Beruflich Dokumente
Kultur Dokumente
4, APRIL 1970
TIE LINES
INTRODUCTION CONTROL ANA K
AN electric energy system must be maintained at a desired
operating level characterized by nominal frequency,
voltage profile, and load flow configuration. It is kept in this
nominal state by close control of the real and reactive powers
generated in the controllable sources of the system. The total real Fig. 1. Interconnection of individual control areas.
and reactive power demands PD and QD will slowly change
throughout the 24 hours of the day, but during time periods
measured in seconds or minutes, they can be considered as detected, since it is the most sensitive indicator that real-
essentially constants with superimposed first-order pertur- power balance is not maintained. If an error exists, "lower" or
bations APD and AQD, respectively. (Throughout the paper, A "raise" commands are sent to the prime mover, resulting in
symbolizes a perturbation of first-order magnitude. The super- decreased or increased real generator power.
script * indicates "nominal" value.) 2) The megavar-voltage control problem, or "Q-V control" for
The perturbations in generated power APG and AQG must short. By means of voltage sensors, bus voltage magnitude
match the load perturbations if exact nominal state is to be deviations are detected; these are the most sensitive indicator -of
maintained. However, due to the statistical nature of the load the reactive power imbalance. If an error exists, "increase" or
fluctuations, this can never be achieved-the best we can hope "decrease" commands are sent to the Q sources.
for is to keep the system within sufficiently small tolerance During major fault situations, the deviations in frequency and
levels. voltage are no longer small, and the noninteraction feature be-
By means of sensitivity analysis [1] the following very im- tween P and Q control channels no longer applies.
portant characteristics of a power system can be proved.
1) A mismatch in the real power balance affects primarily the MEGAWATT-FREQUENCY CONTROL PROBLEM
system frequency f, but leaves the bus voltage magnitudes In this paper the P-f control problem has been selected for
essentially unaffected. critical examination. The problem, of course, is as old as the
2) A mismatch in the reactive power balance affects only the power systems technology itself. Interest in overall system
bus voltage magnitudes, but leaves the system frequency es- stability resulting from some spectacular recent system break-
sentially unaffected. downs has focused renewed interest on this problem area. This
It is reemphasized that these rules apply only when the particular study was actually initiated in an effort to better
changes involve only small percentage, i.e., are first order in a learn the reasons for the relatively unsuccessful attempt to
mathematical sense. In view of the above facts, the nominal synchronize the western United States power-block with the
steady-state control problem can be divided into two independ- eastern block via midwestern tie lines.
ent or "decoupled" control problems.
1) The megawatt-frequency control problem, or "P-f control"
for short. By means of a frequency sensor, the frequency error is Dynamic System Model
The first order of business is development of a system model.
Paper 69 TP 658-PWR, recommended and approved by the Power The following analysis is based on the assumption that the elec-
7Q+flf Engineering Committee of the IEEE Power Group for
at the IEEE Summer Power Meeting, Dallas, Tex.,
trical interconnections within each individual control area (Fig.
<;entation 1) are so strong, at least in relation to the ties with the neigh-
June 22-27, 1969. Manuscript submitted March 28, 1969; made boring areas, that the whole area can be characterized by a
available for printing April 30, 1969. The research reported was
supported in part by a grant from the investor-owned utilities in the single frequency only. In the strictest sense, when the system
State of Florida. deviates from the constant nominal frequency, every bus voltage
0. I. Elgerd is with the Department of Electrical Engineering,
University of Florida, Gainesville, Fla. 32601. will experience its own angular velocity or frequency. The
C. E. Fosha, Jr., was with the Department of Electrical Engi- assumption made means that all generators belonging to an
neering, University of Florida, Gainesville, Fla.. He is now with the area will swing in unison. Should this assumption not be
Department of Mathematics, U. S. Air Force Academy, Colorado
Springs, Colo. 80840. permissible (and one must be careful in the case of an individual
ELGERD AND FOSHA: OPTIMUM MEGAWATT-FREQUENCY CONTROL 557
area covering a wide territory), then one must subdivide the ApDi (s)
area into subareas. The theory that follows will still apply.
