Megawatt-Frequency: Problem

IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, VOL. PAS-89, NO.
4, APRIL 1970 563
Tv turbine time constant REFERENCES

TG speed governor time constant [1] 0. I. Elgerd, Electric Energy Systems Engineering. New York:
R self-regulation of generator McGraw-Hill (to be published).
KI ACE integrator gain [2] L. K. Kirchmayer, Economic Control of Interconnected Systems.
B frequency bias constant. New York: Wiley, 1959.
[3] N. Cohn, "Some aspects of tie-line bias control on intercon-
nected power systems," AIEE Trans. (Power Apparatus and
ACKNOWLEDGMENT Systems), vol. 75, pp. 1415-1436, 1956 (February 1957 sec.).
[4] 0. I. Elgerd, Control Systems Theory. New York: McGraw-
Several figures and graphs are borrowed from [1]. Use was Hill, 1967.
made of the analog computer facilities in the Department of
Electrical Engineering, University of Florida. For Combined Discussion see pp.571-577.
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
Abstract-This paper records the development of a state variable INTRODUCTION

model of the megawatt-frequency control problem of multiarea
electric energy systems. The model is in a mathematical form IN a previous paper [1] the authors developed a dynamic
necessary for application of theorems of modem optimal control system model of the multiarea electric energy system suitable
theory. for study of the megawatt-frequency control problem. Classical
An optimal feedback controller whose structure is radically different optimization theory was used to find the "best" value of pa-
from that considered before is developed. The results of this study rameters KI, gain of the area control error (ACE) integrator,
allow the authors to suggest feasible ways of greatly improving and B, frequency bias, in the sense of minimizing the integral
dynamic response and stability margins of the megawatt-frequency square error (ISE) criterion. The error was defined as the tie-
control system. line power deviation from scheduled value and the frequency
deviation.
Paper 69 C 2-PWR, recommended and approved by the Power In this paper the authors again study the megawatt-frequency
System Engineering Committee of the IEEE Power Group for pre- control problem but from a different point of view. In [1] the
sentation at the IEEE PICA Conference, Denver, Colo., May 18-21,
1969. Manuscript submitted January 31, 1969; made available for control strategy was fixed, being that employed by industry
printing October 8, 1969. This research was supported in part by today [2]. This area controller operates in response to the
a grant from the investor-owned utilities in the State of Florida. integral of the ACE for that area. By dropping the restriction
C. E. Fosha, Jr., was with the Department of Electrical Engi-
neering, University of Florida, Gainesville, Fla.. He is now with the of a fixed control structure, techniques of optimal control theory
Department of Mathematics, U. S. Air Force Academy, Colorado can be applied.
Springs, Colo. 80840. The authors derive a model of the multiarea electric energy
0. I. Elgerd is with the Department of Electrical Engineering,
University of Florida, Gainesville, Fla. 32601. system in state space form necessary for application of theorems
564 IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, APRIL 1970
of modern optimal control theory. The specifications that the System Cost C
system must satisfy are defined mathematically in terms of an The performance of the system is specified in terms of a cost
integral cost that is to be minimized. Minimization of this cost that is to be minimized by the optimal controller. The com-
yields an optimal controller that is a linear combination of the ponents of Q and R are ours to choose in mathematically specify-
states of the system. Study of the optimal controller structure ing the way we wish the system to perform. For example, if we
allows one to suggest new feasible control methods. These let R = 0 but require Q to be nonzero, we are saying in effect
methods could give improved dynamic response and wider that there is no charge for the control effort used; but we penalize
stability margins to the megawatt-frequency control systems in the state for being nonzero. Hence the best control strategy
operation today. would be in the form of infinite impulses. This control would
MODERN OPTIMAL CONTROL THEORY drive the state to zero in the shortest possible time with the
greatest effort.
Probably the most important contribution modern optimal If we let Q = 0 for nonzero R, then we penalize for control
control theory has made to the control engineer is the ability to effort but do not charge for the trajectory the state x follows.
handle a large multivariate control problem with ease. The In this case the best control to use is u = 0; i.e., not to provide
engineer has only to represent the control system in state any control effort at all.
variable form and specify the desired performance mathe- These are two extreme cases, but they emphasize the impor-
matically in terms of a cost to be minimized. Well-proven theories tance in choosing the components of Q and R. The mathematics
a-d techniques are available which generate a unique or "best" bears out the intuitive reasoning presented in these examples.
controller in the sense of minimizing the cost C [3].
That approach is followed in this paper. That is, we represent Optimal Controller u
the multivariate megawatt-frequency control problem in state The optimal controller that minimizes the cost C of the system
variable form, define a system cost to be minimized, and let in state variable form is a function of the present states of the
techniques of optimal control theory develop our controller. system weighted by the components of a constant gain matrix
The reader is referred to the Appendix for a detailed account of K of dimension m X n:
the optimal control theory used. What follows is a general
statement of the problem and the results. u = -Kx. (3)
Problem Statement This optimal gain matrix K is determined by solving a matrix
differential equation, the matrix Riccati equation. For the
The statement of the problem in optimal control theory form infinite time problem, the Riccati equation has a steady-state
is as follows: Given the linear time invariant system represented solution. Since the gain matrix is a constant, the optimally
by the state variable differential equation controlled system can be expressed in the closed-loop form
x = Ax + Bu (1) x = ACx
where
x n X 1 state vector ACA -BK (4)
u m X 1 control vector where ineans defined as.
A n X n state distribution matrix
B n X m control distribution matrix System Conditions-Controllability and Observability
and boldface type indicates the time derivative d/dt, find the What conditions must the system satisfy in order for an opti-
control u, which minimizes the cost mal controller in the feedback form of (3) to exist? To answer
that question we must consider the concepts of controllability
1 f0 and observability. These concepts are defined mathematically
c = (x'Qx + u'Ru) dt (2) in the Appendix. We present here a general discussion to help
2 J
in a first understanding.
where When a system in the state variable form given by (1) is
Q n X n positive semidefinite symmetric state cost weighting said to be controllable, it means that by suitable choice of
matrix control strategy in the form represented by (3), the closed-loop
R m X m positive definite symmetric control cost weighting poles of the system may be specified. Given a set of performance
matrix criteria, i.e., rise time, percent overshoot, etc., a matrix K can be
transpose. found that will allow the closed-loop system to perform in the
desired manner. If the system is controllable, it is said that the
System State x pair (A,B) is controllable.
If the system is linear and time invariant, we are guaranteed Before presenting the concept of observability, we must
that it may be represented in the form of (1). The state of the introduce the matrix D of dimension p X n where
system, represented by the state vector x, is a very important Q A D'D. (5)
concept. The state variables xi, x2, * *, xn are the components
of the state vector x. These state variables are the minimum We define a new variable y, a p X 1 vector p < n, by
number of variables containing sufficient information about the
past history of the system to allow one to compute the future
y = Dx. (6)
of the system, assuming the control inputs are known. The The components of y are a linear combination of the state
state variables are not purely mathematical but have true variables of the system. The system is said to be observable
physical meaning, as will be shown later. from y if, given the information contained in y, we can re-
FOSHA AND ELGERD: MEGAWATT-FREQUENCY CONTROL PROBLEM 565
construct complete information about the state x. This condition

is often stated as the pair (A,D) is observable, or identically, the _/_
\
pair (A,Q) is observable.
With this understanding of the concepts of controllability
and observability, we state the conditions the system must
satisfy for an optimal controller in the form of (3) to exist /
[4]. /.1
If the system described in the form of (1) with cost given by
(2) is unstable (some eigenvalues of A are in the right-hand
complex plane), and the unstable states are observable from the
cost, then it is necessary that those unstable states be control-
lable for an optimal controller in the form of (3) to exist.
A heuristic proof of the necessary condition follows. Assume
that the open-loop system is unstable and that the unstable
states are observable from the cost function. Assume also that
the unstable states are uncontrollable. Hence those unstable
states will always be unstable, independent of the control chosen.
Since the unstable, uncontrollable states are reflected in the Fig. 1. Coherent areas.
cost the cost will be infinite no matter what control we use.
Hence every control will give infinite cost and there is no unique
optimal control. This concludes the proof.
SYSTEM DYNAMIC MODEL

