Settlement

1.0 Settlement without congestion

Let’s consider the base case solution obtained from the notes called
LPOPF2. How would the suppliers and the loads be paid?

To answer this question, we repeat here the solution in terms of the

one-line diagram and in terms of the table of Lagrange multipliers.

Pg2=1.2287pu
Pg1=0.5pu

1 2

PB2=0.0955
PB3
PB1 = =0.3242
-0.0152 PB5=0.4197 Pd2=1pu

PB4 =
4 0.4348 3

Pg4=0.45pu Pd3=1.1787pu

Fig. 1: Result in terms of generation levels and flows for base case

The objective function is Z=2705.8 $/hr.

1
 Table 2: Lagrange multipliers for Pd2=1.0, Pd3=1.1787
and infinite transmission capacity ($/per unit-hr)
Equality constraints Lower bounds Upper bounds
3
Equation Value*10 Variable value variable value
PB1 0.0000 Pg1 96.0000 Pg1 0.0000
PB2 0.0000 Pg2 0 Pg2 0.0000
PB3 0.0000 Pg4 43.0000 Pg4 0.0000
PB4 0.0000 PB1 0 PB1 0.0000
PB5 0.0000 PB2 0 PB2 0.0000
P1 1.2110 PB3 0 PB3 0.0000
P2 1.2110 PB4 0 PB4 0.0000
P3 1.2110 PB5 0 PB5 0.0000
P4 1.2110 θ1 0 θ1 0.0000
θ2 0 θ2 0.0000
θ3 0 θ3 0.0000
θ4 0 θ4 0.0000

The settlement for this case would occur like this:

Amount paid to generators:
Payment g1  Pg1  LMP1  50MW 12.11$ / MWhr  605.50$ / hr
Payment g 2  Pg 2  LMP2  122.87MW 12.11$ / MWhr  1487.96$ / hr
Payment g 4  Pg 4  LMP4  45MW 12.11$ / MWhr  544.95$ / hr
The total payments to the generators will be
605.50+1487.96+544.95=2638.41$/hr.

Now what do the loads have to pay?

Payment d 2  Pd 2  LMP2  100MW 12.11$ / MWhr  1211.00$ / hr
Payment d 3  Pd 3  LMP2  117.87MW 12.11$ / MWhr  1427.41$ / hr
The total payments by the loads will be
1211.00+1427.41=2638.41$/hr, and so we see that the market
settles with total payment to the generators equaling total payment
to the loads.

2
Question: Why does this differ from the objective function of
2705.80 $/hr?
Answer: We optimize on the offers. We settle at the LMPs.
The bus k LMP is the change in the objective function for
increasing the load at bus k by a unit. It is determined by the least
expensive regulating generator. So we are paying generators at the
offer of generator 2.
You can see this clearly by recomputing the total payment if we
paid each generator according to the offers they make: In this case,
it would be
Payment g1  Pg1  s1  50MW 13.07$ / MWhr  653.50$ / hr
Payment g 2  Pg 2  s2  122.87MW 12.11$ / MWhr  1487.96$ / hr
Payment g 4  Pg 4  s4  45MW 12.54$ / MWhr  564.30$ / hr
In this case, if we paid according to the offers, the total payments
to the generators will be 653.50+1487.96+564.30=2705.76$/hr,
which agrees with the value of the objective function (there is a
little round-off error).

So why do we settle at the LMPs rather than the offers? According

to the paper I placed on the web page [i, pg. 26],
“The primary reason for this conclusion is that under the pay-
as-bid settlement scheme, market participants would bid
substantially higher than their marginal costs (since there is
no incentive for participants to bid their operating cost) to try
to increase their revenue and, thus, offset and very likely
exceed the expected consumer payment reduction. As a
result, currently all ISOs in the United States adopt the pay-
at-MCP principle.”

Note to make a comment regarding what happens when you hit

lower limit and constrain-on units as in this example. Otherwise,
we should have objective function=sum of payments.

3
In other words,
 A pay-as-bid settlement scheme incentivizes participants to bid
high since the bid is what they will be paid if their bid is
accepted. The disincentive to bidding high is that their bid might
not be accepted.
 A pay-at MCP settlement scheme provides no incentive to bid
high. The disincentive to bid high because their bid might not be
accepted remains.

2.0 Settlement with congestion

Now let’s consider the case 2 solution obtained from the notes
called LPOPF. This is the case where line 3 was congested. How
would the suppliers and the loads be paid?

To answer this question, we repeat here the solution in terms of the

one-line diagram and in terms of the table of Lagrange multipliers.