The net power surplus in the area following a disturbance
APD equals APG - APD MW, and this power will be absorbed aFi (s)
by the system in three ways:
1) by increasing the area kinetic energy Wki. at the rate
dl d (fY1
dt Wkin = dt [Wkinir* (f1 2 IPtie
~tie i (s)
i
GOTi (s)
AP (,s) + AP
i(s)
LFu(s) AFv(s)
Fig. 3. Incremental tie-line power out of area i. Fig. 5. Block diagram of turbine generator.
t A~~~~~~~
PDI S
STEAM G t( ) LPGI(S) Gpi(Js)
FLOW APci (5F
iI+tlsf51I)(I4ST5i ) \
I < I+sTpi
(s)
'DECREASE TURBINE ktie i(S)
APC; -LZ SPEED CHANGER I APGI
4INCREASE
,CLOSE GENERATOR A ~~~I!;
I S QOPEN
SPEED GOVERNO
HIGHH,,
PRESSURE)
OIL AFj(s) AF vC)
Fig. 4. Typical turbine control arrangement. Fig. 6. Block diagram of single-area perturbation model.
The total increment in exported power from area i (symbolized gard for the time being the dotted portions of the figure.) Before
in block diagram form in Fig. 3) is finally obtained from (9) we proceed, we must explain the presence of the block containing
the transfer function al2. This transfer function equals the
APtiei(s) = - > T AFi (s) - AF,(s)]. (17) following:
z i
V+ AV1 VA AV2
Fig. 8. Mechanical analog of two-area interconnected
power system.
Af (f)
_ sec
<\Af2(t]~~~~~~~~~fsa
Fig. 7. Block diagram of two-area perturbation model. e
i ~ ~ ~ f
at the nominal constant speed v*, and a load change is suddenly sec
applied in one of the assemblies, the speeds of the two assemblies AP4ie stat
will change with the amount Av1 and Av2, and the spring force
(or power) with the amount AFtie. Fig. 9. Uncontrolled response of two-area system to step-load
One can show [1] that the differential equations describing increase in area 1.
an n-assembly train are identical (except for scale factors)
with those we have derived for the electric system in this paper; We introduce here the area frequency response characteristic
i.e., the mechanical system is an analog of the electrical (or vice (AFRC) for each area:
versa). In the two-area case the following variables are inter-
related: - Di + 1/Ri, pu MW/Hz
(21)
Afi ± AV, 02A D2 + 1/R2, pu MW/Hz
Af2 =/\A)2 and can then write (20)
APD2 - al2APD1
APtie ± AFtie. Afstat = - 13-
02- al2ol
(22)
Response of Uncontrolled System: Before we enter into a dis- _1 APD2 -
32APD1
cussion of how to control the two-area system, it is useful to APtie 1 stat
082- a12#1
take a look at the response of the uncontrolled system. For this
purpose we consider the system in Fig. 7 with no command The physical significance of the AFRC can be stated as
inputs fed to the speed changers; i.e., we set AP,1 = APC2 = 0. follows: An area operated alone will, if uncontrolled and subject
A step load is applied to area 1 and the subsequent variations to a step-load change, experience a static frequency drop in-
of Af1, Af2, and APtie are studied. Fig. 9 shows the results as versely proportional to its AFRC. This follows directly from
recorded on an analog computer. We note the following important Fig. 6, which renders (by setting s = 0 and APtie i = 0)
features: P
Afstat = (23)
1) all three variables have nonzero static errors;
2) the two frequency errors will, of course, be equal after Equation (22) becomes particularly simple should we assume
steady state is reached; identical area parameters, i.e.,
3) the system is oscillatory but stable.
D1 = D2= D
All these results could have been predicted by considering the
mechanical analog. Ri = R2 = R
Area Frequency Response Characteristic Concept: Let us compute 31 = 12 = 1
the static frequency error Afl stat = Af2 stat and the static tie-
line power error APtie stat identified in Fig. 9. We can either use a12 = -1.
(3) and set d/dt = 0, or use the block diagram in Fig. 7 and set We then get
s = 0. The analysis gives the result
APD2 213
+ APD1
al2APD1
APD2 -
Afstat = - X Hz
Afstat = -
D2 + 1/R2- a12(D1 + 1/R2) APtie 1 stat = - APtie 2 stat (24)
+ 1/R)
APtD2(Dl 1/20-a APD(D2 + 1/R2) (20) APD2 -
APD1
AP tie 1 stat =-
2 y puMW.