The system model that will be used was derived in [1] as
was mentioned earlier. Briefly we will develop the model again
for completeness. Two assumptions are made. The first is / /
that for incremental changes in demand power, the two problems,
control of real power and frequency, and control of reactive
power and voltage, are decoupled and can be considered sep-
arately. The megawatt-frequency control problem is the first
of the two problems. By maintaining control over the real power,
the frequency deviation is kept within prescribed limits. The
second assumption is that the individual electrical connections
within an area are so strong at least in comparison to the ties
between adjoining areas, that each area may be represented Fig. 2. Incremental tie-line power from area i.
by a single frequency. This characteristic of an area is called
coherency. In other words, all generators in a single area swing
in unison during changes in area load (Fig. 1). Tie-Line Power
The total real power exported from area i equals the sum of
Power Equilibrium Equation all outflowing line powers Ptie i, to adjoining areas v, i.e.,
The net surplus power in an area, represented by the dif- Ptiei = E Ptiei, (8)
ference in increased generation AP, and increased demand
APd, is absorbed by the system in three ways: The real power in per unit transmitted across a lossless line
1) increased kinetic energy represented by (2Wkin/f*) (d/dtAf) of reactance Xi, is
where Wki. is the kinetic energy of the system and f* is the
nominal frequency; Ptiei = vlv sin (6 - a) (9)
2) increased load consumption represented by DAf, where the XipPri
constant D, MW/Hz, is the rate at which system load changes where
with frequency evaluated at nominal frequency f*;
3) increased export of power over tie lines represented by
APtie.
V = IVileJ6i
Expressing what has been said in mathematical form gives the -
v,,-=|V,eiV,e5.
power equilibrium equation in per unit for area i: Assuming small deviations in phase angles 6i = bi* + A6i
2H
d and realizing that
Afi + Di4fi + APtiei = APgi APai
- (7)
A = 2llfAfi dt (10)
where the expression (Fig. 2) for incremental changes in tie-line power
Wkin in area i is
H A inertia constant, seconds
Pri APtiei = E Ti,*(fAfi dt - fAf, dt) (11)
Pri rated power of area i, MW.
DYNAMIC SYSTEM IN STATE VARIABLE FORM

t The three equations (14)-(16) that describe a single area of
DIEESE the multiarea system must be rewritten in terms of the state and
control variables before we go further. We define the state and
I1IAE
control variables in the following way:
AXgv
Xl A fAPtie 1 dt
X2 A Af1 dt
X3sAf,
A
(17)
X4 A AP g0
A
X5 AXg
U1 A AP,.
Fig. 3. Typical turbine control arrangement.
In each area the control input is the generator power. By ma-
nipulating the output of the generator we can maintain the
where frequency deviation within limits following load changes. To vary
the generator power, we must change the speed-changer position.
Hence we define the control variable as the speed-changer
*T- 211 I, cos (3i* - 8*)
position.
The variables that change as a result of surplus power in an
Incremental Generated Power area are the governor valve position, the generator output, the
The real power generated by a synchronous machine is con- frequency deviation, and the tie-line power, which is a function
trolled by means of the prime mover torque. A study of the of the integral of frequency deviation. We define these variables
system for small changes around nominal settings reveals that as states. The integral of tie-line deviation must also be defined
the generator-turbine-governor system may be represented by as a state. One of the specifications we place on the system is that
two time constants, Tt of the turbine and T0, of the governor. the tie-line deviation be zero following a step-load disturbance.
The generator response is considered to be instantaneous in To guarantee this we must have knowledge of the integral of
comparison with the time constants of the turbine and governor. the tie-line deviation.
We can write For each area there is a set of five state variables and one
control input. However in the two-area problem that we will
consider, the tie-line deviation in the first area is proportional to
dt Tt Tt g(12) the tie-line deviation in the second by a constant:
d =
-AX0,,=
1
AX0,,-
1
~Af +
1
Ape (13)
APtie2 = aiNAPtie 1
dt T0v,TTauT where
where APO is the incremental change in generation in pu MW, A
a12 -= PlP2 (18)
AX0,, is the incremental change in the governor valve position
in pu MW, R is the self-regulation of the generator in Hz/pu In this case we need not define an additional state for the integral
MW, and AP, is the incremental change in the speed changer of tie-line deviation in area 2.
position in pu MW. The generator system is illustrated in
Fig. 3. Matrices A and B-Two-Area Problem
For the two-area problem the states and the control variables
System Equations are
Taking the equations for the power equilibrium, the incre- X1 f AP tiel dt
mental tie-line flow, and the change in generation and position
of the speed governor in area i, we have X2 A f Afi dt
X3 A 'Af
2fIf* ddt Afi
Ah + D ZiAf
+DA) + Ti,**(fAfi dt - fAf, dt) X4 A AP gl
=AJ%i - APdi (14) Xs A AX gvl
d AP0i U1 A 'Ape, (19)

= - 'AP01+ AX0,, (15)
dt t Tti X6 A f Af2 dt
d AX0v, X7 A Af2
+T ,a APCj,
1 1
= -
AX1v - Afi (16)
dt Tgvi T0R Tg X8 A AP02
For each area we have this set of three differential equations
describing the dynamic performance of the system to incre-
Xg A AX0,v2
mental load changes. U2 A APC2.
Substituting the definition of the states and the control into the six differential equations that define the
two-area problem places the system in the form x = Ax + Bu + rAPd where
-0 T12* 0 0 0 -1T2* 0 0 0
0 0 1 0 0 0 0 0 0
f*T2* f*D, f* f*Tl2* 0 0 0
2H, 2H1 2H, 2H,
0 0 O T1 0 0 0 0
Tt, lTtl
1
O O -]P o 0 0 0
g ViRl T1
o 0 O 0 0 0 1 0 0
0 0 0 a,2f*Tl2* f*D2 f*
2H2
0
2H2 2H2 2H2
1
o 0 O 0 0 0 0 -T
Tt2 T12
o 0 0 0 0 0 1 0 1
Tg v2R2
1
0 o 0 0 0 0 o]
B' [[o
0 O 0 0 0 0 0 T12
Tgv2-
_
o 0 0 0 0 0
2H,
r- (20)
O 0 0 0 0 f* O
L-
The matrix r, which is of dimension n X p, is called the System Specifications

disturbance distribution matrix. We observe that our model Before we define the Q and R matrices we must define a set
is not in the form we desire for two reasons. First, in the optimal of response requirements the system must satisfy.
control theory we use there is no r matrix, and second, the cost
function requires the states to be driven to zero for it to have a 1) The static frequency deviation following a step-load change
minimum. For a step-load change in area 1, we will require the must be zero.
steady-state frequency deviation in each area to be zero. But 2) The static change in tie-line power following a step-load
the increased generation in area 1 will by necessity in steady change must be zero.
state equal the increased demand, a nonzero quantity: 3) The transient frequency deviation should not exceed ±0.02
Hz under normal conditions.
APgi = APdl* (21)
4) The time error represented by the integral of frequency
Some of the other states will also be nonzero. deviation should not exceed 4 3 seconds.
To clear up this dilemma, we must redefine the states in To define these specifications mathematically requires the
terms of their steady-state values, i.e., sum of the following terms:
Xi Xi - xiss, i = 1,2, n. (22)
from condition 1)
This change of variable puts the system in the form
(Af,)2 + (Af2)2
xI = Ax' + BRu
from condition 2)
xl(0) = -x.8. (23)
By redefining the states in terms of their steady-state values (APtie 1)2 =[T,2*(f Af, dt - f Af2 dt) ]2
we have shifted the reference position of the system. We shall from conditions 3) and 4)
drop the superscript 1 to prevent unnecessary notation problems.
The matrices A and B remain unchanged. [(f Af, dt)2 + (f Af2 dt)]2.
*Af(t) AM(t)
IAPtie i(t) SEC . APiel It)

SEC.
Fig. 4. System dynamic response, case 1; Q and R as defined. Fig. 6. System dynamic response, case 3; coefficients of Q multiplied
by 10, R as defined.
A£f(t)
tie i(t) APtie (t) SEC.
Fig. 5. System dynamic response, case 2; Q as defined, coefficient Fig. 7. System dynamic response, case 4; coefficients of Q multiplied
of R multiplied by 100. by 10, coefficients of R multiplied by 100.
Defining these variables in terms of their respective states, system data were used:
and putting the products in matrix form x'Qx specifies Q to be
Pr, = P,2 = 2000 MW
1 0 0 0 0 0 0 0 0
H1 = H2 = 5 seconds
0 T12*2+1 0 0 0 -T2*2 0 0 0
0 0 1 0 0 0 0 0 Di = D2 = 8.33 X 10- pu MW/Hz