Pg2=1.1803pu
Pg1=0.5pu

1 2

PB2=0.1197
PB3 =0.3
PB1 =
-0.0393 PB5=0.4197 Pd2=1pu

PB4 =
4 0.4590 3

Pg4=0.4984pu Pd3=1.1787pu

Fig. 6: Cases 2 flows

4
 Table 3: Lagrange multipliers for Pd2=1.0, Pd3=1.1787
and infinite transmission capacity except for and
0.3 capacity constraint on PB3 ($/per unit-hr)
Equality constraints Lower bounds Upper bounds
3
Equation Value*10 Variable value variable value
PB1 0.0000 Pg1 63.7500 Pg1 0.0000
PB2 0.0000 Pg2 0.0000 Pg2 0.0000
PB3 0.0860 Pg4 0.0000 Pg4 0.0000
PB4 0.0000 PB1 0.0000 PB1 0.0000
PB5 0.0000 PB2 0.0000 PB2 0.0000
P1 1.2432 PB3 0.0000 PB3 86.0000
P2 1.2110 PB4 0.0000 PB4 0.0000
P3 1.2647 PB5 0.0000 PB5 0.0000
P4 1.2540 θ1 0.0000 θ1 0.0000
θ2 0.0000 θ2 0.0000
θ3 0.0000 θ3 0.0000
θ4 0.0000 θ4 0.0000
The settlement for this case would occur like this:
Amount paid to generators:
Payment g1  Pg1  LMP1  50MW 12.432$ / MWhr  621.60$ / hr
Payment g 2  Pg 2  LMP2  118.03MW 12.11$ / MWhr  1429.34$ / hr
Payment g 4  Pg 4  LMP4  49.84MW 12.54$ / MWhr  624.99$ / hr
The total payments to the generators will be
621.60+1429.34+624.99=2675.93$/hr.

Now what do the loads have to pay?

Payment d 2  Pd 2  LMP2  100MW 12.11$ / MWhr  1211.00$ / hr
Payment d 3  Pd 3  LMP2  117.87MW 12.647$ / MWhr  1490.70$ / hr
The total payments by the loads will be
1211.00+1490.70=2701.70$/hr.

5
Notice: The amount paid by the loads exceeds that paid to the
generators by 2701.7-2675.93=25.77$/hr.

Why is this?

This is due to the congestion charges, denoted as CC and given by:

M
CC    j Pbj
j 1
In our example, since we have only one congested line, this is:
M
CC    j Pbj  86 * 0.3  25.8
j 1
Note that the units of μj are $/per-unit hr and the units of Pbj are per
unit, and so the units of CC are $/hr. Congestion charges are
allocated by the market operator to holders of financial
transmission rights (FTRs). We will discuss more about FTRs
when we talk about (a) risk management and (b) planning.

Comparison to the difference between payment to the generators

and payment by the loads, 25.77$/hr, indicates that this accounts
for it (within roundoff error).
Thus we are led to conclude that
M
CC    j Pbj   LMPj * Pdk   LMPk * Pgk
j 1 kload kgen

Let’s try to prove this.

We show in the Appendix A (see eq. 26) that the LMPs at each bus
are given by:

6
k  load : LMPk   Energy component
Ploss
 Loss component
Pdk
M
   j t jk Congestion component
j 1

If we ignore the loss component, then LMP’s are given by:

M
LMPk      j t jk
j 1

where tjk are called shift factors and give the change in flow on
circuit j to a change in real power injection at bus k, under a
specified slack distribution, according to
F j
t jk 
Pk (1)
If the network is linear over its entire operating range, then (1)
applies even when
F j  F j  0, Pk  Pk  0 (2)
so that
Fj
t jk  (3)
Pk
or
F j  t jk Pk (4)
In matrix form, (4) becomes:
F TP (5)

We show in Appendix B how to compute shift factors.

Then the congestion charge is: (Note there is a minus sign error on
the below somewhere, need to find it).

7
 N N
CC   LMPk * Pgk   LMPk * Pdk
k 1 k 1
N  M  N  M 
      j t jk * Pgk       j t jk  * Pdk
  

k 1 
  
j 1  k 1  j 1 
N  M   M  
       j t jk * Pgk      j t jk * Pdk 
   
k 1 
    
 j 1   j 1 
  
 
N M
       j t jk  Pgk  Pdk
 
k 1 
  
 j 1 
 
   
N M
   Pgk  Pdk    j t jk Pgk  Pdk
k 1 
 j 1 

   
N N M
   Pgk  Pdk     j t jk Pgk  Pdk
k 1 k 1 j 1

   
N N M
   Pgk  Pdk     j t jk Pgk  Pdk
k 1 k 1 j 1

 
N M
 0     j t jk Pgk  Pdk
k 1 j 1
Now interchange the summation to obtain:

 
M N
CC     j t jk Pgk  Pdk
j 1 k 1

 
M N
   j  t jk Pgk  Pdk
j 1 k 1
M N M
   j  t jk Pk    j F j
j 1 k 1 j 1

8
 Appendix A (LMPs)
1.0 Objective function

We make two simplifications on the objective function:

 We assume that demand bids are not price sensitive, i.e., that
demand is fixed, independent of the price. Therefore, the problem
of maximizing the social surplus becomes one of maximizing the
producer’s surplus, which is the same as minimizing the producer’s
cost. So our objective function will be to minimize producers’s
cost.
 We assume that producers cost is expressed linearly as a function
of their generation. You can think about this in two different ways.
o The producers represent their cost functions using a
piecewise linear approximation, or
o The producers are simply making offers of certain quantities,
at fixed prices.
In either case, we may represent the objective function as
N
G( P g )   sk Pgk (1)
k 1
where sk are the $/MWhr offers being made on an amount of
generation of Pgk over 1 hour, and there are N generators making
such offers. If bus k is a pure load bus, then Pgk=sk=0.

2.0 Power balance

If our analysis is based on linearized network representation, then

it is implied that resistance has been assumed zero. Therefore,
losses should be zero, and the power balance equation would be
N N
 Pgk  Pdk   Pk  0 (2)
k 1 k 1
That left-hand summation of (2), when placed equal to 0, says that
the sum of generation is exactly equal to the sum of demand. The

9
right-hand summation of (2), when placed equal to 0, says the sum
of injections is exactly equal to zero.

However, we will, for the moment, give a more general relation

that accounts for losses, i.e.,
N
 Pgk  Pdk  Ploss (3)
k 1

3.0 Line flow constraints

Regarding network representation, we assume that we have a

constraint sensitivity matrix T with elements tjk that give the
change in flow on circuit j to a change in real power injection at
bus k, under a specified slack distribution, according to
F j
t jk 
Pk (4)
If the network is linear over its entire operating range, then (4)
applies even when
F j  F j  0, Pk  Pk  0 (5)
so that
Fj
t jk  (6)
Pk
or
F j  t jk Pk (7)
In matrix form, (7) becomes:
F TP (8)
It is important to note that the sensitivity factors of (6) are
computed under a so-called “slack-bus” assumption which
indicates how the change ΔPk is assumed to be compensated. It
could be compensated from another certain bus, or from several

10
other buses, or from all other buses, and the elements of T will
change depending on which of these is assumed. It is generally
considered best to employ a so-called distributed slack bus
assumption here where the compensation is assumed to come from
all other generator buses.

Given that we know the “normal” flow constraints on every circuit,

then
 F max  F  F max (9)
Substitution of (8) into (9) results in
 F max  T P  F max (10)
We assume at this point that high flows in our network are
unidirectional, i.e., we need not be concerned with high flows in
both directions. This does not prevent bidirectional flows, it merely
enables us to be concerned with reaching the upper bound in only
one direction. Therefore, we may ignore the lower bound in (10) so
that our circuit flow constraint is
T P  F max (11)
In scalar form, (11) is
N
 t jk Pk  F j max, j  1,..., M (12)
k 1
and replacing injection with difference between generation and
load, we obtain:
N
 t jk ( Pgk  Pdk )  F j max, j  1,..., M (13)
k 1

4.0 Optimization problem for LOPF-1

A linearized optimal power flow (OPF) problem minimizes (1)

subject to (3) and (12), that is,

11
 N
min G ( P)   sk Pgk
k 1
s.t.
N
 Pgk  Pdk  Ploss
k 1 (14)
N
 t jk ( Pgk  Pdk )  F j max, j  1,..., M
k 1
We will call (14) LOPF-1.

The Lagrangian function for LOPF-1 is

N N  M N 
L( P g ,  ,  )   s k Pgk    Pgk  Pdk  Ploss     j  t jk ( Pgk  Pdk )  F j max 
k 1  k 1  j 1  k 1 
(15)
The first order conditions for finding the optimum to LOPF-1
include:
L Ploss M
k  gen :  s k   (1  )    j t jk  0 (16)
Pgk Pgk j 1

But we are more interested in the load buses. Consider

L P M
k  load :   (1  loss )    j t jk (17)
Pdk Pdk j 1

Note carefully that Pdk is not a decision variable, and therefore we

do not set it equal to 0.

Let’s consider (17). What is this? To answer this question, we need

to learn a theorem.
5.0 Envelope theorem

Consider the following optimization problem.

