1396

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 21, NO. 3, AUGUST 2006

Minimization of Imbalance Cost Trading Wind Power

on the Short-Term Power Market
Julija Matevosyan and Lennart Sder, Member, IEEE

AbstractPresent power markets are designed for trading

conventional generation. For wind generation to participate in a
short-term energy market, lengthy wind power production forecasts are required. Although wind speed forecasting techniques
are constantly improving, wind speed forecasts are never perfect,
and resulting wind power forecast errors imply imbalance costs for
wind farm owners. In this paper, a new method for minimization
of imbalance costs is developed. Stochastic programming is used
to generate optimal wind power production bids for a short-term
power market. A Wind power forecast error is represented as a
stochastic process. The imbalance costs resulting from this strategy
are then compared to the case when wind power production bids
on a short-term power market are based directly on a wind speed
forecast.
Index TermsForecasting, optimization, power market, stochastic process, stochastic programming, wind power generation.

I. INTRODUCTION

HE PRESENT short-term power markets are designed for

trading conventional (dispatchable) generation. The time
span after the market clearing until the delivery hour can be up
to 36 h (Scandinavia) and up to 38 h (Spain); any deviations
from the submitted production plan are penalized. In order to
avoid high imbalance penalties, many different techniques are
used to forecast wind energy generation (see [1] for a review).
The forecasting techniques have considerably improved over the
last decade but never will give perfect results. Furthermore, the
forecast error increases with increasing prediction horizon. With
forecast being needed 36 or 38 h ahead of time, wind generators
may be subjected to substantial imbalance costs.
In the Nordic countries, all bulk electricity production must
be traded through a balance responsible player. For a wind
power owner, there are, thus, three options available. One
option is to become a balance responsible player, second is
to trade wind power and have a contract with the balance
responsible for balancing any mismatches, and third is to sell
all wind power to the balance responsible player [4]. In the
first case, the wind power owner is paying a market imbalance
price for its imbalances. In the other two cases, the wind power
owner is paying a contracted imbalance price, which is lower
than the market imbalance price, due to the fact that the balance
responsible player has other energy sources in its portfolio and
can balance wind power mismatches internally.
Manuscript received August 25, 2005; revised December 28, 2005. This work
was supported by the Swedish Energy Agency. Paper no. TPWRS-00544-2005.
The authors are with the Royal Institute of Technology, Stockholm, Sweden
(e-mail: julija@ets.kth.se; lennart.soder@ets.kth.se).
Digital Object Identifier 10.1109/TPWRS.2006.879276

When trading wind power on the short-term power market,

the balance responsible player may choose the following
strategies.
1) Assume wind power forecast as certain and bid it on the
market.
2) Bid the amount that minimizes expected costs for imbalances considering possible scenarios of wind power production and imbalance costs.
The latter strategy has been previously studied in [2]. In [2],
power production scenarios are generated using statistical
methods based upon historically observed power production.
The energy output of the wind farm (WF) is divided into several
energy bands and, given the band, where the WF production
resides initially, the probabilities that the WF production resides
in each of energy bands are calculated for each forecasting
delay and form probability tables. There is, however, a difficulty with generating these probabilities as they require a
significant quantity of historical data. The drawbacks of this
method, reported in [2], are as follows: one year of data is not
enough to produce smooth probability tables, and, since wind
exhibits seasonal behavior, probability tables are likely to differ
between seasons and years.
In this paper, bidding strategy 2 is studied further. Here wind
power production scenarios are generated based on day-ahead
wind speed forecast and statistical data about the forecast error.
Any forecasting method can be used. The statistical forecast
error data are required. The advantage of this method is that
the forecast error is modeled and added on top of actual dayahead wind power production forecast that already accounts for
seasonal winds. One year of the forecast error data proved to be
sufficient to calibrate the forecast error model.
The imbalance prices are assumed deterministic in [2],
whereas in this paper, the imbalance prices are represented
by several equally probable scenarios in the planning stage,
and, in the imbalance cost calculation stage, actual imbalance
prices are used. This allows more realistic evaluation of bidding
strategy 2.
The method for minimization of imbalance costs suggested
in [2] is applicable only for a small amount of scenarios. In this
paper, bidding strategy 2 is formulated as a stochastic optimization problem. Mixed integer programming is used here to obtain
the solution for large number scenarios for wind power production and imbalance price. The proposed method can be used to
determine the energy level to bid on the day-ahead market,
to minimize the imbalance costs of the WF;
estimate the value of using better forecasting techniques.
The imbalance cost minimizing strategy is explained with an
example of the Nordic power market, but with some minor adjustments, it can be applied in any market.

