4

ECONOMIC DISPATCH, UNIT

COMMITMENT, AND OPTIMAL POWER
FLOW AS AUCTIONS
Once understood, it is obvious. Why have I missed the obvious?

4.1 Economic Dispatch

The Economic Dispatch algorithm is the most used optimization for real-time and
for planning. Economic Dispatch is investigated to determine if any comparison
could be made with the auction problem by a more appropriate choice of algorithm.
The algorithms which we have evaluated for the Economic Dispatch problem are
presented in order of complexity [Sheble, 1986; Wood, 1996].

The general formulation includes transmission losses modeled as a set of penalty

factors.

N
Minimize PI==L Yn(uJ
n=!

L
N
Subj ect to unPFn == BLD + LOSS
n=!
Where:

(1) Unit cost function:

Yn == inenFn (uJ
where: Yn = production cost
un = production power
in = fuel cost
en = efficiency
Fn = heat rate curve

G. B. Sheblé, Computational Auction Mechanisms for Restructured Power

Industry Operation © Kluwer Academic Publishers 1999
108 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

(2) Unit capacity limits:

Un:?un:?U n

(3) Network Loss Model:

PFn = penalty factor

(4) Power system model: BLD + LOSS

Where: BLD = the base load demand (system load), and LOSS = the transmission
loss.
The unit limitations apply to all units n-I, ... ,N.

The unit models for each the input output conversion curve from generation level
to cost (I/O curve) has not been standardized at this point. Instead, many different
curves are in use as shown in Table 4-1. However, the piecewise linear curve
model is very popular.

Table 4-1. Energy Conversion Curve Models

Input/Output Incremental Heat Rate

Curve Function Curve Function
(Fn{u n)) Un (un))

(I) Polynomial Curve

(2) Cubic Curve/Quadratic Curve

Fn{un)=anu/ +bnun2 +CnUn +dn

In (un) = 3anun2 + 2bnu n + Cn

(3) Quadratic CurvelLinear Curve

Fn{u n)=anUn2 +bnun +c n

In (Un) = 2anun + bn

(4) Linear Curve/Constant

Fn{un) = anUn + bn
In {uJ= an
Computational Auction Mechanisms for Restructured ... 109

The penalty factors are computed from on-line measurements in many energy
control centers. However, this calculation has not been standardized. We have
assumed that the penalty factors are stored for a variety of network conditions and
operating conditions. The format we have assumed is shown in Tables 4-2.

Table 4-2. Penalty Factor Matrix

Network Topology 1 All lines in service
Generation Range (MW) 500-600 600-760 760-925 925-1225
Interchange -225,-125 PFIV #1 PFIV#2 PFIV #3 PFIV #4
-125,-50 PFIV #5 PFIV #6 PFIV #7 PFIV #8
Range -50,50 PFIV #9 PFIV #10 PFIV #11 PFIV #12
50,125 PFIV#13 PFIV #14 PFIV #15 PFIV #16
(MW) 125,176 PFIV #17 PFIV #18 PFIV #19 PFIV #20

Network Topology 2 765 kV line 1024-E out of service

Generation Range (MW) 500-600 600-760 760-925 925-1225
Interchange -225,-125 PFIV #1 PFIV #2 PFIV #3 PFIV #4
-125,-50 PFIV #5 PFIV#6 PFIV #7 PFIV #8
Range -50,50 PFIV #9 PFIV #10 PFIV #11 PFIV #12
50,125 PFIV #13 PFIV #14 PFIV #15 PFIV #16
(MW) 125,176 PFIV #17 PFIV #18 PFIV #19 PFIV #20

Network Topology 3 345 kV line 1036-N out of service

Generation Range (MW) 500-600 600-760 760-925 925-1225
Interchange -225,-125 PFIV #1 PFIV #2 PFIV #3 PFIV#4
-125,-50 PFIV #5 PFIV#6 PFIV#7 PFIV#8
Range -50,50 PFIV#9 PFIV #10 PFIV #11 PFIV #12
50,125 PFIV #13 PFIV #14 PFIV #15 PFIV #16
(MW) 125,176 PFIV #17 PFIV #18 PFIV #19 PFIV #20

Note: The penalty factor information vector (PFIV) contains all of the penalty
factors.

The basic optimization formulation is to include one of the curves from Table 4-
1 into the overall set of relationships. The objective function is augmented with the
constraint, the first derivatives are found and set equal to zero. Mathematically, this
satisfies the necessary condition but not the sufficient condition.

The second-order derivatives would have to be taken to satisfy the sufficient

condition. Most techniques rely only on the necessary condition to find the optimal
solution. If the energy conversion curve has favorable characteristics (as defined
below), then the sufficient condition may be satisfied simultaneously with the
necessary condition. The inclusion of system losses simply requires the addition of
the penalty factors. The penalty factor (PF), the incremental transmission loss
factor (ITL), and the network sensitivity factor are all related. Any technique,
which produces one of the three, may be used. This research assumes that the loss
model parameters (PF, ITL, or NS) are available and included as part of the input
data.
110 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

The type of algorithms considered are based upon directed search techniques.
Directed search techniques use information available from the objective function to
improve the solution either without forcing the solution to be an infeasible solution
along the way or by forcing the solution to become feasible.

The first algorithm considered is the steepest descent gradient search. This
algorithm starts from a known feasible solution and experimentally searches for the
best direction to improve each variable (unit generation). Some of the more natural
solutions to start from include initial conditions, the optimal solution for the
previous stage (hour), or the optimal solution for the previous combination. Any of
the above energy conversion curve models may be used as long as the first
derivative exists. The strength of this approach includes simplicity, easy to model
unique unit operating constraints, will converge if the set of energy conversion
curves is monotonically increasing and sufficiently smooth, solution accuracy, and
some sensitivity information. The weakness of this approach includes slow
convergence, inaccuracies of curve representation for non polynomial models, the
models have to be regenerated whenever a unit status changes (e.g., the system loss
model changes, fuel cost changes, fuel type changes) for piecewise linear models.

(1) Augment objective function with constraint

(2) Find all partial derivatives and equate to zero

d1r = dF -,.1(1)=01 n=l,oo.,N

dUn dUn

d1r= -(
-
dA
LN
n=!
Un -BLD ) =0

The next algorithm considered was the first-order gradient method. This
algorithm can be obtained from the Taylor series expansion [Wood, 1996].

t F"(U")-'{t.
(1) Augment objective function with constraint

A(!.N1f= u" -BLD-LOSS)

(2) Find all partial derivatives and equate to zero

d1r
dUn
= dFn _ A *
dUn
(1- dLOSS)
dUn
°
= 1n = 1,00', N
Computational Auction Mechanisms for Restructured ... III

dJr =
d)'
-(fn=1
un - BLD - LOSS) = 0
Note: The loss coefficient is often rewritten into an equivalent form:

dLOSS [1- ITL n ]= 1/ PFn

dun
(3) Find starting point, !!.

(4) Compute gradient, GRADfu.) = L

N

n=1
df
- ei
dUi
(5) Update generation of each unit (n=l, ... ,N)
un = un + s * GRADfu.n)
un min if un < un min
Un = un =U n
un max if un > un max
PTOT = PTOT + un
(6) If PiOT > GTBD then go to Step 5, else go to Step 6.

(7) s=sI2.,gotoStep3.

(8) Update!!. and PTOT.

(9) If ABS(GTBD - PTOT)~ B, then go to Step 10.

Pu - Pu min if dfu < 0

dpu

(10) Calculate maximum step size:

0= min{o(Pu), o(~)}

(11) Update solution:

pi+1
u
= piu + O., p,i+1
1
= p,i1 + 0

(12) If ABS(Oi - Oi+1 )~ 0 3 , then go to Step 12.

(13) it=it+l;
If IT > ITMAX then flag nonconvergence and return, else go to Step 2.

(14) Flag convergence and return.

112 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

This method starts from any feasible solution. The generating units, which will
change the objective function the most, are selected and the unit generations
changed accordingly. This method is obviously very similar to the previous method
except that feasibility is never lost.

Any of the above energy conversion curve models may be used as long as the
first derivative exists. The strength of this approach includes simplicity, easy to
model unique unit operating constraints, will converge if the set of energy
conversion curves is monotonically increasing and sufficiently smooth, solution
accuracy, and some sensitivity information. The weakness of this approach includes
slow convergence, inaccuracies of curve representation for non-polynomial models,
the models have to be regenerated whenever a unit status changes (e.g., the system
loss model changes, fuel cost changes, fuel type changes) for piecewise linear
models.

(1) Find feasible starting point, l!:

(2) Compute gradient, GRAD&!.) = i

n~1 du; ,
df e·

(3) Check for convergence at a stationary point (local optimum or saddle

point)
If GRAD ~;)I < 02 for all i=I, ... ,N

t {~ ("J}
then go to 12.

(4) Sdoct p~
(5) elf., _ _dj,u_ I For all i=l, ... ,N and i <> u
Calculate relative costs: c. = __
, dp; dp;

L
N
(6) Select: ~ = (en)n <> U
n~1

(7) Calculate generation change of each unit (1 and u):

(8) IT = IT + 1; ifIT > ITMAX then go to Step 9, else go to Step 2.

(9) Flag nonconvergence and return.

(10) Flag convergence, save optimum solution, and return.

If the objective function is additively separable, then each function can be

linearized separately by a set of points for each variable [Lasdon, 1983, Neuhauser,
1988]. If the objective functions are linear functions, then this grid linearization
process can be transformed into the Dantzig-Wolfe decomposition algorithm. The
Dantzig and Wolfe decomposition principle [Luenberger, 1984] is considered to be
the most efficient solution algorithm for convex simplex problems of linear
objective functions with linear constraints. The methods to achieve such
Computational Auction Mechanisms for Restructured ... 113

efficiencies are listed in the text by Lasdon [1970]. The approach is to form an
equivalent "master program," which has only a few rows but many columns. The
technique to achieve efficiency is to generate the columns only when needed by the
simplex algorithm.

The main linear programming (LP) application that has also been most
successful is the application of the Dantzig-Wolfe Decomposition to the Economic
Dispatch algorithm. This is due partially to the predominant use of piecewise linear
curves. Another reason is due to the convexity of the energy conversion curves even
though this convexity is normally forced since the raw data used to generate the
curves are normally generated by a step function and even though there is
considerable measurement error. The convexity is often required for the real-time
control algorithms. This algorithm was tested for this research.

All of the energy conversion curve models may be used, but any LP-based
technique will work best with curves that are nearly linear. Linear Programming
methods are used extensively in optimization. Linear Programming based
algorithms are the techniques used for most commercial optimization packages.
The literature abounds with information from special coding techniques to special
solution algorithms for specially structured problems (e.g., network flows). The
strengths of any LP-based algorithms include simplicity, easy to add models for
unique unit operating constraints, convergence is easily detected, convergence to a
global optimum is guaranteed if the set of energy conversion curves is
monotonically increasing (convex), solution speed, solution accuracy if the curves
are nearly linear, and a wealth of sensitivity information.

The weakness of these algorithms is predominantly due to the energy conversion

curve representation. These weaknesses may include slow convergence if the curves
are highly nonlinear, storage requirement for the linearized models, inaccuracies of
curve representation, and need for regeneration whenever a unit status changes (e.g.,
the system loss model changes, fuel cost changes, fuel type changes).

N
Original problem: MINz="'c
L...J n *u n
n=1

L
N
Such that: An *U n =bo
n=1

En *U n =b n =bo

Un ~O
For all n=I, ... ,N
N ~1

Master problem:
114 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

Such that: L
j&IJ
Vj)=1
L (Ij)= I
j&IJ

Ij ~O
For allj &B

And each subproblem:

MIN z; =(c; - PI * A;)*u;
such that: B;*u;=b;
u; ~O
For all i=I, ... N

All of these techniques, and many others, estimate the system incremental cost
(A.). This system incremental cost is called the shadow price in any operations
research text and is related to the price of the commodity under certain situations. It
is the price if the market has reached ideal conditions and perfect competition is
attained. Otherwise, it is only an indicator of the price. This is because the system
incremental cost does not capture any of the fixed costs that are lost after taking the
derivative. Thus, the system A. is an estimate of the price under ideal conditions.
However, the algorithms show a very strong similarity to the auction methods of the
previous chapter. The system A. is adjusted until feasibility occurs. Alternatively,
think of feasibility as an agreement to trade contracts. An auctioneer and a supplier
agree to the price when supply equals demand. Specifically, conservation of energy
is achieved. If both buyer and seller offer bids, then we have an auction as
originally established by the Florida Brokerage system [Wood, 1996].

4.2 Unit Commitment [Dekrajangpetch, 1998, 1988]

The unit commitment problem has been approached by many techniques but only
acceptably solved by two techniques: dynamic programming and LaGrangian
relaxation. The problem of scheduling electric power units is due to the integer
nature of the problem that a unit can either be off-line or on-line. The modeling of
thermal power plants, for accurate scheduling, is complicated.

