11th PSC Conference, Avignon, Francia, Agosto 1993.

Computation and decomposition of spot prices for transmission

pricing

Michel Rivier, Ignacio J. Prez-Arriaga
Instituto de Investigacin Tecnolgica
Universidad Pontificia Comillas
Alberto Aguilera 23
28015 Madrid, Spain



Keywords: Transmission pricing, Spot pricing, Power systems
economics, Utility regulation.

Abstract:

Marginal cost pricing is widely recognized as the core of any
sound approach to economic valuation of transmission services.
This paper presents results that help to recognize and to compute
the several components of the spot prices, so that the
contributions of each user to the network costs can be identified
and understood.

The paper analyzes, both theoretically and computationally, the
potential use of a decomposition of the spot price
k
of any node
k into a generator component and a network component
k
,
which is a prerequisite to precisely define separate markets for
generation and transmission. The notions of "slack node" and
"system lambda" are revisited and several new properties of the
spot prices, which can be useful in interpreting the numerical
values of network spot prices in large power systems, are
presented.

The results have been checked in large networks and are
illustrated with numerical examples.

1. INTRODUCTION.

The lack of a satisfactory procedure for pricing network services
is a major obstacle in the current debate on transmission
regulation that is taking place in different parts of the world.

The problem being addressed here is the allocation of the total
costs (i.e., capital and maintenance costs) of a transmission
network among all its users. This is different from the "wheeling
problem", see [1 to 4] for instance, where the objective is to
determine the extra costs of (typically) a vertically integrated
utility because of a given power transit within or across its
borders.

The former problem arises when access to a transmission
network (regardless of the number of its owners) is made fully
available to all the system agents, without discrimination, for a
fee. This is the situation in some of the countries with an
independently owned transmission network and a liberalized
electricity market, like Chile, England and Wales, New Zealand
or Argentina. This may also be the case of the Single Electricity
Market of the European Community under some of the schemes
of third party access (TPA) that have been proposed.

Transmission pricing is a complex matter that has to meet several
and often conflicting requirements: financial viability of the
network entity itself, promotion of economic efficiency of the
electricity market in the short and long term and quality of
supply in the provision of transmission services.

According to basic principles of economic theory, optimal
economic efficiency in the short run and the long run is obtained
when conditions for perfect competition exist and both
consumers and generators pay or are paid at spot prices for the
energy consumed or produced, respectively. A generator that
supplies a power g
k
at a node k is paid
k
.g
k
. Any participant
(large consumer or distribution utility) that withdraws a power d
k

from a node k pays
k
.d
k
. The spot price
k
(t) at a given instant
of time t and at a given node k is defined as the short-term
marginal cost of electricity production with respect to a change in
the demand at this node and in this instant of time [1]. Here the
explicit mention of the time t will be dropped for convenience.

Let the transmission network be regulated so that it receives the
surplus of the above transactions:

k

k
(
d
k
- g
k
) (1)

Let also the network regulation be such that the optimal
investment rule is followed: To invest in transmission facilities in
order to reduce as much as possible the operation costs of the
system, but only while the incremental investment cost is smaller
than the resulting savings in operation expenses.

It happens that, when the optimal network investment rule is
followed, conditions for perfect competition exist and there are
no economies of scale in transmission, expression (1) allows the
network to recover exactly its total costs (i.e., investment costs,
with the associated O&M costs), see [5].

One problem with using expression (1) to remunerate the
network entity is that its value increases when the network
performance deteriorates, i.e., more losses and congestion,
therefore creating perverse incentives for the network entity
regarding new investments and maintenance practices. This has
to be dealt with via adequate regulation schemes: standards of
network performance with associated penalties or credits and,
most importantly, a remuneration method that guarantees the
recovery of rightly incurred network costs; some elements of
competition, e.g., via competitive bidding for new investments,
may also be useful.

The second problem with (1) is that in actual systems, because of
the existence of important effects of economies of scale and other
deviations from the ideal conditions stated above, the network
revenues from (1) fall significantly short from recovering the
total costs of the network. Therefore the variable charges in (1)
must be topped with a complementary charge that allows full
recovery. This last charge must take the form of lump sums, so
that they interfere as least as possible with the optimal economic
signals and do not distort them [6].


In this paper it is analyzed, both theoretically and
computationally, the potential use of a decomposition of the spot
price
k
of any node k into a generator component and a
network component
k
:

k
= +
k (2)


The motivation for such a breakdown is multifold. First, because
even in markets that are based on marginal pricing, it may be
decided to price transmission services separately from generation
services (this is currently the case of England and Wales, for
instance). Moreover, in systems where transactions are priced at
the full spot price
k
(such as Chile or Argentina), there is often
a node that, because of its relevance as the centre of mass of the
market (this is exactly the case with Santiago or Buenos Aires),
is chosen as the reference for the computation of spot prices and
transmission charges, therefore suggesting that could be
adopted as the marginal price at the reference node. Also, there is
the case of systems where electricity prices are not based on
marginalistic principles but there is the need to find a convincing
transmission pricing method; this situation may result when
trying to implement third party access to a network. Finally,
fundamental to any transmission cost allocation method that is
based on marginal prices is the ability to recognize and compute
the several components of the spot prices, so that the
contributions of each user to the network costs can be identified
and understood.

More specifically, the paper makes the following contributions:

i) To demonstrate the physical meaning and the numerical
properties of the "generation" and "transmission"
components of the spot price, describing their calculation
procedure and showing the dependence of their numerical
value on the formulation of the computation model that is
adopted.

ii) To generalize the notions of "slack node" and "system
lambda" so that the user can precisely define the meaning of
the generation component of the spot price and adapt it to
the particular type of market regulation.

iii) To present several still unpublished properties of the
spot prices, which can be useful in interpreting the
numerical values of network spot prices in large power
systems.

These concepts are illustrated with numerical examples that have
been obtained using JUANAC, see [7], a computer program that
can calculate spot prices in large power networks.