0885-8950/$20.00 2006 IEEE

MATEVOSYAN AND SDER: MINIMIZATION OF IMBALANCE COST TRADING WIND POWER

1397

This paper is organized as follows. A short overview of the

Nordic power market is provided in Section II. Section III provides modeling details: wind farm model for conversion of wind
speed forecast into power production forecast; brief description
of the forecast error model; and stochastic optimization program for bidding strategy 2 together with some helpful hints
for problem solution and application of the optimization results.
The detailed derivation and analysis of the solution to the optimization problem is provided in Section IV. Bidding strategy
2 is compared to strategy 1 in a case study; the results are presented in Section V. Section VI summarizes the main conclusions and plans for future work.
II. OVERVIEW OF THE NORDIC POWER MARKET
A. Spot Market
On the Nordic power market, Norwegian, Finnish, Swedish,
and Danish actors trade in hourly contracts for the 24 h of the
coming day. The spot market is closing at 12:00 the preceding
day. Purchasing and selling curves are constructed, and the point
where they cross determines the spot market price and the volumes being traded during each hour the coming day [3]. With
such market structure, wind power forecast length should be
1236 h to bid on the spot market.
B. Elbas
Elbas is an adjustment market for the power exchange players
in Sweden, Finland, and East Denmark. The market opens for
trade for the coming power exchange day at 15:00 each day, i.e.,
after the spot market has closed. Trading in hourly contracts is
conducted electronically and can take place up until one hour
prior to delivery.
Trading on Elbas is more beneficial for wind power because
more accurate forecasts can be made for the shorter time horizon
[4]. However, currently, Elbas is not very active, and only small
amounts of energy are traded there. Nevertheless, the bidding
strategy described in this paper may be easily used also for
trading on the Elbas market.
C. Regulating Market
The transmission system operator (TSO) is responsible for
physical balance between production and consumption. Prior
delivery hour actors with power reserves are placing bids for fast
(up to 10 min) production increase or decrease to the so-called
regulating market. The bids are arranged in order of price and
form a staircase for each delivery hour. At the end of each
hour, the regulation price is determined in accordance with the
most expensive upward regulation measure that was taken by the
TSO, or the cheapest down regulation measure that was taken
by the TSO [3].
Wind power could also be bidden for upward regulation if
a certain production margin is kept on the WF for this purpose (see, e.g., [5]), but for simplicity, this possibility is disregarded in this paper, as participation in downwards regulation
means power spillage for the wind power owner (power cannot
be stored to the next hour). This option is, thus, not economically efficient.

Fig. 1. Flowchart for the bidding strategy and its evaluation.