The modeling used in this work is a simplified model of a thermal power unit.
The unit's minimum and maximum capability will be included.

U"' pmin < 0.' < U',£;nmax

-£; - for i=l...N and t=l...T

The unit's minimum uptime and downtime will be included.

The unit's production cost will be represented by a single quadratic curve.

Computational Auction Mechanisms for Restructured ... 115

P_cost = F(Pi) = aiPi2 + biPi + ci

The unit's startup cost and shutdown cost will be treated as a transition cost.

T_cost = Startup_cost (i) if the unit starts this hour,

or

T_cost = Shutdown_cost (i) if the unit shuts down this hour.

The modeling of the power system used in this work is simply the conservation
of energy.

pidemand =" ru'

N

~;
i=l
i
for t= I...T

Other transmission models could be included inter area and intra area flows,
power flow limitations, transmission security constraints, fuel limitations,
environmental limitations, spinning reserve, ready reserve, individual equipment
flows, etc. Spinning reserve will be discussed in the next chapter.

The easiest manner of showing the mathematical formulation is to use a state

variable, U(i,j), to show the status of a unit, i, at a given hour, j. This variable then
has the following values:

U(i,j) = -I ifunit i is off-line during hour j,

U(i,j) = + 1 if unit i is on-line during hour j.

The resulting objective function for this unit commitment problem is as follows:

L L [F; (~I)+ Startup JU: = F(~I ,U: )

T N
costi,1
1=1 i=1

The constraints are rewritten to include the unit status:

N
"plU
pIload-£..,. I =0 for t= I...T
I I
i=1

ulpmin < pI < Ul Dmax

I I - I - II; for i= I...N and t= 1.. . T

The LaGrangian function from the problem statement, excluding unit limitations, is
as follows:
116 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

that is minimized subject to the unit limitations. Note that the unit limitations are
independent of the output or of the status of other units. The conservation of energy
function is a coupling constraint. Changing one of the unit's status or output
changes one or all of the other units.

The LaGrangian relaxation method temporarily ignores the coupling constraints

and solves the problem as if there was not any coupling. A dual optimization
procedure is used to search for the constrained optimum by maximizing the
LaGrangian over the multipliers while minimizing with respect to all other
variables:

q * {A}= maxq{A}
;!

q{A} = minL(P,U,A)
p/.uf

This is a two-step procedure, one step for each optimization. First, find the value of
1.(t) that makes the objective value as large a number as possible. Second, search
for the minimum value of L by adjusting the output and the status of each unit.
Assume for the second step that the value of A is fixed.

Many methods will update the dual variables. This work uses the simplest
technique to compare the procedure with auctions. Consider the following popular
formula based on steepest ascent techniques:

The step length is heuristically changed by the following rules:

d
If -q{A} > 0, then let a = 0.001.
dA

d
If -q{A} < 0, then let a = 0.05.
dA
Note that the dual variables are updated individually for each hour. The normal
solution procedure is to continue updating the dual variables until the duality gap is
"sufficiently small":

J• - q•
--.....!....-:::;;; &
q

Also note that the updates are the inverse value of those values normally shown
[Wood, 1996] because we do not wish to overestimate any of the dual variables
since the outer loop stops whenever the duality gap fails to improve significantly.
Computational Auction Mechanisms for Restructured ... 117

The generators on during this iteration when the dual variable is higher than
necessary will be paid a higher price than necessary. This is shown in more detail
below.

The LaGrangian is rewritten to eliminate the constant term by frrst ignoring the
hard unit limit constraints:

After expanding the formulation, the second term is a fixed constant, not a
function of generation, and can be dropped:

T N T T N
L=:L :L [F;(p;')+Startupcost;"P; +:L A'P;~ad -:L :L A'p/V;
~ ~ ~ ~ ~
Rearrange the terms to show the sequence of optimizations, one inside the other.

The innermost term is solved for each generating unit independently of all other
units.

T
:L ([p; (p;I)+ Startup cost;,1 ]V; - AI p;IV; }
;=1

The minimum of the LaGrangian is found by minimizing the generation cost for
each unit over the complete time horizon of the study.

N T
min q(A) =:Lmin:L ([F;(p;')+ Startup cost;,1 ]V; - AI p;IV;}
;=1 1=1

The unit limits are used to constrain the solution.

u l pmin < pI < ul pmax for t=l...T

I' -, - I I

Dynamic programming very easily solves the solution since only two states are for
each stage. Now the curse of dimensionality has been inherently removed.
Normally, a stage is an hour of operation, but any defmed time interval can be used.
The initial condition of the unit is often obtained from the real-time status of the
unit. Specifically, is the unit on or off now? At each stage, the minimization is with
respect to generation output only.
118 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

Remember that A is a constant, so set the fIrst derivative to zero.

~[F.(p)-A/
dpt pt]=~F.(pt)_
IdptI
.,1/ = 0
I I I
I I

The solution is trivial.

~F.(p
dPti opt
)=A/I
I

Based on our knowledge of optimization, there are three cases of concern:

If p;opt ::;; p;min , then the output is constrained to the minimum:
minlF; (p;) -)} P;t J= F; (p;min )- At p;min

If p;min : ; p;opt ::;; p;max , then the output is set to the equal incremental cost:

If p;opt ::;; p;max , then the output is constrained to the maximum:

The overall procedure can be shown by the data flows in Figure 4-1. Note that
the process is to pick a set of dual variable values, one for each hour, request
generation bids from each unit, add the bids to determine if the system constraints
are satisfIed, increase the value of the dual variables for those hours when the bids
are not suffIcient, decrease the value of the dual variables for those hours when the
bids are excessive, and repeat the process until all constraints are satisfIed for all
hours. This is a one-sided (generation only) auction.
Computational Auction Mechanisms for Restructured ... 119

Figure 4-1. LRUC Data Flow.

This can be changed to a two-side (double) auction simply by enabling the loads
(buyers) equal bidding rights. The only change to this diagram is that the term
"Gen" for generator has to be replaced with "Co" or some other generic name for
company.

Finally, this would be more of a standard auction if the boxes representing the
companies bidding where removed from a central location and enabled to be solved
anywhere. As an example, the bids could be transmitted from the generation
company's computers as Figure 4-3 shows.

Then each company could decide when to bid generation to maximize profit.
The bidding strategy essentially replaces the dynamic programming solution of the
LaGrangian relaxation process. A more complete description, including multiple
rounds, is shown in Figure 4-4. Note that a mUltiple round market emulates the
multiple iterations of the LaGrangian relaxation solution procedure.

Figure 4-2. Double Auction Unit Commitment.

120 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

Figure 4-3. Decentralized Unit Commitment as Auction.

Figure 4-4. Multiple Round Double Auction.

Give model data to central authority, number of parameters (e.g., 32), Bid is not
disclosed, Central authority matches bids, unit commitment, power flow, security
constrained dispatch, Central authority posts matched bids. Pros, Centralized Data,
One Set of Programs, Looks Like Old Environment, Cons, Audit of Data, Incentive
for New Programs, Correction of Bugs, Responsibility for Equipment Constraints,
One-Shot Bidding, Contingencies, TRANSCOs, etc. Matching of Sealed Bids,
Adjust Matched Bids for Transmission Equipment Usage. Ancillary Services,
Losses, Frequency Regulation, Voltage Support, Stranded costs.

4.2.1 Auction Mechanisms for Electricity

One sided and double sided auctions have been investigated. One sided auctions
allow only the GENCOs to bid. Double sided allow the GENCOs and the ESCOs to
bid.

4.2.1.1 Sequential Sealed-Bid and Sealed-Offer Auctions. This approach of

applying auctions to the pricing of electric power uses the following three-step
process:
Computational Auction Mechanisms for Restructured ... 121

(1) A sealed-bid auction optimally allocates supply and demand independent of the
transmission network.
(2) The supply allocations are adjusted for losses.
(3) A sealed-offer auction optimally selects flows on the transmission system.

Sealed-bid auction: demand and supply. An LP solution to a sealed-bid auction

model is presented. Linear programming is used to compute the optimal exchange
of a commodity. Assume there are m generators and the output (MWh) each has to
sell from bus i is as follows:

Pi > 0 for i = 1, 2, ... , m (4.1)

Generators also specify a reservation price, 'IT:i, below which they will not sell
electric power. Note that buyers do not know the reservation price prior to
submitting their bids.

Frequently sellers specify a reservation price. If the market price equals the
reservation price, some portion of the supply may remain unsold. Thus the seller
"sells the goods to himself' at the reservation price. Reservation prices are an
example of price rigidity and can cause the market formation to fail to reach an
equilibrium in which the market clears and all goods are sold [Thompson, 1992].

Also n demand centers desire the following quantity of power.

cJ.j>O for j = 1, 2, ... , n (4.2)

At the auction center, buyer j submits a sealed bid of

cij ::: 0 (4.3)

for one MW of generator i's power. Let each generator i sell

Xij ::: 0 (4.4)

MWs to buyer).

An LP problem that maximizes total surplus uses the following form [Thompson,
1992]:

m n m
maximize ~ ~ CijXij + ~ 7'CjYj (4.5)
j=1 j=1 j=1
122 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

subject to ~>ij +Y; =P; Vi (4.6)

j=l

(4.7)

(4.8)

(4.9)

where:
Yi = the unsold MWs offered by generator i
!:.i = minimum generation of generator i

The objective function (4.5) maximizes total surplus in the system. Constraint
(4.6) ensures that the sum of units sold and bought back by generator i equals the
amount, Pi, originally offered for sale. Equation (4.7) requires generator i to
produce at a level greater than or equal to its minimum generation level. Equation
(4.8) limits the MW units received by node j to dj. Equation (4.9) requires the
decision (exogenous) variables to be positive.

This linear program must be used in an iterative fashion. After the first iteration,
any generator scheduled to supply at prices below its reservation price should be
removed from the LP. Ifmore than one generator is included in this condition, only
the violating generator with the highest reservation price should be removed. The
LP is iteratively solved until the reservation prices of the generators are not violated.

Once system demand and supply is obtained with the above LP, generation must
be adjusted for losses. First, penalty factors are used to determine system losses.
Second, losses are accounted for by increasing generation of the most economical
generator in operation by the amount of losses.

Sealed-offer auction: transportation

The sealed-bid auction, used to determine the optimal supply and demand of
electricity, yields a fixed inelastic demand for delivery services. Transmission
owners submit sealed-offer schedules and the transmission market is cleared at a
delivery price that equalizes the inelastic demand quantity and the supply of
delivery services.
Computational Auction Mechanisms for Restructured ... 123

The LP used to maximize total surplus takes the fonn of the classical
transshipment model with the addition of capacity constraints. This approach
assumes transmission owners are able to control the power flows on transmission
lines. This can be accomplished with flexible AC transmission system (FACTS)
devices.

The LP takes the following fonn:

Minimize (4.10)

Vi
Subject to (4.11)

V i,j (4.12)

fij ? 0 (4.13)

where: ij = transmission line from node i to node j

fu = flow on line ij
fl} = initial flow on line ij
.f[Jax = maximum flow on line ij
Sj = MW supply at node i; Sj < 0 denotes demand
tOij = transmission offer price per MW flow on line ij

The objective function (4.10) minimizes the cost of transmitting electric power
from the generators to the demand centers. The first constraint (4.11) requires the
flow out of node i to equal Sj. The second constraint (4.12) reflects line capacity.
The third constraint (4.13) requires the decision (exogenous) variable to be positive.

4.2.1.2 Sealed-Bid Offer/Double Auction

This approach is similar to that used by McCabe et al. [McCabe, 1989] in the
natural gas industry. Representatives of generation, transmission, and distribution
submit bids/offers to a regional transmission group (RTG) that acts as the
auctioneer. The RTG is responsible for identitying the equilibrium price and
allocation so that total surplus is maximized.
124 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

Players

Generation. Owners of generation sites submit price/quantity offers for the electric
power they are able to supply. Generation may be from independent power
producers (lPPs), qualifying facilities (QFs), exempt wholesale generators (EWGs),
etc.
Transmission. Transmission owners submit price/quantity offers for the
transmission of electric power. To decide the amount of power they are capable of
transmitting and, thus, the price for their offers, transmission companies use the
following approach to determine the validity of their offers [Sheble, 1994a]:

(1) Solve an AC/DC power flow for the power system (including existing
transaction flows but not proposed offers).

(2) Create constraints to restrict flow on all overloaded lines.

(3) Implement step 2 by the use of distribution factors.

(4) If there is no change in system limitations (with respect to base case power
flow) then quit, else continue.