560 IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, APRIL 1970
Afstat = - ,PD2
2$
Hz (25) LAB
AfO(t) 1 01 /
T T /D \
2 /-' sec
251
Z
* 20-
LU 1.
z
0 IC'
z IJt
U
B OOA______ -
0.5 1.0 1.5 2.0
B INTEGRAL GAIN K1
Fig. 11. "Cost surface" showing minimum value of C at Ki opt Fig. 13. Cost function C, penalizing both APti.l (t) and Af1(t).
and Bopt.
20
45070~125 100%25%750s
15
H li
z
0
0 1
I0 I
Optimum Gain Settings: When confronted with a response The "cost-function" C can be plotted versus K, and B in a
of the type shownao in Fig. 10 one naturally asks, what
picture 0~~~~~~~~ three-dimensional coordinate system as shown in Fig. 11. One
combination of the four (two, in the case of identical areas) gain obtains a bowl-shaped "cost surface," the lowest point of which
parameters produces the best response? corresponds to the optimum gain settings, KI opt and Bopt.
Having posed this question one must next clarify one's defini-
tion of "'best." Often a so-called scalar integral error criterion COMPUTER RESULTS
provides the most meaningful measure for the quality or good-
ness of transient response. The integral of squared error (ISE) As we have indicated earlier, analysis of the transient response,
criterion is chosen in this case. It is defined by although straightforward, is very impractical. We have instead
simulated the system on the analog computer and also let the
C-' (Ate + aAlfi2) dt. (30) analog provide the cost integral (30). Two multipliers and one
summer-integrator are required to perform the squaring, sum-
ming, and integrating operations called for.
(Since ASf1 and Af62 have very close resemblances, we have It would, of course, be very impractical to plot on a two-
included only one frequency error Api.) dimensional paper the three-dimensional cost surface. We
The merits of this criterion are described elsewhere [4]. The plot instead the "equi-B" cost contours, two of which are shown
parameter Gin (30) is a weighting factor that determines the dotted in Fig. 11. They are obtained simply by fixing the B
relative penalty attached to the tieline power error and fre- parameter and then computing the C integral for various K,
quency error, respectively. For example, if a is chosen equal to settings. Figs. 12-14 show the results of such a simulation study.
zero, all penalty is attached to the tie-line power swings. The particular system parameters used in the numerical study
In making use of a criterion as in (30) one would proceed as are given in Fig. 12. In each figure is also indicated the particular
follows: The responses, APtie and Af', corresponding to a par- cost integral used. The cost contours in Fig. 12 correspond to a
ticular parameter pair KI and B (we consider identical areas) cost integral where 100 percent of the penalty has been attached
are found. Upon substitution into (30), the integral is evaluated. to APtie. In Fig. 14 all the penalty is attached to the frequency
Clearly, the value thus obtained is a function of KI and B, i.e., error. In Fig. 13 a 50-50 weighting ratio has been used. This
CC=AC(K,, B). means that the two integral portions of (30) are of about equal
(31) size at the optimum point; to find the particular a value that will
That particular value pair of KI and B that renders C a provide this penalty division requires some trial-and-error
minimum represents the optimum parameter settings. investigation.
562 IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, APRIL 1970
For this study the following system data were used: ing upon the KI value, we may be located anywhere on the
cost surface along this curve, even on the vertical portion.
Pri = P,2 = 2000 MW
7) In attempting to synchronize the western United States
H1 = H2 = 5 seconds power block with the eastern block via midwestern tie lines, on
DT = D2 = 8.33 X 10-s pu MW/Hz occasion the low-frequency (about 4-5 cycles/min) power
TTI = TT2 = 0.3 second swings on those lines have shown an unstable tendency of the type
TG2= 0.08 second demonstrated in Fig. 10, and it has been necessary to sever the
R1 = R2 = 2.4 Hz/pu MW connection. It is entirely conceivable that the reason for in-
Ptie max = 200 MW stability is the one given in 6).
51 *-,62*-= 30 degrees 8) The utility industry is presently readjusting its bias setting
l 12* 0.545 pu MW/Hz every year, based upon updated measurements of the AFRC.
APdl = 0.01 pu MW.