D. Balance Settlement
Via balance settlement, the TSO distributes the costs of regulation among balance responsible actors on the power market.
All balance responsible actors pay or are getting paid for their
unplanned deviations from the balance.
If upward regulation alone was activated, the upward regulation price is paid by players with negative imbalance (i.e.,
actual production purchase actual load sold power), while
players with positive imbalance are getting paid according to a
spot price.
If downward regulation alone was activated, the downward
regulation price is paid to players with positive imbalance, while
players with negative imbalance pay according to the spot price.
If no regulation took place, all actors are settled at spot price.
If both upward and downward regulation have been ordered
depending on which regulation had higher volume, upward or
downward regulation price is applied. If volumes for ordered upward and downward regulation are equal, the spot market price
is used [3].
The balance responsible player further distributes imbalance costs among power producers/consumers in his area of
responsibility.
III. BIDDING STRATEGY
The flowchart for the presented bidding strategy is given in
Fig. 1 and is discussed step by step in this section.
A. ARMA Forecast Error Scenarios
A wind speed forecast for the next day is obtained from, e.g.,
numerical meteorological programs (box A1). A wind speed
forecast is never perfect, and the forecast error should be considered when placing a bid to the spot market.
A model for wind speed forecast error (box A2) is developed
in [6]. It is assumed that data concerning accuracy of the forecast are known. Wind speed forecast is assumed available for

1398

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 21, NO. 3, AUGUST 2006

the studied site. The model then simulates possible outcomes

(scenarios) of the wind speed forecast errors that have the correct statistical behavior. The model is based on autoregressive
moving average series (ARMA), [7], defined as

(1)
where
is a wind speed forecast error in -hour forecast,
is a random Gaussian variable with standard deviation ,
and and are parameters.
The wind speed scenario for each hour
can then be caland the wind
culated as the sum of the wind speed forecast
, i.e., the outcome of
speed forecast error scenario

(2)

Fig. 2. Power curve of 2000-kW wind turbine ( ) and its continuous approximation (solid line).

where is number of forecast error scenarios.

, , and are identified using the leastThe parameters
square fitting, minimizing a difference between sample forecast
error variance, based on data from the site, and modeled forecast
error variance (see [6] for details).
Obtained wind speed scenarios should, thus, be converted to
power (box B; see Fig. 1).
B. Wind Speed to Ower Conversion
Power production of the wind farm depends on wind speed,
wind direction, wind farm layout, and availability of the wind
turbines within a wind farm.
1) Wake Effect Model: As the air passes through the wind turbine rotor disc, the speed of the airflow downstream is reduced.
As a result, not all wind turbines within a wind farm meet the
same undisturbed wind speed. The wind turbine wake model on
the flat terrain is assumed as follows [8]:

Fig. 3. Example of power curve of the 160-MW wind farm (80 2-MW wind
turbines) with (solid line) and without (dashed line) consideration of the wake
effect according to equation (3). Note that here, all wind directions were assumed equally probable.

(3)
where
is wind speed in the wake,
is undisturbed wind
speed at the upwind turbine with rotor radius , is a horizontal distance between wind turbines, is wake decay constant
onshore,
offshore [9]), and
is a thrust
(
coefficient that depends on wind turbine (WT) type and can be
obtained from the respective WT manufacturer.
2) Wind Farm Model: The expression for power production
of the wind turbine depending on wind speed is [10]

(4)
where is swept area of the WT rotor, is air density, and
is overall efficiency of the WT, expressed here as a function of
the wind speed.
In practice, the relation between WT power production and
a wind speed for each WT type is given by a so-called power
curve, which is a set of experimentally obtained values available

from WT manufacturer (see Fig. 2, stars). It is convenient to approximate the power curve with continuous function, to be able
to calculate power output at any wind speed. This is achieved
is expressed from (4); then, substituting
as follows. First,
with discrete, experimental values,
is calculated for
corresponding wind speeds. Approximating obtained discrete
by continuous piecewise linear function [11], and
values
substituting it back to (4), an experimental power curve can be
approximated by a continuous function that will allow us to determine the power production at any wind speed. As shown in
Fig. 2 (solid line), the approximation is quite good.
Given a wind speed forecast for undisturbed wind flow for
, wind speed scenarios
for undisturbed wind
hour ,
flow are calculated using (2). Then, given the wind direction,
within a wind
wind speed scenarios at each wind turbine
farm are calculated by (3) (see Fig. 3). Power production scenarios of each WT can be obtained using continuous approximation of the WT power curve described above. Power production
scenarios of individual wind turbines are summed to represent

MATEVOSYAN AND SDER: MINIMIZATION OF IMBALANCE COST TRADING WIND POWER

the power production scenario of the wind farm,

where is the set of all considered scenarios.