0
0 0 0 0 0 0 0 0
0 0 0 0 0
0
0 0 0 (24) Tt, = Tt2= 0.3 second
0 -T12*2 0 0 0 00 0 0 0
Tg VI = Tg = 0.08 second
0 0 0 0 0 1 00
0 0 0 0 0
0
0 00 R, R2 = 2.4 Hz/pu MW
=
0
0 0 0 0 0 0 0 O_ Ptie max = 200 MW
We penalize for large control effort by adding the terms (u1)2 + 61*-62* = 30 degrees
(u2)2, which requires for
T12* = 0.545 pu MW
R = I =[o (25) APdl = 0.01 pu MW.
Figs. 4-7 illustrate the improvement in damping given by the
COMPUTER RESULTS optimal controller to the system. The controller for each area is a
function of the states of both areas. We point this out by writing
A digital computer program was written that, when given the out the controller for case 1:
system and cost, checks for controllability and observability,
and if the conditions are satisfactory, solves the necessary u, = -0.707x1 - 0.300x2 - 0.932X3 - 1.28x4 - 0.296x5
conditions for the optimal control matrix K. The system was
simulated on the analog computer with the optimal gains.
- 0.701x6 - 0.064X7 - 0.030x8 - 0.006x9
Four cases were tried. For the two-area problem we considered, U2 = 0.707x1 - 0.701x2 - 0.064X3 - 0.030X4 - 0.006x5
the A and B matrix did not change. The Q and R matrices were
varied to show their effect on the system response. The following -0.300x6 - 0.932X7 - 1.28X8- 0.296x9. (26)
Af(t) controller 2 and area 1. The control system is essentially non-

interacting. To illustrate this we let the coefficients of APR2
and AX0 v2 in u1 = 0 and the coefficients of APR1 and AX,gi in
U2 = 0. The simulation of this suboptimal system is shown in
Fig. 8. The reader is invited to compare this response with
that of Fig. 4.
In Fig. 9 we show the "best" response obtainable [11 in the
sense of minimizing the tie-line deviation using the control
strategy employed by industry today; that is,
AP eI(t) SEC. U1 = - K.1 (xi + BX2)
U2 = -KI2(-X1 + BX6) (29)
Fig. 8. Suboptimal response. where KI is integral gain and B is frequency bias.
CONCLUSIONS
af(t)
In this paper a state variable model of the multiarea electric
energy system megawatt-frequency control problem has been
presented. Recent optimal control techniques have been applied
to develop optimal controllers for the electric energy system
that significantly improve transient response to perturbations in
load. The form of the optimal controller has shown that
1) by using all information about the system in the controller,
significant improvements in system response can be achieved;
2) the area controllers are essentially noninteracting;
3) area megawatt-frequency controllers should be functions
x-x1
SECS,
not only of the integral of tie-line deviation and frequency
deviation, but also of tie-line deviation, tie-line deviation rate,
frequency deviation, governor valve position, and area genera-
Fig. 9. Best response using existing control strategy. tion deviation.
In this paper we have studied only the two-area problem.
Future work should include study of the effect of size difference
We use the fact that on noninteraction of area controllers.
APtie 1 = T12*(x2 -x6) The megawatt-frequency control problem has been considered
separate from the megavar-voltage control problem for two
reasons. One reason is that for small changes in system load,
dAPtie
dt
1
T12*(x3 -X7) (27)
system voltages remain fairly constant while frequency deviates.
The other is that the megavar-voltage control problem has
where much faster response than the megawatt-frequency control
problem.
APtie 2 =-Ptie 1 The next step is to study system response when the two
dAPtie 2 dAPtie 1 problems are not considered to be noninteracting. Some work
dt dt has been done in this area already, and more is to be expected
[5].
Solving for these variables in our controller and replacing the
other states with their physical meanings gives NOMENCLATURE
U = - 0.7071 (APtie 1 + 1.41Afi)dt + 1.28APtie 1
Optimal Control Theory
x n X 1 state vector
+ 0.117 d -tie
1 0.996Af1 - 1.28APgl u m X 1 control vector
dtIf X n X 1 costate vector
A n X n state distribution matrix
-
0.296AXgvl 0.030OPA2 O.006AXgv2
- -
B n X m control distribution matrix
K m X n optimal gain matrix
U2 = - 0707 f (APtie 2 + 1.41 Af2)dt + 1.28APtie 2 Q n X n positive semidefinite symmetric state cost weighting
matrix
+ 0.117 dttie 2 0.996Af2 - 1.28AP,2 R m X m positive definite symmetric control cost weighting
matrix
-
0.296AX,v2 0.O30APg1
- -
0.006AXgvl. (28) C scalar cost
H Hamiltonian
In this form we see that the controller of area 1 depends P n X n positive definite symmetric matrix Riccati equation
weakly on information received from area 2 and vice versa for solution.
Electric Energy System and X* must satisfy the differential equations

f* nominal system frequency
Wki. kinetic energy of area * = OH (x* * U*)
H inertia constant ax
Ptie tie-line power
IAPtie incremental change in tie-line power J*= OH (x*l X'*, U*)
Oix
(35)
Af incremental frequency deviation
D load frequency constant and the function H(x*, X*, u) must have a minimum at uO.
APdj incremental load demand change We apply these conditions to our problem in the following
APO, incremental generation change manner. The last condition allows for us to determine ua since u
AXgv incremental change in governor valve position is unconstrained
Apc incremental change in speed changer position
area power angle
-= 0 = Ru + B'X
1Tiv tie-line power flow constant au
time constant
Tt turbine
speed governor time constant U* = -R-'B'X. (36)
R self regulation of generator
KI ACE integrator gain Equation (35) gives differential equations that x* and X*
B frequency bias constant. must satisfy:
%* = -A'X* -
Qx*
APPENDIX
CONTROLLABILITY x* = Ax* - BR-lB'X*. (37)
Given the system (1), let F be the n X
in nm matrix defined We would like u as a function of x so we assume
by A
x Px (38)
F _ (B, AB, A2B, A 3B, .*, A -'B). (30)
and try to determine conditions on P. Substitution for the
The system is controllable if and only if the rank of F = n, costate in terms of x into the last two equations gives
or equivalently if and only if there are n linearly independent
column vectors of F. Px* + Px* = -A'Px* - Qx*
x* = Ax* - BR-lB'Px*. (39)
OBSERVABILITY
Given the system in (1) where Substituting x* into the first equation of (39) and factoring
out on the right the vector x gives the n X n matrix differential
y = Dx (31) equation
is of dimension p X n, let G be the n X np matrix defined by P + PA - PBR-'B'P + A'P + Q (40) = 0.
G = (Di A'D', A'2D', *, An-1D'). (32) This matrix differential equation is called the matrix Riccati
equation. For the infinite time problem, the Riccati equation
The system is observable if and only if the rank of G = n, has a steady-state solution for which
orequivalently if and only if there is a set of n linearly indepen-
dent column vectors of G. P = 0. (41)
Once the steady-state solution to the matrix Riccati equation
DERIVATION OF OPTIMAL CONTROLLER is found [6]-[S],
We use the variational approach [3] to give us the necessary u* = -R-'B'X = -R-B'Px (42)
conditions for an optimal controller.
The system is in the form of (1), with cost to be minimized as and if we let
expressed in (2). We augment the cost by adding X'(Ax + Bu -
x) where X is an n X 1 costate vector. K -R-B'P (43)

then
C' 2 (x'Qx + u'Ru) + X'(Ax + Bu - x) dt. (33)
U*= -Kx (44)
Let the Hamiltonian be defined as which is in the desired form.
H - (x'Qx + u'Ru) + X'(Ax + Bu). (34) ACKNOWLEDGMENT
2
Several figures are borrowed from [9]. Use was made of the
The variational approach says: let u* be an admissible control digital computer facilities at the University of Florida and the
that drives the system from an initial point x0, where * in- analog computer facilities in the Department of Electrical
dicates that = u*. For u* to be optimal, the variables x*
u Engineering, University of Florida.
REFERENCES If any study of high-speed control is to be meaningful, the test

