IEEE Transactions on Power Systems, Vol. 10,No.

2, May 1995 1077

A Capacity Based Lagrangian Relaxation Unit Commitment

with Ramp Rate Consbrain&
William L. Peterson Steven R. Brammer
Member IEEE Member IEEE
Stagg Systems Inc.
Houston, Texas

-
Abstract This paper presents a Lagrangian Relaxation Recently, Lagrangian relaxation has emerged as an
based technique for sohting the power system thermal unit effective method of solving the unit commitment problem.
commitment problem. As a general approach, Lagmngian Most research in this area has involved more rigmus
relaration has proven to be an effiient method of solving attempts to optimize the dual objective [2,3,4,5,61.
the thermal unit commitment problem. The spec&
technique described in this paper focuses on fnding a Another approach to the Lagrangian relaxation problem
feasible commibnent, then attempting to improve on the was proposed in [7] where the emphasis is on obtaining a
optimality of the solution. A new method for processing feasible solution and then attempting to improve on that
mmp rate constraints k also introduced. The algorit?tm Iras solution. This approach provides a fast, efficient method for
been implemented on a large scale power system, and results generating near-optimal feasible solutions. However, the
me shown using actual power system dah. method described does not address how to process the ramp
rate constraints within the optimizarion problem.
E;EyIKoRDs: Unit Commitment, Generation Scheduling,
Economic Operation, Lagrangian Relaxation, Interchange This paper describes a Lagrangian relaxation unit
Evaluation commitment which uses the maximum capacity conStraint
method in [7]. The algorithm is extended to incorporate unit
minimum capacity constraints and unit ramp rate constraints.
I. INTRODUCTION The introduction of unit ramp rate constraints requires
modification of the calculation of capacity constraints and
The problem of developing an optimal short term operating reserve constraints. The algorithm incorporatesother practical
strategy is a primary concern in the operation of electric features such as boiler frre-up characteristics and non-linear
utilities. The optimal strategy or unit commitment typically ramp up sequences.
provides the minimum cost of operating for a period of one
week while respecting various generating unit and system
constraints. This operating strategy is further used as the 11. UNIT COMMITMENT PROBLEM FORMULATION
basis of determining economic and reserve based interchange
tranSaCti0m. The unit commitment primal objective function, when
considering the optimization of start-up,dispatch, and shut-
The unit commitmentproblem is a mixed integer nonlinear down of fossil fired generation units, is given by minimization
programmiug problem that has typically been solved in the of the following cost function:
past using dynamic programming techniques with heuristic T NG
algorithms to limit the solution space [ll. F = [ Cl<PI.S + SI, 1 (1)
I-1 1-1

94 SM 590-0 PWRS A paper recommended and approved where cj is the fuel cost function for unit i,
by t h e IEEE Power System Engineering Committee of t h e pj,, is the unit generation for unit i in hour t,
IEEE Power Engineering Soccety f o r p r e s e n t a t i o n a t Si,, is the start up cost for unit i in hour t which is a
t h e IEEE/PES 1994 Summer Meeting, San Francisco, CA, function of the unit state and state hours,
July 24-28, 1994. Manuscript submitted December 28,
1993; made a v a i l a b l e f o r p r i n t i n g June 1 , 1994.
T is the number of study hours,
NG is the number of generators.
The minimization is performed subject to the constraints
outlined below:

(a) System Constraints

0 1994 IEEE
0885-8950/95/$04.00
1078
i) Power Balance solution is made feasible by economically dispatching the

c
NG

1-1
Pi,, = D,
committed generators.

