Pricing of reactive support ancillary services

X.J. Lin, C.W. Yu and C.Y. Chung

Abstract: Good reactive power management is essential for the secure operation of power systems.
In a deregulated power market environment, apart from technical complexities, economic
considerations of compensating reactive power suppliers and charging transmission users add new
challenges to the problem. The problem of pricing reactive ancillary services is addressed. It is
formulated as a joint cost allocation problem. The total reactive support costs are determined to be
the increase in real generation costs from the reactive unconstrained case (base case) to the case in
which voltage security constraints are considered. The incurred costs are allocated to consumers
with the application of the Aumann-Shapley method. The results of the simulation of the proposed
methodology are demonstrated on an eight-bus system.

1 Introduction individual loads. However, the coupling of real and

Reactive power has profound effects on real energy
transfers and on the security of power systems, as it
affects the voltage profile throughout the system. In the
vertically integrated electricity industry, the costs of reactive
support are included in the bundled electricity prices
that retail customers pay. This practice is not suitable for
new power markets, characterised as they are by the
unbundling of generation, transmission and distribution.
The incurred costs of different reactive power suppliers
should be precisely identified and properly compensated
for, and each individual consumer needs to pay for his/her
part of the reactive power requirements. Much of the
literature on reactive power pricing builds on marginal
cost theory, which has been applied in the spot price for
real power [1]. An early attempt at this approach can be
found in [2]. A decoupled optimal power flow (OPF)
model was proposed for active and reactive pricing in [3].
In-depth theoretical discussions on applying the concept of
marginal cost for the real-time pricing of reactive power
was provided in [4]. Detailed cost models of reactive
power support can be found in [5] and a similar approach
based on the opportunity cost of dispatching reactive power
from generators was adopted in [6]. However, as pointed
out in [7, 8], the application of marginal reactive pricing
may not be practical, owing to the volatility and
erratic behaviour of such prices. Moreover, marginal cost
pricing is subject to the problem of reconciling marginal
cost prices with the requirement to recover costs. Another
approach is to formulate reactive power pricing as a
reactive support cost allocation problem. Reactive
power tracing [9], graph theory [10] and modified Y-bus
method [11] are in this category. These methodologies
attempt to charge system participants by determining
the reactive power that each generator contributes to each
r IEE, 2005
IEE Proceedings online no. 20045268
doi:10.1049/ip-gtd:20045268
Paper first received 7th December 2004 and in final revised form 2nd April 2005
The authors are with the Department of Electrical Engineering, The Hong
Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong, People’s
Republic of China
E-mail: eecwyu@polyu.edu.hk
616
reactive flows in a transmission network makes contribution
factor calculations using these methods subjective to some
extent.
Owing to the inductive nature of the transmission system,
it is difficult to transport large amounts of reactive power.
Sources of reactive power need to be located close to the
points where they are consumed. Apart from the produc-
tion of reactive power, additional reactive power capacities
should be reserved to cover the sudden increase in system
reactive power requirements due to a contingency. Reactive
power support may be provided by a variety of devices,
including shunt capacitors, static var compensators and
generators. The consensus is that a significant degree of
obligation and centralised control is necessary in reactive
power management.
In this paper, the pricing of reactive ancillary services is
formulated as a joint cost allocation problem. The ISO
(independent system operator) will procure the required
reactive service, taking technical feasibility and economic
efficiency into account. The incurred costs will then be fully
allocated to different consumers in an equitable and
transparent manner. The paper focuses on reactive support
from generators. Cost recovery for other reactive devices
may be included in the design of transmission tariffs and is
not the subject of this paper. The methodology proposed in
the paper is for reactive power support cost allocation in
short-term (day-ahead) scheduling and only one time
interval (one hour) problem is considered. Real generation
rescheduling is used as a preventive control to satisfy the
reactive support requirements imposed on generators in
both normal and contingent states. The reactive support
cost of an individual generator is measured as opportunity
cost, which is evaluated from its profit loss in the real power
market. Under the unchanged system total real power
output and real power market clearing price assumptions,
the total reactive support costs (the summation of the
reactive support costs of all generators) is consistent with
the difference in real power production costs between the
unconstrained and constrained cases of reactive power/
voltage. These are discussed in detail in Section 2. Cost
allocation factors for real loads and reactive loads are then
calculated with the application of the Aumann-Shapley
method [12–14]. The proposed methodology is tested using
a simple eight-bus system.
IEE Proc.-Gener. Transm. Distrib., Vol. 152, No. 5, September 2005
2 Analysis of reactive support costs
Maintaining a proper system voltage profile in both
normal and contingent states requires sufficient provision
and reserves of reactive power. The effective management
of reactive power is vital to the operation of the power
market. The first issue related to the pricing of
reactive power is to identify the cost of reactive support
requirements. In the paper, they are measured
as the increase in the cost of producing real power
from the base case (with network constraints neglected) to
the case in which reactive power/voltage constraints are
considered.
2.1 Base case real power scheduling
The following simple model is used to solve the base case
real power scheduling problem:
X
F U ðP d Þ ¼ min Cpgi ðPgi Þ ð1Þ
i2NG
subject to:
X X
Pgi
P dj ¼ 0
i2NG j2N
0  Pgi  Pgi max
where NG is the set of all generator buses; Cpgi ðPgi Þ is the
real power production ‘cost’ function of the generator at
bus i constructed from generation bids (it is approximated
as a quadratic function Cpgi ðPgi Þ ¼ agi þ bgi Pgi þ cgi Pg2i
under the assumption that the generator bids at its marginal
cost Bpgi ðPgi Þ ¼ bgi þ 2cgi Pgi Þ; Pgi is the generation of
real power

