Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

Proceedings of the 34th

Conference on Decision & Control


FM07 1:30 -
New Orleans, LA December 1995

ADVANCED GENERATION CONTROL


WITH ECONOMIC DISPATCH
D. Brian Eidson Marija D. IliC
Dept. of Electrical Engineering and Computer Science
Massachusetts Institute of Technology
Cambridge, MA 02139

Abstract veloped. Then, a slower-acting, centralized control


scheme is proposed to effect this tie-line regulation.
A new approach for the Automatic Generation Control (AGC) This control, called tertiary control, acts by sending
of electric power networks is introduced in this paper. This
approach also supports an Economic Dispatch feature which
frequency (set-point) goals to every secondary con-
operates at a slower time scale and interacts with the AGC troller in the interconnected system.
frequency stabilization function. In developing the control models, a recursion for
AGC is separated into two subtasks: frequency regulation the evolution of generator power is also developed.
and tie-line flow control, and a hierarchy of controls are formu-
lated t o realize these two functions. The frequency regulation
h e 1 cost curves are then associated with generator
level, called secondary control, is decentralized, and optimizes power outputs, the result being an equation govern-
with respect to both frequency offset and change in generator ing the evolution of AGC cost. Given this new cost
governor controls. This innovation allows an administrative model, controls can be applied, and an Economic
area to readily choose the cost-performance mix it desires. The
tie-line flow control level, called tertiary control, is centralized, Dispatch-type function realized.
and works on a slower time scale than secondary control. Ter- This version of Economic Dispatch attempts to
tiary control compensates for inadvertent flows between areas minimize overall system fuel cost in meeting load de-
by momentarily offsetting generator frequencies. These fre-
quency offsets shift the phase angles found at tie-line terminals,
mand. Like the tie-line regulator of AGC, Economic
and eventually realize the desired line flows. Dispatch is realized as a tertiary level control, which
The role of Economic Dispatch (ED) is to reschedule the en- sends governor frequency set points to all secondary
tire system to minimize overall generation cost. The proposed level machines participating in frequency regulation.
version of ED is centralized, and invoked at an even slower
rate than tertiary tie-line regulation. When applied, Economic
One difference, however, is that Economic Dispatch
Dispatch sends signals directly to secondary controllers, and is applied at a much slower rate than tie-line control,
establishes new interface flow settings for tertiary level tieline so when invoked, it also sends a signal to the ter-
control. tiary level (tie-line) controller. This control updates
control flow set points used by the tie-line control op-
eration.
1 Introduction
In this paper, we introduce models and control tech- 2 Model Development
niques for Automatic Generation Control (AGC) and
Economic Dispatch on electric power systems. We begin by developing a model relevant for sec-
In what follows, we develop a discrete-time model ondary level frequency control. The model is
which tracks the succession of frequency steady-states constructed by linking machine droop characteris-
evolving from periodic governor set-point changes, tics (section 2.1)-which describe the steady-state
slow load disturbances, and slow movements in tie- frequency-versus-power behavior of generators-with
line flows. With this model, a decentralized frequency network power flow constraints (section 2.2).
control is proposed which allows a regional system
operator to choose a cost-performance mix most suit-
2.1 Droop Characteristic
able to his needs. This frequency control function is
called secondary level control. Assume real/reactive power (frequency/voltage) de-
Next, to address the issue of inadvertent tie-line coupling. For Automatic Generation Control (AGC)
flows between areas, an interconnected system model applications, this implies that governor actions only
relating tie-line flows to frequency set points is de- induce changes in frequencylreal power; no changes

0-7803-2685-7/95 $4.00 0 1995 IEEE 3450


in terminal voltage occur. This decoupling assump- incorporation of a governor-turbine-generator model
tion is commonly made in AGC studies 14,71. different from the one chosen in (1)-is
Consider a single machine consisting of a governor,
turbine and generator. Under the decoupling assump- (4)
tion, a simple state-space model for the linearized'
dynamics of this machine is
For further use, stack the droop relations for a num-
ber of machines to form a vector of generator frequen-
cies:

7
* [ k ] = ( I - cD)g;f[k]- E&[k] , (5)

+I [: - 1
A,,,."