The maximum capacity constraint farces commitment of

where D, is the system demand in hour t adjusted for sufficient generation to meet the demand and reserve
baseloaded generation and scheduled interchange. requirement. The maximum capacity constraint for hour t is
given by
ii) Reserve Requirement NG
..-
NE E P r M 2 D ,+RI (5)
E ri 2 RI
i-1
(3) 1 - 1

where Pi"" is the maximum genemtion contribution for unit

where ri is the unit reserve contribution for unit i, i.
R, is the system reserve requirement in hour t.
The minimum capacity constraint assures that the demand
The reserve contribution for each unit is dehed as can be met with all dispatched generators loaded above their
respective minimum limits. The minimum capacity constraint
(4) for hour t is given by
NO
where PI" is the upper generation limit 0for E Pi"
i-1
5 0,
reserve calculations,
p,, is the reserve ramp mte (MWbh) for where PiwN& the minimum generation contribution for unit
unit i, i.
z is the reserve time frame (min).
B . Reserve Constraint
(b) unit opeaatins constraints
Absolute capacity to satisfy the reserve requirement is
i) minimum up and down times obtained from (5). An additional constraint is necessary to
ii) unit response rate limitations assure that the committed generam have sufficient reserve
iii) minimum and maximum operating limits ramping capacity to satisfy the reserve requirement
iv) boiler fire-uprequirements ..-
N42

(7)

m.LAGRANGIAN
RELAXATION FORMULATION If both (5) and (7) are satisfied, the reserve requirement of
the primal problem may be fulfilled by using an economic
The unit commitment problem is solved by formulating the dispatch of the committed generators that accounts for the
problem as a Lagmngian relaxation problem where the system reserve requirement. Such economic dispatch
constraints are relaxed with Lagrange multipliers. The functions, which essentially dispatch the required reserve
Lasrangian dual is formed by appending the relaxed economically, are standard functions of production Energy
constraints to the primal objective. The Lagrange multiplim Management Systems.
are updated using a subgradient method which drives the
solution toward feasibility. C. Ramp Rate Constraint

A. Capacity Constraints The ramp capacity constraint couples the objectivefunction

to the available unit ramp rate capacity. The constraint is
Formulating the Lagrangian dual objective with (2) as the given by
only coupling to the system demand has been shown to
produce solutions with a very small duality gap. However,
i-1
the solution obtained is not necessarily feasible when the dual
objectiveconverges, and typicald t s show 30-100 iterations where pzj is the normal unit response rate (MWbour) for
are required for convergence 141. unit i in hour t,
P,"" is the r e q u i d change in generation h m hour
This technique introduces new constraints and produces a t-I to hour t.
feasible dual solution in 5-10 itemtions. The feasible dual
solution is a commitment schedule that is a viable The required generation change is determined from the
commitment schedule for the primal problem. The primal hourly change in system load plus fixed changes in
 1079

generation. Fixed changes in generationresult from scheduled

changes in base-loaded generaton and hxn generators
ramping on-line and off-line according to pre-determined
ramp sequences.
b , ' ~ l , l - ~ 4 . 1 Pzj 1 *
The ramp rate constraint ensures that sufficient ramp rate Solution of the unit subproblem is independent of the terms
capacity is committed to accommodate required generation related to D,, R,, and P,"" because these terms are constant
changes. Any generation changes beyond the required with respect to each unit subproblem.
changes are due strictly to economics of the committed
generators. Again, the actual loading, and therefore unit ramp E . Solution of the Dual Problem
rate enforcement, is accommodated by the economic dispatch.
Rather than seek a rigorous maximum of the dual objective,
D . Lagrangian Dual Objective this paper uses a technique that drives the solution toward a
feasible commitment, and then attempts to improve on the
optimality of that solution. Each solution of (11) is referred
to as a mujor iteration. Each major iteration begins with a
zero vector of Lagrange multipliers p and h initialized using
a priority dispatch. It then proceeds with solution of all unit
subproblems, where each subproblem is
min ~ ~ ' ( h , .p ) (13)
After the solution of all unit subproblems, the Lagrangian
dual constraints are checked for violations. When there are
constraint violations, the Lagrange multipliers p are updated
using a subgradient method in order to enforce the dual
constraints. Once a feasible dual solution has been found,
the commitment schedule is economically dispatched to
update h and produce a generation schedule that is feasible
where & is the Lagrange multiplier for the demand for the primal problem. Then the Lagrange multipliers p are
constraint, systematically reduced until the solution of the dual problem
p,, is the Lagrange multiplier for the maximum becomes infeasible. While decreasing the multiplim, an
capacity constraint, economic dispatch is performed to update h whenever the
R, is the Lagrange multiplier for the minimum commitment schedule changes. These three steps in the
capacity constraint, algorithm, that is, solving (13) for all generators, computation
p3, is the Lagrange multiplier for the reserve of the subgradients, and update of the Lagrange multiplien p,
consmint, are referred to as a minor iteration.
p, is the Lagrange multiplier for the ramp rate
constraint. Each step in the solution of (11)is descn'bedin the sections
below.
It is well known that the solution of the Lagrangian dual is
a lower bound on the solution of the primal objective. IV.UNITSUBPROBLEM
Thmfore, the minimum of the primal objective is found by
maximizing the dual objective. Solution of the dual problem The optimum generation pI* is found by evaluating L: at
is stated as: the megawatt breakpoints of the piece-wise quadratic cost
max( min L(k,p) 1 . (10) curves and at the point where