at bus i; N is the set of all of the buses;
P d ¼ Pd1 . . . Pdj . . . PdN is an N-dimensional real
power load vector and its ith element Pdi is the real power
load at bus i; and Pgi ;max is the real power upper limit of the
generator at bus i. The solution
of this model includes the
u
dispatch of real power (P g ¼ Pgu1 . . . Pgui . . . PguNG ),
the minimum real power production cost F U ðP d Þ, and
the market clearing price l (the Lagrange multiplier
associated with the power balance equation) in the
unconstrained case (regarding network limitations).
This base case is used as a reference for an analysis of the
costs of reactive support. The real generation schedule that
is obtained may not be feasible when network constraints
are considered. In this paper, the emphasis is placed on the
reactive power/voltage constraints. The generation of real
power may need to be rescheduled out of merit in order to
satisfy these constraints. In this case the real power problem
and the reactive power problem have to be handled
simultaneously. A modified OPF model is proposed for
this purpose.
2.2 Real and reactive power
co-optimisation model
2.2.1 Objective:
X
F C ðPd ; Qd Þ ¼ min Cpgi ðPgi Þ
i2NG
The objective is to minimise the costs of producing real
power, similar to the objective described in problem (1).
2.2.2 Equality constraints: Two sets of equations,
including the normal and contingent states, governed by
IEE Proc.-Gener. Transm. Distrib., Vol. 152, No. 5, September 2005
Kirchhoff’s laws that characterise flows of power through-
out the system, are expressed as:
  P   
Pgi
Pdi ¼ V_i  V_j Yij  cosðyij þ dj
di Þ
Qgi
Qdi ¼ V_i 
j2N
  P   
V_j Yij  sinðyij þ dj
di Þ
j2N
i
ðnormal stateÞ
  P  0  0   0 
Pgi
Pdi ¼ V_i0  V_j Yij  cos yij þ d0j
d0i
j2N
  P  0  0   0
Q0gi
Qdi ¼ V_i0  V_j Yij  sin yij þ d0j
d0i
j2N

i
ðcontingent stateÞ


where Qd ¼ Qd1 . . . Qdj . . . QdN is the reactive
load vector and its ith element Qdi is the reactive demand at
bus i; Qgi is the reactive power generated at bus i; Yij ffyij is
the element of the admittance matrix; and V_i ¼ Vi ffdi is the
bus voltage at bus i. Variables with superscript represent the
corresponding values in the state of contingency.
2.2.3 Voltage limits:
 