+ -
(5) To determine optimal L1Pi and L1Pi , solve the following constrained
economic dispatch LP:

Minimize L: (a· L1p.+ - a· L1P~)

I I I I (4.14)
't:/.
I

subject to (4.15)

(4.16)

(4.17)

L atAL1P/ - L1P;) ~ "max - f( VI

't:/ i (4.18)

L a/i(L1P/ - L1P;) ;::: _"max - f( VI

't:/ i (4.19)
Computational Auction Mechanisms for Restructured ... 125

where:
ali = generation shift factor (the sensitivity of the flow on line / to a change in

generation at bus i);

AiL
ali = A.Pi
D = load demand
.tP = base case flow on line /
.trax = maximum flow on line /
i=bus index
]= interchange (out of system)
/ = line index
Pi = maximum generation at bus i
f..i = minimum generation at bus i
PP = base case generation at bus i
PL = losses
p2 = base case losses
+
A.Pi = positive change in generation at bus i
A.Pi- = negative change in generation at bus i
The objective function (4.14) maximizes the shift in generation for the submitted
offer prices (maximized since offers are negative). Equations (4.15) and (4.16)
represent the generation shift limits. Equation (4.17) ensures the conservation of
energy. Equations (4.18) and (4.19) represent the line flow constraints. For an
explanation of these constraints and the LP formulation see [Wood, 1996].

(6) Adjust control variables [generation, tap-changing under-load transformers

(TCUL), quadrature phase-shifting transformers (QPS), high voltage DC
transmission lines (HYDC), flexible AC transmission (FACTS)].
(7) Return to step 1 and repeat.

Transmission and distribution owners require all line flows to be within their
limits. This is done by solving the LP problem with constraints for each overloaded
line. After solving it, the generation shifts should be made and a new base case load
flow run. Any "new" overloaded lines should be included when re-executing the
LP. Classical approaches [Wood, 1996] suggest keeping all line flow constraints
from the fIrst LP as well as the "new" overloads as a safety measure. This process,
called an iterative constraint search, is repeated until the network has no overloads.
126 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

ESCOs and representatives of coops, municipalities, and industry submit

price/quantity bids for the electric power needed to meet their demand.

The ISO implements the following procedure [Debs, 1998]:

(1) Collect bids/offers from all players.

(2) Identify maximum potential exchange between areas by the use of an

optimal power flow.

(3) Obtain distribution factors for entire system by the use of a decoupled
power flow.

(4) Obtain system interchange and precontract generation from all players.

(5) Run constrained economic dispatch LP (4.14) through (4.l9) for the entire
system.

(6) Adjust contracts according to Step 5 changes to the base case.

(7) Report results to players.

The two applications of auctions presented in this chapter approach the pricing of
electric power from different angles. The sequential sealed-bid/sealed-offer
approach considers the supply and demand allocations to be independent of electric
power transmission; however, demand centers may submit bids that reflect expected
transmission prices. The sealed-bid/offer double auction simultaneously allocates
the supply, demand, and transmission of electric power. Based on similar
approaches in the natural gas industry [Post, 1993], the simultaneous double auction
is expected to achieve more efficient results.

4.2.1.3 Results

To illustrate the methods discussed above, a 12-bus, 7-generator system will be

used. Table 1 lists the generation and load data of the system. Table 2 contains the
transmission line data.

Seven different companies may own the seven generators of this power system.
Alternatively, generation companies may own two or more generators at various
geographical locations that are scheduled using economic dispatch. The necessary
condition is that enough generation companies exist so that a competitive
environment is realized.
Computational Auction Mechanisms for Restructured ... 127

Table 4-3. Generation and Load Data.

BUS LOAD MAXIMUM MINIMUM PENALTY

(MW) GENERATION GENERATION FACTOR
(MW) (MW)
1 0.0 0.0 0.0 1.00
2 350.0 400.0 100.0 0.97
3 0.0 400.0 100.0 0.97
4 0.0 400.0 100.0 0.96
5 0.0 600.0 150.0 0.95
6 250.0 400.0 100.0 0.96
7 100.0 0.0 0.0 1.01
8 200.0 0.0 0.0 1.01
9 200.0 0.0 0.0 1.02
10 0.0 0.0 0.0 1.00
11 300.0 200.0 50.0 0.98
12 100.0 200.0 50.0 0.99
TOTAL 1500.0 2600.0 650.0 -

Table 4-4 Transmission Line Data

LINE # FROM TO CAPACITY LENGTH

BUS BUS (MW) (MILES)
1 1 2 300.0 40.0
2 1 4 300.0 120.0
3 I 12 300.0 30.0
4 2 3 300.0 60.0
5 3 11 500.0 90.0
6 3 4 100.0 50.0
7 5 7 300.0 50.0
8 5 6 500.0 90.0
9 6 10 500.0 60.0
10 7 8 300.0 30.0
11 8 12 100.0 60.0
12 9 10 500.0 60.0
13 9 12 300.0 70.0
14 11 12 100.0 27.0
15 4 5 500.0 40.0
(example
2 only)
128 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

Sequential sealed-bid and sealed-offer auctions

The example system is solved by the sequential auction approach of Chapter 3.

This approach forms a basis against which other auction mechanisms, such as
double auctions, can be evaluated.

Sealed-bid auction: demand and supply. The sealed-bid auction was used to
optimally allocate the supply and demand for the system. The following
assumptions were made:

• Generator owners offer all available power, subject only to the reservation
price.
• Each load center submits bids, in units ofRIMW, for each generation bus.

Table 4-5 contains the sealed bids and generation reservation prices, 1Ci, submitted to
the auctioneer.

Table 4-5. Sealed-Bids and Generator Reservation Prices.

LOAD LOAD LOAD LOAD

GEN. BUS BUS BUS BUS
BUS 2 6 7 8
2 10.0 9.4 9.0 9.2
3 9.5 9.1 8.9 9.2
4 9.1 8.9 9.0 9.3
5 9.0 10.6 9.5 9.4
6 8.0 11.0 9.4 9.3
11 9.0 10.0 9.3 9.4
12 9.1 10.3 9.4 9.6

LOAD LOAD LOAD RESERVED

BUS BUS BUS PRICE
9 11 12 1Ci
9.4 9.2 9.3 9.4
9.3 9.4 9.4 9.4
9.4 9.3 9.2 9.4
9.4 9.1 9.2 9.8
9.6 9.0 9.3 9.4
9.6 9.4 9.4 9.3
9.7 9.4 9.4 9.3

The bids of Table 4-5 have been arbitrarily assigned by the author. This would
not be the case if auctions were used for pricing electricity. However, the intent of
this example is to show how an auction might be implemented, not how bids should
Computational Auction Mechanisms for Restructured ... 129

be submitted. Thus, the reservation prices in Table 4-5 were set. Next, the "sealed
bids" were assigned so that some were acceptable and others were not. In this
example, bids are submitted as a single-step process. However, bids (offers) may be
piecewise linear functions that reflect different prices for specified blocks of power.

The LP based on (4.5) through (4.9) was used to determine optimal supply and
demand. However, the first iteration results violate the reservation prices of the
generators at bus 4 and bus 5. Since the reservation price of bus 5 generation is
larger than that of bus 4, the LP was solved a second time without generation at bus
5. This results in a schedule that violates the reservation price of generation at bus
4. After withdrawing the generation at bus 4, the LP was solved a third time and,
this time, no reservation prices were violated. Results for this example are
presented in Table 4-6 and Table 4-7. Table 4-6 indicates that all the load was met
except the demand at bus 7. Table 4-7 contains the supply allocations and shows
which bids, of Table 3, were filled.

Table 4-6. Demand Allocation.

BUS # DESIRED SUPPLIED

DEMAND (MW) DEMAND (MW)
2 350 350
6 250 250
7 100 0
8 200 200
9 200 200
11 300 300
12 100 100
TOTAL 1500 1400

Next, generation was adjusted for losses. Each operating generator's output was
multiplied by its corresponding penalty factor (see Table 4-3). The sum of these
values gave the system losses. These losses presented an additional demand which
were assigned to the generator with the lowest, unmatched high-bid (with
consideration of the reservation price). In this example, generation at bus 3 was
assigned an additional load of 40 MW. This was supplied at a price of 9.4 RlMW.
Thus the total value of sales for the system increased to R 16,106.00, and the total
amount of power generated at bus 3 increased to 290 MW.

Sealed-offer auction: transportation. The loss adjustment discussed in the previous

section results in the supply and demand allocation needed for the transportation
model. These values are listed in Table 4-8. Note that the generation supply values
were adjusted by penalty factors and, in the case of bus 3 generation, adjusted for
the increased supply necessitated by losses. (The generators actually produced what
is shown as "amount sold" in Table 4-7, except for generator 3, which produced 290
MW).
130 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

Table 4-7 Supply Allocation

GENERATION AMOUNT AMOUNT PRICE SOLD TO

AT BUS OFFERED SOLD (R/MW) LOAD
(MW) (MW) CENTER
AT BUS
2 400 350 10.0 2
3 400 150 9.4 11
100 9.4 12
4 400 0 0.0 -
5 600 0 0.0 -
6 400 250 11.0 6
150 9.6 9
11 200 50 9.6 9
150 9.4 11
12 200 200 9.6 8
TOTAL before loss after loss
VALUE OF adjustment: 15,730.00 adjustment: 16,106.00
SALES(R)

Table 4-8 Supply and Demand for Transportation Problem

BUS DEMAND SUPPLY TOTAL SUPPLY

(MW) (MW) (supply- demand)
1 0 0 0
2 350 339.5 -10.5
3 0 282.5 242.5
4 0 0 0
5 0 0 0
6 250 384.0 134.0
7 0 0 0
8 200 0 -200.0
9 200 0 -200.0
10 0 0 0
11 300 196.0 -104.0
12 100 198.0 98.0
TOTAL 1400 1400 0

The transportation problem represented by (4.10) through (4.13) was used to

optimally allocate the transmission of electricity. The "total supply" column of
Table 4-8 represents the supply, Si, of (4.11). For the example power system, initial
flows are zero and maximum line capacities are given in Table 4-4. The cost-per
MW-mile, c, for each transmission line was given by transmission companies A, B,
and C. The costs per MW-mile multiplied by the transmission length gives the
transmission offers, to, shown in Table 4-9 below. Offer prices are assumed to be
Computational Auction Mechanisms for Restructured ... 131

independent of the flow direction on a line. This need not be the case.
Transmission owners may wish to encourage the sale of power in a specific
direction on a line. For example, a heavily loaded line transporting power from
node i to node j may be relieved by submission of a low offer price for transmission
from node j to node i.
Example 1
The transportation problem was solved, and the results are shown in Table 4-10.
With the sequential auction approach, the total cost of supplying and transporting
electricity was R 19,412.90.
Example 2
The example system was modified with the addition of line 15 between bus 4 and
bus 5 (data for this line is included in Table 4-4 and Table 4-9).

Company B's addition ofline 15 to the power system increases its market power.
Company A must now compete with two lines, owned by company B, for the
transportation of generation from buses 2, 3, and 4 to the demand centers.
Company C is also affected by the addition but not as much as Company A. In this
example system, Company B has an advantage because it controls nearly all of the
interchange between the areas served by Company A and Company C.

Company B submitted a low offer price for line 15 in an attempt to encourage

use of their other transmission lines. The transportation problem was solved, and
the results for example 1 and example 2 are shown in Table 4-10.

Company B' s addition of line 15 to the power system caused an increase in

power routed through Company B's lines. The addition changed the flows in
Company C' s transmission lines and reduced the use of Company A's transmission
lines. In addition, the cost of transmission was reduced from R 3,306.90 to R
3,030.42. Therefore, the total cost of supplying and transporting electricity was R
19,136.42, a savings ofR 276.48 over example 1.

Example 2 assumes that the supply and demand allocations would not be
affected by the addition of line 15. However, the bids for generation submitted by
demand centers might reflect the expected change in transportation prices caused by
the addition ofline 15.

4.3 Auction Mechanism with Classical Optimization

The two types of auctions presently implemented can be derived from many
optimization techniques (LaGrangian Relaxation Unit Commitment, Dantzig-Wolfe,
and Pricing Mechanism Decomposition). The connection between hourly auctions
and unit commitment is most clearly visualized by examining the data flow as
shown in Figure 4-4. Note that the price (lambda) is fixed for each period, then
each generator unit is requested to provide the amount of generation available for
that price. This is a non discriminating auction since the same price is paid to all
GENCOs.
132 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

Table 4-9. Transmission Costs.

LINE # COST PER LENGTH to (R/MW)

MW-MILE (MILES)
1 0.05 40.0 2.0
2 0.04 120.0 4.8
3 0.06 30.0 1.8
4 0.03 60.0 1.8
5 0.04 90.0 3.6
6 0.08 50.0 4.0
7 0.06 50.0 3.0
8 0.03 90.0 3.6
9 0.04 60.0 2.4
10 0.05 30.0 1.5
11 0.09 60.0 5.4
12 0.03 60.0 1.8
13 0.05 70.0 3.5
14 0.07 27.0 1.9
15 0.02 40.0 0.8
(ex. 2 only)

Table 4-10. Transmission Results (Example 1).

