IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO.

3, AUGUST 2007 985

Supervisory Control of a Wind Farm

Zbigniew Lubosny, Member, IEEE, and Janusz W. Bialek, Member, IEEE

AbstractThis paper presents a scheme for supervisory control Plants output (real power).
of wind farms. We concentrate on reduction of power output vari-
Wind velocity.
ation, but the proposed scheme can also be used for system-wide
controls that are related to power variation, such as frequency Standard deviation.
control or damping of power swings. The scheme may utilize
Pitch angle ( maximum,
either a separate (external) energy storage device or a power
reserve achieved through part-loading of one or more turbines minimum).
in a wind farm. The proposed supervisory control scheme has Rotor speed.
been validated on a model of a wind farm with four turbines.
Simulations have confirmed the usefulness of the scheme. The
proposed control system is general and could also be used for I. INTRODUCTION
other types of intermittent generators such as wave or solar.
Index TermsPower system control, wind farm, wind power.
P OWER systems traditionally have been characterized by a
stochastic energy consumption and a dispatchable energy
production. Increased penetration of wind generation changes
NOMENCLATURE that picture as the variability and stochastic character of wind
causes wind turbines to be intermittent (or rather uncontrollably
Energy ( maximum, minimum).
variable) energy sources. Additionally, the construction of wind
Area swept by rotor blades. turbines (long elastic elements such as blades, shaft, and tower)
, Turbine and generator rotor inertia constants. cause the output of wind farms to fluctuate. That fluctuation is,
Blade length. for higher frequencies of wind variations, partially self-compen-
sating at the system level due to a large number of turbines with
, Shaft stiffness and damping.
non-correlated outputs. Nevertheless, the overall variability of
, Pitch control proportional and integral gain. wind power increases with the number of wind turbines
, Torque control proportional and integral gain. due to a non-algebraic adding of power outputs from indi-
, Pitch compensation proportional and integral vidual turbines as follows:
gain.
Wind farm supervisory controller gain.
(1)
Pitch control time constant.
Power control time constant.
, , Wind farm supervisory controller time However, some low-frequency wind fluctuations are
constants. non-Gaussian, e.g., due to atmospheric turbulent winds, so
that a certain correlation does take place [1][4]. In that case,
, Energy storage controller lead and lag time
the sum of power fluctuations can become higher than that
constants.
resulting from (1).
, Energy storage controller gain and time Power fluctuations caused by wind variability may create
constant. problems for power system operation, especially related to
, Standard deviation controller gain and time power quality, system stability, system control, etc. [5]. Those
constant. problems increase with a high penetration of wind turbines and
Energy storage real power. if large wind farms operate in a weak grid [6][8]. In those
, Generator real and reference power. cases, it may become necessary to take some steps in order to
limit the negative influence of wind farms on the system.
Wind turbines and wind farms face increased technical re-
Manuscript received September 13, 2006; revised November 24, 2006. This
quirements [9], [10] and manufacturers constantly update their
work was supported in part by the Engineering and Physical Sciences Research products, introducing new control features to fulfill Grid Code
Council, Grant GR/S28082/01 SUPERGEN: Future Network Technologies requirements. Those additional features concern mostly voltage
and in part by Komitet Badan Naukowych (KBN) Grant 3T10B 011 27. Paper
no. TPWRS-00631-2006.
and reactive power control [11][16].
The authors are with the Institute for Energy Systems, The University of There has been also some progress toward real power control
Edinburgh, Edinburgh, EH9 3JL, U.K. (e-mail: [email protected]; janusz. such as damping of electromechanical oscillations, frequency
[email protected]). control, improvement of energy quality [17][19], and supervi-
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. sory control of real power by a system operator [16], [20]. How-
Digital Object Identifier 10.1109/TPWRS.2007.901101 ever, while it is relatively easy to implement system require-
0885-8950/$25.00 2007 IEEE
986 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO. 3, AUGUST 2007