Vi;min  V_i   Vi;max 8i 2 N
 
0
Vi;min  V_i0   Vi;max
0
8i 2 N
where Vi;min and Vi;max are the lower and upper voltage
0 0
limits in the normal state, and Vi;min and Vi;max are the lower
and upper voltage limits in the contingent state. Voltage
requirements in the normal state are normally stricter than
those in the contingent state.
2.2.4 Generator capacity limits:
Pg2i þ Q2gi  Sg2i ;max
Pg2i þ Q02gi  Sg2i ;max
Pgi  0
where Sgi ;max is the maximum apparent power of the
generator at bus i. For the sake of clear illustration, a
simplified circular loading capability diagram is assumed for
a generator and the real power up limit Pgi max is equal to the
maximum apparent power Sgi ;max . It is possible to
incorporate more detailed capability diagrams [15, 16] at
the expense of some mathematical complexities. As the
simple circular model enlarges the feasible operating region
of generators (so is the feasible domain of the OPF
problem), the calculated reactive support cost would be
smaller compared with the case when more detailed
capability diagrams are used.
As a result of the adoption of the preventive control
concept, the real power outputs of generators in the
normal and contingent states are the same. Reactive
power capabilities reserved in the generators are sufficient
to maintain system voltages under a specified contin-
gency. Rescheduled real power generation vector P g ¼
C
 C 
Pg1 . . . PgCi . . . PgCNG and the minimum real produc-
tion costs F C ðP d ; Qd Þ taking voltage security constraints
ð2Þ into account can be obtained. The sequential quadratic
programming (SQP) method [17] is used in this paper to
solve this nonlinear optimisation problem (2).
2.3 Real power settlement and reactive
support costs
As discussed previously, real power generations may be
rescheduled owing to the reactive support requirements.
617
Generators’ profits from real power selling are affected
consequently. It is assumed in this paper that the uniform
bidding curve
market clearing price obtained in the base case is used for
real power settlement, even after generation rescheduling.

Bpg , $/MW
The profit change of generator i can be written as:
h  i h  i
Drgi ¼ lPgUi
Cpgi PgUi
lPgCi
Cpgi PgCi ð3Þ market clearing price

i
The effects of rescheduling on a generator’s profits in real
power market are discussed in the following according to its
base case schedule.
PU
g =0
PC
g
Pg , max
i
0oPgUi oPgi ;max :
i i
2.3.1 In this case, the generator’s bid Pg , MW
price (its marginal cost) at PgUi is equal to the market i

clearing price, l ¼ bgi þ 2cgi PgUi . When the generator is Fig. 2 Lost profit illustration when PgUi ¼ 0
requested by the ISO to decrease its scheduled real power
output (PgCi oPgUi , a constrained-off generator), its income
decrease, lðPgUi
PgCi Þ, will be larger than its real power
production cost decrease, bgi ðPgUi
PgCi Þ þ cgi ðPgUi
PgCi Þ market clearing price
ðPgUi þ PgCi Þ. Conversely, when the unused capacity of a

Bpg , $/MW
generator is called on by the ISO to balance the system real
bidding curve
power insufficiency (PgCi 4PgUi , a constrained-on generator),

i
the increase in real power production cost will be larger
than that in income. Hence the generator’s profit decreases
no matter if the generator is constrained on or off. This is
illustrated in Fig. 1. Shaded areas 1 and 2 represent the
0 PC PU
profit losses in the cases of constrained-off and constrained- g i g i = Pg i , max
on, respectively. Pg , MW
i