2. COMPUTATION AND INTERPRETATION OF SPOT
PRICES.

The objective of this section is to extend previously published
results [1,8,9,10] while placing a particular emphasis on the
contributions of the network to the spot prices. Recent activity in
this area is reported in [7,11], for instance. We ignore for
simplicity the reactive component of the spot price. However, the
conclusions, interpretations and properties derived below can be
easily extended to the reactive component.

2.1 Summary of previous results.

The spot price can be defined as the spatially and temporarily
varying short term marginal cost with respect to changes in
demand. The mathematical derivation of the active component of
the spot price leads to the well known expression [1,8]:

k
=

+

L
d
k


-

Z
d
k


, with

m
+

Z
d
k
1
+

L
d
m
(3)
where
k
= spot price at node k.
d
k
= load demand at node k.
m = marginal generator.
L = Total ohmic losses in the system.
Z = Vector of voltage magnitudes and line power
flows.

= Lagrange multiplier associated to the global
energy balance equation.
= Vector of Lagrange multipliers associated to the
corresponding constraints of Z.

m
= Incremental cost of the marginal generator.

The interpretation of the several terms in (3) is not obvious and
related material in the technical literature does not help in
clarifying this issue. In the first place it must be realized that (3)
has been derived from a specific formulation of the composite
generation-transmission optimal dispatch problem. In particular
the formulation in [1] contains a global energy balance equation
(with being its multiplier) and a slack node. Alternative
formulations are possible (see later), resulting in different
expressions for in (3).

In the original references [1,8]

has been loosely referred to as

the 'system ', which is defined in the classical coordination
equations of economic dispatch as the system marginal operation
cost when the total load demand is increased by a unit according
to a predetermined distribution among the nodes [12,13]. This is
in part due to the coincidence of

in (3) with the classical

definition of the 'system ' when is null (the classical
coordination equations ignore the network limit constraints so
that does not exist). On the other hand

has also been

interpreted as the spot price of the node chosen as the slack [1],
which is in conflict with the previous definition even if the slack
is always chosen as the node with the marginal generator. To
these complications we must add the existing controversy about
the right meaning of the 'system ' and the correct method to
compute it [10].

These discussions are very meaningful because, from inspection
of equation (3), one may decompose the spot price into a
component

that only depends on time, and a both time and node
dependent component
k
:

k
=

+
k
, where
k
=

L
d
k
-

Z
d
k

(4)
Since the spot price coincides with

when a single node model is
used, in some instances this component has been identified as the
"generation" component of the spot price and
k
as the
"network" component of the spot price [1]. This decomposition
might be useful in cost assignment schemes, as the one being
discussed later in this paper. In the next subsection the precise
meaning of each term of (3) is made clear, its relationship to the
particular model formulation is established and the validity of the
proposed decomposition of the spot price into a generation and a
network component is discussed in depth.

2.2 Interpretation of the spot price expression.

2.2.1 Basic formulation.


It is important to realize that expression (3) and the definition of
spot price only make sense in the context of an optimization
problem, in which the operation costs are minimized while
subject to network and other operating constraints.

As it was mentioned before, the starting point of the analysis on
the meaning of

in equation (3) is that it is the Lagrange

multiplier corresponding to the total power balance equation in
the corresponding power flow model. Any further interpretation
depends on this specific model formulation. In the context of
geographically differentiated marginal costing, the most pertinent
power flow models are those making use of the Jacobian matrix
[12,13] or simplifications of it such as the DC load flow
equations [12,13], or extended DC formulations that take into
account ohmic losses [2,7]; in all of these models typically a
power balance equation is established for all nodes except for
one that is called the slack node. This is equivalent to
recognizing that all the generator outputs are independent
variables of the model, except for the output of the slack bus.

Let us work first under this premise of an explicit slack bus. It
implies that any partial derivatives will measure the sensitivity of
a function with respect to an independent variable while
maintaining all the remaining independent variables fixed, and
therefore the outputs of all the generators must be kept fixed,
except for the slack one. This rule has a direct physical
interpretation: the slack generator shifts its output in order to
maintain the electric balance of the system. These considerations
are extensively exposed in [12,10,14]. In order to make explicit
this important issue, we will use all along this section a notation
borrowed from [14]:
F
x

|
y
represents that the partial derivative of
a function F with respect to an independent variable x is
computed while letting the dependent variable y to vary as
needed.

This leads us to mathematically precise and physically
meaningful expressions for both the spot price and

k
=

s
+

s
L
d
k

|
s
-

Z
d
k

|
s
;

s
=

m
+

Z
d
m|
s
1 +
L
d
m|
s
(5)
where we have added a subscript s to

indicating that its value

will depend on the selection of the slack node s. The spot price

k
, however, must obviously be independent of the selection of s,
which is just a numerical artifice in the model formulation. This
is proved in the appendix.

Thus we arrive to the conclusion that
k
=

s
+
k,s
, which
clearly shows that the proposed breakdown of
k
into a
generation and a network component depends on the choice of
the slack bus s. Moreover, from (5) it can be written

s
=

s
(1 +
L
d
s

|
s
) -

Z
d
s

|
s
=

s
(6)
which is consistent with the definition of

s
as the Lagrange
multiplier of the global power balance equation which completes
the n-1 individual node power balance equations in the model;
note that a unit change in the demand at the right hand side of the
total power balance equation, amounts to a unit change only in
d
s,
since the remaining demands are explicitly set to a fixed
value in each one of the individual power balance equations.

We can then express
k
as

k
=

s
(1 +
L
d
k

|
s
) -

Z
d
k

|
s
(7)

that is independent of the premise of existence of a slack bus (see
the appendix for a demonstration):

k
1
=

k
2
(1 +
L
d
k
1

|
k
2
) -

Z
d
k
1

|
k
2
(8)
The interest of this equation is that all of its terms are
independent of the way the power flow equations are formulated;
one typically sets up the n-1 node power balance equations plus
the global power balance equation (this is usually done in load
flows), or alternatively sets up the n node power balance
equations (as it is usually done in optimal load flows). Therefore
(8) is independent of the model formulation.