Because (9) is separable by generator unit, (10) can be

written in terms of the solution of NG generator unit
subproblems as shown below. The optimal commitment schedule which minimizes L,' is
NC
found using a forward dynamic programming technique. The
solution space for the dynamic Programming subproblem is
represented by the predetermined feasible states and the
where associated feasible transitions h m each state in hour t to the
states in hour t + l . Fig. 1shows a graphical representation of
feasible state transitions.
1080

UP1 A. Computation of Subgradients

Subgradientsfor each constraint are formulated to drive the

RU 1 dual problem to a solution that can be made primal feasible
by an economic dispatch of the committed units. The
2
rn
BF2 technique for computing each subgradient is outlined below.

BFl 1) maximum capacity subgradient: "%e subgradient for the

maximum capacity constraint in hour t is
DN2

DN1
where R,4'S is the dispatchable reserve nqubment in hour t.
Hour0 1 2 3 4 5 6 7 The dispatchable reserve requirement is the reserve that must
Fig. 1. Example of feasible state transitions. be satisfied by the economic dispatch of the feasl'ble
UP dc!4Iotcrh e pp &ate RU den& tho ramp up state commitment and is computed by
BF dmoter the boiler fm-up state DN dmc4ca the down state
i-1
Physical characteristics of unit start-ups are modeled by
boiler fire-up states, such that the longer the boiler has been where is the high economic limit for dispatchable units
shutdown, the longer it takes to bring the boiler to the and the actual generation for baseloaded units.
minimum heat required for generation. During operation in
a boiler fire-up state, the unit is off-line but consuming fuel. The maximum capacity contribution for a unit is dependent
The boiler fireup function is madeled as an hourly step on the unit operating state. For off-line units the value is
function, allowing direct use of actual operations data. zero. For baseloaded units (including units ramping up to and
down from their low economic limits), the capacity
Nonlinear start-up ramp sequences (ramping from zero contribution is the actual generation. For dispatchable units
output to low economic limit) and shutdown ramp sequences Operating in the dispatchable range the value must be rate
(ramping from low economic limit to zero output) are limited by the normal ramp rate. The rate limiting is
modeled using states with one way transitions. Likewise, the accomplished using a forward and backward capacity wedge
minimum up time and minimum down time are modeled by placed where the unit enters the up state and leaves the up
states. The number of down states must be at least equal to state. The magnitude of the wedge slope is the normal unit
the maximum number of boiler &-up states. In the response rate. The rate limiting is also necessary for units in
algorithm implemented here, units are ramped off-line from the up state at the study boundary hour.
low economic limit using a constant ramp rate.
It is important to note the difference between a unit's
The table of feasible states is constructed using a set of reserve response rate and its normal response rate. This
rules for each unit which give the feasible transitions from difference is accounted for in (16) by use of the rate limited
each possible state. These rules may vary from unit to unit high economic limit.
but m constant fmm hour to hour, with one exception. If a
unit initially has been in the up state longer than the minimum An example of rate limiting the unit maximum capacity
up time, a transition out of the up state is not atlowed for the contn'bution is shown in Fig. 2. 'Ihe rate limited maximum
number of hours it takes the unit to reach its low economic capacity contribution is given by the upper limit of the
limit from the initial generation. This exception occurs only dispatchable region while the unit is in the up state.
at the boundary hour.
2) Minimum Capacity Subgradient: The subgradient for the
minimum capacity constraint is
NG
The solution of (11) starts from a zero vector of Lagrange gz(
= -D, + Pi" .
multipliers, excluding lambda, which is initialized with a i-1