Fig. 3 Lost profit illustration when PgUi ¼ Pgi ;max

bidding curve

2 To ensure that the required reactive support is provided,

Bpg , $/MW

the ISO needs to compensate a generator for the

1 market clearing price
opportunity cost making it indifferent whether generating
i

real power or providing reactive power support. The

total amount that the ISO has to pay for reactive support
is the summation of all generators’ opportunity costs,
which is defined
P as the total system reactive support
costs ðFQ ¼ i2NG Drgi Þ. Substituting (3) into this equa-
0 PC PU PC Pg , max P P P
g (off ) g g (on) i tion, FQ ¼ lð i2NG PgUi
i2NG PgCi Þ þ i2NG Cpgi ðPgCi Þ
i i i
P
Pg , MW
i2NG Cpgi ðPgUi Þ. If transmission real losses are neglected,
i
system total real output remains unchanged. It also implies
Fig. 1 Lost profit illustration when 0oPgUi oPgi ;max that it is not necessary to curtail loads to ensure system
voltage security. Hence, the total system reactive support
cost can be expressed as the increase of real power
production cost:
2.3.2 PgUi ¼ 0: The generator can only be requested to  
FQ ¼ F C P d ; Qd
F U P d ð4Þ
increase its output in this case and its generation bid is higher
than the market-clearing price. The real power production The incurred cost FQ of providing reactive support needs to
cost increase is larger than the income increase. The generator be recovered from system users. The reactive support cost
is subject to loss of profits (shaded area in Fig. 2). allocation problem is described in the next Section.

3 Reactive support cost allocation factors

2.3.3 PgUi ¼ Pgi ;max : In this case, the real power output
hits the upper limit. The generator can only be requested to
decrease its output and its generation bid is lower than the
market clearing price. The generator is also subject to loss of
profit (shaded area in Fig. 3) as the real power production
cost decrease is less than that of the income.
It can be concluded that the generator will be subject to
loss of profit (Drgi 40) if its real output has to be
rescheduled away from the base case. This loss of profit
can also be called the generator’s opportunity cost for
reactive support.
618
The procurement of reactive support services and the
calculation of the corresponding cost have been discussed in
the last Section. In the process of procurement, the ISO on
behalf of all consumers buys sufficient reactive support
based on the least cost criterion. Reactive support is
centrally dispatched and treated as a common service from
the perspective of voltage control. Another aspect related to
reactive power management is how to charge different
consumers according to their needs of reactive support
requirement. It is formulated as a joint cost allocation
IEE Proc.-Gener. Transm. Distrib., Vol. 152, No. 5, September 2005
problem in this paper. Joint cost allocation problems arise where qFQ ðtDÞ=qPdi and qFQ ðtDÞ=qQdi are marginal costs
in many contexts in economics and management science. In of real and reactive loads evaluated at demand level
a typical problem that we have in mind, a decision maker t 2 ½0; 1, respectively. To find these two definite integrals
must decide how to allocate the joint cost of service among analytically is very difficult because the analytical expression
several consumers. Both direct consumption of reactive of the integrands will not be available. However, we can
power and indirect consumption of reactive power are obtain the marginal costs at different demand levels by
accompanied by the use of real power need system reactive solving the optimisation problems (1) and (2):
support and consequently have responsibilities to share the
cost. So there should be two prices corresponding to real qFQ ðtDÞ qF C ðtDÞ qF U ðtP d Þ
¼

load and reactive load of a consumer. Furthermore, these qPdi qPdi qPdi
prices must satisfy the cost recovery requirement, i.e. total ¼ p0Pd ðtDÞ þ pPdi ðtDÞ
lðtDÞ ð8Þ
revenue generated from these prices must be able to cover i

the total cost. Given a loading condition, it can be expressed

P qFQ ðtDÞ qF C ðtDÞ
as: FQ ¼ Ni¼1 ðZPdi  Pdi þ ZQdi  Qdi Þ where ZPdi and ZQdj ¼ ¼ p0Qd ðtDÞ þ pQdi ðtDÞ ð9Þ
qQdi qQdi i
are reactive cost allocation factors (prices) for real load and
reactive load at bus i, respectively. The vector form of this where p0Pd ðtDÞ and pPdi ðtDÞ are the Lagrange multipliers
i
equation is: associated with the real power balance equations at bus i of
FQ ¼ D  Z ð5Þ problem (2) in normal and contingent sets at demand level t,
respectively; lðtDÞ is the unconstrained market clearing
where D ¼ ½P d ; Qd  is a 2N-dimensional demand vector
price (real power) at demand level t; p0Qd ðtDÞ and pQdi ðtDÞ
including both real and reactive loads, and Z ¼ ½ZPd ; ZQd  is i