LINE # FROM BUS TO BUS FLOW

(MW)
I 1 2 -68.0
2 1 4 0.0
3 1 12 68.0
4 2 3 -78.5
5 3 11 204.0
6 3 4 0.0
7 5 7 100.0
8 5 6 -100.0
9 6 10 34.0
10 7 8 100.0
11 8 12 -100.0
12 9 10 -34.0
13 9 12 -166.0
14 11 12 100.0
Total Cost (R): 3,306.90
Computational Auction Mechanisms for Restructured ... l33

Table 4-11 Transmission Results (Examples 1 & 2)

LINE # FROM TO BUS FLOW (MW) FLOW

BUS (example 1) (MW)
(example 2)
1 1 2 -68.0 0.0
2 1 4 0.0 0.0
3 1 12 68.0 0.0
4 2 3 -78.5 -10.5
5 3 11 204.0 172.0
6 3 4 0.0 100.0
7 5 7 100.0 100.0
8 5 6 -100.0 0.0
9 6 10 34.0 l34.0
10 7 8 100.0 100.0
11 8 12 -100.0 -100.0
12 9 10 -34.0 -l34.0
l3 9 12 -166.0 -66.0
14 11 12 100.0 68.0
15 4 5 -- 100.0
Total Cost (R): 3,306.90 3,030.42

Figure 4-6 shows an hourly auction where the competing generating units submit
bids consisting of price and quantity. If a non discriminating auction price were
used, then the results would be the same as for the unit commitment in Figure 4-1.
A discriminating auction would require each unit to receive the bid as the price
amount.

The objective of both the unit commitment and the hourly auction is to find the
optimal solution of operation subject to all operational constraints.

If ESCOs are enabled to bid, then the extensions to the unit commitment
algorithm is to model purchases as negative generation. The same is true for the
hourly auction. As shown in Figure 4-6, the objective of both methods is to find the
optimal solution of the supply demand curves for each hour. The difference
between lambda update techniques can be analyzed with respect to English and
Dutch auctions.

Thus, the first place to identify bidding strategies is to find the best method for
updating the LaGrangian multipliers for the unit commitment problem. However,
each bidder would want to be more conservative since an overbid may destroy all
profits if price discovery occurred before the bid could be corrected. Indeed, many
auctions would not allow an overbid to be corrected.
134 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

Results reported
to market participant

Figure 4-6. Block Diagram Of The Auction Process.

Figure 4-7. LaGrangian Relaxation Unit Commitment.

Instead of detailing the various methods to update the multipliers, which are so
well documented, it is more interesting to determine what the price strategy would
be for each hour if the market is dominated by one of the companies. Indeed, this
has been the case in many actual markets, where one company has achieved
dominance. The essential step is to identify the demand elasticity from previous
market bids made by purchasing companies.
Computational Auction Mechanisms for Restructured ... 135

L...-_~--'n" B 2, ... ,B N
P"P 2 "",P N

for Each Time Period

No

Yes

Figure 4-8. Hourly Auction.

8.5
8
7.5
7

6.5
6
5.5

5+-------~--------~-------r_------~
lst bid 2nd bid 3rd bid 4th bid

Figure 4-9. Auction Market Objective.

4.4 Auction Formulation

Two major types of auctions are single-sided auctions and double-sided auctions.
Either generation companies (GENCOs) or energy services companies (ESCOs)
submit bids to the independent contract administrator (lCA) in single-sided
auctions. GENCOs and ESCOs submit bids to the ICA in double-sided auctions.
The ICA performs auctions with the formulation below. This formulation is for
double-sided auctions.

The supportive services included in this formulation are transmission losses,

spinning reserves, and ready reserves. Transmission losses, spinning reserves, and
136 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

ready reserves are bundled, i.e., traded in the same market as one contract with
energy. Regulation, automatic generation control (AGC), and load following can be
incorporated into the formulation just as the spinning and ready reserves if they are
considered bundled. Regulation, supplemental control (AGC), and load following
are for similar purposes. They differ in time frame. The time frames of regulation,
AGC, and load following are 6 seconds, 30 minutes, and 60 minutes respectively.
The frequency of the auction in the formulation below is every 30 minutes. There
are various possible formulations. Alternative formulations are discussed below.

All of the quantities in the formulation are in per unit, except the bus angles,
which are in radian and the bids' prices which are in $/MWh. The symbols used for
the formulation are shown below. Note that GENCO i is at bus i and ESCO j is at
bus m+j.

Csi price of ith seller's bid for active power

Cbj price ofjth buyer's bid for active power
CSsi price of ith seller's bid for spinning reserve
CSbj price ofjth buyer's bid for spinning reserve
crsi price of ith seller's bid for ready reserve
Crbj price ofjth buyer's bid for ready reserve
L1Psi accepted amount of active power of ith seller
L1Pbj accepted amount of active power ofjth buyer
L1Pmsi amount of active power of all other contracts of
ith seller
amount of active power of all other contracts of
jth buyer
LlQsi change in reactive power of ith seller
LlQbj change in reactive power ofjth buyer
Ssi accepted amount of spinning reserve for ith seller
Sbj accepted amount of spinning reserve for jth buyer
Ssi max maximum limit of spinning reserve for ith seller
Sbt ax maximum limit of spinning reserve for jth buyer
ks s required ratio of spinning reserve for sellers
ks b required ratio of spinning reserve for buyers
Rsi accepted amount of ready reserve for ith seller
Rbj accepted amount of ready reserve for jth buyer
Rstax maximum limit of ready reserve for ith seller
Rbtax maximum limit of ready reserve for jth buyer
RRmax maximum ramp-rate limit
krs required ratio of ready reserve for sellers
krb required ratio of ready reserve for buyers
n number of buyers
m number of sellers
LlOi change in ith bus angle
Ll8"ax maximum limit of change in bus angle
B' matrix of negative bus susceptance matrix
Bsi amount of active power submitted by ith seller
Bbj amount of active power submitted by jth buyer
Computational Auction Mechanisms for Restructured ... 137

p/ original active line flow between buses i and}

LJPij change in active line flow between buses i and}
p.max active flow limit of line between buses i and}
lj
trci transmission cost coefficient for bus i
Kp coefficient for active power flow constraints
Kq coefficient for reactive power flow constraints
PFi power factor of ith bus
PF"'in minimum power factor allowed for each bus

n In II

Maximize LCb)M},) - LCS~i + LCSb)Sb)

)=1 i=1 )=1
m n m
- LCss,ss; + LethA,) - LCr,;R,; (4.20)
i=1 )=1 i=1
m+n
- Ltrqi1b;
;=1

The objective function is composed of surplus in trading power, surplus in

trading spinning reserve, surplus in trading ready reserve, and the negative of the
transmission cost. The surplus is defined as the difference between the buyers'
revenue and the sellers' cost. Transmission costs (trc) are converted from
transmission costs of lines into transmission costs at buses. The negative of the
transmission cost indicates that the transmission cost is minimized. The objective
function (4-20) is maximized subject to the following constraints:

A. Active Power Constraints with Losses. The active power flow constraints and
conservation of active power equation are manipulated to be constraint (4-21). Note
that Kp is calculated from the real power loss coefficients and B I matrix.

m n
LKPiM'.i + LKPm+jMbj = 0 (4.21 )
i=1 j=1

An iterative procedure is normally used.

B. Reactive Power Constraints with Losses. The reactive power flow constraints
and conservation of reactive power equation are manipulated to be constraint (4-23).
Note that Kq is calculated from the reactive power loss coefficients and B I matrix.
An iterative procedure is normally used.
m n
LKqi1Q.;+ LK£k)i1Q,) =0 (4.22)
;=1 )=1
138 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

C. Transmission Line Flow Limits. An iterative procedure is used to limit the

complex flow by limiting the real and reactive flows. Equation (4-23) shows the
flow limit. The change in flow, t1Pij is approximated as follows.

Ip: + Mij I $; Pijrnax

(4.23)
Mij = -B~(t:..Oi -t:..Oj )

Similar constraints are used for reactive flows.

D. Sellers and Buyers' Bid Amount. The bid offered can be used partially or
completely.

o $; ~i $; Bsi (4.24)
o$; Mbj $; Bbj
E. Spinning Reserve Obligations. Spinning reserve obligations of sellers and buyers
are shown and conservation of spinning reserve follows.

Ssi = min~Psi'~ax)
Sbj = min~bM;,l'S;''') (4.25)
m n
~)Si - LSbj = 0
i=l j=l

F. Ready Reserve Obligations. Ready reserve obligations of sellers and buyers are
shown and conservation of ready reserve follows.

R.i = min(a;LlP,i,~j
~j = min(aj,M;,l'R;t) (4.26)
m n

LR.i- L~j=O
i=l j=l

G. Ramp Rate. Equation (4.27) represents ramp-rate constraints for sellers and
buyers. Up-ramp-rate limit is assumed to be equal to down-ramp-rate limit. Note
that t-J indicates the former period before auction.

1(LlP,i + M1'fIs;) -(&;i-I + Mni~1 ~ ~ O.5RR",x

(4.27)
1(M;,j + M"I,) -(Mt-1 + MnJ,jl ~ ~ O.5RR"'x
H Bus Angle Limit. Simply limits the angle separation between any two buses.
Computational Auction Mechanisms for Restructured ... 139

(4.28)

Note that stability margins may be represented with more sophisticated models.

I Relation between Active and Reactive Power. Required power factor correction
may be included.

PF>ppnin
,- (4.29)

Note that all variables are nonnegative except all Llbi are free variables. Constraints
(4.23) to (4.29) are for all i from 1 to m and all} from 1 to n.

4.4.1 Handling Constraints

Some constraints are modified so that they can be solved with LP. Some constraints
that require modification are as follows.

A. Absolute Constraints. Many constraints are in absolute form. The general form
of an absolute constraint is shown in (4-30) and it is modified to two constraints.
Note that F(P) indicates a function of P and U indicates a constant.

iF(J1 :s;U
(4.30)
-U:S;F(P):S;U

B. Reserve Constraints. Reserve constraints can be generalized as (4-31). Sis

forced to be either kLIP or !!"ax. This concept can be explained according to Figure
4-10. S is equal to a constant ratio, k, of LIP until S is equal to !!"ax. In other words,
LIP is equal to !!"ax/k. After this, S is equal to !!"ax no matter how much LIP is. Such
modifications can be implemented by LP.

S =minW,S"'ax)
S=W (4.31)
S :S;S"',x
140 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

Figure 4.10. Reserve Constraint.

c. Non-Negativity Constraints. Free variables, LI~are changed.

(4.32)

4.4.2 Structure Options

The ICA must know all transactions each player has to operate the system
economically and securely. This makes the net power at each bus a known quantity
in the formulation. Thus, the ICA should supervise the implementation of every
contract between parties by incorporating all of them as preconditions into the
auction formulation. This will be effective only if the ICA knows all transactions
and is enabled to enforce all transactions with sufficient penalties.

4.4.2.1 Single-sided auctions vs. double-sided auctions. Both sellers and buyers
can specify the prices and amounts of bids as they desire for double-sided auctions.
They can adjust the bids to maximize their own profits. For one-sided auctions,
such as seller bidding only, only sellers submit price and quantity bids to the auction
and buyers specify only the amount of power desired. Sellers might have to sell
cheap power because all sellers submitted low price bids depends on the pricing
method. This problem can be lessened if a seller can specify a reservation price.
However, this is inflexible to buyers. This is true for all one-sided auctions, since
one of the two sides does not submit a price. These arguments indicate the
inefficiency of one-sided auctions.

4.4.2.2 Bundled vs. unbundled ancillary services. If an ancillary service is

unbundled, the supplier and the buyer have to be identified for each ancillary
service contract. This will allow the ICA to identify who is entitled to the ancillary
service. If the ancillary service is bundled, the service is included with the energy
contract that clearly lists the sellers and buyers. This enables the ICA to monitor the
use of all ancillary services. For transmission loss service, it is hard to know the
power flow amongst specific lines without knowledge of all contracts.
Additionally, it is hard to monitor and accumulate all flows on an individual
contract basis, especially with inadvertent flows. Thus, it is difficult to bid the
losses separately.
Computational Auction Mechanisms for Restructured ... 141

4.4.2.3 Spinning reserves and ready reserves. In the formulation, bids' prices of
power, spinning reserve, ready reserve are different. This is for showing the
formulation generally. If bids' prices are different, players might bid in a way that
they do not have to sell or buy the reserves. Should reserve transactions that have
been used be the same price as the transactions that have not been used? If they are
not the same, there will be more markets. The question is whether it is worthy to
have more markets? Constraints for spinning reserve and ready reserve neglect
losses due to these reserves. Additionally, the transmission capacity for reserves is
not included. This author selects this approach because losses are small and only
seldom will reserves be used.

4.4.2.4 Frequency of auctions. Auctions can be conducted every 15, 30, or 60

minutes. Frequent auctions cause more transaction costs while they give more
flexibility in scheduling for each party.

4.4.2.5 Ramp rate. Ramp rate constraints are endemic to specific equipment.
Some feel that the owners within the bids offered should implicitly handle
equipment constraints. Thus, the ICA would not supervise the equipment use by
every party.