C. Scenario Reduction
The computational effort for solving scenario-based optimization models depends on the number of scenarios.
Therefore, it is necessary to obtain the subset of wind power
production scenarios that has a smaller number of scenarios
but still is reasonably close to the original set. The scenario
reduction approach that is used here (box C) is described in
detail in [12]. The scenario reduction algorithm determines a
subset of preserved scenarios that is the closest to the original
set of scenarios using the Kantorovich metric. The distance
trades off scenario probabilities and the distances between
scenario values.

1399

where is a variable, and the rest are parameters. is a bid of

the WF operator to the spot market, is wind power production
according to scenario , is probability of wind speed scenario
,
is probability of the imbalance price scenario , and
is spot price forecast. The terms
are equivalent to
of the given set of price scenarios .
mean imbalance price
There is no coupling between subsequent hours of operation
of WF; the optimization problem can be solved separately for
each hour. This also minimizes computational efforts.
For the small number of the production scenarios, the optimization problem (5) can be solved analytically, as in [2]. With
ARMA series, thousands of scenarios can be generated; this improves the solution but makes it impossible to solve (5) by hand.
In this paper, mixed integer programming is applied to solve (5);
thus, the hourly optimization problem is expressed as follows
(box D):

D. Optimization Problem
As discussed in Section II, the imbalance price can be different depending on if the balance responsible actor is in positive or negative imbalance and if upward or downward regulation was undertaken by the system operator during the hour in
question.
As an illustration, in this paper, the imbalance prices from the
previous days are used as price scenarios for the day in question
(see Fig. 1, box D1). The price model does not effect the optimization method presented in this paper. The analysis of the optimization problem, provided in Section IV, is general and does
not depend on the chosen price model. In the real application of
this bidding strategy, the stochastic model (see, e.g., [15]) of the
regulating prices can be used to achieve better results.
The following imbalance price model is used:
, where
is a spot price at hour

, if the actor is in positive imbalance in hour and no

downward regulation is undertaken;

, where
is a price for downward
regulation at hour , if the actor is in positive imbalance in
hour and downward regulation is undertaken with
;

, where
is a price for upward
regulation at hour , if the actor is in negative imbalance in
hour and no upward regulation is undertaken or
;

, if the actor is in negative imbalance in hour and upward regulation is undertaken with
.
The negative sign means that the balance responsible player
is paying the price for imbalance, and the positive sign means
that the balance responsible player is getting paid.
Note that the wind power producer is assumed to be a price
taker here; it means its bidding strategy does not affect power
prices. In countries with small amounts of wind power (less than
5% of total power production according to the study in [16]), this
is a reasonable assumption.
The stochastic optimization problem is formulated for each
hour to maximize WF owners expected profit and, consequently,
minimize the imbalance costs

(5)

(6)
is the total installed capacity of the WF, is a biwhere
nary variable,
is a large positive number that exceeds any
maximum feasible value of , and is a set of preserved scenarios after scenario reduction (see Section III-C).
, if
it is optimal to keep
(underproduction), and the third
, if it is
term of the objective function would disappear;
optimal to keep
(overproduction), and the second part
of the objective function would vanish.
The detailed derivation and analysis of the solution to the
optimization problem (6) is provided in the next section. Also,
several special cases are analyzed.
E. Evaluation of Bidding Strategy
The suggested bidding strategy can be tested against actual
data for the same period, for which forecast error statistics and
imbalance prices are available. Real wind speed measurements
(box E1) should be converted to power following the same procedure as wind speed scenarios above (box B, see Fig. 1). If
actual power production measurements are available, no conversion is needed. The actual power production data and bids,
obtained from the optimization, are then used to calculate imbalance costs (box E)