a price vector of the same dimension (ZPd ¼ ½ZPd    are the Lagrange multipliers associated with the reactive
1 power balance equations at bus i of problem (2) in normal
ZPdi    ZPd , ZQd ¼ ½ZQd    ZQdi    ZQd ). and contingent sets at demand level t, respectively. It should
N 1 N
The well-known Aumann-Shapley (A-S) pricing method be noted that as reactive power demands have no effects on
falling in the category of axiomatic approaches is adopted in the base case dispatch (qF U =qQdi ¼ 0), the reactive
this paper to calculate the price vector Z. The A-S method marginal costs in (9) only use information of problem (2).
has been applied to many areas such as transportation cost Hence, a numerical approach can be used to solve (6) and
allocation [18], risk management [19] and transmission (7). Numerical approximations of ZPdi and ZQdi are given by:
loss cost allocation [20]. It has become a powerful tool for K 



joint cost allocation problem due to the properties of 1X 0 k k k
ZPdi ¼ p D þ pP d i D
l D
economic efficiency, fairness and robustness. Billera, Heath K k¼1 Pdi K K K
and Raanan [12] first applied the value concept of
nonatomic games studied by Aumann and Shapley to solve ð10Þ
an internal telephone-billing problem. It has been shown in K 


[13, 14] that A-S prices are uniquely determined by a set of 1X k k
ZQdi ¼ p0Qd D þ pQdi D ð11Þ
neutral, and in a sense equitable, axioms imposed on pricing K k¼1 i K K
methods. The axioms and relevant economic interpretations
are given by: where K is the number of numerical integration steps.
Problems (1) and (2) have to be solved K times sequentially
(A1) (Cost sharing): The generated revenue exactly recovers to calculate A-S prices. Indeed, A-S prices are average
the total cost. marginal costs concerning all level of the demand vector,
(A2) (Additivity): The price based on the total cost function but a constant mix of each component (real loads and
is equal to the sum of the prices computed with respect reactive loads increasing at the same speed) is needed.
to the (additive) components of the cost function. Normally, reactive support cost FQ has a positive relation
(A3) (Positivity): The demand of a consumer that actually with the level of demand. FQ is zero when there is no
contributes to total costs should have non-negative voltage congestion under light loading condition. There
price. exists a critical demand level satisfying FQ ðtDÞ ¼ 0; t  t0
(A4) (Rescaling): A (linear) change in the units of and FQ ðtDÞ40; t4t0 . This is illustrated in Fig. 4 using the
measurement of the demand results in a concomitant test system described in Section 4.
change in the price. Demand changes under t0 have no effect on the reactive
support costs. The interval [t0, 1] is the effective integration
(A5) (Consistency): Two consumers that have the same region. A preliminary identification of t0 would improve the
effect on the cost function should be charged the same computation efficiency in practice. The bisection method,
price. which is commonly adopted in polynomial root finding [21],
is used for this purpose. Given this new start point t0 of
Rather than using game theoretic notions, these axioms are numerical integration, (10) and (11) can be rewritten as:
K h
stated in purely economic terms, hence providing an 1
t0 X
economic justification for using A-S prices. According to ZP d i ¼ p0 ðt0 þ khD
Þ
K k¼1 Pdi
[13], A-S prices related to the real and reactive loads of the i
consumer at bus i can be expressed as: þ pPdi ðt0 þ khD
Þ
lðt0 þ khD

Þ ð12Þ
Z 1
qFQ ðtDÞ
ZPdi ¼ ð6Þ K h i
qPdi
dt 1
t0 X  
0 ZQdi ¼ p0Qd t0 þ khD þ pQdi t0 þ khD ð13Þ
K k¼1 i
Z 1
qFQ ðtDÞ
ZQdi ¼ dt ð7Þ where h ¼ ð1
t0 Þ=K is the step size. Obviously, the
0 qQdi approximation accuracy of these two equations is higher

IEE Proc.-Gener. Transm. Distrib., Vol. 152, No. 5, September 2005 619
 1.6 Table 1: Generator data
1.4 Generator G1 G2 G3

1.2 Bus number 1 2 8

Maximum apparent 2.0 3.0 5.0
1.0
power (p.u.)
FQ , $ (× 103 )

0.8 Active power cost 500Pg1 þ 550Pg2 þ 600Pg3 þ

function ($/h) 300Pg21 400Pg22 450Pg23
0.6

0.4
Table 2: Consumer data
0.2

Consumer Bus number Real load (p.u.) Reactive load (p.u.)