priority dispatch. The multipliers are then increased using a For off-line units the minimum capacity contribution is
subgradient technique until the solution of all subproblems zero. For baseloaded units (including units ramping up to and
results in a feasible dual solution. The multipliers are then down h m their low economic limit), the contribution is the
decreased in an effort to improve the solution. The actual generation. Again, for dispatchable units operating in
techniques for updating the multipliers are described below.
 1081

HEL C . Update of System Lambda

A feasible dual solution is a feasible commitment schedule

for the primal problem. However, the generation dispatch of
that solution is not necessarily a feasible solution of the
prirnal problem, because the Lagrange multiplier 2. is not
updated and the demand constraint is not n e c d y satisfied.
Lambda can be updated and the &mind constraint satisfied
by economically dispatching the committed generators. The
economic dispatch used must have reserve and unit response
Hour 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 rate constraint functions. Also, the ecoIlomic limits used in
State U U U U U U U U U D U U U U U the economic dispatch must be dynamically modified to
reflect the rate limited capacity limits in each study hour, as
Fig. 2. Feasible dispatch region with cspabiiily wedge. depicted in Fig. 2.
HEL ia the high economic limit LEL is the low economic limit
Uistheupstate D is the down state
Although the ramp rate constraint (8) provides sufficient
ramp capacity to satisfy the hour to hour required generation
the up state, the minimum capacity contribution must be rate change, it does not ensure that the economic dispatch wiU be
limited, but only at the boundary of the sNdy. ramp rate feasible. For that reason forward and backward
dispatches are used to satisfy the ramp rate constraints on a
The minimum capacity contribution is rate limited at the unit basis. In any hour that the solution is infeasible, the hour
normal response rate until the unit reaches the low economic is redispatched without ramp rates, and the dispatch continues
limit. An example of the minimum capacity rate limiting is forward, rate limited to the next hour. If a violation occurs in
shown in Fig. 2 and is given by the lower bound on the the forward dispatch, a backward dispatch is performed. The
dispatchable range while the unit is in the up state. hour of the last violation becomes the boundary condition for
the ramp rate imposition. While there is potential for further
31 Reserve Subgradient: The subgradient for the reserve violations, we have experienced no cases using actual
constraint is operating data where the forward / backward dispatch could
NG not resolve the unit ramp rate constraints.

D. Optimizing the feasible solution

4)RampRateSubgradient: Thesubgradientfortheramprate After finding the initial and m g the

constraint is Lagrange multipliem p are systematically decreasedby a fixed
percentage as shown below.
g,, = P,'" -
NO
P2,l * c
1-1
(19) CPj,,Ik+' = rCpj,Ik (21)

where y is a positive tunable constant that is generally on the

B. Searching for a Feasible Solution order of 0.95.

If the solution of the subproblems in an me are using the upaated

infeasible solution, the Lagrange multipliers multipfiem, and the subgradients are computed to Check for
updated
using the subgradient method shown below. constraint violations. This process is repeated until the
reduction of the multipliers results in an infeasible solution at
[ U j , l f + l = Cu,,lL + 0ljgj, (20)
which time the solution process is stopped. If the solution of
where is a positive %ding fac. for each the unit subproblems r e d 6 in a new feasible Codtment
multiplier that is generally on the order of 0.1. schedule, an economic dispatch is performed to update A,.