Another important result of this subsection is to show that the
term

appearing in definition (3) of the spot price is model

dependent, and that, under the two modeling options described
above, it coincides with the spot price of the slack node.
Therefore, in these cases

could hardly be interpreted as the

'system ' or as the generation component of the spot price. We
can point out, however, that it will coincide with the improperly
called 'system ' that is obtained from the classical coordination
equations, when they are based on a loss formula (B loss
coefficients) derived from the inverse of the Jacobian [12]. The
losses in this formula are expressed in function of n-1 injections,
with a slack node being in charge of maintaining the electric
balance.

2.2.2 The spot price and the 'system '.

Our objective now will be to express the spot price as a function
of the "total system incremental cost" (TSIC), defined as the
increment in the operation cost when the total demand is
increased by a unit according to a predeterminated distribution t
i

among the nodes (t
i
=1). Note that this is a generalization of the
procedure to compute the 'system ' in the classical coordination
equations when a loss formula (B loss coefficients), function of
all injections, is used [12]. Here, besides the flexibility in the
definition of the distribution factors t
i
, the network limits (line
power flows and voltage magnitudes) are considered. From the
definitions of spot price and TSIC one can write:

TSIC = t
i
.
i
, (9)


where an obvious choice for t
i
is t
i
= d
i
/
j

d
j

This is an obvious extension of the system incremental cost
derived in [9,14], when network limits are considered. If now,
according to previous notation, we define

t
1
,...,t
n
= TSIC, it can
be proved (see the appendix) that:

k
=

t
1
,...,t
n
(1 +
L
d
k

|
t
1
,...,t
n
) -

Z
d
k

|
t
1
,...,t
n
(10)

where we have generalized the notation (F/x)
|
y
so that y may
represent several variables which respond all together, according
to a predetermined distribution in order to maintain the global
power balance.

If m is a marginal generator, we can then express the total system
incremental cost (the extended 'system ') as

t
1
,...,t
n
=

m
+

Z
d
m|
t
1
,...,t
n
1 +
L
d
m|
t
1
,...,t
n
(11)
where the multipliers are null when the network limits do not
constrain the solution, leading then again to the well known
expression of the classical 'system ', but this time with all the
terms being unambiguously defined. Equations (10) and (11) can
be seen as an extension of (5) where a fictitious node t
1
,...,t
n
is
used as slack. All the properties that have been derived for the
standard spot price formulation also apply here (see the
appendix). In particular from this expression we can also recover
(8), as it could be expected since (8) was independent of the
model formulation.

2.2.3 Decomposition analysis of the spot price.

The following conclusions concerning the proposed
decomposition of the spot prices, can be derived from the results
obtained in the previous subsections:

i) The spot price decomposition in distinct terms is rather a
mathematical artifice than a physically meaningful reality. It is
not possible to extract a generation and a network component,
not even a purely loss component or a component that captures
all the contribution related to the network limits. The more
general expression is (8) where

k
2
L
d
k
1

|
k
2
measures the contribution of the losses to the
difference between the spot prices at nodes k1 and k2.

Z
d
k
1

|
k
2
measures the contribution of the network limits
to the difference between the spot prices at nodes k
1
and k
2
.

One can conclude that this is still far from the concept of a global
generation component, a global loss component or a global
network limit component.

ii) In order to obtain a more significant decomposition of the spot
prices we can refer them to some more or less arbitrarily defined
"average marginal cost", which we decide characterizes an
electric system. The first candidate is obviously the previously
defined total system incremental cost TSIC (which corresponds
to an extension of the classical 'system ' as shown before),
which is the spot price of the fictitious node t
1
,...,t
n
. Then
equation (10) applies, and the terms

t
1
,...,t
n
(
L
d
k

|
t
1
,...,t
n
) and

(
Z
d
k

|
t
1
,...,t
n
) measure the respective contribution of the
losses and the network limits to the difference between the spot
price at node k and the spot price at the fictitious node. These
terms, as well as

t
1
,...,t
n
, become meaningful if we agree in
chosing this fictitious node as the standard reference point of an
electric system.

Still it will not be accurate to speak of a generation component, a
loss component and a network limit component because

t
1
,...,t
n
,
as any spot price, contains in itself information on generation,
losses and network limits in an inseparable meshed form.
However we are now able to distinguish a term that is common
to all the nodes and is sufficiently characteristic of the electric
system,

t
1
,...,t
n
, and a node specific term
k
=

t
1
,...,t
n
(
L
d
k

|
t
1
,...,t
n
) -

(
Z
d
k

|
t
1
,...,t
n
), where the mechanisms for the
computation of each term and the relationships with the model
formulation have been unambiguously defined.

2.2.4 Some additional properties of spot prices.

i) The preceding expressions make use of two kinds of terms:

- The spot prices
k
and the Lagrange multipliers , which
depend on both the operation point (i.e., the current values of
generator outputs, flows, etc.) and the status of the constraints
(i.e., generation limits, line capacity limits, etc.). They are
univocally defined.