PG+ [i ] W a c ]

M , and D , are, respectively, the moment of inertia


.(1)
where * [ I C ] , g z f [ k ] , and &[k] are vectors de-
scribing the frequencies, frequency set points, and
electrical power outputs of all the generators in a
subsystem; b = diag[ D' D2 D" 1, C =
diag[ o1 u2 ... ( T 1~, and I , an identity matrix of
dimension m.
of the generator and its damping coefficient. T, and
These droop relations are coupled through &,
Tg are the time constants of the turbine and gover- which in turn is governed by network interconnection
nor. The state variables are the generator frequency,
constraints.
W G ; the turbine mechanical power, Pt,; and the gov-
ernor valve opening, a, which serves to regulate tur-
bine power. w z f [ k ] k, = 0,1, . . . is the (frequency) set 2.2 Network Constraints
point value for the governor-a set point which can Consider now the connections between machines
be updated every kT, seconds. PG, the real (electri- within an area, or sub-network, of a large intercon-
cal) power output of by the generator, appears here nected system. Network constraints are typically
to be an exogenous variable; however, it is is not an expressed using nodal equations which require the
independent input since it depends on constraints im- -N
posed by interconnections with other generators.
complex-valued power into the sub-network, S , to
Assume from here forward that the machine dy-
be equal to the complex-valued power 3 = +j -
Q
injected into each node. In other words,
namics of (1) are stable for the anticipated range of
PG. This is normally the case if the governor (primary
control) is designed properly, such that the system
-N
matrix designated APT;has eigenvalues with negative where 5 = - PN + j Q N is the vector of net
real parts. complex power injections into all nodes, and ?ius
Consider now the steady state to which the system is the admittance matrix of the sub-network.
settles between updates for the governor set point. 9 = [ Vlejsl, V2ej62, . . . ] is the vector of all nodal
-
Setting the derivatives in (1) to zero and solving for voltage phasors, each with magnitude V , and phase
WG yields
6; (collectively, with magnitude and phase 4).
W G [ ~=
] (1 - ( ~ D ) w ; ~ [-k(]T p ~ [ k,] (2) The real part of ( 6 ) can be further partitioned into
real power network constraints at the generator and
where (T is the (steady-state) droop constant, which, load terminals
for the model in (1) is
(7)

where & represents injections into the network from


To emphasize that the WG here is the steady state generator buses, EL represents power removals from
response at t = kT,, note that WG in (2) is expressed load buses, and .FGand EL represent tie-line flows of
as w ~ [ k ] . power from adjacent systems into the network at the
An equivalent definition for the generator and load buses, respectively.
droop characteristic-a definition which allows facile Linearizing (7) under the decoupling assumption
lNote that the linearization assumption is not restrictive. (af"/ay = 0), one obtains
Since the models developed within are causal, the our tech-
niques can be extended to to stepwise linearization-for the PG+& JGG JGL
case of 1-step finite-horizon control.

345 1
where J;j = al’r fahj, i, j = G,L. Observe that, Another useful generator power model is one that
upon linearization, the variables appearing in (8) are relates generator output power to generator fre-
all incremental quantities: their values indicate off- quency. This model, derived by evaluating (12) at
sets from the nominal value at which the linearization two sample points and making the backward Euler
occurred. approximation of (14), is
J L L should be invertible under normal operating
conditions. Make the definitions
&[k + 13 = &[k] + KpT&G[k + 13
A
- + DPdPl .
-f[kI (20)
Dp = -JGLJ;i (9) This model will be manipulated to develop tertiary
A
K P = JGG+DPJLG (10) level control schemes for both tie-line flow manage-
A ment (section 4.1) and, when combined with the gen-
-
F = &+Dp& . (11) eration costing methods of section 2.5, Economic Dis-
Solving the partitioned expression (8) for & yields patch (section 6.3).