The unit subproblems are solved using the updated

multipliers. If constraint violations remain, the multipliers are VI. RESULTS
again updated and the process repeated until a feasible dual
solution is found. m e Lagrange multiplier x remains constant me relaxation dgdw has
during the search for an initial feasible dual solution. been implemented at the IES Utilities System Control Center
located in Cedar Rapids, Iowa. It underwent extensive testing
before and during that installation.
1082
Table II
For testing purposes, a test database was devised to exercise 168 Hour Study Reaults
the function. Detailed unit results are shown for a 24 hour
test case along with summarized results of larger cases ___---Minor Itdons--
showing the ability of the algorithm to quickly find a feasible No.Units Incrcascp -p
solution. 12 3 2
17 2 2
22 1 2
The detailed test case consists of 9 units and a study period 26 2 2
of 24 hours. Unit data for the case is given in Tables A.1, 35 1 2
A.& and A.m. AU units shown are cycling (available for
commitment). Dispatch results for the detailed case are
shown in Table I. The load is given by the sum of the Table m
generation and represents an actual load profile with a peak 72 Hour Transaction Evaluation R e d
to valley ratio of 1.6. The system reserve requirement is 10
M W and the resme t h e frame is 10 minutes. Evaluated WdI Qodc
No.Units Trau8acti01~ UCSolutim The(%)
12 5 5 47
The ability of the algorithm to find a feasible solution for 17 5 5 61
a large scale study case is shown in Table II. The data for 22 5 5 71
this case is the actual IES generating unit data and load
profile. The data includes multi segment (7-9 segments)
incremental heat rate curves. The overaU timing performance of the algorithm is shown
in Table III. In these examples the algorithm is evaluahg
Table I
Generating Unit Output (MW)
five interchange transactions for a 72 hour study period. Each
transaction evaluation requires a complete unit commitment
___---____-_________-----
Generating Unit----- solution. The wall clock times shown m for all five unit
______________-_---I-------