(7)
and
are actual imbalance prices,
where
known ex-post for the studied hour for the balance settlement
(see Section II-D). As it will be shown in the case study, the
difference between the expected imbalance prices and actual

1400

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 21, NO. 3, AUGUST 2006

imbalance prices influences the total profit from the suggested

strategy.

and, taking into account that

if
if
if

F. GAMS/MATLAB
The developed model has been implemented in the General
Algebraic Modeling System (GAMS) [13], which is specifically designed for modeling linear, nonlinear, and mixed integer
optimization problems. The scenario reduction algorithms described in [12] have been contained in the library SCENRED
since 2002. The DICOPT GAMS solver was used to solve the
optimization problem. In order to obtain optimal bids for the
day-ahead spot market, the stochastic mixed integer optimization problem should run 24 times, separately for each hour of
the day. Obtained results could then be used for further calculations, e.g., calculation of imbalance costs or testing against historical data. It is convenient to perform such calculations using
GAMS/MATLAB interface [14].
IV. ANALYSIS OF THE OPTIMIZATION PROBLEM
In this section, the general solution for the optimization
problem (6) is derived. The solution is also analyzed in detail
for several special cases. Let us, for simplicity, introduce the
,
,
,
following notation:
,
,
, and
; then (6) can then
be rewritten as

Let us denote with the index each combination of the solutions

of (8),
,
. Then, from (8),
, where

, we get

(11)

In general case, for any fixed

,
is a linear function [see (10)].
, and, thus, maximum is attained at the
ends of the segment
if
if
if
(12)
would lead to optimal
and
. Several special cases can
be analyzed in detail.
Case 1)
Substituting
and
with
in (9) and after sim,
plification, the objective function is
. Thus, the obwhere is the mean value of
jective function does not depend on or , and these
variables can take any values allowed by (8).
Case 2)
and
, since
and
; thus, from (11),
.
Case 2a) If
, then, from (11),
; as
, we
. Thus,
can be chosen
must have all
arbitrary from
as follows from (8).
The maximal value of the objective function (9)
.
in this case is
,
would lead to a
Case 2b) If
higher value of (9). From (12),
;
.
thus, from (8),
From the latter relationship, it follows that
can be chosen arbitrary from {0,1} if
and
if
.
The maximal value of the objective function (9)
.
is in either case
Summarizing, the optimal solutions are
arbitrary

(8)
or

arbitrary

if
if

Let us rewrite the objective function as follows:

lead to the same maximum of the objective

, when
function
.
(9)

Case 3)
. Since
, we have

Denote

and
. Thus

(10)

and
.

Case 3a) If
,
. Since
then from (11), we have
and
, all
, and thus,
and can be chosen arbitrary from
all
as follows from (8).

MATEVOSYAN AND SDER: MINIMIZATION OF IMBALANCE COST TRADING WIND POWER

1401

The maximal value of the objective function (9)