4.4.3 Linear Programming or Mixed-Integer Programming

If the bids are required to submit in blocks and one block of bid is big, auctions
have to be formulated in mixed-integer programming. This will cause a lot more
difficult to solve. In fact, the power the buyers receive may be not in block due to
sales to cover losses. Thus, submitting bids in blocks when the block size is big
may not facilitate loss charges. However, if the size of bid is of same size as
solution tolerance, then the auction can still be formulated in LP and solved very
efficiently.

4.4.4 Transmission Loss Allocation

After the solution is found, we know the optimal transactions. The amount and cost
of the loss depends on each transaction. One frequent question is how to allocate
the cost of the transmission loss? One possible way is to use the information from
the simplex method. While the simplex method implements the auction, the
variables moving into and out of the basis at each iteration can be used to assign
loss charges. This enables the identification of each seller and buyer of each
transaction, as well as the cost of the transaction loss. The cost of the loss may be
paid by either the seller or the buyer or by both the seller and the buyer in any
proportion.

4.5 Utilizing Interior-Point Linear Programming Algorithm as Auction

Method

4.5. I Introduction

The purpose of this work is to illustrate application of IPLP to auction methods. An

extended IPLP algorithm is developed and used in the following. This extended
142 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

IPLP algorithm can find the exact optimal solution (i.e., exact optimal vertex) and
can recover the optimal basis. The optimal basis can be recovered in the
straightforward way. The great benefit is that sensitivity analysis can be performed
after the optimal basis is found. This extended algorithm is expanded from the
affine-scaling primal algorithm. The concept used in this extended algorithm to
fmd the optimal vertex and optimal basis is simple. Note that the exact optimal
vertex solution is desired for auctions because the solution must be fair and
uniquely identify assignments between buyers and sellers. Thus, fairness is
measured by the exactness of the solution.

4.5.2 Extended IPLP Algorithm

The extended algorithm is expanded from the affine-scaling primal algorithm. The
explanation of the algorithm is divided into two parts. The first part gives the basic
concept of the affine-scaling primal algorithm. The explanation of the basic
algorithm is based on Arbel [1993]. The basic algorithm is described along with the
expanded algorithm in the second part to be the complete algorithm of IPLP used in
the following auction. The expanded algorithm in the second part comes from the
addition of a section to the basic algorithm so that IPLP can find an exact solution
and can recover an optimal basis.

A. Basic Concept

Two major components of the affine-scaling primal algorithm are centering and
projective gradient direction. Movement is made through projective gradient
direction for maximizing the objective function or opposite to projective gradient
direction for minimizing objective function. The projective gradient direction is
used instead of gradient direction for the purpose of maintaining feasibility.
Centering is performed to achieve the potential to improve objective function in
each iteration.

LP using big-M method is used so that the unity vector, [1 1 ... 1], can be used
as the starting solution [Arbel, 1993]. This LP has one additional variable. If this
additional variable is driven to zero at the end of the algorithm, the primal problem
is feasible. Otherwise, the primal problem is infeasible.

Three quantities, duality gap, primal feasibility, and dual feasibility are used as
the stopping criteria for terminating the basic algorithm. These quantities are
defmed in (4-14) ([Arbel, 1993]). Note that x'is (n+I)-component column vector
of variables. c if is (n+ I)-component row vector of cost coefficients. c if signifies
the transpose of vector c~ A 'is m x (n+ 1) matrix oftechnological coefficients. b is
m-component column right hand side vector. If all three criteria are met
simultaneously, the basic algorithm is terminated. In other words, the solution
found is very close to the optimal solution. This type of solution is called E-optimal
solution. Although dual feasibility is desired in terminating the algorithm,
sometimes it cannot be achieved. In the implemented algorithm, the dual
feasibility-stopping criterion is neglected and the algorithm still works quite well
with test problems.
Computational Auction Mechanisms for Restructured ... 143

( __ ,T , bT )
· gap = _n_o_rm---'-c
d uaIIty x_---:=---'-y.:....
1 + norm(c,T x')
. I fieaSI'b'I'
pnma norm(b - A'x')
1 Ity = -_-'.......----'- (4.33)
1 + norm(b)
c' _A,T y-z
dual feasibility = - - - - - ' - - -
1 + norm(c')

B. Extended Algorithm

The extended algorithm for IPLP, affine-scaling primal algorithm, is described in

Table 4-12. The left column is step number. The right column is the algorithm.
Note that (iter) indicates the iteration number that the variable's value belongs to.
Step 0 to step 4 of the algorithm is based on Arbel [1993], and step 5 to step 7 of the
algorithm is developed by the authors. Two symbols, a and p as used in step 4,
explain interior point algorithms.

At step 4, a is the maximum allowable step size that maintains feasibility.

Changing by step size a will make at least one variable hit the boundary of the
feasible region. Thus, a factor p is used to make the new solution remain inside the
feasible region. The algorithm uses pas 0.95 for the results presented.

, {-x;(iter)
a=mm ,. ,
:Vdxj(iter)<O, l:S;i:S;n+l }. (4.34)
dXj(ller)

The extended algorithm is expanded from the basic algorithm so that the
algorithm can find the exact optimal vertex. Not only can this extended algorithm
fmd the exact solution, but also sensitivity analysis can be performed after the
optimal basis is found. The algorithm is composed of two parts: the basic IPLP
algorithm and the movement from the interior solution to the vertex solution.

The basic IPLP algorithm is terminated when the stopping criteria, duality gap
and primal feasibility, are small enough, usually in the range of 1e-6 to le-8 [Arbel,
1993]. The value of the stopping criteria is denoted as &1. If &1 is set too small,
numerical instability might occur. The authors handle this problem by adding an
additional stopping criterion. A quantity, z (defined at (7)) is calculated at every
iteration and z is the estimate of the reduced cost coefficient vector. When the
current solution is very close to the optimal vertex, the components of z, which
belong to the basic variables of the optimal vertex, are very close to zero. Thus, the
additional stopping criterion is to measure z and check to see if all the components
of z that belong to the basic variables are less than a value denoted as &2. In
addition, the algorithm has to check that the number of components of z which are
less than &2 is equal to the number of constraints because the number of basic
variables is equal to the number of constraints, Otherwise, degeneracy is indicated.
The value of &2 is set as 1e-6 for the extended algorithm. By adding this criteria, we
144 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

can use bigger value of &j to avoid the numerical instability problem. Thus, the
value of &j is set to 1e-4, which is bigger than the suggested value, 1e-6 to 1e-8 in
[Arbel, 1993].

Table 4-12. Extended Algorithm.

0 Initialize iteration counter, iter=O.
Initialize 8j and 82 as le-4 and le-6 respectively.
Initialize the starting solution vector:

f
x '(iter)=[1 1 .. , 1

1 Increment the iteration counter, iter=iter+ I.

Define the scaling matrix D(iter):

D(iter)=diag( [x 'liter) x](iter) ... x ~,+ liter)] ),

where diag(x? means diagonal matrix of vector x'and x/iter)

is the ith component of the present x' (iter).
2 Calculate the dual estimate, y(iter), where y(iter) is an m-
component column vector:

[A 'D2(iter)A ;rJ[y(iter)]=[A 'D 2(iter)c].

3 Find the estimate of the reduced cost vector, z(iter) and then
use it to find the primal step direction vector, dx '(iter), where
z(iter) and dx'(iter) are n+ J component column vectors:
z(iter)=c' -A ,T y(iter),
dx '(iter) =_D2 (iter)z(iter).

4 Update the solution vector:

x '(iter+ J)=x '(iter) +pa dx '(iter).

5 Test with two criteria.

First criterion: duality gap and primal feasibility are less than
8j.
Second criterion: the number of components of z that are less
than 82 are equal to the number of constraints.
If both criteria are satisfied, go to step 6. Otherwise, go to step
J.
6 The expected optimal basic variables are the variables having
z less than 82. Test the KKT conditions. If the KKT
conditions are satisfied, the optimal basic variables have been
found, go to step 7. If not, go to step 5 and reduce 8j and 82 to
be one-tenth of the value previously used in step 5.
7 Find the optimal solution. B is the optimal basis and XB
contains values of basic variables.

xB=B-jb
Computational Auction Mechanisms for Restructured ... 145

The parameters 6} and 62 are important criteria for terminating the basic IPLP
algorithm in step 5. The values of these two parameters are needed to be set
properly so that the algorithm can frod the solution. The procedure for changing
these two parameters are explained in step 6.

At step 6, &} and &2 are repeatedly reduced to be one-tenth if the KKT conditions
are still not satisfied. Actually this situation is unlikely because the original values
of &} and &2 (1 e-4 and 1e-6 respectively) are reasonable for most applications. These
values have been tested with many examples and the test shows that these values
can be used as criteria for terminating the algorithm's interior point process (step 0
to step 4) to frod the exact solution at step 7.

At step 5, the number of components of z that are less than &2 might be greater
than the number of constraints although these components of z belong to the optimal
basic variables for IPLP (i.e., variables which are greater than zero). This might
occur for 2 causes. First, free variables exist in the problem and each free variable,
Xi, is implemented as x/-x/, x/ and Xi- are non-negative. The problem occurs
because both of x/ and Xi- are greater than zero for the interior-point solution while
either each of x/ and Xi- is greater than zero and the other must be zero for the
simplex solution. Second, the optimal solution is degenerate and IPLP reaches the
optimal solution that is not on a vertex. The number of basic variables must be
equal to the number of constraints to use (10).

These two causes are handled as follows: For each of free variable, Xi, only either
each of x/ or Xi- with greater value is selected. For degenerate optimal solutions, if
p variables are selected at step 5, and there are m constraints, p>m, there are more
than one possible optimal vertex. Some of these are feasible and some are
infeasible. Trial and error is used to frod one feasible optimal vertex. Trial and
error might consume time when the number of possible optimal vertices is large.
However, this drawback does not happen when this algorithm of IPLP is applied to
auction methods because we will use the following procedure to handle degeneracy
instead of trial and error. The main concept is that the degenerate solutions carmot
be used for auction methods. If one of degenerate solutions is used, it will be unfair
to the bidders who are not selected. Indeed, it would be hard to select a set of
bidders without arbitrary allocation. Thus, when the number of components of z
that are less than &2 is greater than the number of constraints at step 5, the program
will report that degeneracy occurs. The basic variables having z less than &2 are
reported, which will tell the possible bidders to be selected. The independent
contract administrator can use some criteria to select the bidders. An example
criterion is to select the bidders who submitted bids first.

The movement from the interior solution to the vertex solution can be performed
just after the termination of the basic IPLP. Step 6 moves the solution point from
the interior solution to the vertex solution. The estimated optimal basic variables
are those variables satisfying the requirement that values of the reduced cost
coefficients are zero. These estimated optimal basic variables must be verified with
the Karush-Kuhn-Tucker (KKT) conditions for optimality. The KKT conditions are
elegantly developed in Bazaraa [1996]. If the KKT conditions are satisfied, the
146 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

estimated optimal basic variables are correct and the optimal solution can be
calculated as explained in step 7.

At step 7, values of non-basic variables are zero. The value of objective function
can be calculated from substituting the values of all the decision variables into the
objective function. To avoid fmding B- 1, LU or QR factorization can be used to
solve (10).

The extended algorithm also can check for infeasibility, unbounded, and
degeneracy of primal problem. The solution is infeasible if the selected variables at
step 5 contain the artificial variable added for the extended problem. The solution is
unbounded if for any iteration, all of the components of dx '(iter) that are found at
step 3 are all greater than or equal to zero. The solution is degenerate if the reduced
cost coefficient of any of the non-basic variables while testing with the KKT
conditions at step 6 is calculated to be zero. Actually, degeneracy can be detected at
step 5. If the number of components of z that are less than &} is greater than the
number of constraints, excluding the effect of free variables, indicates that
degeneracy has occurred.

4.5.3 Formulation

The formulation of the auction methods used is fully developed in [Dekrajangpetch,

1997]. The transmission line flow constraints are added and always treated as hard
limits. The framework of the energy market used can be seen in [Sheble, 1994b].
The auction is double-sided. Additionally, we assume that real power and reactive
power are bid in different markets. In other words, the real power and reactive
power are unbundled. To operate the market securely, there must be the linking
between the real and reactive markets. The linking in an unbundled set of markets
is to use and estimated real power limit in the real power market and use another
estimated reactive power limit in the reactive power market. Actually such series
equipment should be represented with only an MYA limit. The unbundled market
requires that the real and reactive power limits are estimated from the MYA limit.
We will not discuss this estimation detail. We are convinced that this linking issue
demonstrates the need for a market where real and reactive power are bundled. The
details of such an argument is not in this section. This section demonstrates only
the auction of real power market. Because of the unbundling of the real and
reactive power market, the fast-decoupled power flow can be solved independently.
Although the real power and reactive power markets are unbundled, the effect of the
reactive power market is reflected in the shadow prices of the real power market
based on the estimates of the real power flow limits. The interactions of shadow
prices of both markets occur through the linking between the real and reactive
power limits.