l&=Kp&+-F+DpPL . (12) 2.5 Generation (Fuel) Cost


2.3 Frequency Model AGC cost can be formulated using (19) and marginal
fuel cost. Typically, curves for the fuel cost of indi-
To derive a state-space description for frequency dy- vidual generators are available. These curves are, in
namics, the network constraints on & in (8) are first general, nonlinear, and due to the variability in the
substituted into the multi-machine droop character- price of &el, time-varying:
istic (5):
I?= h(PG,t) , (21)
@G[k] = (1- C D ) G f [ k ]-
where PG is a large signal generator power output-
( K P b [ k l - @I + DpP,[kl) 413) rather than the incremental quantity PGwith which
Then (13) is evaluated at two successive AGC sam- we have dealt previously-h(.) is a nonlinear function
+
pling instances, kTs and (k l)Ts, these results sub- which maps generation into a cost, and t is time.
In this paper we choose to model h(.) as time in-
tracted, and the approxiniation
variant, although our methods can easily be extended
WG[k + 1]Ts = &[k + 13 - &[k] (14) to account for slow temporal variability (e.g. season-
incorporated. Note the backward Euler approxima- ality) in the cost curves.
tion in (14), which is important because this leads to Typically, generator fuel cost curves are modeled
a (causal) model different from that found in [3]. as quadratic functions of P G : for a single generation
The subsystem frequency model which evolves from unit, this implies
these operations is t = a0 + alPG + a z P , , (22)
WG[k + +
11 = (1 CKPTS)-~{%[k]+ where the a0 parameter accounts for the nonzero cost
(I - ED) u [ k ]+ (f[k]- Dpdlk])} . (15) of maintaining spinning reserve.
Since the frequency control model is linearized, we
where linearize this cost curve about an operating point Pz.
The incremental model for the offset in generator cost
-u[k] &f[k+ 13 -&f[k] (16) from this operating point is
A
-f[k] = E[k + 11 - F[kl (17) C=mcPG , (23)
A
d[kI = P L [ k + 11 - P L [ k l . (18) where c is the incremental cost, and

2.4 Generator Power Models


A model which follows the evolution of generator out- is the marginal fuel cost.
put power, P G ,(in response to governor controls) can The incremental cost for all of the generators on the
be derived using a similar approach. The resulting system is found by stacking the equations governing
model is cost into vectors, c. The resulting expression is
-
P,[k + 11 = (I + KpCT,)-l{ P&]+ c=Mc& , (25)
KP (1- ED) ~ s L G f [ k 1- (@I - D,d[kI)}(lS) where M , is a diagonal matrix of marginal fuel costs.

3452
2.6 AGC non-participation the objective of the finite horizon control is to mini-
mize
Not all generators in an area need participate in fre-
quency regulation. Define a participation matrix, C J [ k ]= E { g T [ ki-
1]Qg[k + 11 + uT[k]Ru[k]}
(27)
as a matrix with dimension numpartgen x numgen
composed of Ones and zeros such that only partici- with respect to 14, subject to the model constraint
pating generators are selected. For example, the par-
ticipation matrix for a three-generator system where
the-second generator doesn't participate would be the E {.} in (27) represents the expected value operator,
2 x 3 matrix and

A f (I+CKpTs)-l (29)
Thus, the participating controls for such a model be- B e A(1-Cb) (30)
come g p = CZL;in the model of (15), 21 is replaced
with CTgP.
d,[k] 2 AX (f[4 .
- D,d[k]) (31)
Note that the frequency at non-participating gen- The weighting matrices Q and R in (27) are diago-
erator buses will still be directly regulated given this nal, with nonnegative entries. Their relative values
adjustment. If no direct stabilization is desired at trade off-for each participating generator-the im-
non-participating buses, the controlled states of (15) portance of frequency regulation versus the coupled
are reduced via appropriate multiplication by C and fuel cost and undesirability of changing a governor set
CT. point.
The optimal g [ k ] is of the form
~[
- k=] -K, (A%[k] - g g t [ K ] ) , (32)
3 Secondary Level Control:
Regulating Area Frequency where
K , = ( R +B ~ Q B ) - ' B T Q . (33)
The goal of secondary level control is to stabilize gen-
erator frequencies, %[k +
11, within a region so that 3.2 Infinite Horizon
they reach desired set values, g g t [ K ] despite
, the in-
For the infinite horizon case, 24k] is chosen so that
troduction of a slow load disturbance, &c], which is
the performance criterion
assumed to have zero mean. For this application, we
examine two types of control: finite (1-step) and infi-
nite horizon control. Finite horizon control facilitates J, = J[kI (34)
frequent parameter updating or relinearization, while k=O
infinite horizon control should offer better asymptotic is minimized. The control which minimizes J , is
performance if the model parameters are quite accu-
rate. The effects of plant uncertainty are not studied ~ [ k=] -Ks(gG[k]- &3'[K])+B-' ( I - A ) )@'[IT] ,
in this paper. (35)
In both approaches, the expected increment in tie where
line flows, f[k],is considered small, and is neglected +
K , = ( R BTSB)-lBTSA, (36)
in the computation of the controls. A decentralized and S is the solution to the familiar (see [5, 61) dis-
control which incorporates some flow dynamics is cur- crete algebraic Riccati equation
rently being developed, but will not appear here.
Non-interactive flow compensation like that found o =A As +Q - A ~ S B (+
~ S- RB T S B ) - ' B ~ S A .
in [3] cannot performed here since measurements for (37)
+ +
E[k 11 = f [ k ] E[k] are not available at the time
the control computed.
4 Tertiary Control: Regulat-
3.1 Finite Horizon ing Tie Line Flows
Given the assignment In the secondary level frequency control application,
arealsubsystem control is decentralized. As a con-
z[k
- + 11 = W G [ k + 11 - ,=t[K], (26) sequence, the dynamics of tie-line flows between the