- Terms such as (L/d
k)|

and (Z/d
k)|

,

where may
represent a node or an ensemble of nodes with the associated
distribution factors t
i
, which only depend on the present
operating point, regardless of the status of the constraints. They
depend on the choice of and t
i
.

ii) When several generators m
i
, iI
m
are at their margin it is
immediate from (10), since the spot prices at these nodes must be
equal to their marginal generation cost, that:

t
1
,...,t
n
=

m
i
+

Z
d
m
i
|
t
1
,...,t
n
1 +
L
d
m
i
|
t
1
,...,t
n
, i, iI
m
(12)
Note that, even if there are several generators at their margin and
widely different values of

k
's, there is a single value of

t
1
,...,t
n
,
whose easiest interpretation is probably obtained from (9).

iii) Some interesting properties can be derived when no network
constraints are binding. The spot price equations become

k
1
=

k
2
(1 +
L
d
k
1

|
k
2
) =

t
1
,...,t
n
(1 +
L
d
k
1

|
t
1
,...,t
n
) (13)

where only a multiplicative factor, that is not node-specific and
does not depend on the status of the constraints, accounts for all
the differences between the two spot prices. Therefore this spatial
difference will change smoothly with the operating point. The
spatial distribution of spot prices will be maintained
proportionally even when the spot prices experience discrete
changes, for example because of a substitution of the marginal
generator. Moreover, since the operating point immediately
before and after the change (respectively without and with
apostrophes in the expression below) is the same, the
multiplicative factor remains unmodified, resulting in:

k
1
-

'
k
1

k
1
=

k
2
-

'
k
2

k
2
=

t
1
,...,t
n
-

'
t
1
,...,t
n

t
1
,...,t
n
(14)
showing that when there are no binding network constraints, the
node-specific component
k
of the spot price is not directly
affected by the changes in location of the marginal generators.

2.3 Spot price computation.

Since the spot prices and their components are marginal costs,
they can usually be derived from dual variables of the solution of

the specific optimization model being employed for a particular
application [7,15]. Thus spot prices can be derived, at least
conceptually, from most optimal power flow programs.

For example in JUANAC [7] the spot prices are computed
directly as a subproduct of a modified DC optimal power flow,
through combinations of the dual variables associated to the
optimal solution.

To obtain the decomposition between the chosen

t
1
,...,t
n
in its
more general formulation) and
k
it suffices with computing

t
1
,...,t
n
through (9) in any model and then simply substracting it
from each
k
to obtain the node-specific component of the spot
price
k
.

2.3.1 Case example 1.

Examples are provided to illustrate the major points that have
been made in the previous subsections. The data and the system
configuration are summarized in fig. 1 and tables 1 and 2. The
results have been obtained with JUANAC [7].

Figure 1

1
2 3
4
5
6
7
8
9


Table 1
LINE Reactance Resistance
Capacity
From To
(P.U.)
Limit
(P.U.)
1
1
2
2
2
2
3
3
4
5
5
6
6
7
2
4
3
4
5
6
5
8
6
6
8
7
9
9
0.2913
0.2041
0.1695
0.4
0.2
0.4
0.099
0.4
0.6
0.2
0.4
0.6
0.2
0.2
0.0777
0.0544
0.0424
0.1
0.05
0.1
0.0248
0.1
0.15
0.05
0.1
0.15
0.05
0.05
(MW)
500
500
500
500
500
500
500
500
500
500
500
500
500
500

Table 2
Bus
Hydro
generation
(MW)
Thermal generation Load
Unit 1 Unit 2 Unit 3
(MW) (Cost) (MW) (Cost) (MW) (Cost)
Demand Unserved
(MW) (Cost)
1 300 75 65 125 70 100 75 1 15000
2 - 100 59 50 67 50 74 240 15000
3 160 100 61 50 76 50 80 40 15000
15000
15000
15000
15000
15000
15000
-
-
150
-
100
-
4
5
6
7
8
9
160
240
80
100
15
100
$/MW $/MW $/MW
$/MW


Four different cases have been run: In the first one losses and
network limit constraints have been ignored; losses have been
included in the second run; the operating points of cases 2 and 3
are identical, but the capacity limit constraint in line 1-2 is now
just binding while in run 2 it was not; finally in case 4 the line
flow 3-5 is strongly limited. Two solutions are given for each one
of the cases 2 and 3, which only differ in the choice of the slack
bus.

Table 3 summarizes the main results that have been obtained. All
the values presented have been computed independently:
-

defined as the Lagrange multiplier associated to the global
energy balance equation coincides with the spot price of the
slack node.
- All the terms in the table except for

are independent of the
choice of slack.
- The partial derivatives in the table only depend on the
operating point.
- Expressions (5) to (12) and (A-1) to (A-12) have been
numerically verified.
- In the first case all the spot prices are identical, as expected.
The values of the spot prices are equal to the marginal cost of
the generating unit that is on the margin, which is the second
unit at node one.
- In the second case the spot prices vary spatially due to the
losses, although the same generating unit is still at the margin.
The direction of the flows clearly indicates in which way the
spot prices increase. Observe that in this case with no binding
network constraints only the losses are accountable for the
spatial differences between spot prices, see (13). For instance,

2
=
1
(1 + 0.1924).
- The differences between the spot prices of cases 2 and 3, even
if they have the same operating point, are due to the just
activated capacity constraint of line 1-2, which in case 3
forces the dispatch to deviate from the optimal unconstrained
solution of case 2. Now the partial derivatives of the power
flow through line 1-2 measure the contribution of this
saturated line to the spatial differences between spot prices,
see (8). For instance,
2
=
1
(1 + 0.1924) + 2.2132x0.66.
Notice that the partial derivatives maintain their values.
- In the fourth case there is a generalized increase in the spot
prices because the system is dispatched far from the
unconstrained optimum, due to line capacity limits. An
important amount of power that node 3 cannot generate now
because line 3-5 is at its limit must be generated by node 1.
The spot price at node 3 has come down to 61 because the
marginal unit at this node is now a less expensive unit. On the
contrary, the spot price at node 1 has increased because a
more costly unit is now generating at the margin.
- From cases 3 and 4 it is important to note that strong
differences may exist between spot prices even in the absence
of non supplied power, because of the limitations imposed to
the optimal dispatch by the active operating constraints. We
have even found realistic examples where some spot prices
were negative.