.
in this case is
, then from (12), we
Case 3b) If
have
. In this case, all
cannot
.
be 1. From (12), it follows that
according
Thus,
to (8);
.
As
in
the
general
case,
would lead to opand
.
timal
Then,
should be compared
, and the optimal solution could then
with
be chosen.
V. CASE STUDY
The bidding strategy has been applied in case studies. Real
wind speed data and forecasted wind speeds for 2003 for East
Denmark were available. For each hour of the year, there is one
wind speed and one wind speed forecast. Note that these are
accumulated wind speeds for all of East Denmark, and thus,
the smoothing effect due to geographical dispersement of WT is
included in the series. For a single wind farm, a larger forecast
error could be expected.
The data flow is as follows: at 6:00, the Danish Meteorological Institute delivers a forecast for wind speeds in Denmark.
Elkraftsystem gets the forecast at 10:00 since it takes 3 h to calculate forecasts. The forecasts are used in order to put bids to
Nordpool before 12:00 for all hours for the coming day. This
means that the forecast for hour 0:00 of the coming day is an
18-h forecast and for hour 23:00a 41-h forecast.
Based on these data, the standard deviation of the forecasts for
24 different hours was calculated. By using the method defined
in [6] and briefly described in Section III-A, it is possible to
estimate the parameters of the ARMA series that minimize the
difference between the standard deviation of the actual forecast
,
error and the forecast errors of the ARMA series:
, and
. The standard deviation of the
actual forecast error and the forecast error of the ARMA series
are shown in Fig. 4. In this case study, 1000 ARMA forecast
error scenarios were generated for each simulation.
Six days from 2003 are tested, January 14 and 31, March 20,
April 27, September 23, and December 28. During these days,
according to data, the forecast errors were large. Spot and regulating prices from respective hours of 2003 are used. Imbalance
prices from days before and after the tested day were used as
price scenarios. The results of the bidding strategy are compared
against the alternative strategy, where the wind power forecast
is bidden on the spot market.
Wind direction data and data about WF layouts were not
available; therefore, the characteristics of one V80 (Vestas, 2
MW) wind turbine is used to test the bidding strategy.
Only a demo version of the GAMS solver DICOPT with a
limited number of variables and constraints was available; therefore, 1000 forecast error scenarios were generated and then reduced to 50 scenarios, which still corresponds to 72% of the
information from the original scenario set.

Fig. 4. Standard deviations of the real wind forecast error ( ) and standard
deviations from the ARMA series with  = 1:0073,  = 0:0327, and  =
0:1372 (solid line).

Fig. 5. Power production bid, according to cost minimizing bidding strategy,

real power production, and wind power forecast according to data for January
14, 2003 (top); profit for imbalance cost minimizing strategy and bidding forecast strategy (bottom).

The results for one of the simulated days are shown in Fig. 5.
Fig. 5 (top) shows bidden power, according to cost minimizing

1402

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 21, NO. 3, AUGUST 2006

Fig. 6. Power production bid, according to cost minimizing bidding strategy,

real power production, and wind power forecast according to data for January
31, 2003 (top); profit for imbalance cost minimizing strategy and bidding forecast strategy (bottom).

Fig. 7. Power production bid, according to cost minimizing bidding strategy,

real power production, and wind power forecast according to data for September
23, 2003 (top); profit for imbalance cost minimizing strategy and bidding forecast strategy (bottom).

strategy, real power production, and wind power forecast according to data for January 14, 2003. The expected (average)
imbalance prices and the spot price are also placed on the same
figure. Fig. 5 (bottom) illustrates profits for the suggested bidding strategy and for the case when the wind power forecast is
bidden on the spot market. The profit is calculated as income
from selling wind power on the spot market minus imbalance
costs (see Section III-E). Imbalance costs are calculated based
on actual imbalance prices that are placed on the same figure.
For the first hour expected
, the expected value
of the objective function (6) is equal to
, i.e., does not
depend on bidden wind power , that can be equal to any value
between
and
(see Case 3(a) in Section IV). The actual imbalance prices are both equal to the spot price. The profit
is
according to (7) and is, therefore, equal for both
bidding strategies. Between the 2nd and the 6th hour, the forecasted wind power is less than actual wind power production. As
, the profit from the biding forecast strategy is
according to (7). For the imbalance cost minimizing
strategy, the bidden levels are chosen in accordance with the
general solution derived in Section IV in (12). The bidden power
is higher than real wind power production; thus, the hourly profit