conditions, model and control criteria must be extended beyond that
[1] 0. Elgerd and C. E. Fosha, Jr., "Optimum megawatt-fre-
I.
discussed in this paper. With regard to the model, nearly all dynamics
quency control of multiarea electric energy systems," this issue,
pp. 556-563.
in the system are important, e.g., measurement and telemetering
[2] L. K. Kirchmayer, Economic Control of Interconnected Systems. lags, data and error sampling techniques. Governor dynamics are
New York: Wiley, 1959. nonlinear and more complicated than used in the model. Valves are
[3] M. Athans and A. Falb, Optimal Control. New York: Mc- also nonlinear. Generation dynamics are also important; they will
Graw-Hill, 1966. vary with the type of generation and may vary with load level.
[41 R. E. Kalman, "When is a linear control system optimal?" Load and generation are not lumped but often spread over large
Trans. ASME J. Basic Engrg., ser. D, pp. 51-60, March geographical areas. Generation changes are generally rate limited
1964. for safety of the equipment and to prevent malfunction due to a
[5] L. Yu, K. Vongsuriya and L. Wedman, "Application of optimal telemetering outage. This generation rate limiting would prevent an
control theory to a power system," IEEE Trans. Power Ap-
paratus and Systems, vol. PAS-89, pp. 55-62, January, 1970. "optimal" controller from functioning during large disturbances,
[6] R. W. Bass, "Machine solution of higher order matrix Riceati where it is needed, and would fruitlessly chase small random dis-
equations," presented at the 1967 Winter Institute on Optimal turbances, where it is not needed.
Control, University of Florida, Gainesville. Any control strategy that effectively deals with the high-speed
[7] D. R. Vaughn, "A negative exponential solution for the linear control problem must take into account the various dynamics in the
optimal regulator problem," Proc. 1968 JACC (University of system. For example, all tie-line flows must be measured in phase
Michigan, Ann Arbor), pp. 717-724. relative to each other to prevent improper control. The control
[8] T. R. Blackburn and J. C. Bidwell, "Some numerical aspects of
control engineering computations," Proc. 1968 JACC (Univer- error ACE must be formed to give the proper phase relation. This
sity of Michigan, Ann Arbor), pp. 203-207. includes phasing the frequency deviation, or tie-line deviation,
[9] 0. I. Elgerd, Electric Energy Systems Engineering. New York: to correspond to one another dynamically. Finally, the ACE must
McGraw-Hill (to be published). be phased properly to compensate for the remaining dynamics in
the control loop. This phasing depends on the configuration of the
system and is subject to change as the system is reconfigurated.
It is the feeling of this discusser that the results arrived at by the
authors are largely due to the effective phase shift in ACE resulting
from the adjustment made on frequency bias B. The results would
Combined Discussion" 2 undoubtedly have been different if other dynamics in the system
had been considered and if the relative size of the area under control
to the interconnected system had been made more realistic, e.g.,
1:10. Consideration should also be given to the control activity tak-
Charles W. Ross (Leeds and Northrup Company, North Wales, ing place in the area that did not produce the disturbance. With
Pa.): The authors are to be commended on their continuing efforts to regard to the weighting factor a in (30) of Elgerd and Fosha', it is
evaluate and improve techniques for load frequency control. How- not clear how the numerical value should be obtained and what its
ever, this area of endeavor is beset with many pitfalls resulting from effect will be on the "optimum" value of B.
the practical complexity of interconnected systems dynamics and its
multiplicity of desired control criteria. This discusser has com- With regard to the size of the area being considered during a
mented previously on some of the factors that should be considered control study relative to that of a large interconnected system, it is
for an objective evaluation of load frequency control [see discussion clear that most of the power flow will come from interconnecting
to Fosha and Elgerd2]. The discusser would like to reemphasize ties. This is the primary advantage of interconnection. Since this is
certain of these points, and to raise other points, which may also be generally the case, why does it matter what value of frequency bias
controversial, with regard to the subject paper.' is used for LFC? From the discusser's point of view the answer is
Load frequency control (LFC) of the form or "strategy" described clear-the frequency bias setting mainly represents the contribution
in this paper, is not generally intended for controlling disturbances of the area under consideration to the integrity of the interconnected
that have their power spectral density (PSD) in the range of syn- system as determined by the only common denominator, system
chronizing power swing frequencies. Indeed, when only reset control frequency. Any control action by the LFC that would reduce gen-
is used it implies a trimming or adjustment for low-frequency eration when system frequency is low, e.g., due to a small bias
disturbance components. This control action also normally includes setting, certainly will not enhance the security of any system.
economics, in contrast to the statement made in the paper to the A well-designed LFC must take into account disturbances that
contrary. cannot or should not be controlled, and these disturbance com-
In practice, when the higher frequency disturbances are con- ponents should be removed from the control error to be used for
sidered, with regard to LFC, the model considered in this paper is area supplementary control. Effective techniques, consistent with
far from adequate. This is evidenced by the fact that one rarely the actual system controllability and observability, have been
sees reponse characteristics in operating records as shown in the developed to accomplish this [1], [2]. If fast control is required to
paper; exceptions are usually due to electrically long tie lines in an damp synchronizing power swings, this may be better accomplished
interconnected system. Considerable care must be taken in designing by a properly phased control signal to select fast-responding units.
the LFC, particularly in the exceptional case of relatively large syn- In this case it may be preferred to control the generator's excitation
chronizing power components occurring in the area control error rather than the governor servo, since the response may be faster
(ACE). The first, and by no means trivial, question to be answered and the control action more reproducible.
is: Should any attempt be made to control synchronizing power The answer should now be clear to the authors' question in 5)
swings? The answer is generally, no. of their summary: "On what basis then does NAPSIC base its
1) Power systems are generally designed to be stable under any present policy of recommending to industry to set B = # but saying
expected operating conditions. nothing about Ki?" B and K1 serve entirely different functions. B
2) Synchronizing swings are usually uncontrollable, and any is that factor required to form the proper ACE and must be specified
attempt to control them may make the situation worse. to insure uniformity throughout the interconnection. KI is the
3) It is uneconomical to move generation unnecessarily in the controller gain used by the area to accomplish the required control
fruitless chase of synchronizing power swings. action on the error ACE. K1, of course, can vary considerably be-
tween areas because of differences in dynamics of an area and its
disturbances.
Manuscript received July 9, 1969. The adjustment of K1 is made by a systems engineer to obtain near-
0. I. Elgerd and C. E. Fosha, Jr., "Optimum megawatt-frequency
control of multiarea electric energy systems," this issue, pp. 556- optimum control performance (assuming these authors' control stra-
563. tegy) for his area. In essence, the engineer is carrying out an optmiza-
C. E. Fosha, Jr., and 0. I. Elgerd, "The megawatt-frequency tion similar in principle to that described in the paper for a simulated
control problem-a new approach via optimal control theory," system. The systems engineer, however, is experimenting on the real
this issue, pp. 563-570. "beast" and must exercise his knowledge and experience to deal
with many control conditions and criteria not covered here. In due to a telemetering outage. This generation rate limiting would
order to cope with this complex control problem, the systems prevent an "optimal" controller from functioning during large
engineer must have a control scheme considerably more complex disturbances, where it is needed, and would fruitlessly chase small
than described [2]. random disturbances, where it is not needed.
In summary, the test results for evaluating control and frequency A well-designed LFC must take into account disturbances that
bias cited may be misleading due to inadequacies in the model, cannot or should not be controlled and these disturbance com-
control strategy, control criteria, and test conditions. It is suggested ponents should be removed from the control error to be used for
that a more detailed evaluation be made by the authors with regard area supplementary control. Effective techniques, consistent with
to these areas before proceeding with further simulation. With the actual system controllability and observability, have been
regard to bias settings, a very simple but interesting method for developed to accomplish this [1], [2]. If fast control is required to
consideration may be that each area in the interconnection have the damp synchronizing power swings this may be better accomplished
same relative value based on the area's load, i.e., an agreed-to by a properly phased control signal to selected fast-responding units.
value for contribution to the system's integrity. This criterion In this case it may be preferred to control the generator's excitation
seems more equitable in light of inadvertent interchange that may rather than the governor servo, since the response may be faster
occur when an area is contributing relatively more to the integrity and the control action more reproducible.
of the system. Any objection to this operating procedure, e.g., on With regard to the test conditions, if D and T8V are intended to be
the part of an area having large governor and auto-regulating typical values, then they could be off by a factor of 20 or more. Is
characteristics, should be offset by the benefits of the interconnection this due to the computer scaling of the problem, e.g., frequency?
and new control techniques that reduce unnecessary control activity If not, the actual controllability is a matter entirely different
by LFC. from that simulated, i.e., the relative dynamics of the generation to
that of swing frequency. If, on the other hand, the values are for
REFERENCES convenient computer scaling, either the swing frequency is far
[1] C. W. Ross, "Error adaptive control computer for intercon- from being typical or the control activity u must have been large.
nected power systems," IEEE Trans. Power Apparatus and The authors should have recorded the activity of u and included a
Systems, vol. PAS-85, pp. 742-749, July 1966. time scale on the figures.
[2] ,"A comprehensive direct digital load-frequency control- As for the "good neighbor policy" it is not clear from the control
ler," Proc. 1967 IEEE PICA Conf. (Pittsburgh, Pa.), pp. equations (28) of Fosha and Elgerd' what control action (dynamic
231-238. and steady state) would be taken in each area if one or both areas
are not able to control optimally. Simulation of contingencies of
this sort, together with control actions in each of the areas, would
be helpful for evaluating the control strategy. One of the primary
advantages of an interconnected system is to permit tie-line power
flow to enhance the security of the system as a whole, for example,
Charles W. Ross (Cont'd) when one area is deficient in generation. Any successful control
For a system whose characteristics and control criteria under all scheme should be able to deal with this type of problem in a clearly
contingencies are well defined, the state variable technique may defined manner.
lead to optimal, or near-optimal, control as is illustrated for a Another control criterion not considered by the authors is the
postulated interconnected electric power system described in this economic loading of the regulating units. If this criterion is not
paper2. However, if the system or criteria are not invariant and not considered in the design of the control strategy [2], the consequences
well defined, a given control strategy may actually make certain will be unnecessary movement of generation and the creation of
situations worse. control errors, which may lead to cycling in the attempt to get
In practice, when the higher frequency disturbances are con- optimum control.
sidered with regard to LFC, the model considered in this paper is Finally, if one could assume that system characteristics and
far from adequate. This is evidenced by the fact that one rarely control criteria could be idealized, there still remains the practical
sees response characteristics in operating records as shown in the problem of system constraints, e.g., units reaching regulating
paper; exceptions are due to electrically long tie lines in an inter- limits, units being constrained because of equipment and safety
connected system. There are a number of reasons for this discrep- considerations, and units being blocked from direct control because
ancy, including: of telemetering outages. Constraints will lead to additional control
1) interconnection of power flows between the various loads, gen- degradation due to controller windup problems if they are not
erators, and tie lines that exhibit different characteristics; provided for in the control strategy [2].
2) damping of the high-frequency components due to measuring, Assuming that the authors would recalculate the optimum control
telemetering, and system characteristics. strategy in the event of constraints or system reconfiguration, it
Considerable care must be taken in designing the LFC, par- would be of considerable interest to know what is required in the
ticularly in the exceptional case where relatively large synchronizing way of input data and computation time. In answer to this ques-
power components occur in the ACE. The first, and by no means tion, of course, the authors should indicate specifically how these
trivial, question to be answered is: Should any attempt be made to requirements increase if the model were made more realistic, i.e., if
control synchronizing power swings? The answer is generally, no. generation and loads are not lumped together and more dynamic
1) Power systems are designed to be stable under any expected parameters and nonlinearities are considered.
operating conditions.
2) Synchronizing swings are uncontrollable by conventional
LFC, and any attempt to control them will probably make the
situation worse.
3) It is uneconomical to move generation unnecessarily in the
fruitless chase of synchronizing power swings. P. E. Mantey (IBM Corporation, San Jose, Calif.): This paper'
If any study of high-speed control is to be meaningful, the test properly addresses the control of tie-lines and frequency as a dynamic
conditions, model, and control criteria must be extended beyond control problem. This is in contrast to conventional load frequency
that discussed in this paper. With regard to the model, nearly all control strategies, which are formulated on the basis of static
dynamics in the system are important, e.g., measurement and considerations alone.
telemetering lags, data and error sampling techniques. Governor I commend the authors for bringing to this problem area the more
dynamics are nonlinear and more complicated than used in the powerful tools of modern control theory. The principal difficulty in
model. Valves are also nonlinear. Generation dynamics are also applying modern control theory to the problem of interest is that of
important, they will vary with the type of generation and may vary obtaining both the range of validity of the models used and the
with load level. Load and generation are not lumped but spread values of the parameters of the model. With good operating data
over large geographical areas. Generation changes are generally from the system it should be possible to estimate the appropriate
rate limited for safety of the equipment and to prevent malfunction values for the parameters of the model.
Manuscript received June 3, 1969. Manuscript received June 18, 1969.