ments related to voltage (and reactive power) control, real power

control requires using some type of energy storage or power re-
serve. That increases the complexity and cost of wind energy.
In the case of a single turbine, energy storage can be me-
chanical, e.g., flywheel, or electrical, e.g., battery [21]. Another
possibility is to apply a technology for energy transfer from a
wind turbine to a generator that can play the role of energy
storage, e.g., a hydraulic drive system [22]. Generally, the re-
quired storage capacity depends on wind characteristics at a par-
ticular location and the desired control effect.
For a wind farm, real power control can take one of the fol-
lowing forms:
use of individual energy storage schemes like the ones de-
scribed above;
use of a separate (standing alone, external) energy storage
based on any (non-wind-based) technology; Fig. 1. Wind farm with external energy storage supervisory control (WFC
operate one or more of turbines in the farm as part-loaded wind farm controller, ESCenergy storage controller, ESenergy storage,
Kfrequency control).
(with respect to currently available wind capacity) and use
the available power reserve for real power control.
Minimization of power fluctuations in a power system can be re- For power quality improvement (i.e., for smoothing out
alized in two ways, either by active control (using controllers) or wind power fluctuations), fast and short- (or medium-) term
by utilization of rotors. Rotors act like power (and frequency) fil- energy storage has to be used. For frequency control, fast and
ters in a power system, but their filtering effectiveness is limited. medium-term energy storage seems to be suitable. For economic
It can be increased by running the turbines slightly part-loaded dispatch, slow and long-term energy storage can be used.
when compared with the current wind capacity [23], [24]. This paper presents a general control scheme for the uti-
Larger fluctuations of power have to be compensated by ac- lization of a fast and short-term energy storage and fast output
tive control of a turbine or wind farm. As mentioned before, this control of a wind turbine or a farm. We concentrate on the
can be realized by running a turbine (or a farm) part-loaded or by improvement of power quality, i.e., reduction of power output
utilizing any energy storage. When part-loaded, wind turbines fluctuations in the range of a few to tens of seconds. This is
(or whole farms) can operate with constant power (balance con- higher than can be absorbed by turbines rotors and lower than
trol) or with a certain constant reserve capacity (delta control). can be eliminated by a system operator action. The proposed
Operation in the balance control mode (i.e., at a constant control scheme can be treated as a supplementary control loop,
power lower than that which could be harnessed from the with respect to existing (or considered in literature) supervi-
wind) effectively eliminates power fluctuations for DFIG and sory power controllers, dedicated to the reduction of power
synchronous generators-based wind turbines [15], [16]. In such fluctuations [14][16], [25], [29]. The proposed supervisory
case, no additional, supervisory control loop is necessary. control scheme has also the ability to contribute to system-wide
Operation in the delta control mode causes higher frequencies controls such as frequency control or damping of power swings.
of wind variation to be effectively eliminated by utilizing the The control algorithm is local, i.e., it uses locally observable
power reserve. However, low-frequency variations of wind are signals and acts only on the farm under control. The control
passed to the plant output, and some additional control may be may utilize either a separate (external) energy storage device
necessary to minimize them. or reserve power due to part-loading of one or more turbines in
Part-loading of wind turbines means a variable financial loss a wind farm.
to the owner, as not all available wind energy is converted into The proposed supervisory control scheme has been validated
electricity. This is the cost of increasing controllability and on a model of a wind farm with four turbines. Simulations
quality of wind power. The use of external energy storage for have confirmed the usefulness of the scheme for controlling
the same purpose reduces the variable loss but attracts a higher the farm output power. It should be noted that the proposed
capital and maintenance cost. It should also be noted that no control system is general and could also be used for other types
energy storage is 100% efficient, so using it also incurs some of intermittent generators such as wave or solar.
losses. Economics of wind energy are very important, but the
cost-benefit analysis of the proposed supervisory control is II. SUPERVISORY CONTROL OF A WIND FARM
beyond the scope of this paper, which concentrates on technical
aspects only. A. Supervisory Control With External Energy Storage
The type of energy storage chosen depends on demands of Fig. 1 shows a wind farm with external energy storage (ES)
the control process. From the point of view of access time to the connected to a power system. The wind farm can be connected
energy stored, there are fast and slow energy storage devices. to the high-voltage (HV) network through a step-up transformer
From the point of view of capacity, and therefore the length or can operate in a medium-voltage (MV) network. The energy
of operation, there are short-, medium-, and long-term energy storage considered here is a storage battery connected to the
storage devices. power system through a power electronic inverter and a step-up
LUBOSNY AND BIALEK: SUPERVISORY CONTROL OF A WIND FARM 987