0

L1 3 0.50 0.31
−0.2
0 0.80 t 0 0.90 1.00 L2 4 3.00 1.85
t L3 5 1.50 0.93
Fig. 4 Reactive support cost against demand level L4 7 2.00 1.24

than that of (10) and (11) with the same computation

burden (K times base case and reactive optimisations). Table 3: Transmission-line data

From To Resistance Reactance Susceptance

4 Case study (p.u.) (p.u.) (p.u.)

A simple eight-bus system (Fig. 5) is used for computer 1 3 0.00 0.037
0.06
simulation studies. The system has eight buses and nine tie 1 8 0.00 0.03
0.06
lines. There are three generators and four consumers in 2 4 0.00 0.031
0.06
the system. Their data is summarised in Tables 1 and 2,
2 7 0.00 0.03
0.06
respectively. Table 3 gives transmission line data. The
system base capacity is 100 MVA. The voltage limits are 3 5 0.00 0.037
0.06
all within the range 0:95  Vi  1:05 (p.u.) 4 5 0.00 0.05
0.06
(i ¼ 1; 2; . . . ; 8). Bus 8 is selected as the angle reference 4 8 0.00 0.03
0.06
bus. The base (unconstrained) case real power dispatch is 6 7 0.00 0.03
0.06
given in Table 4. 6 8 0.00 0.015
0.06
When reactive/voltage constraints are taken into account,
the generators’ real power outputs will need to be adjusted
owing to system voltage control requirements both in Tables 5 and 6, respectively. The voltage profile shows that
normal and contingent states. The preventive control no bus voltage is constrained in the normal state. But in the
concept is adopted in this paper and only one potential contingency case, the voltage set point of G3 at bus 8 has to
contingency is considered. In this example, only the outage be adjusted to its upper limit and the voltage at bus 5 hits its
of the line 4–8 will force the base case real power dispatch to lower limit.
be rescheduled owing to the voltage constraints. The voltage Adjusted real power generations are shown in column 2 of
profile and generation rescheduling results are given in Table 6. This column shows that G1 and G2 have to
decrease their real power output for the provision of reactive
power support. Owing to the no-load curtailment assump-
L1 tion, the most expensive generator G3 has to generate more
G3 G1
than its base case schedule. All three generators are subject to
losses of profit as listed in the last column. The total real
8 1 3 power production costs (Fc) rises up to 11.871  103($/h).
Using (4), the system reactive support cost is 1509.88 $/h,
which will be recovered from consumers. Column 3 shows
6 5 the corresponding reactive capabilities given the real power

L3

Table 4: Base case real power dispatch

Real power generation (p.u.) Total real Market clearing

7 2 4 production price
cost FU (103$/h) l(103$/p.u.h)
L4 G2 G1 G2 G3
L2

2.000 2.676 2.323 10.361 2.69

Fig. 5 Eight-bus test system

620 IEE Proc.-Gener. Transm. Distrib., Vol. 152, No. 5, September 2005
Table 5: Voltage profile

Bus number 1 2 3 4 5 6 7 8

Normal voltage (p.u.) 1.0165 1.0000 0.9780 0.9633 0.9516 1.0025 0.9692 1.0188
Contingent voltage (p.u.) 1.0402 1.0133 0.9868 0.9507 0.9500 1.0338 1.0042 1.0500

Table 6: Generation rescheduling results

Generator Real power output (p.u.) Reactive power capability (p.u.) Reactive power output (p.u.) Loss of profit ($/h)
normal contingency