The primary advantages of the approach taken in this paper- interchanged power deviations are not so strict. Indeed some con-
that of designing a control strategy based on a linearized model of sideration is being given, since it is probably well-known, to the ad-
system operation and a quadratic measure of performance-are: vantages of adaptive controllers, which strongly decrease the correct-
1) Stability of the closed-loop system so designed can be directly ing action by the regulating plants under normal operating con-
assured for the range of validity of the model. ditions, while increasing it in emergencies. On the other hand, the
2) By specifying performance in terms of the cost matrices Q and implementation of noninteracting optimal controls, which may de-
R [(3) of Fosha and Elgerd2] the closed-loop performance can be mand each member of the interconnection to send continuous infor-
flexibly and intelligently modified via these parameters in order to mation on the state of his own system to all other partners, may pose
obtain quantitatively what is qualitatively desired of the system. some practical problems. Other aspects of system frequency control
The quadratic form of the cost function also has some intuitive that should be considered if theoretical studies are to be made
appeal in penalizing large deviations and controls more severely. It practically applicable, concern the distribution of secondaryalmost and
should be pointed out that it is possible to make a slight reformula- primary control among the generators operating in the system;
tion of this problem with possibly greater appeal; that is, to specify
the desired trajectory for x and/or u in order to reach the required
all units cooperate in primary governing, but usually only a part
are in charge of load frequency control.
state and to penalize deviations about this desired trajectory. Such A further question relates to the sensitivity of optimum controls
an approach could result insmaller fluctuations of variables such as to the power system mathematical model. The model chosen by the
unit outputs. authors is a simplified one, but it would certainly be possible to
Information that would improve the communication of the apply the same methods to a more elaborate system matrix, and I
ideas of this paper are time scales on the curves and also plots of will not insist here upon the advantages of classical frequency
the speed-changer motion during the transient studied. response techniques for dealing with transcendental transferhydro- func-
I am also surprised that the authors choose to formulate this tions, such as may arise in the case of governors for high-head
control problem as an analog problem. The trend in the utility electric generators. However, the power system as a whole may be
industry is strongly to digital control, using the digital computer for defined by a transfer function which, on one side, certainly does not
calculating generation changes. A discrete formulation of this describe its behavior to a great approximation, since low signal
problem would thus seem ofmore practical interest, while somewhat nonlinearities, due, e.g., to governor deadbands, are quite relevant,
simplifying the problem by eliminating the possibility ofimpulses but on the other side also varies during the day, depending on
when R = 0. The computer requirements for this algorithm do not the system configuration.
appear to be appreciable, but comments on these requirements
by the authors would be of interest.
Experimental trials for deterministic and statistical identification
of power system dynamics well prove this fact. The question is, how
As a final comment, I anticipate that this paper is but a beginning much may the computed optimum vary with the variable model of
in the application of modern control theory to the dynamic control the power system, or conversely how wide may the performance
and regulation of a various quantities associated with the operation variations be when controller parameters are kept fixed at the
of interconnected power systems. optimum computed for a given situation? Finally, it is true that the
computed optimum depends on the definition of Q andR matrices,
and the question arises of how to choose them; but proper choice of
vector u and matrixR may allow for including control power rates
within the cost function, as it is customary in standard optimization
G. Quazza (ENEL, Milan, Italy): I would like to congratulate the studies for load frequency controllers.
authors for their plain and clear presentation of linear optimal REFERENCES
control theory as applied to load frequency control of intercon- Cuenod and G. Quazza, "Dynamic statistical analysis of
nected power systems. [3] M.
electric power system controls," presented at the IFAC Symp.
I wonder at this point whether I can still add any interesting on System Dynamics and Automatic Controls in Basic In-
comments on the paper,2 after such exhaustive discussions. How- dustries, Sydney, Australia,nonAugust 1968. di reti intercon-
ever, there are a few questions on which I would like to hear the [4] G. Quazza, "Regolazioni interangenti
authors' opinion. nesse" (in Italian), L'Energia Elettrica, August 1963.
First, the type of disturbance chosen for the search of optimum [51 ,"Noninteracting controls of interconnected electriit
controls is a step variation in load. Although this is probably ade- power systems," IEEE Trans. Power Apparatus and Systems,
vol. PAS-85, pp. 727-741, July 1966.
quate to qualify the system response to large perturbations, such as
loss of a big generator or a heavily loaded tie line, the performance [6] WorkingduGroup Aldans of CIGRE Committee 13, "Critbres de
of the controlled system is usually analyzed also with reference to qualit6 reglage les r6seaux 6quip6s de dispositifs de
normal operating conditions, when load changes about their average reglage frequence-puwssance," Paris, 1966.
G. Quazza, "Criteria for equitable participation of areas in tie-
value are better described as random variations having a Poisson
[7] line
powerAutomazione
and frequency control of an interconnected power
distribution, or the like. Is the optimum control theory applicable
also to random inputs? Can the authors show how? And, finally,
system," e Strumentazione, March 1966.
what are the significant differences with respect to step inputs?
It must be recalled at this point that standard optimization
techniques have been extensively employed to find the optimal
parameters of load frequency controllers for a given control struc-
ture, under Poisson-distributed random load variations [31. The A. J. Pennington (Drexel Institute of Technology, Philadelphia,
choice of the best control structure has also been the object of Pa.): This interesting paper' begins with a presentation of the
analytical studies based upon multivariable control system theory standard simplified dynamic model for a two-area power system. It
[4], [5], and of discussions at the CIGRE Working Group on Power then proposes a "cost function"-(30) of Elgerd and Foshal on
System Automation [6]. the transient performance after a disturbance involving the time
Noninteracting control schemes, such as those insuring that single integral of the weighted sum of squared frequency and scheduled
partners of an interconnected system could be capable of taking tie-line power deviations. This cost function is minimized with
care of their own load changes-with due contribution to main- respect to loadan frequency control simulation.
gain and bias settings by trial and
taining system frequency and fulfilling interchange power obliga- error, using analog computer With this background
tions-are found to be more equitable than the single criterion of the following observations can be made:
minimizing errors, which may require a greater effort from some 1) In spite of its rigorous appearance, the definition of the "opti-
members than from others [7]. mum" controller is still a matter of engineering judgment; here, the
Actually, the performance that can be usually achieved by such judgment has been exercised in the choice of an integral squared
separate frequency and tie-line power controllers appears to be error criterion and the value of theweighting coefficient a. For
more than satisfactory, although difficulties may arise with overly another set of choices another "optimum" controller would result
weak interconnecting lines. The requirements on frequency and whose response might be more or less appealing from the point of
Manuscript received July 29, 1969. Manuscript received May 27, 1969.
view of engineering judgment. This is not to say that the proposal formance of a given system after specific disturbances. In conclusion
of a quantitative measure of transient performance is of no value; it appears that at its present stage of development, optimal control
it simply must be put in perspective. theory is not capable of solving the engineering problem of load
2) In practice the specific nature of the recovery from a system frequency control system design. It may, however, provide some
disturbance is not of primary importance. As long as it is reasonably valuable insights, and on this basis the present paper is certainly to
rapid and well damped, without danger of instability or a sustained be welcomed.
limit cycle system performance can be considered to be satisfactory
from a dynamic point of view. Much more important is that the
combined system be able to cope with the situation in which one
area has insufficient generation or load to return frequency to
normal, or in other words, has insufficient "control authority."
In this case it is necessary for the frequency bias to be set equal to L. K. Kirchmayer and D. N. Ewart (General Electric Company,
the area response characteristic in order to prevent the control from Schenectady, N. Y.): The authors have conducted a linear analysis
actually forcing system frequency in the wrong direction in a sus- of the performance of a two-area power system and have attempted
tained fashion. This point is treated thoroughly by Cohn [8]. The to optimize the dynamic recovery of tie-line flow and frequency
question of the associated dynamic response is treated by Kirch- following a sudden upset in the balance between generation and
mayer in [2] of Elgerd and Foshal and he concludes (also on the load in one of the areas. Based on this study, they have recom-
basis of an analog simulation) that a bias setting close to the area mended' that the frequency bias setting used in load frequency
response produces good dynamic performance; here "good" is control equipment be set at about half the value currently being
defined via judgments about typical responses, rather than through used by most utilities.
minimization of a cost function. They referred to a book by one of the discussers-published in
On the basis of the above reflections on the results obtained, and 1959 [21 of Elgerd and Fosha.' Contrary to statements made by the
on a consideration of the adequacy of the dynamic model itself authors, the referenced work did address itself to the selection of
(it is really only a first approximation) the dramatic recommendation K1, did investigate the optimum value of frequency bias, and was
that the industry change its frequency bias settings by a factor of based on dynamic rather than static considerations. The results
two seems premature, to say the least. I trust the authors did not presented in this book show an optimum value of frequency bias
really intend an engineering recommendation at this time. somewhat less than the system regulating characteristics.
The presentation of a new approach to analysis of load frequency The work described in the book, (as well as that of the authors),
control systems, i.e., the use of an integral cost function, is to be was based on nonreheat turbines having relatively fast response. In
welcomed. If the method is applied with proper judgment to an the ensuing years, large reheat turbines have come to have a dom-
adequate model, and with proper consideration for steady-state inant influence on the performance of systems. More recent dynamic
performance during the important case of deficient or excess genera- studies by the discussers have recognized these changing system
tion after a disturbance, then further work may, in fact, suggest characteristics. Considerations of response-rate limitations and
some changes that will improve system performance. I look forward governor deadband have resulted in load frequency controllers
to the authors' promised future work on this problem with great having much smaller integral gains K1 then considered in these
interest. studies. Values of 0.05-0.25 are more typical today. Unit response
rate limitations are typically on the order of 3 percent per minute.
REFERENCES The result is that even for moderate generation/load unbalances,
1 percent of system capacity, the units in the area that sustains
[8] N. Cohn, Control of Generation and Power Flow on Intercon- say the trouble require many seconds to bring ACE back to zero. Larger
nected Systems. New York: Wiley, 1961.
unbalances may require several minutes for recovery. Since frequency
bias normally corresponds to frequency droop, the other areas in the
interconnection do not see a generation requirement and they do
nothing beyond governor action to assist in the restoration of fre-
quency. The resulting slow rate of frequency recovery can be seen
A. J. Pennington (Cont'd) on the frequency charts in every dispatch office in North America
As indicated in the Introduction, this paper2 builds on a previous following the sudden loss of a large unit anywhere in the inter-
one.1 Likewise, this discussion builds on the one provided for the connection.
referenced paper. Both deal with the application of "optimal" If rapid restoration of frequency were of primary importance, it
control theory to load frequency control of power systems, where would obviously be advantageous to increase rather than decrease
"optimal" means that a stated integral cost function is minimized the frequency bias so that all areas will participate. Of course, as the
by the controller. Two general points should be made: area in trouble picks up generation, the others will back off again, and
1) Many "optimal" controllers exist; in fact, for a controllable this unnecessary change in generation is often considered undesirable.
system there will be one for each choice of cost function. The author's results substantiate the fact that bias should be in-
2) These controllers do not necessarily satisfy standards of engi- creased to restore frequency more rapidly. Examination of Fig. 14
neering judgment other than the one represented by minimization of Elgerd and Foshal will reveal that when all the weight is given to
of the chosen cost function. frequency error, the optimum value of frequency bias is greater than
With regard to the representation of load frequency control 100 percent at small values of Ki.
treated in this paper in particular, we see that the radically dif- We are pleased to note the academic interest in this problem area.
ferent "optimal" responses of Figs. 4 and 5 of Fosha and Elgerd2 The work of the authors would be greatly enhanced by recognition
are obtained via a change in the cost function; the long-term error of the system characteristics that depict today's problems.
of Fig. 5 results from the heavy penalty put on control action in the
cost function for that case. The cost of control is not really measured
as squared integral, and the practical reason for this term is to limit Manuscript received June 30, 1969.
the amplitude of control effort; as pointed out, without this weighting
"the best control strategy would be in the form of infinite impulses."
The rather loose connection between what the control engineer
"really" wants (as determined by engineering judgment) and the
"optimal" control is even more strikingly illustrated in terms of the
stated system specifications on frequency and time error. Inequality Nathan Cohn (Leeds and Northrup Company, North Wales, Pa.):
constraints of 4± 0.02 Hz and 3 seconds are stated. Aside from the New analyses of the power systems control problem are always
lack of conformity to industry practice in this matter, it is clear that welcome, and so it is with this paper.1 Its content and conclusions,
inequality constraints cannot, by their very nature, be implemented however, suggest to this discusser that the authors are not fully
via a cost function, except in terms of case by case testing of per- informed on important aspects of interconnected systems operation,
Manuscript received May 27, 1969. Manuscript received July 10, 1969.
including points that are of particular significance in determining perimentation, resides in its ability to eff ectively control, as presently