transformer (not shown in Fig. 1). A choice of appropriate The WFC provides a control signal for the energy storage
storage capacity should be preceded by a statistical analysis changing power fed to the power system or consumed from
of wind data at a given location. By weighing the investment the wind farm.
cost, maintenance cost, and storage cycle efficiency against The control process can lead to a full energy storage
the required control result, an optimal storage capacity would charge/discharge. This is undesirable, as it makes the proper
be obtained. It would be economically inefficient to use a operation of the WFC impossible. To prevent that, the energy
storage of very large capacity capable of smoothing all possible storage is equipped with its own energy storage controller ESC.
fluctuations. ESC is a simple controller that is relatively slow in order not
The wind farm feeds the network with power , which is to degrade the effectiveness of the WFC but fast enough to
equal to sum of power generated by all the wind turbines minus prevent full energy storage charge or discharge for a defined set
power loss in the wind farm grid. Output power depends on the of disturbances.
wind velocity . Its variation depends on the wind velocity vari- ESC acts on the storage (signal ) by charging or dis-
ation, the type of a wind turbine, the operating point, and a type charging the energy storage, which in turn influences powers
of disturbance (e.g., turbine emergency disconnection). The en- and simultaneously .
ergy storage injects or consumes power that theoretically The amount of energy stored is defined by
is equal to zero in steady-state, but in practice, it varies due to
wind velocity variation.
(3)
The total power fed to the power system by the wind farm and
energy storage is equal to .
The supervisory control of the considered system is executed When energy reaches its limits ( or ), the appro-
by two controllers: priate limit of the storage power (see Fig. 1), i.e., or
wind farm controller (WFC)its aim is to minimize the , is decreased to zero. When the energy stored is within
variations of real power output ; the limits , the control signal and the storage
energy storage controller (ESC)its aim is to keep the output is limited by . In our simulations,
defined energy storage charge level , where we have assumed that MW. The energy storage con-
is the rated (or maximum allowable) energy of the troller signal is limited to . The gain is also
storage. limited, and we have assumed that is between 2 and 50.
The wind farm controller consists of the wash-out element, with The standard deviation is calculated as
time constant , and a compensating element with time con-
stants and . The aim of the former is to remove the constant
component from the plants output , while the latter com- (4)
pensates for the phase shift introduced by the object (transfer
function from to ) and transducers.
from samples , by using a moving window, mea-
For properly defined compensation time constants and ,
sured with time step . Hence, the window width is
the effectiveness of the WFC depends only on the gain .
equal to 10.24 s. Power is the mean value calculated
The higher the gain, the better minimization of the plants output
from the samples .
power fluctuations. In general, that gain can be set and kept con-
In the simulations, an ideal power electronic inverter was as-
stant. However, the proposed control scheme includes a possi-
sumed, so no compensation was necessary, i.e., . In
bility of modifying the gain during operation. The gain value is
practice, an inverter can be also treated as a non-inertia element
controlled by a PI controller that acts on the standard deviation
(i.e., a gain), but lags introduced by measuring elements and real
of the plants output power . Standard deviation is used
energy storage have to be compensated for.
here as an easy-to-compute measure of output power variability
The considered energy storage can also be used for other con-
although it is not directly considered in power quality standards
trol purposes, e.g., frequency control. This is shown in Fig. 1 by
(grid codes). However, from the Parsevals theorem, the total
a frequency control block and signal , both drawn with dashed
harmonic distortion (THD) is related to the standard deviation
lines. However, the possibility of controlling frequency with en-
and the average by [26] as follows:
ergy storage is limited. The limiting factor here is the amount
of energy that can be stored. Such storage can, in fact, mini-
(2) mize frequency variation in the short term but cannot influence
the steady-state value of frequency. Full, i.e., long-term, fre-
Thus, by controlling the standard deviation, we can control quency control would require using an additional energy source,
THD, which is a variable used in standards related to power e.g., fuel cell, gas turbine, or diesel engine, which could supply
quality. The presented control scheme can be also easily con- power system directly or via the storage. Such source should be
verted to a direct control of THD. started and would achieve a desired level of power generation
In the considered setup, an increase of causes an increase until the storage is fully charged/discharged.
of the gain , and that leads to a decrease in power variation
and, through the feedback, reduction in . This control B. Supervisory Control of Power Reserve
loop has to be fast in order to be effective in controlling fast The idea of an external battery storage control is simple. The
changes in the plants output power. storage just adds energy when plants output power decreases
988 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO. 3, AUGUST 2007