12
 max f ( x, )
x
s.t. g ( x, )  0 (18)
where x is the decision variable and θ is some parameter that is
influential in the problem, but it is not a decision variable, i.e., we
may not select its value. We desire to find how the optimal value
of f changes with respect to θ.

Let’s give a name to the optimal value of f. Let’s call it V; it is a

function of θ. That is,
V ( )  f ( x * ( ), ) (19)
Then what we are trying to find is
V ( )
 (20)
Note that V will change both because θ affects f and because it also
affects the optimal choice of x.

The Lagrangian function is

L( x, ,  )  f ( x, )  g ( x, ) (21)
Envelope theorem: The total rate of change in the optimal value of
the objective function due to a small change in the parameter θ is
the rate of change in the Lagrangian L evaluated at the optimal
value of x. That is,
V ( ) L( x( ), ,  )

  x  x*
(22)
The proof requires several lines of calculus that we omit here.

6.0 Locational marginal price

Need to review following paper to consider updating this part.

13
T. Organogianni and G. Gross, “A General Formulation for LMP
Evaluation,” IEEE Trans. On Power Systems, Vol 22, No 3, Aug
2007.

Armed with the envelope theorem, we may now identify the

meaning to (17), which is repeated here for convenience:
L P M
k  load :   (1  loss )    j t jk (17)
Pdk Pdk j 1

Equation (17) gives the change in the optimal value of the

objective function due to a small change in the parameter Pdk.

In other words, if we solve the optimization problem with

Pdk=Pdk0, obtaining G*(Pdk0), and then resolve the optimization
problem with Pdk=Pdk0+1, obtaining G*(Pdk0+1), then
L
G * ( Pdk 0  1)  G * ( Pdk 0 )  (23)
Pdk
L
We call the LMP for bus k, that is,
Pdk
Ploss M
k  load : LMPk   (1  )    j t jk (24)
Pdk j 1

Written slightly different, it is

Ploss M
k  load : LMPk        j t jk (25)
Pdk j 1

And (25) show us a very useful way to think about LMPs. They
consist of three components:
k  load : LMPk   Energy component
Ploss
 Loss component
Pdk (26)
M
   j t jk Congestion component
j 1

We discuss each one of these terms in what follows.

14
7.0 Energy component

We are considering the components of the LMP at a particular bus

k. The first component is the energy component, represented by λ.

To gain better understanding of exactly what this is, we will

neglect losses in our original formulation (14), resulting in
N
min G ( P)   sk Pgk
k 1
s.t.
N
 Pgk  Pdk  0
k 1 (27)
N
 t jk ( Pgk  Pdk )  F j max, j  1,..., M
k 1

Rewriting the equality constraint in (27) so that the function of

decision variables is on the left-hand-side and constants on the
right-hand-side, we have
N
min G ( P)   sk Pgk
k 1
s.t.
N N
 Pgk   Pdk  PD,tot
k 1 k 1 (28)
N
 t jk ( Pgk  Pdk )  F j max, j  1,..., M
k 1

15
Now write the Lagrangian function:
N N N  M N 
L( P g ,  ,  )   s k Pgk    Pgk   Pdk     j  t jk ( Pgk  Pdk )  F j max 
k 1  k 1 k 1  j 1  k 1 

(29)
or
N N  M N 
L( P g ,  ,  )   s k Pgk    Pgk  PD,tot     j  t jk ( Pgk  Pdk )  F j max 
k 1  k 1  j 1  k 1 
(30)

Notice that λ is the Lagrange multiplier (or dual variable) on the

power balance equality constraint. This immediately gives us an
interpretation of λ.

The energy component λ of the LMP is the increase in the

objective function (in this case, cost per hour) if demand
PD,tot increases by 1 unit.

Without losses, the LMP expression becomes (from (29)):

L M
k  load : LMPk       j t jk (31)
Pdk j 1

The summation is the congestion component. If there is no

congestion, then
k  load : LMPk   (32)
Equation (32) makes the interesting point that, if we ignore losses,
and if there is no congestion, then the LMP will equal to λ, and this
will be true for every load bus in the network.

One last comment here. It is worthwhile to identify what

determines λ. We may gain insight to this via the first order
condition (16) which, without losses, becomes
L M
k  gen :  s k      j t jk  0 (33)
Pgk j 1

16
Solving for λ, we obtain:
M
k  gen :   s k    j t jk (34)
j 1

Under the condition of no congestion, then

k  gen :   sk (35)
What does this mean?...

To understand what this means, it is important to understand that

Pgk for which we differentiate to obtain (35) must be “regulating,”
i.e., it cannot be at its limit. We could have exposed this idea more
clearly by including constraints on Pgk in the optimization problem
formulation, in which case we would have obtained corresponding
terms in the objective that would have vanished for regulating units
and would have contributed for non-regulating units.