HowA B C D E P G H I commitment studies and include the time for the transaction
0 0.0 0.0 17.4 52.0 0.0 109.3 530.0 0.0 321.6 evaluation. These timing studies were executed on an IBM
1 0.0 O.O? 20.1 70.0 0.0 91.3 530.0 5.0T 321.6
2 0.0 0 . 6 120 52.0 0.0 73.3 530.0 5 . 6 301.7 FOWERserver 580 workstation in a multi-user development
3 0.0 O.O? 12.0 50.0 0.0 70.0 530.0 lO.O? 254.0 environment.
4 0.0 0.d 12.0 50.0 0.0 70.0 528.2 41.8? 194.0
5 0.0 O.O? 120 50.0 0.0 70.0 530.0 4 2 . 6 176.0
W.CONCLUSIONS
6 0.OT 0.d 12.0 50.a 0.0 70.0 530.0 42.0 185.0
7 23.0 0.d 12.0 50.0 0.0 79.2 523.7 44.7 175.4
This paper has shown the development of a Lagrangian
8 23.0 0.d 120 50.0 0.0 97.2 521.8 66.5 235.4
9 23.0 0.d 120 50.0 0.0 115.2 530.0 88.4 295.4 relaxation unit commitment with ramp rate constraints. The
10 23.0 0.d 17.5 64.4 0.0 133.2 530.0 110.2 321.6 Lagrangian dual objective is formulated with conslraints
which drive the solution quickly to a feasible commitment
11 28.9 O.O? 20.6 77.7 0.0 151.2 530.0 132.1 321.6 schedule. The generation of the suboptimal commitment is
12 41.5 O.O? 29.4 78.0 0.0 169.2 530.0 144.2 321.6
78.0 176.5 530.0 144.2 321.6
made feasible by economically dispatching the committed
13 43.7 9 . 6 30.0 0.0
14 41.3 31.0? 29.4 78.0 0.0 166.6 530.0 144.2 321.6 generators. This technique has been implemented at the IFS
15 41.3 46.6 29.4 78.0 0.0 166.6 530.0 144.2 321.6 dispatch center. It provides an efficient capability for the
solution of multiple commitment studies requiredfor real-time
16 41.6 5 5 . 6 9.4 78.0 0.0 175.2 530.0 144.2 321.6 Economy B evaluations.
17 48.0 6 0 . d 30.0 78.0 0.0 193.2 530.0 144.2 321.6
18 44.2 9.0 30.0 78.0 0.0 178.0 530.0 144.2 321.6
19 28.7 120.0 20.5 77.6 0.0 160.0 530.0 131.7 321.6
20 23.0 126.5 17.5 64.4 0.0 1420 530.0 109.9 321.6 m.REpwwcEs
21 23.0 125.4 17.1 57.0 0.0 124.0 530.0 106.9 321.6 [l] CK Pang, Shcble. and F. Albuysb, "Evaluation d Dynamic
22 23.0 127.5 17.9 66.4 0.0 106.0 530.0 1126 321.6 hgmmming Based Methods and Mddple Area Rcpwmtdm f a
23 23.0 108.6 120 50.0 0.0 88.0 530.0 90.8 321.6 T h e d Unit Commitment," IEEE Tmnr. on Powrr Appamiur and
24 23.0 78.6 12.0 50.0 0.0 70.0 530.0 68.9 280.4 Systerrp, PAS-100, NO.3, March 1981. p ~ 1212-1217.
.

How 0 is the boundary b u r and ? indicates the boiler fire-upstate when the [2] Merlin, A.. and Sandrin, P.. "A New Method for Unit Gmnnitmmt at
output ie zcro, or a unit in the ramp up state when the output is non-zero. Eledricitede Francs," IEEE Tram. on Paver Appamnu and Systems,"
VoL PAS 102, NO. 5, khy 1983, p ~ 1218-1225.
.
 1083
Niev4 R Inda, A., and Guillen, I., "LagrangianReduction of Search- Table A.II
Range for Large Scale Unit Commitment,"IEEE Tram. on Power Boiler pin-Up Step hctims for Test Case
Systems, VoL PWRS 2, No. 2, May 1987, pp. 465-473.

ZhaMg, F. and Galinn4 F.D., "Towards a Mom Rigomus and Practical

Unit Commitment by Lagrangian Relaxation Based Unit Commitment
Pmhlem,"PICA Conference Proceedings, Montreal, 1987,pp. 385-393. A 2 1 31
24 10 192
Muckstadt. John A., rmd Kcenig, Shcny A., "An Apj&ation of
Lagrangian Relaxation to Scheduling in Power Generation Systems," B 2 2 62
Opendzotu Ruearch, VaL 25. No. 3. May-June 1977. 24 12 225
Bard, J.F.. "Short-Tern Scheduling of Thermal-Electric Generators
using Lagraugian Relaxation," Opratiom Research, VoL 36, No.5. Table A.IIl
Septanber-october 1988. Non-LinearRamp Up S e q n s n ~ efor
~ Test h e