Table 3

Case 1 Case 2 Case 3 Case 4
Slack Node 1 1 3 1 2
Marg. Nodes 1 1 1 1,3 1,3

65 65 74.73 65 78.84

1-2

0 0 0 -2.21 -2.21
L/d
2
|
1
0 0.192 0.192 0.192 0.192
L/d
1
|
2
0 -0.161 -0.161 -0.161 -0.161
L/d
3
|
1
0 0.149 0.149 0.149 0.149

L/d
3
|
2
0 -0.036 -0.036 -0.036 -0.036
Z
1-2
/d
2
|
1
0 0.63 0.63 0.63 0.63
Z
1-2
/d
1
|
3
0 -0.514 -0.514 -0.514 -0.514
Z
1-2
/d
1
|
2
0 -0.522 -0.522 -0.522 -0.522
Z
1-2
/d
3
|
2
0 -0.01 -0.01 -0.01 -0.01

1

65 65 65 65 65 70

2

65 77.48 77.48 78.84 78.84 86.54

3

65 74.73 74.73 76 76 61

4

65 78.77 78.77 79.41 79.41 88.75

5

65 81.54 81.54 82.91 82.91 113.76

6

65 86.24 86.24 87.57 87.57 109.62

7

65 102.58 102.58 104.16 104.16 130.38

8

65 72.25 72.25 73.47 73.47 80.39

9

65 98.75 98.75 100.27 100.27 125.52

1....9
(Eq. 9)
65 83.95 83.95 85.23 85.23 102.81

Table 4 illustrates the property (iii) described in section 3.2.4.
The first column shows the spot prices for the system loaded at
87.76% of the load refered in table 2 (3 is the marginal
generator), and the second column for a load of 87.77% (2 is the
marginal generator). The change of the marginal generator results
in a general increase of the spot prices. However as it is shown in
the third column the relative change is constant for all the nodes.

Table 4

Node
' (' - ) /
1 54.53 56.73 4.033
2 64.40 67.00 4.035
3 61.00 63.46 4.037
4 64.62 67.24 4.039
5 66.10 68.77 4.039
6 69.14 71.93 4.035
7 80.49 83.74 4.038
8 58.67 61.04 4.039
9 77.85 81.00 4.046


2.3.2 Case example 2.

A realistic example, corresponding to a 251 nodes and 361 lines
approximate representation of the peninsular Spanish system, is
presented now to illustrate the breakdown of spot prices into
"generation" and "network" components.

The values shown in fig. 2 and 3 correspond to annual time
average values obtained through a deterministic simulation of 42
scenarios representing different load conditions, scheduled
maintenance of generators and also forced outages. Both the
optimization of scenarios and the computation of yearly results
are performed by JUANAC [7].

Fig 2 represents the annual average in time of spot prices for all
generating and/or load nodes (
t

k
(t) /
t
t ). The nodes have been
ordered according to their annual energy injection to the system
(in descending order), that is according to
t
(G
k
(t) - D
k
(t)) where
G
k
(t) and D
k
(t)

respectively represent the total energy generated
and withdrawn at node k for a given period t. This value is null
around the node 65.

When the values in Fig. 2 are referred to the average spot price
for all nodes (represented by a horizontal line) which can be
chosen as the value of , the figure specifically shows the
corresponding annual average value for the "network"
component
k
. Observe that according to (2) the points below
the reference value correspond to a negative
k
and those above
it to a positive
k
. It is very interesting to observe that, according
to (15) below, typically all the nodes contribute to the network
income. As it will be explained in detail in the next section, the
nodes that are net energy importers should typically have a
negative
k
, while those that are net importers have typically a
positive
k
.

Figure 2
4,2
4,4
4,6
4,8
5
5,2
5,4
1
2
1
4
1
6
1
8
1
1
0
1
1
2
1
1
4
1

Figure 3
3,5
4
4,5
5
5,5
6
6,5
7
7,5
1
2
1
4
1
6
1
8
1
1
0
1
1
2
1
1
4
1

In Fig. 3 the annual value for the spot price of each node has now
been obtained by weighting the spot price of each scenario by
the sum of the absolute values of the energy retired and
generated at the node .for the considered scenario (
t
[

k
(t) (G
k
(t) + D
k
(t)) ] /
t
(G
k
(t) + D
k
(t)) ).
The dispersion in the values is now much larger, but the pattern
characterizing importer and exporter nodes is also very clear.
The outliers on the generation side can be easily explained: they
correspond to purely generation nodes with expensive generation
that is only rarely used, precisely when the marginal cost of the
system is high.

3. IMPLICATIONS ON TRANSMISSION PRICING.

In power systems where the bulk electricity market is based on
actual buying and selling of electric energy at spot prices with
spatial discrimination, there is no need in principle to decompose
the spot price into the and components. In this case the
network entity receives as variable charges the surplus (1) of the
bulk power transactions; what remains to cover its total cost must
be obtained from the users of the network as complementary
charges.


While the variable charges are automatically assigned to the
network users, an allocation criterion is needed for the
complementary charges. An approach that appears to meet the
requirements for long term economic efficiency is to allocate the
complementary charge of a given facility in proportion to the
benefits (reduced cost for consumers and increased revenues over
operation costs for generators) that each market participant
obtains from the existence of the considered facility. This topic is
beyond the scope of the present paper, see [16] for a detailed
discussion.

This prototypic market model with full utilization of spot prices
has not been realized or even approached in most actual power
systems. In some cases with a liberalized regulation and an
independent transmission entity, it may be found convenient to
deal separately with generation and transmission issues, so that
transactions that are mostly related to generation services may be
handled in a more conventional way if so desired (and not be
necessarily based on spot pricing). It is here where the properties
of spot prices that were developed in the last section can make a
contribution.

Regardless of the particular power system organization, power
exchanges in most electric systems are based on a single price,
the already referred to 'system ', which is therefore independent
of the relative positions in the network of the participants in the
transaction. If this 'system ' is made to coincide with a
satisfactory definition of the node independent component

of
the spot price, this leaves the following balance of the monetary
exchanges for the bulk power transactions:

k

k
(
d
k
- g
k
) = .
k
d
k
- .
k
g
k +

k
(
d
k
- g
k
) (15)

therefore indicating that the transmission utility charges
k
to
each unit of energy that is supplied from the network at every
node k, and pays
k
to each unit of energy that is injected to the
network at any node k.