Fig. 3. GE 1.5- and 3.6-MW wind turbine control system model [27], [28].

normal mode of operation, when the plant maximizes the use

of wind, whenever the wind speed changes, the output power
changes, too.
If the plant operates with a power reserve, the pitch angle is
set to a higher value than minimal, e.g., .1 Then the
characteristic does not cross the peak of the turbine charac-
teristic , but it crosses a point, say, 4. Power difference
is a measure of the power reserve. If the wind speed
drops from to , then, in the absence of the proposed super-
visory control and for a constant pitch angle , the transition to
a new operating point will progress similarly as before, i.e., in
two stages along points 4-5-6curve B. Point 6 is at the inter-
section of the characteristics and . The plant char-
acteristic is shaped by the turbine governorblock
in Fig. 3so the operating point will tend toward .
The idea of the proposed supervisory control is to control the
variations of the output power. When wind speed changes from
Fig. 2. Characteristics of a wind turbine. to , it is necessary to move the electrical characteristic
to the right2 to so that the characteristic intersects
with with unchanged power (point 7). As the result, the
turbine output power does not change, despite the drop in the
and consumes surplus of energy when the plants output power
wind velocity.
increases. The plant characteristics are not important for that
The turbine governor, acting according to the turbine charac-
type of operation. On the other hand, utilizing power reserve
teristic , will reduce the pitch angle while simultaneously,
due to a part-loaded wind plant is more complex because turbine
the generator controller will shift back the electrical character-
characteristics influence the overall system properties.
istic to the left. As a result, the operating point will move to
The idea of utilizing power reserve is shown in Fig. 2, where
the left (to point 4) with constant output power . The overall
a normal operation of a wind plant, i.e., with full utilization of
plant response is shown by line C.
the available wind power, will be compared with its part-loaded
In reality, the ability to eliminate power fluctuations due to
operation. We will use standard turbine control models.
wind velocity variation depends mainly on the control system.
The operating point of a turbine is determined by the intersec-
The ability is limited by various mechanical and electrical con-
tion of the turbine aerodynamic power characteristic, dependent
straints. The change of energy extracted from wind, to keep the
on wind velocity , pitch angle , and turbine speed , with the
constant output power, has to be temporarily stored in the wind
electrical power characteristic that, in the case of DFIG, can be
turbine, where the rotor is the main but limited storage. There-
controlled by changing the rotor current. When a wind turbine
fore the farm response follows curve C characterized by some
maximizes the capture of wind power, the resultant wind plant
power transient (variation).
characteristic is made up largely from peak parts of turbine
The described operation corresponded to the case when a
characteristics denoted by , , etc. Assume that
single plant compensated its own power output variations due
the plant operates at a not-rated point 1 at the peak of the char-
to a wind speed drop. When such a plant operates in a farm with
acteristic , where constant is the minimum value
a number of turbines, its reaction (power output) has to compen-
of the pitch angle. A step drop of wind velocity from to
sate power variation of the remaining turbines, too. In that case,
causes a two-stage transition to a new operating point (curve A)
when the wind velocity drops, the electrical characteristic has
corresponding to the peak of . In the first stage, when the
to be shifted to a proper location (i.e., not necessarily to point
rotor speed is still unchanged due to inertia, there is a step
7). The proper location means here intersection of the electrical
reduction of power from to (i.e., transition from point 1
to point 2). In the second stage, the operating point moves on 1# is defined by a control signal c shown in Fig. 4 and discussed later.
characteristic from point 2 to point 3. Therefore, in that 2This can be done in the DFIG by controlling the rotor current.
LUBOSNY AND BIALEK: SUPERVISORY CONTROL OF A WIND FARM 989