Vim&. S.. Adrian. E.. lmhof. K. and Mukherjeq S.. " Implementation Rate Rate
of a Lagrangian Relaxation Based Unit Commitment Problem," IEEE unit Hoar W/min) Unit Hoar (MW/mia)
Trans. on Power Systems, Vol. 4, No. 4. octobcr 1989, pp. 1373-1379.
B 1 0.15 F 1 0.25
2 0.367 2 0.0
APPENDIX 3 0.25 3 0.917
4 0.15
5 0.053
Table A.1
Generating Unit Operating Parameters for Test Case C 1 0.133 H 1 0.083
2 0.0 2 0.0
LEL Low Eamomic Limit 0 VOM Variable O&M (!$/MW) 3 0.0 3 . 0.083
HE!L High Economic Limit 0a cost Curve Coeff. (S m 4 0.067 4 0.583
HOL High opersting Limit b Cost Curve Coeff. (!$/MW)
H a High capebility Limit 0 c Cost Curve Coefficient (S) D 1 0.283
MUT Min Up Time (Hours) &-Up h e 1 Cost ( $ w T u ) 2 0.067
MDT Min Down Time (Houra) ISH InitialStateHoura 3 0.033
S H R shutdown Rate (MW/miIl) FSC Fixed Startup Cost (10%) 4 0.580
PZ b p o n s e Rate (MwIHour) p1 Emerg. Req.Rate Whin)
_-___---__c---------__ -Gmeratorpanund+nr--------------------------
BIOGRAPHIES
A B C D E P G H I
LEL 23.0 60.0 12.0 50.0 25.0 70.0 200.0 42.0 86.4 WUUm L. Peteraon (M' 1982) received his B.S.m Electrical Engineering
HE!L 48.0 128.0 30.0 78.0 58.0 207.0 530.0 144.2 321.6 in 1982 and B.S. m Applied P ~ Y Bm ~ C1987
S from the Ge~rgiaInstitute of
HOL 49.5 130.0 32.5 77.0 58.6 207.0 545.0 144.2 323.5
HCL 50.0 135.0 33.0 Technology. He M e m o r e received a M.S. d e p in Electrical
83.0 58.9 207.0 550.0 144.2 324.0 Engineering from C l e m ~ University
~n in 1988. Pram 1982 to 1986 he was
MUT 10 12 employed at C e r o l i ~Power & Light company in System Operations m
12 12 1 48 48 48 24 snpport of the Energy Cmtrol center. Prom 1989 to 1990 he wan employed
MDT 2 2 8 48 1 4 8 48 4 8 2 4
at Fenanti Intedonal Controls. H o ~ t o a ,m the Power Applications
Softwan department,where he worked on network analysis frmctione. Since
Pa 18. 30. 18. 18 24. 18. 199. 21.6 60
0.4 1991 he has been employed at Stagg Systtms. Honston, in the Applications
P1 0.7 0.4 0.4 0.5 0.4 4.0 0.4 2.0 Development department, where he bas worked m ~Chednleanalysir and
FSC 0.3 0.5 0.3 0 0.2 10.7 20.0 10.0 6.0 netwodc analysis functions. Mr.Peterson is a registeredprofeadoaal engineer
VOM 1.7 in North C a r o h and Gemgin.
1.5 0.1 1.7 1.1 1.0 2.3 3.0 0.7
ppc 3.89 3.89 - - - Steven R. Brmmer (M' 1974) received the BSEE ('74) and MSEE (7' 5)
a .039 .008 .065 .008 .062 .0005 ,011 .001 014 iiwn the University of Texas at Arlington. He received the PhD. degree.
- ..
b 9.504 7.02 10.52 9.105 from UTA m 1978. where he m s d e d load modeling at the Energy
48.19 5.746 5.683 7.813 8.95
Syatans R u d C-. He has been with Stagg Systam aiace 1978,
c 43.33 129.6 32.2 74.3 885.2 448.5 0.0 116.2 480.8 whem he has worked on a number of projects and in pmmtly semiug
ISH Applications DevdopmentManagerat the Houston Devclopmsnt Labonrtory.
9 99 99 99 99 99 99 99 99
SHR 99 99 99 99 99 99 99 99 99 His specialty is energy management ~ p t e mapplications,mduding power
~ l e network analyiia, and simulation hctim. Dr.
cantml. ~ c h ~ danalysis,
Brammer is a registered pfeiaonal engineer in the State of Texas.
 1084