3453
areas are not modeled. Each area uses its own sec- 3. K p and Dp are interconnected system equiva-
ondary control to regulate frequency, and this action lents of the secondary level Kp, and D p , , a =
changes the phase angles at its end of tielines. Since 1, 2, . .. , R (but, due to area interconnections,
the party at the other end of a tie-line is performing a neither can be be easily assembled from the Kp,
similar, uncoordinated action, tie-line flows can drift or DR).
from their scheduled values.
An equivalent expression for interconnected system
One responsibility of tertiary level control should
dynamics can be obtained by stacking the state-space
be to occasionally update the governor set points
grt [K] in each otherwise autonomous area such that
equations for each area's secondary dynamics (15)one
atop the other:
prescribed tie-line agreements are maintained. An-
other goal of tertiary control is to reschedule tie- PG[K + 11 = PG[K] + KgkTtQzt[[K 11 +
line flows in response to the expected load variations,
or new contracts between areas. At present, this + Dzkd[K].
-f[K] (39)
rescheduling task is not done in a systematic way; it For this case, f[K]= F[K+ 11 - F[K]is a vector
is, instead, carried out by agreement between several concatenation of the increments in flow measurements
areas at a time when needed. from all areas, and DFk and Kgk are block diagonal
Since tertiary level models are by necessity central- matrices formed from the Db and Kj, associated with
ized, and thus quite large in scale, the time intervals, each area.
Tt, between updates should be much longer than the Equating (38) and (39), and solving for F[K+ 11 =
Tsfor secondary control. This presents no problem f[K] + F[K]yields the tie-line flow model for which
however, because in practice, tie-line rescheduling is tertiary level controls will be designed:
done on a fairly slow time scale.
F[K+ 11 = F[K] + (Kgk - Kp) TtC?zt[K+ 11
4.1 Centralized Model for Tertiary + (D!k - Dp) d[K] . (40)
Control
4.2 Tertiary Level Control Design
We now derive a model relevant for tie-line flow con-
trol. The goal is to find a relationship between gen- The goal of tertiary control is to derive a frequency
erator frequency and tie-line flows for the entire in- set point vector OFt[K]-which is later broken up
terconnected system. and sent to the secondary level controllers in each
Assume that the interconnected system consists of area-such that a desired (mapped)2 flow schedule,
R areas, enumerated 1, 2, . . . , R. Assemble a vec- . eventually realized, perhaps over several Tt
p e t is

tor of the frequencies of all participating generators, sampling periods. Fset itself can be occasionally reset
from every area, within the interconnected system; to reschedule interface flow agreements between the
call this vector fig'. Denote the associated vector of administrative areas, perhaps made by means of a
generator power outputs PG.Take the new sampling recent Economic Dispatch calculation.
rate to be Tt, and index these longer time increments The finite and infinite horizon proportional control
with [K], rather than [IC]. solutions for the feedback gain matrix Kt are struc-
Now treat this new global, interconnected system turally identical to those found at the secondary level:
as a very large subsystem, and apply (20)-which the only difference is that, for this control problem,
describes the power dynamics of a subsystem. This A ---+ I and B + (Kjtfk - Kp) Tt. Relative values
yields of the matrix weights, R and Q, determine the trade-
off of frequency deviation versus the rate at which
tie-line flows settle to their scheduled values.
The structure for the infinite horizon controller is

Note that although the form of (38) may resemble


+
Ogt[K 11 = - K t (F,[K] - F s e t ) -
(20), some important differences do exist. Since this
is the entire interconnected system, where E { d [ K ] }is the load deviation expected over
1. no external tie-line flows exist, so f[K] the tertiary sampling period. Notice the addition of
0;
an noninteracting (anticipative) control term.
2. d[K] is a vector describing the increment in load 'We denote actual flows as FG and F L ; mapped flows are a
demand throughout the interconnected system; +
function of the actual flows: F = FG D r k F h .