output power variations in the positive direction (i.e., requiring

increasing the output of the selected turbine) is capped by the
value of power reserve . The compensation of output
power variations in the negative direction (i.e., requiring reduc-
tion of the output of the selected turbine) is restricted by the
present mean output of the selected turbine(s).
The PRC control signal has to change rel-
atively slowly but should ensure a permanent reserve of real
power. Therefore, it should be introduced in the control loop
acting directly on the turbine (pitch angle). Fig. 3 shows that the
signal is introduced after the pitch angle compensation block.
The signal is a constrained negative number. The upper
limit of the signal is equal to zero, and that corresponds to op-
eration of the wind turbine without part-loading, i.e., operating
at maximum output for a given wind. The lower limit of is
Fig. 4. Wind farm with supervisory control of power reserve (WFCwind equal to minus half of the present mean output of the turbine(s)
farm controller, PRCpower reserve controller, Kfrequency control). selected to maintain a reserve.
The upper limit of is also forced by a nonlinear element,
denoted as in Fig. 4, when the power output drops rapidly
characteristic with turbine characteristic at a nec- due to a temporary drop of wind speed or an outage of turbine(s).
essary output power (curve D). In that case, the power dif- It will be discussed in the next section.
ference has to be equal to the power change of the When more than one turbine is selected to maintain a reserve,
remaining turbines in the farm. in Fig. 2 represents the sum of outputs of those turbines.
Fig. 4 shows a block diagram of a wind farm that would ex- The PRC output signal has to be then spread over the
ecute the control strategy described above. The wind farm can all turbines maintaining a reserve, e.g., proportionally to their
be connected to the HV network through a step-up transformer actual output power.
or can operate in the MV network. The considered setup can be also used for frequency control,
The farm feeds the network with power , which is equal as shown by the dashed part of the diagram. The control signal
to the sum of power generated by all the wind turbines minus should be then added to and/or control signals. When
power loss in the wind farm grid. The power reserve is pro- a steady-state frequency change has to be achieved, the signal
vided by one or more wind turbines selected to be part-loaded. should be added to signal (see Fig. 4).
Thus, the unused power reserve that could be additionally har- The standard deviation of the measured signals is calculated
nessed from wind is the equivalent of energy stored in a con- similarly as before, i.e., from samples by using a moving
ventional, described earlier, storage system. Power output of the window and the time step equal to 0.01 s.
part-loaded turbine(s) is denoted as , while power output of Fig. 3 shows the control diagram of a GE wind turbine main-
the remaining turbines is denoted as . taining a reserve power. In that case, the main element whose
The supervisory control system consists of two controllers: transfer function should be compensated for is the torque con-
WFCits aim is to minimize the variations of real power troller. To achieve effective operation of the WFC, the compen-
output ; sating element parameters can be set as: , ,
PRCits aim is to set and keep value of power reserve .
that could be additionally fed to the grid from the The value of gain that controls the power reserve was set
selected turbine(s). to unity in simulations. Higher gain increases the power reserve
WFC shown in Fig. 4 has a structure identical to that shown but simultaneously decreases the power generated by the wind
in Fig. 1. The WFC acts on the selected turbine(s) (signal ) turbine. Its value should be chosen as a compromise between the
changing their output so that the farms output is smoothed plants ability to minimize output power variation and efficiency
out. Power is a part of farm power . Signal is fast, as of the wind farm.
its aim is to minimize the output power variations. The compensating element with time constants and in
Signal has to cause fast changes in the output of the se- the PRC should ensure a proper, from the dynamic point of view,
lected turbine(s), and therefore, it should be introduced in the operation of the controller. Its response should be relatively slow
plant control loop acting directly on the generator. in order not to cause oscillations of the turbine blades. We have
The power reserve controller PRC acts on a measured signal assumed time constants equal to , and
equal to the difference between the farm output and obtained good results.
, i.e., output power of the turbines selected to maintain a re- It is worth considering a question about which form of the
serve. In other words, is equal to the output of turbines proposed control is more effective from the technical point of
not controlled by the proposed supervisory control system. Its view: controlling a single wind plant or a whole farm. Both
standard deviation characterizes , the require- schemes are similar, but the advantage of controlling the whole
ment for the power reserve provided by the selected turbines to farm is in the simplification of the supervisory control system.
smooth out the overall farms output. Thus, the compensation of Moreover, the power reserve is used to compensate fluctuations
990 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO. 3, AUGUST 2007