Now consider how an electricity market works. Each generation

owner offers in their sk with a corresponding range. The algorithm
selects the lowest offer, and takes the full range of that offer, and
then selects the next lowest offer, and then the next, and so on until
the demand is met. Figure 1 illustrates.
9
8
7
6
5
C’(P)
4

2
1

Fig. 1
The only unit that is selected, and is regulating, is unit 5. This is
the unit for which λ=sk. It is the unit that will pick up the extra

17
demand when the demand is increased by 1 unit. We say that unit 5
is “on the margin.”

8.0 Loss component

Consider the expression for LMP again, from (25)

Ploss M
k  load : LMPk        j t jk (25)
Pdk j 1

Assuming no congestion, we have

Ploss
k  load : LMPk     (36)
Pdk
When we increase the demand at bus k by one unit, the losses will
increase due to more current flowing through the network.
Ploss
Therefore the term will be positive. This results in each bus
Pdk
seeing a higher LMP than that set by the energy component λ.

For a particular bus k, the increase in LMPk beyond λ will depend

on how an increase in that buses demand Pdk would be
compensated. The way it would really be compensated is that the
marginal unit would increase its generation. This would require
Ploss
to be recomputed each time the marginal unit changes which
Pdk
Ploss
is very frequent. What is really done is that is computed for
Pdk
each bus under an assumed compensation strategy. For example,
Ploss
reference [ii]1 shows how to compute relative to a
Pdk
distributed slack bus reference. We will not cover this but will
Ploss
simply assume the availability of .
Pdk
1
I have placed this reference on the web page. It is an excellent paper on LMP calculation.

18
9.0 Congestion component

Finally, we reconsider the expression for LMP once again, from

(25)
Ploss M
k  load : LMPk        j t jk (25)
Pdk j 1

At this point, our interest is the last term. Let’s ignore the losses,
resulting in
M
k  load : LMPk      j t jk (37)
j 1

The summation in (37) will contain zero terms for all circuits j for
which flow is not at the rating, i.e., the only non-zero terms in the
summation will be for circuits that are at their rating, i.e., that are
congested. Let’s consider that there is only one such circuit in the
network, circuit 5. Then
k  load : LMPk     5t 5k (38)
The Lagrange multiplier (dual variable) μ5 is on the flow constraint
for circuit 5, and it will always be nonnegative. On the other hand,
t5k, the generation shift factor, representing the change in flow on
circuit 5 for an increase in injection at bus k, may be positive or
negative. Thus we see that congestion, although usually increasing
LMPs for most buses, can also decrease LMPs under certain
conditions.

We will study the effects of congestion on LMPs in some depth in

the next set of notes.

19
 Appendix B (Sensitivities)
1.0 Introduction

Operation of the Eastern Interconnection has become heavily

reliant on using the so-called Interchange Distribution Calculator
(IDC). This is an internet-accessed system that interfaces with
OASIS and allows market participants and network operators to
efficiently, but approximately, determine the change in MW flow
on a flowgate given a set of changes in MW bus injections.

A flowgate is a circuit or set of circuits that interconnect different

regions of a network that can be limiting under some condition.

The IDC does not represent buses but rather represents control
areas, and there are 97 of them in the eastern interconnection.
Therefore the flowgates often represent interconnections between
these control areas; however, a flowgate may also be internal to a
single control area as well.

For our purposes, a control area is a bus, and the flowgates are
interconnections between the buses.

One of the most important uses of the IDC is in the coordination of

Transmission Loading Relief (TLR) actions. TLR procedures are
in place to guide operators in mitigating flows that exceed
operational security limits. TLR levels, summarized in Table 1 [1]
have been defined that correspond to different types of actions that
may be taken for which curtailments must be made. When a TLR
level 5 is declared, all ongoing transactions including those with
firm transmission service are subject to curtailment.

20
What we desire to obtain, then, is an expression for computing the
change in flow on a branch in a network for a given change in MW
bus injection.

21
 Table 1: Summary of TLR Levels [1]
TLR RELIABILITY COORDINATOR Action Comments
Level

1 Notify RELIABILITY COORDINATORS of potential

OPERATING SECURITY LIMIT violations

2 Hold INTERCHANGE TRANSACTIONS at current levels to Of those transactions at or above the

prevent OPERATING SECURITY LIMIT violations CURTAILMENT THRESHOLD, only those under
existing Transmission Service reservations will

System
Secure
be allowed to continue, and only to the level
existing at the time of the hold. Transactions
using Firm Point-to-Point Transmission Service
are not held. See Section B.1.