from this strategy is

from (7).
This value is lower than the profit from the bidding forecast
strategy
, because the actual difference between the spot
price and the negative imbalance price is higher than expected.
, and thus, the bids are chosen
In hours from 7 to 19,
in the same way as for the first hour. The actual profit from the
imbalance cost minimizing strategy is higher than or equal to
the biding forecast strategy, in hours when
,
and lower than the profit from the bidding forecast strategy, in
hours when
. The latter result is due to the fact
that actual difference between the spot price and the negative
imbalance price is higher than was expected. In the hours from
19 to 24, the expected imbalance prices coincide with the actual imbalance prices. Due to this reason and as the imbalance
cost minimizing strategy accounts for the forecast uncertainty,
the profit from this strategy is higher than for the biding forecast
strategy.
The results for other simulated days are in accordance with
general or special case solutions from Section IV. It can be
concluded that the profit from the imbalance cost minimizing
strategy is higher than or equal to the profit from the biding
forecast strategy (all hours on January 31 (see Fig. 6), March

MATEVOSYAN AND SDER: MINIMIZATION OF IMBALANCE COST TRADING WIND POWER

1403

TABLE I
PROFIT FOR TWO BIDDING STRATEGIES

Fig. 8. Power production bid, according to cost minimizing bidding strategy,

real power production, and wind power forecast according to data for December
28, 2003 (top); profit for imbalance cost minimizing strategy and bidding forecast strategy (bottom).

20, and April 27). The exceptions are cases where the difference
between imbalance price and spot price is other than expected,
e.g., hour 13 and hour 12 on September 23 (see Fig. 7) and
hours 13 and 5 on December 28 (see Fig. 8).
From the results in Figs. 58 and from the solution derived in
Section IV (cases 1 and 3), it follows that in some situations, installed capacity is bidden in imbalance cost minimizing strategy,
although the forecasted power production is not that high, and
the actual power production turns out to be very low as well.
This is because the wind farm is assumed to be a price taker, and
its behavior does not affect the regulating market prices. From
the derivation in Section IV, it follows that in cases 1 and 3, the
bid can be any value in the given range. One can, for example,
choose the lower limit in order to cause less imbalance in the
system.
Profit evaluation is also made for all of January 2003, and
results are shown in Table I. One can see that during some days,
bidding of forecasted power gives higher profit than imbalance
costs minimizing strategy. As in the examples above, this is due
to the fact that the difference between imbalance price and spot
price is other than expected.

VI. CONCLUSIONS AND FUTURE WORK

In this paper, a new bidding strategy minimizing imbalance
costs of a wind farm was described. The bidding strategy was
simulated for one Vestas V80 wind turbine using real wind
speed data and wind speed forecasts from Denmark.
The results of simulations have shown that imbalance cost
minimizing bidding strategy generally results in higher or equal
profits than the strategy where a player bids the forecasted wind
power production and then pays imbalance costs. The calculation of optimal bids with suggested strategy does not require any
additional data inputs and is fast with regards to computational
efforts. The bidding strategy was designed having the Nordic
electricity market in mind; however, it can be easily adjusted to
fit other markets, e.g., with a different market closure delay. The
following improvements can be made in the future.
As the outcome of bidding strategy is sensitive to the
choice of imbalance price scenarios, in the future, they
will be substituted by a stochastic price model as in [15].
A similar stochastic model can be used, even for spot
prices.
Here, the wind power producer is assumed to be a price
taker, but as it can be seen from the simulation results, the
use of this bidding strategy may lead to higher imbalances
in the system and thus will affect imbalance prices. This
effect should also be considered in the future.