DISCUSSION conditions. Generally, these are the conditions that are desired
to be replicated in a unit commitment study.
Alva J. Svoboda and Raymond B. Johnson, (Pacific Gas and
Electric, San Francisco, CA): The author's representation of With respect to the comment on the heuristic nahm of the
system ramping constraints is an very tempting one: it attaches a algorithm, the actual ramp rate procesSing is decomposed into
multiplier value (a "price") to having enough ramping capability
two processes. The Grst commits sufficient units to satisfy the
committed to be able to dispatch the system flexibly subject to
required energy balance change. This: requirement is
ramping constraints. It promises to be more computationally
efficient than other common approaches such as dynamic accommodatedas arelaxed constraint similar to other typically
program state space discretization of generation levels but it is a relaxed constraints in the Lagrangian Relaxation problem and
suboptimal approach. Have the authors compared their technique is not considered a heuristic characteristic of the algorithm.
to that of associating distinct Lagrange multipliers with each The second process is to dispatch the committed generation,
resource's ramping constraints between each subperiod? satisfying the famp rate constraints with a forward and
The benefits of the authors' representation would presumably be backward dispatch. The forward and backward dispatch
even more substantial as the number of resources with significant technique is a heuristic aspect of the algorithm. However, this
ramping constraints increased and/or as the length of scheduling apparent limitation must be put into perspective with other
subperiods was reduced. We have not been able to determine, Lagrangian Relaxation techniques. Even noncapacity based
however, whether solution suboptimality or dispatch infeasibility formulations of the Lagrangian Relaxation problem do not
would also be more likely as ramping constraints came to
guarantee that the solution will converge to a feasible primal
, dominate the complexity of the scheduling problem. We would
solution when the dual solution is found. These techniques
therefore like to know whether this paper demonstrates an elegant
ensemble of heuristics applied to one particular type of system
also reportedly use heuristic dispatch methods to obtain primal
configuration and problem formulation, or provides methods with feasible solutions [Al.
more general application.
We have not tested the algorithm against techniques that
Manuscript received August 22, 1994. attempt to completely resolve the ramp rate constraint as a
relaxed resource constraint. We consider these techniques
more rigorous formulations of the ramp rate problem.
However, we have tested the algorithm against a prototype
algorithm that continuously updates the lambda multiplier.
With or without ramp rate constraints, we recognize that these
techniques produce smaller duality gaps. In general, this
W. L. Peterson, S. R. Brammer: The authors thank the occws because the lambda multiplier is continuously updated,
discussers for their interest in this paper. The algorithm while in our approach the multiplier is held constant until a
presented was developed as a generalized, practical new commitment is found. We have found that the duality gap
implementation of ramp rate constraints in the Lagrangian improvement is not necessarily due to a bansposition of units
Relaxation problem and has been in field use at a specific site as one might surmise. Instead the impvement in optimality
for about one year. In this implementation we have found that usually is due to the shut down of units at the end of the study.
the majority of the units are not ramp-rate-limited within the 1 Even though this produces an apparently better suboptimal
hour steps used. However, we have tested the algorithm with solution, it is not necessarily a more practical or useable
30 ramp-ratelimited cycling units and observed no problems solution.
in finding a suboptimal feasible solution.
References:
Certainly there are pathogenic cases that are difficult to make [A] X. Gw, P.B. Luh, H. Yan, and J.A. - , "Short-
ramp rate feasible. These cases do not necessarily involve Term Scheduling of Thermal Power Systems",
numerous ramp-rate-constrained units, and may in fact involve Proceedings of Seventeenth PICA Conference,
only a few units with skewed unit operating parameters and Baltimore, MD, May 1991, p ~ 120-126.
.
initial conditions. In actual operations ramp rate feasibility is
readily maintained under normal generation and load Manuscript received October 26, 1994.