The other way to implement auctions is to bundle real and reactive power
markets. The formulation for this type of auctions is illustrated in the next chapter.
The comparison of unbundling and bundling real and reactive power markets are
also discussed in the next chapter. For this bundled market, the interactions of
shadow prices of both markets occur directly.
Computational Auction Mechanisms for Restructured ... 147

In this work, generation companies (GENCOs) and energy service companies

(ESCOs) submit bids for auctions. For the test cases included, the interactions of
transmission companies are ignored, but could easily included by incorporating the
transmission costs in the objective function. The constraints are taken from the
OPF. The formulation is described below:

m n
min
AI'bj,APsf,1l0j,Alj f::t CoM
"'" 51 SI
° -
f:t cbo M'.bo
"'"
'1 '1

subject to

-M-B'118=0
--
~ Mo - ~ M'. -lsc*118
~sl~bJ __
T
=0 (4.35)
;=1 j=1

PijO _ B; (118; - 118j ) :s; p'jmax

- Pijo + B; (118; - 118j ) :s; Pt'
O:S;M" :S;B"
O:S; Mbj :s; Bbj

The symbols are clarified as follows:

Cbj price ofjth buyer's bid

c.; price of ith seller's bid
&'bJ accepted amount of power ofjth buyer
&'.; accepted amount of power of ith seller
n number of buyers
m number of sellers
L18 change in bus angle
B' matrix containing the negative of susceptance
of the bus admittance matrix (Y matrix)
B.; amount of power submitted by ith seller
Bbj amount of power submitted by jth buyer
lsc loss coefficient vector
Pij0 original flow of line between buses i and j
Ptax flow limit of line between buses i and j

Note the underscored variables indicate vectors. All the quantities in the
formulation are in per unit, except that the bus angles are in radian and the bids'
prices are in $/MWh. The fast-decoupled power flow is used and thus the B' is a
constant matrix. The formulation above includes only real power relationships
because the auction has been unbundled. All the variables in the formulation are
non-negative except that L1~ and L1b! are free variables; thus, they are implemented
as explained in section II. The objective function is the negative of trading surplus.
The trading surplus is the difference between the revenue from ESCOs' bids and the
cost from GENCOs' bids. The first constraint is the active power flow relationship.
The second constraint restricts the slack injection. The next two constraints restrict
148 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

the flow on network branches. The last two constraints restrict the bids that sellers
and buyers submit.

4.5.4 Results

The results are divided into two main parts. The first part illustrates the accepted
bids from performing an auction. The second part illustrates the sensitivity analysis
of bids' prices and lines' flow limits. A six-bus system (based on [Wood, 1996]) is
used to demonstrate the implementation of network constraints. It is shown in
Figure 4-11. There are three GENCOs and three ESCOs. GENCOs 1,2, and 3 are
at buses 1, 2, and 3 respectively and ESCOs 1, 2, and 3 are at buses 4, 5, and 6
respectively.

#2

#3
# 10
Bus 5
Bus 4
(ESCO 2)
(ESCO 1)

Figure 4-11. Six-bus System.

Computational Auction Mechanisms for Restructured ... 149

Table 5 shows line data for line resistance, line reactance, and half of line
charging admittance. P tax
is the flow limit of line. The MW-base and voltage-base
are 100MW and 230 kY. Table 6 shows bus data. vn
is the original bus voltage
(Le., before trading). tf is the original bus angle. pO is the original bus real power
generation (for bus 1 to 3) and load (for bus 4 to 6). QO is the original bus reactive
power generation (for bus 1 to 3) and load (for bus 4 to 6). Note that this work only
developed the real market. Thus, the reactive market has to be solved separately.

The interaction of the real and reactive markets is hard to separate as shown in
[Dekrajangpetch, 1998]. This work assumed that the reactive power market was
partially solved before the real market was executed. Technically, the reactive
market would have to be re-executed to properly reflect the interaction of the
shadow prices through the change in real power limits based on the actual MYA
limits. The real market is thus reduced to a DC Power Flow model using the B'
matrix.

A. Accepted Bids

Table 4-15 shows the submitted and accepted bids by GENCOs and ESCOs.
GENCO 3 can sell all power offered for sale and ESCOs 1 and 2 can buy all power
they bid on. GENCOS 1 and 2 and ESCO 3 can sell and buy only part of the
desired power. The reason is not only because GENCOs 1 and 2 offer the higher
bids and ESCO 3 offers the lowest bids. It also depends on the network constraints.
Sometimes even the GENCO that bids the lowest or the ESCO that offers the
highest bid will have their contracts limited due to the network constraints.

Table 4-13. Line Data.

Line Fran To r X Y/2 p,!!'"
J
No. Bus Bus 1M 1M (pu) (MijJ
1 1 2 0.100 0.200 0.020 30.00
2 1 4 0.050 0.200 0.020 60.00
3 1 5 0.080 0.300 0.030 53.00
4 2 3 0.050 0.250 0.030 30.00
5 2 4 0.050 0.100 0.010 76.00
6 2 5 0.100 0.300 0.020 35.00
7 2 6 0.070 0.200 0.025 60.00
8 3 5 0.120 0.260 0.025 30.00
9 3 6 0.020 0.100 0.010 60.00
10 4 5 0.200 0.400 0.040 15.00
11 5 6 0.100 0.300 0.030 12.00
150 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

Table 4-14. Bus Data.

Bus V" (pu) 8' (deg.) P"(MW) (! (MW)

1 1.0500 0.00 112.62 34.79

2 1.0500 -2.53 140.00 75.07
3 1.0700 -5.15 60.00 112.34
4 0.9754 -4.68 100.00 70.00
5 0.9677 -6.58 100.00 70.00
6 0.9930 -7.27 100.00 70.00

The optimal surplus is $129.13. The flows of transmission line number 5

(between buses 2 and 4) and number 9 (between buses 3 and 6) are at the limits.
The shadow prices of lines numbers 5 and 9 are 4.39 $IMW and 0.09 $IMW
respectively. The shadow price in this context indicates the incremental
improvement in the objective function of increasing flow limit of the associated
transmission line. Note that the flows and the shadow prices of lines numbers 5 and
9 that will be discussed below are based on flows in direction from bus 2 to bus 4
(for line number 5) and in direction from bus 3 to bus 6 (for line number 9

An upper-bound linear programming (UBLP), simplex program (written by the

authors) is used to implement the auctions to compare to IPLP. The CPU time
UBLP used to find the solution is 9.43 seconds while the IPLP uses only 0.80
seconds. The IPLP uses 8.63 fewer seconds than the UBLP even for this small
problem. This shows IPLP to be very efficient.

Table 4-15. Base Case: Gencos' Bids.

Bids GENCO 1 GENCO 2 GENCO 3

Price 9.70 8.80 7.00
Submitted Amount 20.00 25.00 20.00
Accepted Amount 5.30 20.71 20.00

B. Sensitivity Analysis

1) Case A: Increase in ESCO 3 's Bid Price

From Table 4-15 (base case), only part of the desired power of ESCO 3 is
accepted. Case A here increases the bid price of ESCO 3 to 12.00. The new
accepted bids are shown in Table 4-16. Comparing to the base case (Table 4-15),
the bids of GENCO 1 and ESCO 1 are accepted less and the bids of GENCO 2 and
ESCO 3 are accepted more. The bids of GENCO 3 and ESCO 2 do not change.

The optimal surplus is $154.89. The flows of transmission line number 5

(between buses 2 and 4) and number 9 (between buses 3 and 6) are at the limits.
The shadow prices of lines numbers 5 and 9 are 5.75 $IMW and 6.37 $IMW
respectively. These shadow prices are higher than those of the base case, especially
the shadow price of line number 9. This indicates that it is worthier to increase the
flow limits of lines numbers 5 and 9. In other words, the value of the objective
Computational Auction Mechanisms for Restructured ... 151

function can be improved with the increase of the flow limits of lines numbers 5 and
9 with these new bids' prices. Note that the shadow price ofline number 9 is higher
than that of line number 5 in this case while the shadow price of line number 9 is
lower than that of line number 5 in the base case. This indicates that increasing the
flow limit of line number 9 can give higher benefit than increasing the flow limit of
line number 5 in this case.

Table 4-16. Base Case: ESCOs' Bids.

Bids ESCO 1 ESC02 ESC03

Price 12.00 10.50 9.50
Submitted Amount 25.00 10.00 20.00
Accepted Amount 25.00 10.00 10.29

2) Case B: Increase in Line Number 5 's Flow Limit

The flows of lines numbers 5 and 9 are at the limits for the base case. Case B
here increases the flow limit of line number 5 from 76 MW to 80 MW. The new
accepted bids are shown in Table 4-17. Comparing to the base case (Table 4-16),
the bids of GENCO 1 and ESCO 3 are accepted less and the bid of GENCO 2 is
accepted more. The bids of GENCO 3 and ESCOs 1 and 2 do not change. The
optimal surplus for case B is $135.06, which is bigger than the optimal surplus of
the base case, $129.13. This shows the benefit of increase in flow limit of line
number 5.

There are not any lines operating at their limits. The spare capacities of lines
numbers 5 and 9 are 2.64 MW and 0.62 MW respectively. This shows that it is not
beneficial to further increase flow limit of line number 5 with these bids.

This result is useful for transmission network expansion. The result indicates the
benefit from expanding the flow limit of line number 5. The same procedure of
case B can be performed on line number 9. Simulation of the procedure in case B
on various bidding scenarios can be used to make decisions of transmission network
expansion.

Table 4-17. Case B's Bids.

Bids GENCO 1 GENC02 GENCO 3
Price 9.70 8.80 7.00
Submitted Amount 20.00 25.00 20.00
Accepted Amount 0.00 25.00 20.00
Bids ESCO 1 ESC02 ESC03
Price 12.00 10.50 9.50
Submitted Amount 25.00 10.00 20.00
Accepted Amount 25.00 10.00 9.48
152 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

4.5.5 Conclusions

This work applies IPLP to auction methods. An extended algorithm of IPLP is

developed and used. This extended IPLP algorithm can find the exact optimal
solution (i.e., exact optimal vertex) and can recover the optimal basis. The concept
used in this extended algorithm is simple. The result shows that IPLP is very
efficient. The results of sensitivity analysis are very useful for adjusting the bid's
price to get more amount of bid to be accepted. In addition, the results of sensitivity
analysis are useful for transmission network expansion.

The comparison with UBLP has been extended to investigate the linking of
buyers and sellers. If the market rules require the identification of buyers and seller
uniquely, then the IPLP method has to be extended to include matching of buyers
and sellers. Note that the UBLP method inherently matches buyers and sellers at
each iteration. This is an interesting result of requiring the injection at the slack bus
to be zero. Thus, the authors suggest that the slack bus be a tie bus without any
injection for generation or demand. Specifically, the incremental losses are made
up by the buyer and the seller directly. This resolves the problem of loss allocation
without an arbitrary loss allocation algorithm to be executed after the fact. That is
after the real power market has been resolved.

4.6 Market Power

4.6.1 Market Reach

The goal of a fmn is to maximize its profit subject to constraints. There are two
major types of constraints-production constraints and non-production constraints.
Production constraints include input supply limit, input capability, etc. For
electricity, production constraints are comprised of fuel supply limit, generator's
capability (e.g., minimum up- and down-time.), crew constraints, and other
constraints depending on generating unit types. Non-production constraints could
be in various forms depending on products and market rules. Transmission network
constraints are the important non-production constraints that play a major role for
electricity. Transmission network constraints are comprised of operational and
security constraints. Security constraints are composed of reliability, voltage
stability, and transient stability constraints. Thermal limits for transmission lines,
voltage limits for electrical buses, power flow constraints, are examples of
transmission network constraints.

Market reach is the study of how a firm can get additional high-profit customers.
The fmn may earn more profit from these customers either in short-term or in long-
term. The fmn may consider losing profit in short-term to gain profit in long-term,
or the fmn may consider losing some current customers to gain additional high-
profit customers. This is acceptable as long as the firm can increase its total profit.
In the electricity market, firms include generation companies (GENCOs) and energy
services companies (ESCOs). ESCOs emerge in the deregulated market structure as
the sole source of electricity to the consumer. In addition, ESCOs can sell or buy
Computational Auction Mechanisms for Restructured ... 153

electricity to or from GENCOs. In the traditional regulated paradigm, GENCOs

supply electricity to consumers following regulations specified by the regulatory
commission. In the deregulated competitive market, GENCOs and ESCOs can
freely select the customers they desire. The term that will be used commonly to
represent both GENCO and ESCO hereafter will be "the firm".

There are factors that will limit the capability of freely selecting customers.
Electricity prices in different areas are different because different inputs and
production processes are used by local firms and different factors are encountered
on local firms. This causes difficulty for the firm with higher cost to compete with
the local firm or other firms with lower cost. This problem is more pronounced
when transmission cost (i.e., transportation cost in other products) is taken into
account. Transmission cost could limit the competition capability of the more
distanced firm to the local firm or even a nearby firm. Even the firm with lower
cost might have difficulty competing with the local firm or other firms with higher
cost when the effect of the transmission cost are considered. Moreover,
transmission network constraints could segment market in the way that firms in one
area cannot compete with other firms in other areas. Another important factor that
affects competition is market power.