Fig. 5. Wind farm with external energy storage in quasi steady-state operation. Fig. 6. Response of the wind farm with energy storage to a step increase of wind
(a) Power output. (b) Standard deviation. (c) Energy stored. velocity leading to 1 MW output power increase. (a) Power output. (b) Standard
deviation. (c) Energy stored.

of the sum of fluctuating outputs of individual plants that are

partly self-compensating. Consequently, the sum of required Figs. 57 show examples of test results. is power fed to
reserve capacities of individual plants in the farm is higher than the grid, while is the farms production. Wind farm power
the required capacity of a single reserve maintained for the in the diagrams is marked as to signify that was
whole farm necessary to achieve the same control effect. That increased by 0.5 MW in order to make the signal more visible.
means that the revenues lost due to maintaining power reserve is the energy storage power equal to the difference between
are smaller in the latter case. and . is a standard deviation of , while
is a standard deviation of .
Figs. 5 and 6 show response of the system with constant
III. VERIFICATION OF EFFECTIVENESS OF THE
gain value , while Fig. 7 shows response of the
SUPERVISORY CONTROL SYSTEM
system with an active controller of the variable standard devia-
tion .
A. Wind Farm With Supervisory Control of Energy Storage
Fig. 5 shows quasi steady-state operation of the considered
Verification of the proposed control schemes has been car- system. The high suppression (practically elimination) of farms
ried out on a model of a wind farm with four turbines. Wind output power fluctuations is clearly visiblecompare to
velocity time series, achieved from real measurement made in in Fig. 5(a). The same can be seen by comparing standard
the Swedish system (time step equal to 39 ms), have been used deviation of the output power in Fig. 5(b). The stored en-
as inputs. The wind farm initial output was equal to ergy deviates from the reference value equal to
MW. [see Fig. 5(c)] as the result of a decrease in the mean value of
LUBOSNY AND BIALEK: SUPERVISORY CONTROL OF A WIND FARM 991

Fig. 7. Response of the wind farm with energy storage to a step change in Fig. 8. Response of the wind farm with power reserve to a step increase in wind
the reference value  of standard deviation. (a) Power output. (b) Standard velocity. (a) Power output. (b) Standard deviation.
deviation.