1404

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 21, NO. 3, AUGUST 2006

REFERENCES
[1] D. Milborrow, L. Harrison, J. Jackson, S. Knight, N. Martin, M.
OBryant, and M. McGovern, Forecasting output, Wind Power
Monthly, pp. 3746, Dec. 2003.
[2] G. N. Bathurst, J. Weatherill, and G. Strbac, Trading wind generation
in short term energy markets, IEEE Trans. Power Syst., vol. 17, no. 3,
pp. 782789, Aug. 2002.
[3] Svenska Kraftnt, The Swedish Electricity Market and the Role
of Svenska Kraftnt, Sweden, Dec. 2003. [Online]. Available:
http://www.svk.se/upload/3187/elmarkn_eng_2001_20031212.pdf.
[4] H. Holttinen, The impact of the large scale wind power production on
the Nordic electricity system, Ph.D. dissertation, Dept. Eng. Physics
Math., Helsinki Univ. Technol., Helsinki, Finland, 2004. [Online].
Available: http://www.vtt.fi/inf/pdf, VTT Publications.
[5] P. Christiansen and J. Kristoffersen, The wind farm main controller
and the remote control system of the Horns Rev wind farm, in Proc.
Int. Workshop Large-Scale Integration Wind Power, 2003.
[6] L. Sder, Simulation of wind speed forecast errors for operation planning of multi-area power system, in Proc. PMAPS, 2004.
[7] P. J. Brockwell and R. A. Davis, Time Series: Theory and Methods,
2nd ed. New York: Springer-Verlag, 1991.
[8] I. Katic, D. Hjstrup, and N. O. Jensen, A sample model for cluster
efficiency, in Proc. Eur. Wind Energy Assoc. Conf., 1986.
[9] Wake effect model, WAsP Users Guide Denmark, Ris National
Laboratory, 2004. [Online]. Available: http://www.wasp.dk.
[10] J. F. Manwell, J. G. McGowan, and A. L. Rogers, Wind Energy Explained. New York: Wiley, 2002, p. 60.
[11] J. Sveca and L. Sder, Wind power integration in power system with
bottleneck problems, in Proc. IEEE PowerTech Conf., 2003, vol. 2.
[12] H. Heitsch and W. Rmisch, Scenario Reduction Algorithms in Stochastic Programming, in Computational Optimization and Applications. Norwell, MA: Kluwer, 2003, pp. 187206, Nr. 24.
[13] A. Brooke, D. Kendrick, A. Meeraus, R. Raman, and R. E. Rosenthal,
GAMS A Users Guide, GAMS Development Corporation, 2004. [Online]. Available: http://www.gams.com.

[14] M. C. Ferris, Matlab and GAMS: Interfacing Optimisation and Visualisation Software, Comput. Sci. Dept., Univ. Wisconsin-Madison, Mathematical Programming, Tech. Rep. 98-19, Nov. 1998. [Online]. Available: http://www.cs.wisc.edu/math-prog/matlab.html.
[15] M. Olsson and L. Sder, Generation of regulation power price scenarios, in Proc. PMAPS, 2004.
[16] B. Kirby, M. Milligan, Y. Makarov, D. Hawkins, K. Jackson, and
H. Shiu, California Renewables Portofolio Standard. Renewable
Generation Integration Cost Analysis. Phase I. California Energy
Commission/California Public Utilities Commission, 2003. [Online].
Available: http://www.cwec.ucdavis.edu/rpsintegration/.
Julija Matevosyan was born in Riga, Latvia, in
1978. She received the B.Sc. degree in electrical
engineering from Riga Technical University in 1999
and the M.Sc. and Tech.Lic. degrees in Electrical
Engineering from the Royal Institute of Technology,
Stockholm, Sweden, in 2001 and 2003, respectively.
She is currently pursuing the Ph.D. degree at the
Royal Institute of Technology, concentrating on
large-scale integration of wind power in areas with
limited export capability.

Lennart Sder (M91) was born in Solna, Sweden,

in 1956. He received the M.Sc. and Ph.D. degrees
in electrical engineering from the Royal Institute of
Technology, Stockholm, Sweden, in 1982 and 1988,
respectively.
He is currently a Professor of electric power
systems at the Royal Institute of Technology. He
also works with projects concerning deregulated
electricity markets, distribution systems, and integration of wind power.