Since
k
is typically positive in areas where the load exceeds the
generation and negative when the opposite conditions prevail, as
it can be observed in the examples in section 2, both the
consumers and the suppliers will normally make positive
contributions to the network costs. The volume of each
contribution will be proportional to the amount of power being
transferred in or out of the network and to the magnitude of the
deviation
k
of the spot price
k
, at the considered node k, with
respect to the adopted node independent component

. In section
2 these deviations
k
were shown to be directly dependent on the
contribution to system losses and to network limits saturation of
an additional unit of power being withdrawn or injected (negative
sign) at node k.

In the less frequent cases of a load in a predominantly generating
area, or the opposite situation, the variable charges become
credits, since these participants actually alleviate the losses and
the saturation of the network. In every case the value of the
corresponding
k
will precisely determine the magnitude and
sign of the per unit contribution of each node. The
complementary charges must be separately accounted for.

Note that in (15) the expression that is used for the network
variable charges exceeds expression (1) by an amount equal to
.L, where L represents the total system losses

L =
k

(
g
k
- d
k
)

This is no problem in practice, since the complete economic
signal = +
k
has not been changed and the total network
revenues are fixed a priori. The choice in the value of does
affect the required total amount of complementary charges.

An important feature of the proposed scheme is the simplicity
and precision with which the allocation of network costs and the
application of optimal economic signals with spatial
discrimination are jointly dealt with. One very desirable
characteristic of this method is that it is node based rather than
transaction based. This eliminates the need for decomposing
(more or less arbitrarily) the complete ensemble of power flows
that take simultaneously place in the network into bilateral
transactions.

The concept of long term price insurance contracts (or "contracts
for differences" in the regulation of England and Wales) for
generation services, can now be directly extended to transmission
network services, based on the nodal prices
k
, see [4] for
details.

An important issue that still remains to be discussed is the precise
definition, within this transmission pricing context, of the node
independent component

, with respect to which the values of the

k
's are directly determined from the
k
's. First one must realize
that the applicability of (15) is independent of the actual value of

, within a reasonable range. In order to obtain meaningful and

acceptable values of the
k
's, it seems that

should be roughly
centred within the distribution of values of the
k
's, so that the
charges (and eventually the credits) for use of the network are
not biased either towards the generation or the load. Therefore, in
the absence of an externally provided value of

, a sensible
choice could be:

=

k

k
.
d
k
+
k

k
.
g
k

k
d
k
+
k
g
k
, (16)
which is an obvious extension of (9), but now including both
load and generation. Equation (16) might even be used only for
the computation of the
k
's, regardless of the value adopted for

in the generation-based transactions. In other situations

may be
chosen as the marginal cost of the current marginal generator, or
as the spot price of the load centre of the market, if one does
exist.

In practical applications of these ideas, a different value of

may
be used for generators and consumers. This is due to the fact that
the spot price
k
does not account (although in theory it might)
for certain supply costs, typically named secondary services:
reactive compensation, load frequency control, operating
reserves, etc. These costs are directly paid to the generators
providing the services and they may be uniformly distributed to
the consumers. It is important not to include into these charges
the extra operation costs that are incurred because of the network
(e.g., out-of-merit generation dispatch) since they are already
taken care of by the
k
components.

When a network is congested and transmission constraints
become active, large differences between spot prices may exist
with the creation of "economic islands", i.e., sets of nodes with
spot prices that only depend on their local generators and thus are

not related to the rest of the system. When these conditions exist,
the differences
k
between the average obtained from (16) and
most of the spot prices
k
may be large, therefore resulting in
large individual contributions and credits of most of the
individual market agents to the network entity. Although a large
part of these contributions and credits will typically cancel out,
the total net revenue from variable charges will be also large
normally under these circumstances.

A possible solution to this problem, which does not change the
total value of expression (1) or the total economic signal
k
to
each participant, but does change the individual contributions to
the network and the total net value of the variable network
charges

k
(
d
k
- g
k
) in (15), consist of defining local values
of for well identified "economic islands", therefore resulting in
a redefinition of the
k
's without changing the
k
's. The
reduction in the network variable charges should be compensated
by the corresponding increment in complementary charges, in
accordance with the regulatory principles stated above.

4. CONCLUSIONS.

The mathematical and computational properties of decomposing
spot prices of electricity into a generation component that is
common to all nodes and a node-dependent network component
have been examined in depth. It is concluded that this
decomposition is possible and useful within certain types of
regulation of the electric industry. Caution must be exerted to
avoid the ambiguities of the two components when separately
handled. Numerical examples have been presented that confirm
the mathematical properties of the decomposition and the
heuristic interpretation of the results.

5. ACKNOWLEDGEMENTS

The first author gratefully acknowledges the support provided by
IBERDROLA during the realization of this work. The second
author was partly supported by Red Elctrica de Espaa while he
was spending a sabbatical year at the Massachussetts Institute of
Technology and working on this topic; he is also thankful to R.D.
Tabors, M. Caramanis and R.E. Bohn for helpful discussions.