the magnitude of bias settings. used concepts do, real power in real time in the very dynamic en-
Under the subheading Control Specifications, the paper tabulates vironment of a real interconnected power system.
"'minimum requirements," identified by the authors as suggestions
of the North American Power Systems Interconnection Committee. REFERENCES
There are significant omissions, and hence errors, in the authors' in- [91 F. R. Schleif, L. L. Lloyd, R. W. World, and W. B. Gish,
terpretation of NAPSIC recommendations. "A swing relay for the east-west intertie," IEEE Trans.
For example, in point 1), the authors state that "the static fre- Power Apparatus and Systems, vol. PAS-88, pp. 821-825,
quency error following a step-load change must be zero." This June 1969.
specification is true only if the area in which the load change occurred
can adjust its generation fully to accommodate this change. If it
cannot, the system operating objective is not to reduce the static
frequency error to zero, but instead to permit it to persist to a Nathan Cohn (Cont'd)
degree sufficient to permit, or cause, other areas to provide assistance I am in disagreement with many of the points made in this paper.
to the area in need, for the full duration of the need. I will, however, confine my remarks in this discussion to the systems
Under point 3) the authors state that "the static change in the specifications defined by the authors and to other general condi-
tie-line power flow following a step-load change in either area must tions discussed in the paper.2
be zero." Again, this is true only if the area in which the step-load In the section Dynamic System in State Variable Form the authors
change occurred can adjust its generation, in the steady-state tabulate response requirements which they say the system must
sense, to accommodate the change. If it cannot, the operating satisfy. These correspond to a comparable tabulation in their earlier
objective is not to reduce the static tie-line power flow to zero, but paper' in which they identify these requirements, incorrectly, as
instead to permit such flow to persist so that other areas, operating recommendations of the North American Power Systems Intercon-
on tie-line bias control, will provide sustained assistance to the area nection Committee.
in need, for the full duration of the need. Unless we are dealing with a generation gap in terminology, it
These qualifications to the authors' control specifications are of seems to me that of the four specifications, two-the first two-are
considerable importance. They define a major function of tie-line incomplete and hence are incorrect. The third is an improperly
bias control, and are the dominating factors in establishing the identified constraint, and the fourth-although a correct statement-
desirable magnitude of bias settings. They are the basis for the would seem not to have pertinence in the control analysis. I will
bias setting analysis in [3] of Elgerd and Fosha.' Their omission comment on each of the four in terms of present-day operating
from the present paper may well account for the authors' summary philosophy.
comment that "Nowhere in our extensive computer analysis has it 1) It is stated that the static frequency deviation following a
been confirmed that the bias setting B = ,3, which presently is used step-load change must be zero. This is true only if the step-load
by the U. S. utilities, offers any advantages of a particular nature." change can be accommodated in the area in which it occurred. If it
It is not surprising that this requirement was not encountered since cannot be, for whatever reason, then the system strategy is to
the conditions that make it desirable were apparently not considered. permit the static frequency deviation to persist, so that assistance
To underscore this point more fully, one might examine Fig. 9 of will remain programmed into the area of need.
Elgerd and Fosha.1 It illustrates system conditions after a step-load 2) It is stated that the static change in tie-line power following a
change in area 2, without supplementary control having been applied step-load change must be zero. Again this is true only if the step-
in either area. load change can be accommodated in the area in which it occurred.
Assume that the step-load change represents an appreciable loss To the extent that it cannot be, for whatever reason, a static change
of generation in area 2, and that area 2 is unable, itself, at that in tie-line power will persist, as intended by system control phi-
point in time, to accommodate the full amount of the change. What, losophy, constituting assistancefrom the other area or areas to thearea
in this circumstance, would the authors have area 1 do? Lower its in need.
generation, push an already low frequency lower still, and cut back 3) It is stated that the transient frequency deviation should not
on its assistance to area 2? That is precisely what a bias setting of exceed ±0.02 Hz under normal conditions. By "transient deviation"
B <B, including B = 0.5f, which the authors recommend, would do. I assume the authors mean the "swings" on either side of the steady-
This would constitute totally undesirable operation. Avoiding such state frequency schedule that are typical of a normal system fre-
withdrawal of frequency response assistance under such conditions quency. I am not aware of any effort made with present-day control
is a basic objective of power systems operations. This objective systems to achieve such a quantitative objective, nor am I aware of
can be achieved only with a bias setting of B > ,B. any conscious desire to do so. These "swings" are almost entirely a
It is of interest to note that bias settings of about B = 0.53, function of the power system itself, and it is the longer term "drifts"
which the authors recommend, were actually (though not inten- from frequency schedule rather than the shorter term "swings"
tionally) standard practice prior to the presentation in 1956 of the that load frequency control equipment seeks in general to contain.
analysis of [31 of Elgerd and Foshal; this can be seen by reference to On the other hand, what NAPSIC does recommend is an offset in
the Mollman discussion (p. 1430). The frequency chart of Fig. 11 of frequency schedule of 0.02 Hz for purposes of time-error correction.
the reference illustrates, in the interval B to C, an actual operating The normal "swings" are then "based" at this offset level, and vary
example of undesirable withdrawal of initial frequency response above and below it, as determined by the characteristics of the
assistance, due to biases of about B = 0.53, by remaining areas of the power system itself.
interconnection following a large loss of load in one area. An already 4) It is stated that the time-error represented by the integral of
excessive generation was additionally increased, and a high system frequency deviation should not exceed ±t3 seconds. This has been a
frequency pushed higher still. Subsequently when system operating correct statement, although NAPSIC at present is working on a
guides were revised, and areas biases were increased to B > B, such ±2-second limit. It is not clear, however, that this has any relevance
noncooperative operation was eliminated. It hardly seems desirable, to the analysis in the paper, or is utilized therein. From an operating
to return to a practice found to be objectionable in actual operation view, it simply defines the time error that will stimulate an offset in
and discarded thirteen years ago. frequency schedule to achieve time-error correction.
The authors also suggest that bias settings of B = 0.5,B would I would invite the authors' comments about whether the fore-
eliminate east-west tie separation. Considering the causes of such going discussion reflects differences in terminology usage, or whether
separation, as identified in current literature [91, there is no basis for we are indeed concerned with basic differences in the desirable
believing that the separations are related to bias settings or that operating philosophies of an interconnected system.
they would be significantly influenced if bias settings were decreased. In any event, I should like to add that the control systems cur-
rently in use although improved and refined over the years, are
Despite my disagreement with the paper's conclusions and based on concepts introduced more than 20 years ago; thus it by no
recommendations, I would commend the authors for seeking a new means follows that they are necessarily the ones that will provide
technique for power systems control. I think it appropriate to
faodd, however, that the real test of a new concept, whether derived
from intuition, theory, mathematics, simulation, or empirical.ex- Manuscript received June 2, 1969
optimum system regulation. New analyses are welcome, and the control problems-the already widely extended interconnection of
utilization of advanced control theory in such analyses is to be power systems.
applauded. The authors are to be commended on both counts. On the In exploring such a problem, the use of simplifying assumptions
other hand, an understanding of prevailing practice ought to be a is certainly appropriate in order to quickly obtain a perspective.
prelude to new operating recommendations. Such understanding However, in this process, emphatic cautions are also appropriate:
appears to be lacking in the present paper. results from such a preliminary stage should be kept tentative, and
In any case, there remains the ultimate pragmatic test of no attempt should be made to apply conclusions until the basis is
evaluating proposed new techniques in the real world of present shown to be adequate for the problem.
day power systems. This is a world not of two areas, but of over 120 In setting up the dynamic model the authors assume2 that control
areas, tied together with hundreds of tie lines. There is just as of real power and frequency is decoupled from control of reactive
likely to be lack of "coherency" within areas as between them. In power and voltage. The assumption is reiterated in the conclusion
the aggregate there are probably 3000 or more generating units on the grounds that the megavar-voltage problem has much faster
operating in these areas, most of which are under the control of their response.
individual speed governors. These governors have varying time Actual system behavior has offered powerful evidence to the
constants, deadbands, and sensitivities, and their characteristics contrary. The evidence acquired is that the interrelation between
are far from being linear. Time constants of boilers as well as of voltage, load, and power flow dominates the dynamic response of the
turbines and generators need to be considered. This is the dynamic real system rather than the tie-line control problem. Tests conducted
world in which present day control systems operate, and which in August 1967 disclosed no influence from the tie-line control upon
would be the testing ground for new techniques. a somewhat oscillatory tendency of the real system at the time.
The authors should certainly be encouraged to continue their Another step in the test program did disclose a prominent influence
studies, and the projection of their findings to practical systems between voltage regulators and the dynamic behavior of the system.
operation, if planned, will be awaited with interest. With reference The authors refer to a previous paper' and employ similar models.
to their use of the frequency deviation rate in an optimum controller, In that model the system has been made rather underdamped by
they may be interested in the disclosures of U.S. Patent 2 923 832, assuming a gain for the tie-line control that appears to be higher
which describes the use of the rate of change of frequency to improve than reality. From this, the inference was drawn that the NAPSIC
control performance. recommendation of matching frequency bias to the area frequency
response characteristic is the cause of oscillation-seemingly with-
out proper understanding that the NAPSIC recommendation was
for the purpose of avoiding unnecessary interaction between prime
mover governors and tie-line control. No oscillatory tendency comes
D. N. Ewart (General Electric Company, Schenectady, N. Y.): from this function if the tie-line control gain is properly reduced.
The authors have presented an interesting application of an op- Thus, we propose that the system model would be made much more
timization procedure to derive a "best response" load frequency reliable by introducing a realistic amount of underdamping through
controller. They found that performance could be improved with the action of voltage regulators, rather than by using a high gain for
introduction of control variables other than the ones normally used. the tie-line control.
Although it was not made clear in the paper,2 it is assumed that the This is not to ignore the instructional value of "decoupling"
authors intended the conclusions to apply only to the simulated to explore one effect at a time. It is to propose caution with inter-
system and not necessarily to actual systems. I make this point pretation of results until all pertinent influences have been included
because unit rate limits and governor deadband have major effects and properly scaled.
on the performance of load frequency control, and the authors did
not simulate either phenomenon.
As an example, control signals sent to a thermal unit under load
frequency control typically might be limited to move the unit at a
maximum rate of 3 percent per minute. At this rate, the boiler Olle I. Elgerd and Charles E. Fosha, Jr.: The lively response to this
usually can change steam flow without causing large temperature sequence of two papers 1, 2 on the topic of automatic load frequency
and pressure excursions, and the turbine will not be subjected to control (ALFC) confirms our convictions that, right or wrong, we
undue thermal stresses. Governor deadband might be on the order are discussing a very important problem in power systems operation.
of 1 percent of unit MW rating. (A 0.06-percent frequency dead- In this brief closure we shall attempt to answer some of the questions
band is normally allowed. Translated to megawatts at the 2.4-Hz raised by the discussers. In two forthcoming papers [101, [11] we
per pu MW governor droop the authors used, this would result in a shall answer others that do not lend themselves to brief commen-
deadband of 1.5 percent of unit rating.) Combining the effects of taries.
rate limiting and governor deadband, a unit being sent signals to As we see it, the ALFC systems in our networks can be divided
move at 3 percent per minute would require 20 seconds or more just into two essentially different and presently decoupled loops:
to unwind the deadband each time the demand reversed before any 1) A fast primary ALFC loop that controls (or should do so) the
change in unit output would occur. This long time delay materially fast "power sloshings" between control areas. The main component
influences the response of load frequency control. of this loop is the standard speed governor. The speed of this loop is
Other effects, such as nonlinear response at valve points, and of the order of magnitude of seconds.
varying number and type of units under control, also tend to compli- 2) A slow secondary ALFC loop that executes via speed-changer
cate the problem of selecting the best gains and control variables. commands slower control actions to satisfy certain tie-line require-
The extension of the authors' present work to include the effects I ments and also, of course, together with the primary loop helps
have enumerated would be very useful. It would be helpful if the maintain the frequency. The speed of this loop is of the order of
authors would indicate specific time scales for Figs. 4-9 of Fosha magnitude of minutes. It is important to realize that this secondary
and Elgerd.Y ALFC loop has no significant effect on the response of the primary
loop, because of its low loop gains.
We focused our principal attention on the response of the primary
Manuscript received June 3, 1969. or fast ALFC loop. We made the hypothetical approach in one
paper' to adjust the control parameters that are available in the
present type of controller in order to "make a dent" in the primary
response characteristics. When we found that this approach offered
only very slight improvement possibilities, we suggested an entirely
F. R. Schleif (U. S. Bureau of Reclamation, Denver, Colo.): The new control strategy in the second paper.2
authors demonstrate a commendable interest in applying refine- In the order of importance we now will attempt to answer the
ments of modern control theory to one of the most complex of main questions brought up by the discussers:
Manuscript received June 3, 1969. Manuscript received August 25, 1969.