3a Reallocation Transactions using Non-firm Point-to- Curtailment follows Transmission Service

Point Transmission Service are curtailed to allow priorities. Higher priority transactions are
Transactions using higher priority Point-to-Point enabled to start by the REALLOCATION process.
Transmission Service See Section B.3.

3b Curtail Transactions using Non-firm Point-to-Point Curtailment follows Transmission Service

Violation

Security
Limit
Transmission Service to mitigate Operating Security priorities. There are special considerations for
Limit Violation handling Transactions using Firm Point-to-Point
Transmission Service. See Section B.4.

4 Reconfigure transmission system to allow There may or may not be an OPERATING

Transactions using Firm Point-to-Point Transmission SECURITY LIMIT violation. There are special
Service to continue considerations for handling Transactions using
Firm Point-to-Point Transmission Service. See
Section B.5.

5a Reallocation Transactions using Firm Point-to-Point Attempts to accommodate all Transactions using
Transmission Service are curtailed (pro rata) to allow Firm Point-to-Point Transmission Service,

System
though at a reduced (“pro rata”) level. Pro forma

Secure
new Transactions using Firm Point-to-Point
Transmission Service to begin (pro rata). tariff also requires curtailment / REALLOCATION on
pro rata basis with Network Integration
Transmission Service and Native Load. See
Section B.6.

5b Curtail Transactions using Firm Point-to-Point Pro forma tariff requires curtailment on pro rata

Security Limit
Transmission Service to mitigate Operating Security basis with Network Integration Transmission

Violation
Limit Violation Service and Native Load. See Section B.7.

6 Emergency Action Could include demand-side management, re-

dispatch, voltage reductions, interruptible and
firm load shedding. See Section B.8.

0 TLR Concluded Restore transactions. See Section B.9.

System
Secure

22
TLR “Risk” Criteria Transaction criteria RELIABILITY Comments
Lev COORD
IMMINENCE State Action

Forsee possible condition Notify

1 resulting in violation
Expected to approach, is Hold Not > 30 minutes before going
2 approaching, SOL to higher levels so xactions may
Secure be made based on priority.
Expected to approach is Some non-firm ptp at or Reallocate Curtailments made at top of
3a approaching, SOL above curtailment thres – hour.
holds, higher priority ptp
reservation approved
Existing or imminent SOL Some non-firm ptp at or Hold and Hold on onfarm; Curtailments
3b violation or will occur on
Insecure
above their curtailment Curtail made immediately.
or about
element removal to be thresholds.

Existing or imminent SOL Insecure Hold and Hold on onfarm.

4 violation or about Reconfigur
to be e
At SOL, no further All non-firm ptp at or Reallocate Curtailments made at top of
5a reconfig possible above curtailment upcoming hour.
thresholds curtailed;
Secure xaction request for
previously arranged firm
xmission service.
Existing or imminent SOL All non-firm ptp at or Curtail Curtailments made
5b violation or one will occur Insecure above curtailment immediately.
on element removal, no or about thresholds curtailed.
further reconfig possible to be

Existing SOL violation or Emergency Could include redispatch,

6 one will occur upon Insecure Action reconfiguration, voltage
element removal or about reductions, interruptible and
to be firm load shedding.

23
2.0 Calculation of Generation Shift Factors

The desired quantity is referred to as the generation shift factor and

will be denoted by tb,k. It gives the fraction of a change in injection
at bus k that appears on branch b. The Power Transfer Distribution
Factor (PTDF) is a generalization of the generation shift factor.

This calculation of generation shift factors is relatively

straightforward based on what we have done using the DC power
flow model.

Recall the DC power flow equations and the corresponding matrix

relation for flows across branches.
P  B'  (1)
P B  ( D  A)  (2)
Inverting eq (1) yields:
  B' P 1
(3)
Substitution of (3) into (2) yields:

P  ( D  A)B' P
1
B (4)
As we have previously defined in the notes on DC PowerFlow:
 PB is the vector of branch flows. It has dimension of M x 1.
Branches are ordered arbitrarily, but whatever order is chosen
must also be used in D and A.
 D is an M x M matrix having non-diagonal elements of zeros;
the diagonal element in position row k, column k contains the
negative of the susceptance of the kth branch.
 A is the M x (N-1) node-arc incidence matrix.
 B’ is the DC power flow matrix of dimension (N-1)x(N-1),
where N is the number of buses in the network, obtained from
the Y-bus as follows:

24
 1. Replace diagonal element B’kk with the sum of the non-
diagonal elements in row k. Alternatively, subtract bk (the
shunt term) from Bkk, and multiply by -1.
2. Multiply all off-diagonals by -1.
3. Remove row 1 and column 1.
 P is the vector of nodal injections for buses 2, …, N

The calculation of eq. (4) provides the flows on all lines given the
injections at all buses.