Market power is the ability of a firm or a group of firms to control market in

favor of the firms. The firms having market power have advantages over other
firms in market reach capability. The firms having market power can utilize this
advantage to enhance their ability to get customers. It is evident that the study of
market power is useful for the study of market reach. Thus, the study will begin
with market power study and then follow with a market reach study.

The theme of this work is to determine how generation companies (GENCOs) or

energy services companies (ESCOs) can acquire additional customers or additional
transactions, given the auction results. In addition, this work is to determine how
GENCOs or ESCOs can utilize their market power (if they have any) to enhance
their chances of success.

4.6.1.1 Measuring Market Power

Deregulation of electricity market breaks up the vertically integrated monopoly and

this in tum reduces the firm's market power. However, market power still exists in
horizontal market. Market concentration can lead to market power. The general
case for market concentration is when there are a few large firms and these firms
have large market shares in the market. In addition, a congested transmission
network can segment market and can lead to market power even when the market is
not concentrated. This is because the firms in some strategic locations can
manipulate transactions to beneficially create market power for themselves. The
firms, who provide critical ancillary services, e.g., reactive power for maintaining
system voltages also have market power.

This section focuses on market concentration. Other causes of market power

will be discussed more thoroughly as part of this work. The number of GENCOs
154 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

and market share of GENCOs determine market concentration in electricity market.

The indexes used to measure market concentration are described as follows.

The Four-firm Concentration Ratio (14). This index is a linear summation of the
market share of the four largest GENCOs in the market. The equation for the four-
firm concentration ratio is shown.

Market share is denoted by si and the formula is shown.

The output power of GENCO i and the total output of all GENCOs are denoted by
qi and Q, respectively. Note that the market share and the four-firm concentration
ratio are in percent unit. Because the summation is linear, this index cannot
differentiate between each GENCO's market share provided that the summation is
the same. For example, suppose the market is such that the four largest firms each
has 15% of the market share, then the market has a four-firm concentration ratio of
60%, which is the same a market having four largest firms with market shares 57%,
1%, 1%, 1%. This gives rise to problems in measuring market power because the
latter market actually has more market power than the former market. A detailed
discussion can be found in Shy [1996].

A more generalized form of the four-firm concentration ratio is m-firm

concentration ratio, where m is the number of the largest GENCOs in the market
used to calculate the index. The number of the largest GENCOs needed to make the
index indicate the right concentration level of the market is questionable. In
addition, the number of largest GENCOs needed depends upon the market and
varies according to time even within the same market. For example, ifthere are five
large GENCOs having similar sizes in the market now, the four-firm concentration
ratio cannot be used and it may be necessary to use five GENCOs to calculate the
index instead. Talking this further, if another firm with similar size to those five
largest firms enters the market, the five-firm concentration ratio can no longer be
used and six GENCOs may be needed to calculate the index.

The Herfindahl-Hirshman index (HHI).

This index is the convex summation of all firm market shares. The equation for
HHI is shown in (10). Because the summation is convex, HHI can distinguish
between the effect of unequal market share among GENCOs even though the linear
summation is the same. Thus, HHI can solve the problem with the four-firm
concentration ratio. This is why HHI is used more widely than the four-firm
concentration ratio in measuring market concentration. Note that one problem of
HHI is that the data of all GENCOs is not known. This problem is also found in m-
Computational Auction Mechanisms for Restructured ... 155

fIrm concentration ratio. This will defInitely be the case when new generating units
are installed after the deregulation. A detailed discussion can be found in [Shy,
1996].

HHI=~::CsY
;=1

The other related index is the Lerner index [1934], which is the index for
measuring the market power of a monopoly by determining the percentage that the
price deviates from the marginal cost. The Lerner index is symbolized by t/J and the
formula is shown below. Price in the market is denoted by P and marginal cost is
denoted by Me. The Lerner index can be generalized to measure market power of
an oligopoly and of GENCOs in a market with multi-GENCOs and multi-ESCOs.
In this case, the index, ('od is the ratio of the difference of the price in the market
and the competitive price to the price in the market and the index is shown below.
Price in the market is denoted by P and the competitive price is denoted by

t/J=(P-MC)IP
(,od=(p_p*)lP

Note that the m-frrm concentration ratio, the HHI, and the Lerner index are
special cases of the gradient index and can be acquired by simple transformations
on the gradient index. The gradient index is calculated for measuring the rate of
potential improvement in the welfare performance of a market. Its value is sensitive
to the behavior of the GENCOs in the market. The detail of gradient index is
described in Dansby [1979].

There are plenty of controversial issues for calculating market share. One of
them is that the GENCOs' outputs change over time. For example, the outputs
might be high during peak periods and low during off-peak periods. In addition, the
GENCOs' outputs change with the condition of the network. The outputs at the
same period of each day may vary widely from each other. This is because the
network condition might favor a GENCO in one day but might favor another
GENCO a different day. This variance indicates that market power is dynamic. A
GENCO that has market power one period might not have market power the next
period. This indicates that the concentration measure must be calculated
dynamically.

4.6.1.2 Measuring Market-reach Ability. Market-reach ability suggests to and

from which ESCOs a GENCO can sell power, and with which of these ESCOs the
GENCO can make higher profIt. This concept is related to the market power and
acts like a complement to market power. GENCOs that have high market power
have high market-reach ability. GENCOs with market power have an advantage in
reaching the customers they desire. By knowing their market-reach ability, the
GENCOs can decide what to do to get higher profIt. Market-reach ability can be
determined by investigating past results, by calculating the transferred marginal
benefIts (TMB), or by performing sensitivity analysis to fInd the potential market-
156 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

reach ability. The meanings of these terms are explained separately in each method
below.

Past Result Investigation

By investigating the auction results, GENCOs can determine to which ESCOs they
can seII power. For example, one bidding scenario in the past is that GENCO 1
increases the bid price and the bid amount is accepted more. Resulting from
GENCO l's increased bid price, GENCO 2 sells less power; ESCO 1 buys more
power; and other GENCOs' and ESCOs' accepted bids stay at the same amount.
This result implies that GENCO 1 can sell more power to ESCO 1 and GENCO 2
can seII less power to ESCO 1 because of increased bid price of GENCO 1.

The investigation and analysis must be performed very carefully because there
are many factors that can affect results. Let us modify the above example so that
this time not only GENCO 1 increases the price, other GENCOs might also change
the bid prices and amount, and so do the ESCOs. This time the result is affected
from the changes of every bidder, not only by GENCO 1. This prevents us from
drawing our former conclusions. In addition, it is difficult to conclude which
ESCOs each GENCO will reach when the results of more than bidder change. This
is because it is difficult to distinguish the path of each of the power injections on the
network. The past result investigation is useful for giving us an idea about the
market-reach ability; however, it does not always yield any specific conclusions
because of the disadvantages described.

Transferred Marginal Benefit (TMB) Calculation

This approach is to calculate the marginal benefit of each ESCO with respect to the
location of a GENCO. In other words, the value of the marginal benefit of each
ESCO is transferred to the value corresponding to the location of a GENCO. The
concept is to calculate the amount of power that a GENCO has to supply to each
ESCO for given amounts of loads at each ESCO, assuming that the remaining
GENCOs and ESCOs are inactive. The TMB at the GENCO's location is higher
than the marginal benefit at the ESCO's location because of the transmission cost
and the transmission loss. The transmission cost and the transmission losses must
be determined and then the TMB can be calculated as follows.

MB'=MB+(L1TRC+t1LC) / L1P

The TMB can be used to infer the market-reach ability in the sense that it tells
the GENCO the lowest price to bid. If the TMB of an ESCO is lower than the
present average cost of the GENCO, this indicates that the GENCO cannot reach
this ESCO. Otherwise, the GENCO does have a chance to reach this ESCO. This
TMB is important information for bid pricing in the next period. The bid price for
the next period can be equal to the TMB or a little bit lower. However, this bid
price might not be accepted because other GENCOs might bid lower and get their
bids accepted instead. To develop the suitable bid price, the TMBs of the same
ESCO at other GENCOs' locations are needed. The same method is used to find
TMBs at other GENCOs' locations.
Computational Auction Mechanisms for Restructured ... 157

The additional great benefit from determining TMBs of an ESCO at all

GENCOs' locations is able to determine dynamic market power of all GENCOs to
the interesting ESCO. TMBs of all GENCOs are plotted together and the plot,
together with the cost curves of all GENCOs complete the analysis. The analysis
can tell which GENCO has advantage for the amount of changed load of the ESCO
under consideration. Note that the TMBs of all GENCOs may not be needed
because not all GENCOs can affect the bid acceptance. The GENCOs unable to
affect acceptance could be the GENCOs that are located very far from the
interesting ESCO or those that have more expensive costs. Trimming these
GENCOs from the analysis will help speed up the analysis without loss of accuracy.

This approach can be performed for any individual ESCO that is of interest to
GENCOs. The resulted TMBs are calculated under the assumption that other
GENCOs and other ESCOs will not change their power. If they change their power,
the result will be affected. Although other GENCOs might change their power, the
resulted TMBs' curves can give the big picture of how GENCOs can affect each
other and which GENCO has a greater advantage. The effect of changing other
ESCOs should be included in the analysis. It can be done by estimating the reaction
function of the ESCOs. However, this may result in a complex problem and could
make the calculation slower. This indicates the major drawback of this method
when a specific result is needed. I suggest neglecting the effect other ESCOs
changes and combining the result with the sensitivity analysis. The details will be
explained below.

Sensitivity Analysis
Many parameters can be used with sensitivity analysis to determine how to raise the
market-reach ability. They will be explained later in this work. This section
explains how to determine the potential market-reach ability, which is defined as the
ability to reach customers when the submitted bid price is at the average cost, given
that other bids do not change. This should be an indication of the potential market
reach, which is the best that this GENCO can achieve all other things remaining
constant. Average cost is utilized here instead of marginal cost to ensure that
GENCOs can recover their fixed costs in bidding. The potential market-reach
ability can tell a GENCO about its ability to best reach ESCOs. In other words, it
will tell a GENCO which ESCOs would buy power from the GENCO if the
GENCO bids its average price. Note that the potential ability in this context is
based on the submitted bid price. There are other factors that the GENCO might be
able to change to achieve a better result. However, I choose to focus on the price
because it is the main parameter used for bidding. In addition, the potential ability
in this context assumes that other bidders do nothing to affect the bidding results.
Other bidders might enhance the GENCO's chances of bid acceptance. However,
this is not considered for calculating the potential market reach because of the
uncertainty involved.

The procedure for determining the potential market reach is based on the
sensitivity analysis on the cost coefficient. Note that the submitted amount must
exceed the new accepted amount. If the submitted amount is accepted entirely, it
indicates that better result might be achievable if an additional submitted amount is
158 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

provided. If the new accepted amount is equal to the submitted amount, the
sensitivity analysis on the right-hand-side must be performed. The complete
algorithm for determining the potential market reach will be added in future work.
Note that the potential market-reach ability is dynamic because it depends on the
based condition.

After the potential market reach is found, the market-reach ability can be studied
further. The effect of other bidder's changes can be directly incorporated into the
sensitivity analysis. After the incorporation, the new result tells the potential
market-reach ability including the effect of other bidder's changes. An illustrated
example will be shown in future work. Sensitivity analysis can give a more specific
result for the market-reach ability analysis. However, it does not give us a handle
on the big picture about the market-reach ability. Past result investigation and the
TMB calculation are superior to sensitivity analysis in the context of giving the big
picture but they are inferior in the context of providing the specific information.
The sensitivity analysis results, combined with the past result investigation and the
TMB calculation, can enhance the ability to determine the market-reach ability such
as the accuracy. As suggested above, the TMB of an ESCO is calculated without
incorporating the effect of ESCOs' changes. This is reasonable because the TMB
will be used as information to the sensitivity analysis. The specific effect of
changes of other ESCOs can be incorporated directly in the sensitivity analysis.

4.6.1.3 Enhancement of Market-reach Ability After the market-reach ability is

known, the next step is to enhance the market-reach ability. In the short-run, this
can be done by adjusting the bidding parameters. Some states in the United States
have not deregulated the electric power industry. For the long-run, GENCOs in
these states can install the equipment e.g., a phase-shifter transformer and/or build
the new transmission lines that will increase their market reach ability.

This section uses sensitivity analysis to illustrate how to change parameters to

enhance the market-reach ability. In addition, the procedure of how to take other
bidders' reaction functions into consideration is also illustrated. The detail will be
added infoture work. The examples of changing parameters are described below.