G1 1.588 71.215 0.961 1.215 459.01

G2 1.801 72.399 0.438 2.399 306.06
G3 3.610 73.459 3.070 1.471 744.81

generations listed in column 2. Reactive outputs in normal 7

MCP: marginal cost of real loads
and contingent cases are shown in columns 4 and 5, MCQ: marginal cost of reactive loads
respectively. For G1, its reactive output is 0.961 p.u. in the 6 MCQL3
normal case and hence its reserved reactive capacity is
5
0.254 p.u. This reserve would be used up if a line outage MCQL2
happens. A similar phenomenon applied to G2. But the $/p.u.h( ×103 ) 4
reactive output of G3 in the contingent case is less than that MCQL1
in the normal case. It is a result of reactive support co- 3 MCPL2
ordination among the three generators. MCPL3
Following the cost allocation procedure described in 2 MCQL4
Section 3, the bisection method is first used to determine the MCPL1
critical demand level t0. With a computation tolerance 1 MCPL4
e ¼ 0.01, t0 is found to be 0.875 after seven iterations. Then,
0
the demand vector is divided into K (K is set as 200 in this 0.86 0.88 0.90 0.92 0.94 0.96 0.98 1.00
example) equal parts in the effective demand level interval t
[0.875,1]. Costs allocation results are given in Table 7.
Fig. 6 Illustration of real and reactive marginal costs

Table 7: Costs allocated to consumers

Consumers Allocation factor Cost allocated

($/h) 5 Conclusions
Real load Reactive load
($/MWh) ($/MVarh)
Reactive support from generators is critical to voltage
security. Technical and economic issues related to this
L1 0.7708 1.8536 96.00 ancillary service have been discussed and a framework for
L2 1.1332 2.1247 733.04
allocating reactive support cost has been established in this
paper. Real generation rescheduling is used as a preventive
L3 1.0798 2.9351 434.93
control to maintain feasible system voltages both in normal
L4 0.7468 0.8601 256.01 and emergency states. Reactive opportunity cost of an
individual generator is evaluated by its real profit loss after
real generation adjustment. The total reactive support cost
is deduced as the real power production cost increase caused
It can be observed from Table 7 that allocation factors of by generation rescheduling under unchanged system total
reactive loads (column 3) are larger than that of real loads real power output and real power market clearing price
(column 2) for all four consumers. It means that the direct assumptions. Reactive charges for consumers are calculated
reactive consumption of each consumer is responsible for a using the Aumman-Shapley cost allocation method and the
larger part of the reactive support costs than the indirect bisection search algorithm is used to improve the computa-
consumption (reactive loss) caused by its real load. The tion efficiency. The numerical example shows that the
locations of consumers determine their different effects on proposed pricing scheme ensures the total revenue from
reactive support costs. In this example, the largest allocation loads is equal to the total reactive support cost. A-S prices
factor among real loads appears at bus 4 (consumer L2) are also economically efficient as the economic signal
whereas the largest allocation factor among reactive loads provided by marginal cost is included. For the application
appears at bus 5 (consumer L3). This is illustrated more of the proposed methodology to practical power markets,
clearly in Fig. 6. Summation of the costs allocated to detailed generator capability models, reactive power and
different consumers
 (column 4) is 1519.98
($/h). The voltage controls from transmission system (SVCs and
allocation error 1519:98
1509:88
1509:88
 ¼ 0:0067 is less than 1%. capacitors) and bilateral transactions modelling may need
This computation error will decrease if a larger K is used. to be considered.