Bus this is not what we want. What we want is the fraction change
in flow on all lines given a change in injection at one bus.

In other words, given a change in injection vector ∆P:

 P2   P20   P2 
 P   0   P 
 3   P3   3 
      
P      0      PP
0

 Pk   Pk   Pk  (5)
      
   0  
 PN   PN  PN 
Therefore,

PB  PB  PB
0

 ( D  A)B' P  ( D  A)B' P
1 1 0

 ( D  A)B' P  P
1
 0
 (6)
 ( D  A)B'  P
1

25
Now let the ∆P vector be all zeros except for the element
corresponding to the kth bus, and assign this bus an injection
change of 1.
 P2   P2  0
 P   P  0
 3  3  
      
P     
P P
 k   k  1 (7)
      
     
PN  PN  0
Then
 PB1   t1,k  0 
 P   t  0 
 B 2   2 , k   
      1   
P B      ( D  A)B'  
 t
 Bb   b,k 
P 1 (8)
      
     

 BM   M ,k 
P t 0
Question: Does the above equation imply that the injection is
changed at only one bus? Explain.

Definition: The generation shift factor tb,k is defined as

PBb
tb, k 
Pk Reallocation
P olicy
Example 1:
Consider the example that we started in the “PowerFlow” notes.
Compute the generation shift factors for all branches
corresponding to an increase in bus 2 injection and a decrease in
bus 3 injection.

26
 t1,23  10 0 0 0 0   0 0 - 1
t    - 1 0 0  20  10 1
 2 , 23   0 10 0 0 0    0  1
t 3,23    0 0 10 0 0   1 - 1 0   10 30  10  1
       
t 4,23   0 0 0 10 0   0 - 1 1   0  10 20   0 
t 5,23   0 0 0 0 10  0 - 1 0 
 
 0 0  10
 10 0 0  0.0625 0.025 0.0125  1 

  10  10 0   0.025 0.05 0.025   1
 
 0  10 10  0.0125 0.025 0.0625  0 
 0  10 0 
 0 0  10  0.125 
 10 0 0   0.0375    0.375 

  10  10 0    0.025    0.2125
   
 0  10 10   0.0125  0.125 
 0  10 0   0.25 
Note that the above generation shift factors are for a “double shift.”

You can think of it like this. A generation shift factor for branch b,
bus k would be tb , k and another generation shift factor for branch

b, bus j would be tb 2, j . If we have an injection increase at bus k

of ∆Pk and an injection increase at bus j of ∆Pj, then
Pb  tb,k Pk  tb, j Pj (9)

Increase Pk, Decrease Pj,

Decrease P1 Increase P1
Therefore, if ∆Pk=-∆Pj, then

27
 
Pb  tb, k  tb, j Pk  (10)

3.0 Generation Shift Factors with Distributed Slack

Equation (8) shows how to compute the generation shift factors for
the case when a single specified slack bus corresponds to bus 1.

Example 1 above shows how to compute the generation shift

factors for the case when a single specified slack bus corresponds
to some other bus in the network (not the bus corresponding to the
reference by way of omission from its corresponding row and
column in the B’ matrix).

What we are interested in here is computation of generation shift

factors for the case when we would like to distribute the slack, or
the compensation, throughout the network. The key criterion to
guide this is that the elements in the nodal injection vector should
correspond to the percentage of desired compensation for each bus.

This criterion is illustrated below:

 PB1   t1,k   c2 
 P   t  c 
 B 2   2 , k   3
      1   
P B   
    ( D  A)B '  
 t
 Bb   b,k 
P 1 (11)
        
     

 BM   M ,k 
P t c N 
where

28
  
N  N 
c1    ci  1   ci 
i2  i2  (12)
 ik 
 
is the allocation desired for the reference bus.

One way to distribute the slack is to distribute equally to all buses.