A. Change in submitted bid amount

B. Change in submitted bid price
C. Change in line limit
D. Add or remove line (maintenance)
E. Sensitivity to other players' bidding strategies
F. Tap changing under load
G. Phase shifter control offlow
H Series compensation of line flows
I. Shunt compensation of line flows

4.6.1.4 Market Power Integration to Market-reach Ability Enhancement As

mentioned earlier in this work, some GENCOs will have market power even after
deregulation. This section discusses using market power to enhance the market-
reach ability. The basic form of market power studied in this section assumes a
GENCO owns the generating units in multi-locations. This work studies how this
Computational Auction Mechanisms for Restructured ... 159

multi-location GENCO can modify the bid in each location to favor the bidding. In
other words, this GENCO can find a combination of the bids in all its locations to
enhance its net profit. The GENCO might lose money in bidding in some locations
but gain money in other locations. This is acceptable as long as the net gain
exceeds the net loss. In addition, this GENCO can use the advantage to take market
power away from other bidders.

This is conducted to enhance the market-reach ability of GENCOs by

incorporating the market power they have. However, the ISO can conduct the same
study to investigate market implementation to reduce or eliminate the bidding
advantage of using market power.

4.6.2 Application ofAuction Methods to Power System Expansion

Auctions can be conducted for durations of any length, e.g., every 15, 30, or 60
minutes. Auction results in different periods are different because of different bids
and power system conditions. Different generator and/or transmission network
conditions are coupled to different contingencies. For example, one transmission
line out of service and one generator out of service are considered different
contingencies. If auctions are conducted hourly, there will be about 720 different
auction results in one month, which are coupled to different scenarios. Auction
results in a large number of periods can be aggregated by an on-peak/off-peak or
weekend/weekday model. In addition, auction results can be aggregated through
some other rational combination from hourly to monthly model.

There exist many choices for power system expansion. In other words,
additional generators and/or lines can be added to the system in many different
ways. The auction results in a large number of periods can indicate candidate
generators and/or transmission lines for power system expansion. For example, if a
transmission line is often congested in many auction results, the location of this
transmission line can be a candidate for building a new transmission line. The dual
value of the congested line indicates the benefit of increasing line capacity
[Dekrajangpetch, 1998c] and it can be used for selecting the candidates for new
lines. Note that the case of additional lines is different from the case of increasing
transmission line capacity because additional lines change the effective impedance
between buses while increasing transmission line capacity does not.

After all the new generator and/or line candidates have been identified, each of
these candidates is introduced to the power system network and new auction results
are calculated for each case. The new auction results are compared to the existing
auction results and a justification can be made that a particular line is the optimal
decision for expansion. Decision analysis method is used to justify which
additional line is the optimal choice. It can be seen that a great amount of
calculation is needed for calculating the new auction results for each of the new line
candidates. Thus, the method of calculating auction results must be very fast and
the interior-point linear programming method is very suitable with this need.

This work is intended to impact power system planning. The planning engineer
can use this information combined with other power system planning methods for
160 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

optimal decisions in planning. Methods of power system planning are well

illustrated in Wang [1994].

4.7 Supporting Services

As stated in the introduction, the term supportive services are used instead of
ancillary services. This is because these services are necessary for the proper,
timely, reliable operation of the composite power system. It is impossible to
decouple the real and reactive generation limits of the generator or transmission line
as shown in many textbooks [Wood, 1996]. Indeed, the transmission limits are a
function of weather and other factors. It is also impossible to schedule the energy
over the period without regard to the underlying change of demand from one period
to the next. The demand is constantly changing at least as a ramp between period
with random fluctuations around this ramp. Thus, it is very hard to separate the
reactive limits of generation from the real, as it is with all transmission and
distribution equipment.

4.7.1 Ancillary Services

This work focuses on four ancillary services: spinning reserve, ready reserve,
transmission losses, and load following. The purpose and time frame of these
services are summarized in Table 4-18. The proposed model considers transmission
losses and ready reserves as bundled ancillary services. The spinning reserve and
load following are treated as unbundled ancillary services with certain obligations
on part of each market agent. The sellers of load following contracts are obligated
to participate in area regulation contract defined later in this report. The buyers and
sellers may be obligated by the ICA to buy and sell certain minimum amounts of
spinning reserves in proportion to their accepted buy and sell bids respectively.
This is required because an aggregate system spinning reserve cannot take care of
transient and voltage instability problems. Some units need no spinning reserves
whereas others may do. The ICA is to determine how much of reserves is needed
and where they are needed.

The unbundled ancillary services are treated as separate commodities. In this

case, there exists a separate brokerage market for unbundled ancillary services.
However, the energy market and unbundled ancillary services market are coupled
by their operational characteristics. Hence, the same broker governs both markets
and solves the co-ordination problems between the coupled markets. The bundled
ancillary services can not be traded. However, the broker procures the bundled
ancillary services by single bid auction for a reliable system operation. The broker
must compute the cost of such services. The cost of bundled services are jointly
incurred.
Computational Auction Mechanisms for Restructured ... 161

Table 4-18. Time Frame and Purpose of Ancillary Services.

Ancillary Time frame Purpose

service
Spinning reserve Instantaneous Transient instability
Voltage instability
Load following 5 - 10 minutes Tie flow control
Frequency control
Ready reserve 15 minutes Post-contingency management
Transmission Instantaneous Embedded with power flow
losses

4.7.2 Assumptions

This work aims to recommend pricing mechanism for reserve margins, transmission
losses, and load following in electric power transaction. To simplify the discussion,
the problem is restricted with the assumptions below; however, the theory
developed is flexible enough to allow these assumptions to be relaxed if desired.

Assumption 1: The auctioneer will establish interchange schedules (multilateral

transactions) between the participating agents on hourly basis. Bilateral
transactions may be established outside the auction but are still subjected to scrutiny
with respect to operating constraints. Furthermore, it is assumed that TRANSCOs
are not economic agents, i.e. they are exogenous to the model. This assumption can
be relaxed by including Transco bids and evaluating them on the $IMW-Mile basis.
Post [70] has shown an auction simulation with the transmission bidding. However,
the transmission bid evaluation should include the cost of control (such as phase
shifters) to direct the power flow in accordance with the established contract. Since,
transmission is still considered to be monopoly, this work has not included Transco
bidding.

Assumption 2: The agents (ESCOs and GENCOs) are obligated to buy or sell the
binding bids declared by the auctioneer. Hence, all local operational constraints
(such as ramp rate constraints, emission constraints, fuel constraints, minimum
down times, minimum up times, start-up procedure curves, etc.) must be considered
by the agents while generating bids for the next trading session or round.

Assumption 3: The auctioneer is the sole authority to verify that the network
remains in operation with the new bids in place. The control center operators are
responsible to implement the contracts given by the auctioneer.

Assumption 4: GENCOs and ESCOs both are obligated to bid for electric energy,
spinning reserves, and load following contracts. Additionally, GENCOs are
obligated to bid for ready reserves.
162 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

Assumption 5: GENCOs are constrained to provide a certain minimum amounts of

spinning reserves based on their accepted bids.

Assumption 6: The amount of spinning reserve margins needed is decided by

market mechanisms. The amount of ready reserves needed is a fixed ratio of total
energy bids accepted.

Assumption 7: The cost of reactive power, energy imbalances, and redispatch are
computed as after the fact analysis and are added to industry cost. The allocation of
industry cost among different market agents is an another issue and is assumed to be
beyond the topic of this research.

4.7.3 Development ofAuction Model

The representation of electric power network for energy brokerage system consists
of models for the GENCOs, ESCOs, TRANSCOs and the ICA. The model for
transmission network includes the individual transmission lines, transformers and
the current distribution of load and generations. The selected models for GENCO
and ESCO include the specification of various attributes of a bid offer as shown in
Figure 4-19.

Bus Number
Block description for sellibuy bids:
Transaction order (priority of blocks)
Block size (MW)
Block price for power/spinning reserve
Price for ready reserves (seller)
Minimum biding quantity (sellerlbuyer)
Maximum biding quantity (sellerlbuyer)

Figure 4-19. GENCO and ESCO Bid Data.

The proposed formulation is based on the idea of linearizing the power flow
equations with respect to the current power flow state and solving the problem in
successive incremental steps. The broker maximizes the consumer surplus in the
energy market and spinning reserve market, as well as, minimizes the cost of
required ready reserve for the system. This is accomplished by including the price
information of energy bids and reserve margin bids in the objective function. This
approach is equivalent to maximizing the amount of transactions indicating that all
the potential trade gains are realized.
Computational Auction Mechanisms for Restructured ... 163

4. 7.4 Structure Options

The alternative structures for auctions have been outlined. The leA must know all
transactions each player has to operate the system economically and securely. This
requires that the net power at each bus is a known quantity in the formulation.
Thus, the leA should supervise the implementation of every contract between
parties by incorporating all of them as preconditions for contract acceptance. Power
system operation will be effective only if the leA knows all transactions and is
enabled to enforce all transactions with sufficient fmancial penalties. The leA must
be authorized to terminate contracts or player actions contrary to the security and
the safety of power system operation.

4.7.4.1 Singled-Sided Auctions vs. Double-Sided Auctions Both sellers and buyers
can specify the prices and amounts of bids as they desire for double-sided auctions.
They can adjust the bids to maximize their own profits. For one-sided auctions,
such as seller bidding only, only sellers submit price and quantity bids to the auction
and buyers specify only the amount of commodity desired. One of the concerns is
that sellers might have to sell cheap power because all sellers submitted low price
bids. The bidding price does depend on the pricing method. This problem can be
lessened if a seller can specify a reservation price. However, this is an inflexible
auction for buyers. This is true for all one-sided auctions, since one of the two sides
does not submit a price. These arguments indicate the inefficiency of one-sided
auctions. Such inefficiencies and the desire to implement competitive markets
support the conclusion to use double sided auctions.

4.7.4.2 Bundled vs. Unbundled Supportive Services If supportive services are

unbundled, the supplier and the buyer have to be identified for each supportive
service contract. This will allow the leA to identify who is entitled to the
supportive service. If the supportive service is bundled, the service is included with
the energy contract that clearly lists the sellers and buyers. This enables the leA to
monitor the use of all supportive services. For transmission loss service, it is hard
to know the power flow amongst specific lines without knowledge of all contracts.
Additionally, it is hard to monitor and accumulate all flows on an individual
contract basis, especially with inadvertent flows. Thus, it is difficult to bid the
losses separately.

4.7.4.3 Spinning Reserves and Ready Reserves The formulation to include bids'
prices of power, spinning reserve, ready reserve are possible. If bids' prices are
different, players might bid in a way that they do not have to sell or buy the
supportive services. Should reserve transactions that have been used be the same
price as the transactions that have not been used? If they are not the same, there
will be more markets. The question is whether it is worthy to have more markets?
The above equations, which show inclusion of spinning reserve and ready reserve,
neglect losses due to the flow resulting from the use of such supportive services.
Additionally, the transmission capacity for reserves is not reserved. This author
selects this approach because losses are small and only seldom will reserves be
used. However, when reliability constraints are added, this author believes that
such transportation reserves will be required.
164 Economic Dispatch, Unit Commitment, and Optimal Power Flow as Auctions

4.7.4.4 Frequency of Auctions Auctions can be conducted every 15, 30 or 60

minutes. Frequent auctions cause more transaction costs while they give more
flexibility in scheduling for each party. Some stock market auctions occur once
each day. Others occur every minute of each day. The selection is based on the
need to refme the contracts as other markets and supply availability change.

4. 7.5 Ramp Rate

Ramp rate constraints are endemic to specific equipment. Some feel that the owners
within the bids offered should implicitly handle equipment constraints. Thus, the
ICA would not supervise the equipment use by every party. This author believes
that the owners should be responsible for the use of their equipment not the ICA.

4.7.6 Reactive Power

There are two possible approaches to implement reactive power. The first approach
is direct implementation in the formulation. This approach has difficulty that the
nonlinear equations have to be linearized. The second approach allows players to
bid for reactive power. The objective function will have additional terms of reactive
power bids for this approach. This approach can avoid linearizing constraints but
there is one more market for the auctions that causes more complexity. This
complexity is not warranted based on the models developed.

4.8 Linear Programming or Mixed-Integer Programming

If the bids are required to submit in blocks and one block of bid is big, auctions
have to be formulated in mixed-integer programming. This will cause a lot more
difficult to solve. In fact, the power the buyers receive may be not in block due to
sales to cover losses. Thus, submitting bids in blocks when the block size is big
may not facilitate loss charges. However, if the size of bid is of same size as
solution tolerance, then the auction can still be formulated in LP and solved very
efficiently.

4.9 Transmission Loss Allocation

After the solution is found, we know the optimal transactions. The amount and cost
of the loss depends on each transaction. One frequent question is how to allocate
the cost of the transmission loss? One possible way is to use the information from
the simplex method. While the simplex method implements the auction, the
variables moving into and out of the basis at each iteration can be used to assign
loss charges. This enables the identification of each seller and buyer of each
transaction, as well as the cost of the transaction loss. The cost of the loss may be
paid by either the seller or the buyer or by both the seller and the buyer in any
proportion.