REFERENCES

1. Schweppe, F.C., M. Caramanis, R.D. Tabors, R.E. Bohn,
Spot Pricing of Electricity, Kluwer Academic Publishers,
1988.
2. Caramanis, M., N. Roukos, F.C. Schweppe, "WRATES: A
Tool for Evaluating the Marginal Costs of Wheeling", IEEE
1988 PES Summer Power Meeting, paper 88 SM 649-6,
July 1988.
3. Kelly, K., Henderson, J.S., Nagler, P.A., "Some Economic
Principles for Pricing Wheeled Power", National
Regulatory Research Institute, NRRI 87-7, August 1987.
4. Prez-Arriaga, I.J., "A Conceptual Regulatory Framework
of Transmission Access in Multi-Utility Electric Power
Systems", 15th Annual International Conference of the
International Association of Energy Economists, Tours,
France, May 1992.
5 Prez-Arriaga, I.J., "A Conceptual Model for Pricing
Analysis of Transmission Services", Working Paper IIT-91-
006, Instituto de Investigacin Tecnolgica, Universidad
Pontificia Comillas, October 1990.
6. Trans Power, "Principles for Pricing Electricity
Transmission", Trans Power Limited of New Zealand,
1989.
7. Rivier, M., Prez-Arriaga, I.J., Luengo, G., "JUANAC: A
Model for Computation of Spot Prices in Interconnected
Power Systems", 10th PSCC Conference, Graz, Austria,
August 1990.
8. Caramanis, M., R.E. Bohn, F.C. Schweppe, "Optimal Spot
Pricing: Practice and Theory", IEEE Transactions on
Power Apparatus and Systems, vol. PAS-101, no. 9,
September 1982.
9. Ponrajah, R.A., Galiana, F.D., "Derivation and Applications
of Optimum Bus Incremental Costs in Power System
Operation and Planning", IEEE Transactions on Power
Apparatus and Systems, vol. PAS-104, December 1985.
10. Galiana, F.D., Vojdani, A.F.,"A Comparison of the
Classical and Modified Coordination Equations in
Economic Dispatch", Paper No A79079-5, Presented at the
IEEE, PES Winter Power Meeting, New York, February
1979.
11. Gorenstin, B.G., Campodnico, N.M.,Costa,J.P., Pereira,
M.V.F., "A Framework for Marginal Cost Evaluation in
Hydroelectric Systems with Transmission Network
Constraints", 10th PSCC Conference, Graz, Austria, August
1990.
12. Wood, A.J., Wollenberg, B.F., Power Generation,
Operation & Control, John Wiley & Sons, 1984.
13. Stevenson,Jr., W.D., Elements of power system analysis,
Fourth Edition, McGrawHill,1982.
14. Vojdani, A.F., "Analysis and Continuous simulation of
Secure Economic Operation of Power Systems", Ph.D.
Thesis, McGill University, August 1982.
15. Baughman,M.L., Siddiqi, S.N., "Real-time Pricing of
Reactive Power: Theory and Case Study Results", IEEE
Trans. on Power Systems, Volume 6, Number 2, February
1991.
16. Prez-Arriaga, I.J.,"Pricing of Transmission Services",
Working Paper IIT-92-030, Instituto de Investigacin
Tecnolgica, Universidad Pontificia Comillas, June 1992.


Appendix: PROOFS.

The following general expressions that relate partial derivatives
of losses in different nodes and with respect to different power
injections have been derived in [14]:

(1 -
L
J
i

|
j
) = -
J
j
J
i
(A-1)
(1 -
L
J
i

|
j
) (1 -
L
J
j

|
k
) = (1 -
L
J
i

|
k
) (A-2)

where J
i
is any positive power injection at node i. Applying
superposition in the limit of increments of line flows and bus
injections in the network and using (A-1):

Z
J
i

|
j
=
Z
J
i

|
k
+ (
Z
J
k

|
j
)(-
J
k
J
i
) =
Z
J
i

|
k
+ (
Z
J
k

|
j
)(1-
L
J
i

|
k
) (A-3)


From (5), (A-2) and (A-3):

i
=

s
(1 -
L
J
i

|
s
) +

Z
J
i

|
s
= (A-4)

=

s
(1 -
L
J
i

|
j
)(1 -
L
J
j

|
s
) +

[
Z
J
i

|
j
+ (
Z
J
j

|
s
)(1 -
L
J
i

|
j
)] =


= [

s
(1 -
L
J
j

|
s
) +

(
Z
J
j

|
s
)] (1 -
L
J
i

|
j
) +

Z
J
i

|
j

And with (A-4) this becomes (6):

i
=

j
(1 -
L
J
i

|
j
) +

Z
J
i

|
j
q.e.d. (A-5)

The independence of the spot price with respect of the node
chosen as slack will be proved next. Since

m
=
m
, and starting
again from (5):

i
=

m
-

Z
J
m|
s
1 -
L
J
m|
s
(1 -
L
J
i|
s
)

+

Z
J
i

|
s
,

and making use again of (A-2) and (A-3)

i
=

m
-

[
Z
J
m|
s'
+ (
Z
J
s'|
s
)(1-
L
J
m

|
s'
)]
(1 -
L
J
m|
s'
)(1 -
L
J
s'|
s
)
(1-
L
J
i|
s'
)(1-
L
J
s'|
s
)

+

[
Z
J
i

|
s'
+ (
Z
J
s'

|
s
) (1 -
L
J
i

|
s'
)] =

=

m
-

Z
J
m|
s'
(1 -
L
J
m|
s'
)
(1 -
L
J
i|
s'
) -

(
Z
J
s'

|
s
) (1 -
L
J
i

|
s'
)+

+

Z
J
i

|
s'
+

(
Z
J
s'

|
s
) (1 -
L
J
i

|
s'
) =

=

m
-

Z
J
m|
s'
(1 -
L
J
m|
s'
)
(1 -
L
J
i|
s'
) +

Z
J
i

|
s'
,

which is again (5) with the slack s'.

Expression (10) that extends the standard spot price equation so
that the fictitious node t
1
,...,t
n
can be used as the slack bus will
be proved next. Now a change J
i
in J
i
will result in a response
J
k
of all the remaining injections in order to maintain the global
balance and according to distribution factors t
k
, with a total
injection increment TJ
t
1
,...,t
n
such that:

J
i
+ TJ
t
1
,...,t
n
= L

It immediately follows in the limit that

1 -
L
J
i

|
t
1
,...,t
n
= -
TJ
t
1
,...,t
n
J
i
=
1
-
J
i
TJ
t
1
,...,t
n
=
=
1

k=1
n
( -
J
i
J
k
t
k
)
=

1

k=1
n
(1 -
L
J
k|
i
) t
k


(A-6)

which is independent of the choice of the slack bus. Similarly


Z
J
i

|
t
1
,.....,t
n
= (
Z
TJ
t
1
,...,t
n

|
i
) (
TJ
t
1
,...,t
n
J
i
) =

= - [
k=1
n
(
Z
J
k|
i
) t
k
] (1 -
L
J
i

|
t
1
,...,t
n
) (A-7)

which is also slack independent. From (A-6) and (A-2)