IEE Proc.-Gener. Transm. Distrib., Vol. 152, No. 5, September 2005 621
6 Acknowledgment
Supports from the Research Committee of the Hong Kong
Polytechnic University and Research Grants Council of
Hong Kong (5215/03E) are gratefully acknowledged.
7 References
1 Caramanis, M.C., Bohn, R.E., and Schweppe, F.C.: ‘Spot pricing of
electricity: practice and theory’, IEEE Trans. Power Appar. Syst.,
1982, 101, (9), pp. 3234–3245
2 Baughman, M.L., and Siddiqi, S.N.: ‘Real time pricing of reactive
power: theory and case study results’, IEEE Trans. Power Syst., 1991,
6, (1), pp. 23–29
3 Ei-Keib, A.A., and Ma, X.: ‘Calculating short-run marginal costs of
active and reactive power production’, IEEE Trans. Power Syst., 1997,
12, (2), pp. 559–565
4 Baughman, M.L., Siqqiqi, S.N., and Zanikau, J.W.: ‘Advanced
pricing in electrical systems Part II: implications’, IEEE Trans. Power
Syst., 1997, 12, (1), pp. 496–502
5 Lamont, J.W., and Fu, J.: ‘Cost analysis of reactive power support’,
IEEE Trans. Power Syst., 1999, 14, (3), pp. 890–896
6 Dai, Y., Ni, Y.X., Wen, F.S., Han, Z.X., and Wu, F.F.: ‘A study of
reactive power marginal price in electricity market’, Electr. Power Syst.
Res., 2001, 57, (1), pp. 41–48
7 Hao, S., and Papalexopoulos, A.: ‘Reactive pricing and management’,
IEEE Trans. Power Syst., 1997, 12, (1), pp. 95–104
8 Hao, S.: ‘A reactive power management proposal for transmission
operators’, IEEE Trans. Power Syst., 2003, 18, (4), pp. 1374–1380
9 Bialek, J.J.: ‘Topological generation and load distribution factors for
supplement charge allocation in transmission open access’, IEEE
Trans. Power Syst., 1997, 12, (3), pp. 1185–1193
10 Wu, F.F., Ni, Y.X., and Wei, P.P.: ‘Power transfer allocation for open
access using graph theory F fundamentals and applications in
622
systems without loopflow’, IEEE Trans. Power Syst., 2000, 15, (3),
pp. 923–929
11 Chu, W.C., Chen, B.W., and Liao, C.H.: ‘Allocating the costs
of reactive power purchased in an ancillary service market by modified
Y-Bus matrix method’, IEEE Trans. Power Syst., 2004, 19, (1),
pp. 174–179
12 Billera, L.J., Heath, D.C., and Raanan, J.: ‘Internal telephone billing
rates-a novel application of non-atomic game theory’, Oper. Res.,
1978, 26, (6), pp. 956–965
13 Billera, L.J., and Heath, D.C.: ‘Allocation of shared costs: a set of
axioms yielding a unique procedure’, Math. Oper. Res., 1982, 7, (1),
pp. 32–39
14 Mirman, L.J., and Taumam, Y.: ‘Demand compatible equitable cost
sharing prices’, Math. Oper. Res., 1982, 7, (1), pp. 40–56
15 Rosehart, W.D., Canizares, C.A., and Quintana, V.H.: ‘Effect of
detailed power system models in traditional and voltage-stability-
constrained optimal power-flow problems’, IEEE Trans. Power Syst.,
2003, 18, (1), pp. 27–35
16 Rider, M.J., and Paucar, V.L.: ‘Application of a nonlinear reactive
power pricing model for competitive electric markets’, IEE Proc.,
Gener. Transm. Distrib., 2004, 151, (3), pp. 407–414
17 Gill, P.E., Murray, W., and Wright, M.H.: ‘Numerical linear
algebra and optimization’ (Addison Wesley, Old Tappan, 1991),
Vol. 1
18 Samet, D., Tauman, Y., and Zang, I.: ‘An application of the Aumann-
Shapley prices for cost allocation in transportation problems’, Math.
Oper. Res., 1984, 9, (1), pp. 25–42
19 Denault, M.: ‘Coherent allocation of risk capital’, J. Risk., 2001, 4, (1),
pp. 7–21
20 Tan, X.H., and Lie, T.T.: ‘Allocation of transmission loss cost using
cooperative game theory in the context of open transmission access’.
Proc. IEEE Power Engineering Society Meeting, Columbus, OH,
USA, 2001, Vol. 3, pp. 1215–1219
21 Bachanan, J.L.: ‘Numerical methods and analysis’ (McGraw-Hill,
New York, USA, 1992), pp. 49–54
IEE Proc.-Gener. Transm. Distrib., Vol. 152, No. 5, September 2005