If the industry were to adopt a parameter setting policy out-
lined here, no such readjustments would be really needed, be-
SUMMARY, COMMENTS, AND RECOMMENDATIONS cause the change in cost function C as a function of the gain
variables around the optimum settings is very small. (This is the
1) Nowhere in our extensive computer analysis has it been benefit derived from wider stability margins.)
confirmed that the bias setting B = /, which presently is used by 9) This analysis confirms what many years of operational
the American utilities, offers any advantages of a particular experience already has told us; the control strategy defined by
nature. (27) is a workable one. Like many other workable control
2) If the tie-line swings are taken as sole indicator of the quality strategies it was born out of intuitive thinking.
of control (see Fig. 12) then the results would indicate that B i. However, the strategy is not the "best" one. Modern control
0.53 is the best bias setting. This is half the value recommended theory tells us that there exists one, and only one, strategy that
by NAPSIC! is "better" than all others, including the one presently in use. If
3) In all cases studied so far, the optimum bias setting is "better" means as measured in the ISE sense, this unique
considerably below that used by the industry. optimum strategy is a linear one also. The authors are presently
4) The study reveals that specification of the gain parameter investigating the nature of this "supreme" strategy. Whether its
K1 is equally important as specification of B. advantages warrant its adoption has yet to be judged. Future
5) On what basis then does NAPSIC base its present policy papers will report the status of this research.
of recommending to industry to set B = ,B but saying nothing- 10) It is worthwhile to remember that the cost curves given
about KI? The reasons seem to be found in vious works in are valid strictly only for the particular system parameters that
this area [2], [3], both of which have ha rerable impact were given in Fig. 12. Furthermore, they are valid only for a
on present utility practice. A careful study of these references two-area system.
confirms that both base their conclusions on results obtained by Further work is carried on to increase our knowledge in the
static analysis. Their reasoning goes something like this. following respects:
Each area if operated alone and subject to a step-load in- 1) How sensitive is the cost function to system parameter
crease APD will experience, in accordance with (23), a static changes?
frequency drop of Af = APD/I3 Hz before the controller goes 2) How should one best approach the optimization problem
into operation. Their reasoning then continues: By commanding for an n-area system?
the area generators to increase their output by the exact amount 3) Is it possible (as is intuitively felt) when optimizing the
/31 f , the original steady state will be restored. They then parameters of an individual area to treat the rest of the system
conclude that, by setting the frequency bias B = /, "minimum as having infinite inertia? This then would mean that the in-
tie-line oscillations will ensue." This reasoning has a consider- dividual pool members could optimize their parameters based on
able amount of intuitive appeal, but must at closer scrutiny an "individual" C criterion that would be based upon the in-
be judged wrong. We have, in fact, no right to judge dynamic dividual errors only. If this assumption were to prove true then
performance on the basis of static characteristics. We proved the optimization procedure would be simple indeed.
in connection with (28) that in fact any value for B will in the
end result in zero static tie-line power deviation. NOMENCLATURE
6) A parameter optimization study as presented here has its
greatest value as a means of optimizing the system stability f* nominal system frequency
margins. As indicated Fig. 11,
in the cost function C may assume Wkin kinetic energy of area
infinite values for certain finite KI and B values; i.e., the "cost Pr rated area power
bowl" will have vertical walls in certain places. An unbounded H inertia constant
cost integral indicates an unstable system, and by keeping Ptie tie-line power
away from those vertical portions of the cost surface, we thus APtie incremental change in tie-line power
have a guarantee for stability. By specifying a centain value for f incremental frequency deviation
B but disregarding the other equally important variable KI, D load frequency constant
we have really no guarantee of our stability margins. In fact APd incremental load demand change
our margin may be zero. Ape incremental generation change
This point is demonstrated both in Fig. 11 and the graphs in APe incremental change in speed changer position
Fig. 10. The dotted curve labeled B = 100 percent represents area power angle
the B recommendations presently used in the industry. Depend- Ti* tie-line spring constant
IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, VOL. PAS-89, NO. 4, APRIL 1970 563
10
SFr
FItST PAGE
The Megawatt-Frequency Control Problem:
A New Approach Via Optimal Control Theory
CHARLES E. FOSHA, JR., MEMBER, IEEE, AND OLLE I. ELGERD, SENIOR, MEMBER, IEEE