1 -
L
J
i

|
t
1
,...,t
n
=

(1 -
L
J
i|
s
)

k=1
n
(1 -
L
J
k|
s
) t
k

, then

(1-
L
J
i

|
s
) = [
k=1
n
(1 -
L
J
k|
s
) t
k
] (1-
L
J
i

|
t
1
,..,t
n
)

(A-8)

From (A-7) and (A-6)

Z
J
i

|
t
1
,...,t
n
= -

k=1
n
(
Z
J
k|
i
)t
k

k=1
n
(1 -
L
J
k|
i
)t
k



=
=
Z
J
i

|
s
-
Z
J
i|
s
[
k=1
n
(1-
L
J
k|
i
)t
k
]
+

k=1
n
(
Z
J
k|
i
)t
k

k=1
n
(1 -
L
J
k|
i
)t
k
=
=
Z
J
i

|
s
-

k=1
n
[
Z
J
i|
s
(1-
L
J
k|
i
) +
Z
J
k|
i
] t
k

k=1
n
(1 -
L
J
k|
i
)t
k



and using (A-3) and (A-6)

=
Z
J
i

|
s
- [
k=1
n
(
Z
J
k|
i
) t
k
] (1 -
L
J
i

|
t
1
,...,t
n
) , so that

Z
J
i

|
s
=
Z
J
i

|
t
1
,...,t
n
+ [
k=1
n
(
Z
J
k

|
i
) t
k
)] (1 -
L
J
i

|
t
1
,...,t
n
) (A-9)

Now all the pieces are put together to obtain (10). Starting again
from (A-4) and using (A-8) and (A-9):

i
=

s
(1-
L
J
i

|
s
) +

Z
J
i

|
s
=

s
[
k=1
n
(1-
L
J
k|
s
) t
k
](1-
L
J
i

|
t
1
,...,t
n
) +

{[
k=1
n
(
Z
J
k|
i
) t
k
] (1 -
L
J
i

|
t
1
,...,t
n
) +
Z
J
i

|
t
1
,...,t
n
}
=

s
{
k=1
n
[(1-
L
J
k

|
s
) +

Z
J
k

|
t
1
,...,t
n
]t
k
} (1-
L
J
i

|
t
1
,...,t
n
)
+

Z
J
i

|
t
1
,...,t
n
and using (A-4)

= (
k=1
n

k
t
k
) (1 -
L
J
i

|
t
1
,...,t
n
) +

Z
J
i

|
t
1
,...,t
n


=

t
1
,...,t
n
(1 -
L
J
i

|
t
1
,...,t
n
) +

Z
J
i

|
t
1
,...,t
n
,

q.e.d. (A-10)


Now it will be proved that (A-2) and (A-3) can be extended so
that the fictitious node t
1
...t
n
can be considered as a regular node.
From (A-6) and using (A-2)


(1-
L
J
i

|
t
1
,...,t
n
) =

1

k=1
n
(1-
L
J
k|
j
)(1-
L
J
j|
i
)t
k



=

1

k=1
n
(1 -
L
J
k|
j
)t
k


.

1
(1 -
L
J
j|
i
)


, so that

(1 -
L
J
j

|
i
) (1 -
L
J
i

|
t
1
,...,t
n
) = (1 -
L
J
j

|
t
1
,...,t
n
) (A-11)

From (A-7) and using (A-3)

Z
J
i

|
t
1
,...,t
n
= - {
k=1
n
[
Z
J
j

|
i
(1-
L
J
k

|
j
)+
Z
J
k

|
j
]t
k
}(1-
L
J
i

|
t
1
,...,t
n
)
= -
Z
J
j

|
i
[
k=1
n
(1 -
L
J
k|
j
) t
k
] (1 -
L
J
i

|
t
1
,...,t
n
) -

- [
k=1
n
(
Z
J
k

|
j
) t
k
] (1-
L
J
i

|
t
1
,...,t
n
) = ( see A-6,11)

= -
Z
J
j

|
i
(
1
1 -
L
J
j|
t
1
,...,t
n

) (
1 -
L
J
j|
t
1
,...,t
n
1 -
L
J
j|
i
)
- [
k=1
n
(
Z
J
k|
j
) t
k
] (
1 -
L
J
j

|
t
1
,...,t
n
1 -
L
J
j|
i
) = ( see A-7)
= -
Z
J
j|
i
1 -
L
J
j|
i

+
Z
J
j|
t
1
,...,t
n
1 -
L
J
j|
i

; then

Z
J
i

|
t
1
,...,t
n
(1 -
L
J
j

|
i
) +
Z
J
j

|
i
=
Z
J
j

|
t
1
,...,t
n
(A-12)

Comparing (A-11) and (A-12) with (A-2) and (A-3) we notice
that these general properties of real nodes also apply to the
fictitious node t
1
...t
n
. Next it will be proved that (8)

still holds
when t
1
...t
n
is used as the "slack bus". Starting from (A-10) and
employing (A-11) and (A-12):

i
=

t
1
,...,t
n
(
1 -
L
J
j|
t
1
,...,t
n

1 -
L
J
j|
i

) +

(
Z
J
j|
t
1
,...,t
n
-
Z
J
j|
i
)
(1 -
L
J
j|
i
)



=

t
1
,...,t
n
(1 -
L
J
j|
t
1
,..,t
n
) +

Z
J
j|
t
1
,...,t
n
(1 -
L
J
j|
i
)


-
-

Z
J
j|
i
(1-
L
J
j|
i
)


=

j

(1 -
L
J
j|
i
)
-



Z
J
j|
i
(1 -
L
J
j|
i
)
, so that

j
=

i
(1 -
L
J
j

|
i
) +


Z
J
j

|
i
, q.e.d.