FOSHA.ANDELGERD: MEGAWATT-FREQUENCY CONTROL PROBLEM 577
1) Is it necessary or indeed desirable to suggest changes in the primary a) The analysis suggests that each ACE need only contain signal
ALFC loop? Several discussers do not share our opinion in this information from the area in question. No signal transmission be-
matter. Mr. Ross motivates his "no" by stating that theloops are tween areas is therefore needed.
"stable under any expected operatingconditions." He furthermore b) Although presently used sensors of frequency and tie-line
makes the statement that the synchronizing swings are "uncontrol- powers that are used in the slow secondary ALFC loops of today are
lable" and any attempts to "chase" these fast swings are "fruit- definitely too slow for the proposed strategy, we see no reason why
less." signal measurement should represent an insurmountable problem
Consider first the matter of stability. It was an indisputable (private communication with Dr. G. Parks of Michigan State
case of primary ALFC loop instability (the East-West intertie) that University, who has studied this problem confirms this assumption).
initiated our curiosity in this problem. Subsequent discussions with c) The value of optimum analysis as presented2 is that it tells us
operating people, and the accumulation of a large number of pub- in no uncertain terms the best strategy to use based upon a chosen
lished reports on actual observed real power instability in the range criterion. In reality, practical constraints always force one to chose a
of 0.1-1 Hz seem to indicate that our primary ALFC loops do not suboptimal strategy. The theoretically best strategy guides us in the
possess the inherent stability that we would like to see. search of this.
As indeed the question of stability of the primary ALFC loop is 5) Analog versus digital formulation? Dr. Mantey wishes to see the
basic not only for the relevance of our work but really for every problem formulated as a digital control problem. This is not surpris-
systems operator, we have continued our work by making an ex- ing in view of his association with IBM.
tensive stability investigation of the present type of systems. The We chose the analog formulation for the very simple reason, con-
results of this work [10], using the best systems data we can muster firmed by years of teaching experience, that this approach has very
at this point, indicate that we are presently operating our systems definitive tutorial advantages. However we are the first ones to
with very poor stability margins. Our computations indicate that agree that when and if the proposed strategy is practically imple-
the primary ALFC loops are characterized by two dominant conju- mented, the use of digital controller would be the obvious choice.
gate complex eigenvalue pairs that loom very close to the jw axis in 6) Effect of rate limitation? Several discussers have pointed out
the s plane. If careful attention is not given to the regulation pa- that the rate limitations for the generator permit only slow speed-
rameter R, one of these eigenvalue pairs can easily "glide over" to changer movements. The theory as presented still applies equally
the right-hand s plane, resulting in just the kind of oscillations that well, but we must now modify the optimum control strategy-
have been observed in many of our networks. (26) of Fosha and Elgerd2 -in the following way: From the computed
We believe strongly that the problem is real, and equally strong optimum strategy we must deduct the terms defined-by (29) of
that it can be solved if we take some new control approaches. Our Fosha and Elgerd,2 the latter representing the slow speed-changer
second paper2 contains some reasonable suggestions. commands. The optimum strategy thus modified represents now an
2) Accuracy of our models? The model used in this study was optimum supplementary control signal to be added after the speed
characterized by linearity, and furthermore the turbine plus con- changer, and which has an optimum damping effect on the fast
troller was modeled by two time constants only. Even so-our power swings. (The curves of Figs. 4-8 of Fosha and Elgerd2 do not
overall two-area system turns out to be of seventh order. We felt apply in this case, since the frequency and tie-line power deviations
that the tutorial value of our paper would be impaired by complicat- will now only slowly return to zero final values.)
ing the models any further at this stage. 7) Effect of voltage regulator? The entire analysis was based upon
However, the digital computer programs developed for this and the assumption of noninteraction from the volage regulator; i.e.,
subsequent studies have no such limitations [12]. We can accom- we assume the latter to be very fast compared with the ALFC
modate (and have done so in [10]) any linear model we wish to use. dynamics.
Since we are discussing a perturbation phenomenon, we hold that If the latter assumption is not true, then the magnitudes of the
the linearity assumption is a permissible one at this analysis stage. terminal voltages of the tie line will vary. This effects the electric
Not only does this assumption permit us to use classical stability stiffness of the line T12*, and thus the dynamics.
theory (eigenvalue methods, etc.) but it also permits us to make use It is very simple to prove [10] in the mechanical analog in Fig. 8
of well-developed modern control theory. of Fosha and Elgerd2 that if we vary incrementally the stiffness of
Once we have decided upon an optimum linear control strategy the connecting spring in direct proportion to the incremental ve-
we can later adjust this strategy by simulation to account for non- locities of the masses, then we introduce in effect a positive damping
linear effects. (This approach has been very successful in space into the system.
booster control.) This confirms observations made by Schleif and others that if we
3) Deterministic or stochastic analysis? Dr. Quazza brings up a add to the voltage regulators signals essentially proportional to
very pertinent question. Why do we chose to base our analysis on a the frequency deviations then we should obtain positive system
step function type of load change, when, in reality, the load changes damping.
are of stochastic nature? This phenomenon is very interesting and research is being con-
This, of course, conforms with a standard control practice. If ducted [11] aimed at finding the optimum combination of voltage
we know how a system performs in response to a step input (which is and turbine control.
the simplest of all inputs), we can compute by superposition the
response to any other input, including a stochastic one. We claim REFERENCES
therefore that the step-input response conveys the most compact
and meaningful information. Optimum control theory, in minimizing [10] 0. I. Elgerd and C. E. Fosha, Jr., "Load frequency control of
the cost of control in changing the state of a system from one specified power systems-a reassessment," presented at the SEEE Fall
value to another, is thus essentially a deterministic theory. Meeting, Washington, D. C., 1969.
4) Can the proposed control strategy be instrumented? We have not
[11] 0. I. Elgerd and C. Durick, "Optimum primary ALFC utilizing
both turbine and field control input variables," research in
discussed to any important extent the possibilities of hardware progress.
implementation of the suggested strategy. Many important factors [12] C. E. Fosha, Jr., and 0. I. Elgerd, "Optimum linear control of
enter such a discussion-and we are aware of some possible dif- the multivariate megawatt-frequency control problem,"
ficulties. Let us make a few pertinent comments: Proc. 1969 JACC (Boulder, Colo.), pp. 471-472.