3454
Another subtlety associated with tertiary control is point); in essence it keeps the generation of any par-
+
that F = FG DF'FL maps the actual tie line flows ticipating unit from falling beyond a preset limit. For
FG and FL back to the generators. This means that the infinite horizon controller,
one cannot use our models to directly control FG and
FL. However, these flows can be indirectly controlled (43)
(but not uniquely specified) via F. Most important,
though, is that the number of mapped flows to be The weighting matrices Qed and Red used in calculat-
controlled should never exceed the total number of ing Ked allow a system operator to pick the speed of
tie lines. The control problem is overconstrained oth- cost reduction to the deviation in frequency necessary
erwise. Therefore, multiplication by a participation to achieve that amount of reduction.
matrix C of ones and zeros must be used to reduce Since Economic Dispatch perturbs area interface
the number of flow states to be controlled to the num- flows to minimize cost, a new tie-line set point, FE;,
ber of tie lines. The number of controls, though need must computed and sent to tertiary flow control, since
not be reduced. it is invoked much more often. The set point is com-
puted using (40), where one solves for FZL := F[K+
+
11 with F[K] := F;;: and C4gt[K 11 := C4Et[N 11. +
Unlike the tertiary level tie-line control, cost states
5 Economic Dispatch: Mini- can be minimized throughout the system; none need
mizing Cost be ~ m i t t e d . This
~ is a desirable characteristic, be-
cause the power industry tends to dispatch all units
Denote another form of tertiary control, which again when attempting to minimize costs. As far as the
sends frequency set point signals C4gt, down to sec- authors know, this formulation is the first which has
ondary controllers, Economic Dispatch. Economic treated Economic Dispatch as a non-static optimiza-
Dispatch reschedules the generation within the en- tion process, and the first that has tied Economic
tire system to minimize total cost, while at the same Dispatch to the frequency control function found in
time attempting to minimize the frequency deviation power systems. Moreover, through participation ma-
incurred on the system to realize this function. Con- trices, not all generators need participate in Economic
ventional (static) Economic Dispatch is performed at dispatch. This is an important feature, since not all
a much slower rate than AGC, typically every 15 min- generator owners in a deregulated environment-for
utes in the United States. example, Independent Power Producers-may want
to participate in a coordinated cost reduction scheme.
If the number of states to be stabilized is reduced
5.1 Economic Dispatch Model to less than the total number of generators, for exam-
The cost model used for Economic Dispatch is simply ple, by using a state selection matrix, C, of dimension
derived by evaluating (38) at a sampling rate of T e d , numpartstates x numge,, then a noninteracting antic-
and multiplying both sides by the diagonal matrix ipative control term may be added4 The controlled
of marginal fuel costs, M,, of the generators in the state then becomes cp = C c p ,and the controls are
interconnected system. The derivation of M, is found
in section 2.5. The resulting model for c , the cost of QE'[[K + 11 = -Ked ( c p [ K -
] cPfIoor)-
generation throughout the system, is (CMcx:KpTed)-lMcDpf { d [ ~1 ] (44)

c [ +~11 = C[N] + M , X K ~ T , ~ C +~ 11~ ~ [ N


+M,Dpd[N] . (42)
6 Simulation Study
The 3-generator, 5-bus system depicted in Figure 1
Note that [NI is chosen as a time index. This distin- was used in our simulations. In this system, there
guishes the slower rate at which Economic Dispatch are two tie lines, one between generators #2 and #3,
is invoked from the (relatively) faster rate associated and another between loads #4 and #5.
with the tie-line regulation function of AGC.
The generator parameters and initial operating
points for this system are found in Table 1. All intra-
5.2 ED Control Design area transmission lines had impedance 0.01 +jO.l pu;
3Performance tends to improve, however, if one of the di-
Controls are similar to tie-line flow design, except the rectly regulated states is omitted.
cost set point is cfloor.This set point determines the 4the interconnected system version of Kp has a rank-one
floor for the deviation in cost (from the linearization deficiency because phase angles must have a reference.