from value to value . Then, at

power extracted from wind at about 50 and 80 s. Later on, for , the reference value returned to the initial value equal
, the stored energy level is rebuild by ESC. to . That increase of the reference value between
Fig. 6 shows the plants response to a step increase in the and caused increased output power
wind velocity, which corresponds to an increase of 1 MW in the [see Fig. 7(a)] variations and reduction of power quality [see
plants output at . The response is multi-stage. Fig. 7(b)].
The first part is connected with charging of the storage, Fig. 6(c). The simulation results confirm the ability of the proposed
Stored energy is increasing by an amount equal to the inte- control scheme to influence the level of output power varia-
gral of . The storage smooths out the farms output by con- tions and the quality of energy supplied. Effectiveness of the
suming some energy produced by wind turbines. Consequently, proposed controller in elimination of power fluctuations of var-
at , almost no step change in the power fed to the grid ious frequencies depends highly on the wash-out time constant
is visiblesee curve. Note, by comparing the farms . Low-frequency fluctuations require a higher value of .
production with power fed to the grid , that the power The authors achieved also positive effects by using two parallel
variations are significantly reduced. wash-out functions in the WFC with time constant equal to
At time , the energy storage becomes fully charged and .
( reaches ) so that it is unable to consume more energy, and
becomes equal to zero. This causes a step increase in B. Wind Farm With Supervisory Control of Power Reserve
and its variations because of WFC inability to smooth out the
output [see Fig. 6(a)]. The standard deviation of the plants Figs. 8 and 9 show the responses of a farm in which a single
output obviously increases [see Fig. 6(b)]. turbine is selected to be part-loaded and maintain a power re-
Further reaction of the system is related to the action of the serve for the whole farm. The first part of the responses (up to
storage controller ESC. The controller slowly discharges the ) is related to the activation of the PRC, which de-
storage to the reference value , reaching it at creases power generated by the selected wind turbine and
. Decrease of the stored energy below the max- simultaneously decreases the plants output power . Power
imum value activates the WFC again at so that represents output of the remaining three wind turbines in
becomes not equal to 0. This causes a reduction in the output the wind farm.
power variation, confirmed by a reduction in the standard Fig. 8 shows the system response to a step increase, at
deviation . , in the wind velocity, which corresponds to an increase of
Fig. 7 shows the farms response to a step change in the ref- 1 MW in the output of the three turbines. The signal Pres is
erence value of the standard deviation at time increased by 5 MW to make the signals more visible.
992 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO. 3, AUGUST 2007