Megawatt-Frequency: Problem

Uploaded by

Copyright:

Available Formats

Megawatt-Frequency: Problem

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Megawatt-Frequency: Problem

Uploaded by

Copyright:

Available Formats

IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, VOL. PAS-89, NO.

4, APRIL 1970 563

Tv turbine time constant REFERENCES

Abstract-This paper records the development of a state variable INTRODUCTION

construct complete information about the state x. This condition

SYSTEM DYNAMIC MODEL

DYNAMIC SYSTEM IN STATE VARIABLE FORM

=AJ%i - APdi (14) Xs A AX gvl

d AP0i U1 A 'Ape, (19)

The matrix r, which is of dimension n X p, is called the System Specifications

IAPtie i(t) SEC . APiel It)

tie i(t) APtie (t) SEC.

0 0 1 0 0 0 0 0 Di = D2 = 8.33 X 10- pu MW/Hz

Af(t) controller 2 and area 1. The control system is essentially non-

Electric Energy System and X* must satisfy the differential equations

x) where X is an n X 1 costate vector. K -R-B'P (43)

REFERENCES If any study of high-speed control is to be meaningful, the test

Manuscript received June 3, 1969. Manuscript received June 18, 1969.

Manuscript received June 3, 1969. Manuscript received August 25, 1969.

You might also like