3455
all tie lines were weaker, with impedance 0.1 jl + .-N. 4 /*
pu. Secondary control’s sampling time, T,, was 2 3
seconds; tertiary control, when invoked, acts every
20 ~ e c o n d s . ~

-.-...._...
..../._.... -
-,, 3
(1

- dash..senl,sol&gen2,dots=gen3

k!
0 10 20 30 40 50 60

Tie-line flows into Area I:solid=gen lie, dots=loadlie

-
S OM
P

Figure 1: An interconnected 5-bus power system. Figure 2: Bus #4 0.2 pu step disturbance t = 30 sec:
Secondary (infinite horizon) control, Q = R = I .

Figure 3 (top) compares the frequency regulation


I Load Flow Data 1 performance at generator #2 (which is representa-
bus 1 #1 I #2 I #3 1 #4 I #5
tive) for finite and infinite horizon secondary control.
V I 1
I
I
I
1 1 1 I .9909 1 .9820
I I I
Note that infinite horizon approach offers a slightly
P I .424 I .9995 I .9992 I -1.200 I -1.199 quicker stabilization response. Figure 3 (bottom) de-
I Generator Parameters I
~

picts the case where no secondary control (only pri-

HzH
mary/mechanized control) is applied. Notice that
#1 5% a longer time scale is used to demonstrate that all
#2 5% generator frequencies approach the same equilibrium
#3 5% point, but this equilibrium is nonzero.

Table 1: Per-unit data for the 5-bus example. 6.2 Tertiary (Tie-Line) Control
We next demonstrate that tertiary control can be
used to drive the sum of inadvertent flows to zero.6
6.1 Secondary (Frequency) Control Figure 4 displays the result of invoking tertiary con-
trol on top of secondary control to regulate the
In Figure 2, we demonstrate the results of applying mapped flows for generators #2 and #3. As the top
secondary control (only) to a situation where a step (frequency) plot reveals, this is achieved by increas-
load increase of 0.2 pu occurs at load bus #4 (Area I) ing the frequencies (phase angles) in Area I, while
30 seconds into the simulation. For this simulation, decreasing the frequencies (phase angles) in Area 11.
Q = R = I , where I is the identity matrix. Plots of This intentional frequency offsetting is particularly
frequency offset, generator power outputs, and devi- apparent in the 40-60 second interval.
ation in actual (not mapped) tie line flows into Area
I are found in this figure.
Note that at the outset of the load disturbance in 6.3 Economic Dispatch
Area I, the frequency offset decreases in both areas, We will assume quadratic generator fuel cost curves
but less in Area I1 due to high impedance of the tie + +
c = .5 PG P& for generators #1 and #2, and
lines between the areas. Note that there are net flows + +
c = 1 .5pG 2Pz for generator #3. These curves
into Area 11, however.
6Note: for these simulations, we choose the rightmost term
51n actual practice, tertiary control would invoked on the of (35) to be zero although in this simulation, the set point fed
time scale of minutes; however, for illustrative purposes here, secondary control is non-zero. Including that term also drives
we invoke it much faster. the sum of inadvertent flows to zero.

3456
x lo4 Frequency offset 0 gen U2. inf and 1-step horizon 5 x 10.' Frequency deviation from 60 Hz
- 1 -

U
_J
-10; io 20 j, 40 io 60 o; eo 90 too

-8 ' -a
ec o 1 - --
26 28 30 32 34 38 38 d a h . ; ( a f l l 3 d i U = g ~ 2 W p n-~
Time (sec)
Y jn-' Frequencyoffset (all m s ) for no secondary control
B
\
-1c \ 4

i1-- 23 1

35 30 35 40
---4
(close to same)

45 50 55 60
ii -0.01 L
o io 20 30 40 50
Time (sec)
BO 70 80 m
l
loo

Figure 3: Top: Frequency offset at gen #2 for infinite Figure 4: Same as Figure 1, with tertiary (Q = R =
and 1-step horizon secondary control. Bottom: Fre- I) and secondary Q = R = I . Both controls infinite
quency offsets at all generators, no secondary control horizon.
applied.