In this case,
1
ci 
N 1 (13)
where we use N-1 in the denominator because one bus, bus k, is the
bus for which the computation is being made (and therefore ck=1).
If we use (13), then we can substitute into (12) to obtain:
   
N  N   N
1 
c1    ci  1   ci   1   
i2  i2   i 2 N 1 
 ik   ik 
   
 1   N 1 N  2  1
 1  ( N  2)     
 N 1   N 1 N 1  N 1
Example 2:
Using the system from Example 1 above, compute generation shift
factors for all branches corresponding to an increase in bus 2
injection, when the slack is equally distributed to all buses.
 t1,2all  10 0 0 0 0   0 0 - 1
t    - 1 0 0  20  10 1
 2 , 2 all   0 10 0 0 0    0   1 
t 3,2all    0 0 10 0 0   1 - 1 0   10 30  10  0.333
       
t
 4,2all   0 0 0 10 0  0 - 1 1  0  10 20 
    0.333
t 5,2all   0 0 0 0 10  0 - 1 0 
 

29
  0 0  10
 10  0.0625 0.025 0.0125
 0 0   1 
  10  10 0   0.025 0.05 0.025   0.333
 
 0  10 10  0.0125 0.025 0.0625  0.333
 0  10 0 
 0 0  10  0.1664 
 10 0 0   0.05    0.5001

  10  10 0   0    0.4999 
 
  
 0  10 10   0.0166  0.1666
 0  10 0   0.0002
It is of interest to compare the answer from the example where the
slack was distributed entirely to bus 3 and the example where the
slack was distributed to all buses.
 t1, 23   0.125   t1, 2 all   0.1664 
t    t   
 2 , 23    0 . 375   2 , 2 all    0 . 5001
t 3, 23    0.2125 t 3, 2 all    0.4999 
       
t 4 , 23   0. 125  t 4 , 2 all    0 .1666 
t 5, 23   0.25  t 5, 2 all   0.0002
   
Clearly the assumption on slack distribution is important!

There are other ways to distribute the slack. For example, we may
distribute the slack equally to all generation buses. Or we may
distribute the slack equally to all load buses. Or we may distribute
the slack to all generation buses in proportion to the MVA rating of
the generation that is located there (this approach conforms best to
reality, as we will see when we study AGC).

4.0 Generation Shift Factor Matrix

Given a specified slack distribution, we may compute a matrix of

generation shift factors according to

30
  t1,1 t1, 2  t1,k  t1, N 
t t  t  t 
 2 ,1 2 , 2 2 , k 2 , N 
       
T  
t
 b,1 t b, 2  t b ,k  t b, N 
       
 
 M ,1 M , 2
t t  t  t ,N 
 M ,k
M 
M N
c 1 c c c c c
c c 1 c c c c
 
c c c 1 c c c
( 
D   A )  B 1 
' 
M M M ( N 1) ( N 1)( N 1) 
       
c c c c c  c
 
 c c c c c  1 

( N 1) N

The above assumes that we desire generation shift factors for every
branch (a row of T) and every bus (a column of T). Note that the
first column of T is for a shift at the bus 1, which is the one
assumed to be deleted from the B’ matrix.

However, we need not include every branch. There may be some

branches that we know from experience will never overload, or
there may be policy that requires a particularly application to only
monitor certain branches. The latter is the case for NERC’s IDC
described at the beginning of this document.

31
Example 3: Let’s compute the T-matrix for Example 2. We assume
a distributed slack bus, where, c=-1/3. Therefore
10 0 0 0 0   0 0 - 1
 0 10 0 0 0  - 1 0 0   20  10 0   0.333
1
1  0.333  0.333
 
T   0 0 10 0 0   1   
- 1 0   10 30  10  0.333  0.333 1  0.333
  
 0 0 0 10 0   0 - 1 1   0  10 20   0.333  0.333  0.333 1 
 0 0 0 0 10  0 - 1 0 

0.3333 0.1667 0  0.5 

0.3333  0 .5 0 0.1667 
 
T  0 0.5  0.3333  0.1667
 
 0  0.1667  0.3333 0.5 
0.3333 0  0.3333 0 
The one-line diagram is shown below to facilitate understanding of
the relation between increased injection at bus k (the columns) and
how branch flows are affected (the rows).

Pg2=2pu
Pg1=2pu

1 2

2
1
5 Pd2=1pu 3

4
4 3

Pg4=1pu Pd3=4pu

32
Remember: each column is the set of shift factors for a unit
increase in injection (generation) at a certain bus. Column 1 is
when the injection at bus 1 is increased (there is no “1” in that
column because that is the one corresponding to the bus that was
deleted in the B’ matrix). Column 2 is when the injection at bus 2
is increased, and so on.

References:
[1] North American Electric Reliability Council (NERC)
Operating Manual, Appendix 9C1, May, 2004, available at
www.nerc.com.

[i] J. Yan, G. Stern, P. Luh, and F. Zhao, “Payment versus bid cost,” IEEE Power and Energy
Magazine, March/April 2008.
[ii] Eugene Litvinov, Tongxin Zheng, Gary Rosenwald, and Payman Shamsollahi, “Marginal
Loss Modeling in LMP Calculation,” IEEE Transactions On Power Systems, Vol. 19, No. 2, May
2004.

33