computer Applications in
incorporate them in some approximate ways into the adjustment processes? Large &e Power Systems,-New Delhi, August, 1979:

You might also like