are linearized about the operating point for the sys-


7 Conclusions
tem, yielding marginal costs of 3.0, 1.8, and 4.5 $/pu
power. We have introduced models and optimal control tech-
We choose to directly minimize only two of the cost niques for the tasks of Automatic Generation Control
states (associated with generators #2 and #3)-so (AGC) and Economic Dispatch on electric power net-
our results can be contrasted with those for tie-line works. Unlike current implementations of AGC [l,21,
control. The cost deviation floor for all generators is the our AGC design was composed of a hierarchy
set at 0. A step disturbance at t = 30 sec is initiated of controls, which we called secondary and tertiary
as was done in the previous sections, and like tie- control. Secondary control, which was decentralized,
line control, Economic Dispatch is invoked every 20 demonstrated the capability to adjust the power gen-
seconds. erated within an administrative area so as to drive
Figure 5 displays the frequency offsets, power out- frequency offsets to zero. Tertiary control, which was
puts, and tie-line flow response resulting from an Eco- centralized, returned the sum of interface power flows
nomic dispatch simulation. Observe that since gen- with neighboring systems to their contract values.
erator #3 has the highest marginal cost, its power
output is eventually driven to its (linearization) floor Another, even slower-acting tertiary control
value. Also, as one would expect the net tie-line paradigm, called Economic Dispatch, was introduced
flows into Area I do not sum to zero: since power to reschedule generators so as to minimize overall sys-
is (marginally) cheaper in Area I, power flows into tem fuel cost. The efficacy of this technique, along
that area to compensate for losses. with that of the AGC techniques, was demonstrated
via simulation.
Figure 6 compares the generation costs of using
various control strategies to respond to a load dis- Future work includes demonstrating the perfor-
turbance. Remark that not only does use of no sec- mance advantages of this advanced generation con-
ondary control result in poor frequency regulation, it trol over conventional AGC, comparing this new, dy-
also is more expensive. As one can see, combining namic version of Economic Dispatch with static Eco-
Economic dispatch with secondary control offers the nomic Dispatch approaches, and incorporating gen-
lowest system cost. Tertiary-level tie-line regulation eration constraints into 1-step controls for the var i-
is cheaper than just secondary control only because ous algorithms. Also, some efforts will be undertaken
it minimizes tie line flows of power from the more to design controls more robust to errors in the plant
expensive generator in Area 11. (system) models.

3457
Frequency dewaton from M)Hz Sum of generation costs usmg dlfterem controls

D
U -5
p! sec only
U

20 40 60 80 100 120 ________---


05
----___
. ldevlatm
Generator real Dower outDuts . from lmearmtronf dot dash. Eco Disp
0.2, I
.-..
-n
a
L

g 0- genr;&jfi&ipa;Was=sen5........................... -
a

Tie-line flows into Area I: s o l i tie, &&=load tie


3 0.02
n
Y

%
B
n 0- .....................
B
E -0.02
0 20 40 60 80 100 120 o OO 20 40 . 60 80 100 l 120

Figure 5: Economic Dispatch with (Q = R = I) and Figure 6: Total (systemwide) incremental cost asso-
secondary Q = R = I . Both controls infinite horizon. ciated with different controls.

8 Acknowledgements [5] H. Kwaakernaak and R. Sivan, Linear Optimal


Control Systems, Wiley: New York, 1972.
The authors gratefully acknowledge support for this
research from EPRI/PECO collaborative project [SI R. F. Stengel, Stochastic Optimal Control, Wiley:
RP3954-1 and DOE project grant DEF641-92ER- New York, 1986.
110447. Assef Zobian of MIT also deserves special
[7] J. Zaborszky and T. Chiang, “EconomicAreawise
thanks for supplying a load flow program and assist-
ing in its integration into our simulation software. Load F’requency Control,” Report No. SSM7402,
Parts I and 11, Dept. of Systems Science and
Mathematics, Washington University, St Louis,
References MO.

[l] J. Carpentier, “ ‘To be or not to be modern’ that

is the question for automatic generation control


(point of view of a utility engineer)”, International
Journal on Electric Power & Energy Systems, Vol.
7, No. 2, April, 1985.
[2] N. Cohn, “Research Opportunities in the Control
of Bulk Power and Energy Transfers on Intercon-
nected Systems”, EPRI EL-377-SR, Special Re-
port on Power System Planning and Operations:
Future Problems and Research Needs, February
1977.
[3] B. Eidson and M. IliL, “Advanced Generation
Control: Technical Enhancements, Costs, and
Responses to Market-driven Demand,” to be
presented at the American Power Conference,
Chicago, 11, April 1995.
[4] D. Ewart, “Automatic Generation Control: Per-
formance Under Normal Conditions”, Proc. of
the Systems Engineering for Power: Status and
Prospects, Henniker, NH, August 1975.

3458

You might also like