a discontinuity will appear in the farms output. In other words,

disturbances of amplitude higher than the power reserve will ap-
pear as discontinuities in the power output.
At the same time, because the storage reaches its limit, the
WFC practically stops to influence the controlled processthe
variation of the plants output power increases.
When is forced to be equal to zero, i.e., until ,
the WFC forces signal . Such a high increase of causes
that some additional energy, taken from the kinetic energy of
the selected turbine, is fed to the grid. In the considered case,
the rotor speed dropped to 20% of the rated speed. The kinetic
energy initially supported the wind farm by decreasing the
drop. Then the rotor speed dropped leading to a decrease of
power extracted from wind by the turbine. That is visible in
Fig. 9(a) as an initial increase of followed by a decrease.
From , when the PRC released power reserve (
becomes nonzero), the WFC returned to a proper oper-
ation and the plants output power variation was reduced again.
The system moved aperiodically to a new operating point.
Keeping a power reserve by part-loading one of the turbines
involves a loss of revenues, as not all available wind energy was
converted to electricity. Integration of the farm response with
and without the proposed supervisory control revealed that the
Fig. 9. Response of the wind farm with power reserve to a disconnection of loss of revenues was very small indeed in quasi steady-state con-
one wind turbine. (a) Power output. (b) Standard deviation. ditions. The average generated power over 240 s was 12.3 MW
when the farm was run without the proposed control system and
12.22 MW when the farm was run with the proposed system.
The PRC controller, as a result of increase after the The average reserve was 0.08 MW, which corresponded to the
wind velocity change, increases the value of (see loss of revenues of only 0.65%. Such a high efficiency of the
Fig. 4). That leads to a decrease of power produced by system was due to the controller automatically choosing the
the wind turbine maintaining the reserve. Consequently, the right level of reserve for given wind conditions. With more tur-
farms output is changing smoothly, almost without a step bulent winds, the amount of power kept in reserve would obvi-
change. At the same time, the output power variation is reduced ously have to be higher.
(compare with ).
IV. CONCLUSIONS
That additional action of the PRC is slowly withdrawn and
the plant comes back to a quasi steady-state operation where This paper has presented a supervisory control system of
the power reserve margin (defined by ) depends on the a wind farm. The focus was on controlling the plants output
level of variation in the output power . power fluctuations, but the control system could also be used for
the frequency control and for the damping of electromechanical
Fig. 9 shows the system response to an emergency discon-
oscillations. Two types of systems have been considered, one
nection, at time , of one of the three turbines, with the
with energy storage and one with utilizing a power reserve
corresponding loss of 3 MW.
achieved by part-loading of selected turbine(s).
For such a rapid change, the reaction of the PRC controller
The system with energy storage exhibits the following
depends partly on the nonlinear function measuring the deriva-
characteristics.
tive of the output power (see Fig. 4). The function is used here The proposed structure of the supervisory control is
to prevent improper (i.e., opposite to the needs) operation of the general and independent of the characteristics of power
PRC as a reaction to an increase of deviation resulting sources in the farm. It can also be used for other types of
from the decrease of power generated by the wind farm. renewable energy farms such as marine (wave or tidal) and
After the power generated drops, the plants output power also for any subsystem (e.g., micro-grid) with intermittent
derivative falls below a set value, and the function decreases energy sources and/or loads. The system is independent of
signal to zero [see Fig. 9(b)]. The power reserve is then the type of power sources and the type of disturbances.
limited to zero, and the controller reaches its limit. The proposed system is very efficient. It is possible to con-
Reduction of the reserve to zero, which actually corre- trol the standard deviation of output power and therefore
sponds to an increase to zero in Fig. 9(b), causes an increase the quality of power provided by a farm. The ability to
in generation of the controlled plant [ in Fig. 9(a)]. This smooth out output power variation in the case of signif-
reduces the farms power drop, . If the power increase re- icant changes of power generated by wind turbines de-
sulting from the reduction in is lower than the disturbance, pends mainly on the storage capacity. Higher capacity of
LUBOSNY AND BIALEK: SUPERVISORY CONTROL OF A WIND FARM 993

the storage results in smoothening of higher changes of the , MW, kV,

power generated. , , ,
Synthesis of the supervisory control system is relatively , , , , ,
simple because of practically non-inertial access to energy , , , , ,
stored in battery storage. The lags introduced by power , , , , ,
converter are not significant and can be neglected. , ,
The system utilizing power reserve exhibited the following
features. Data of supervisory control system with external energy
The control structure is simple, general, and highly plant- storage
independent, but it requires controllability of the wind tur- , , , , ,
bine and the generator. The controlled wind turbines (vari- , , .
able speed type) can operate in parallel with constant speed Data of system with supervisory control of power reserve
wind turbines whose ability to minimize power fluctuations , , , , , ,
are limited to filtering by their rotor inertia. , , , , .
High effectiveness of the system is limited by dynamics of
the controlled object, i.e., by the wind turbine(s) operating REFERENCES
with power reserve. [1] N. H. Lipmann, E. A. Bossanyi, P. D. Dunn, P. K. Musgrove, G. E.
Setting of parameter values is relatively simple because Whittle, and C. Maclean, Fluctuations in the output from wind turbine
clusters, Wind Eng., vol. 4, no. 1, pp. 17, 1980.
double fed induction generator (DFIG) introduces very [2] P. Srensen, N. A. Cutululis, J. Hjerrild, L. Jensen, M. Donovan, L. E.
short time lags. Hence, the compensating block settings A. Christensen, H. Madsen, and A. Vigueras-Rodrguez, Power fluc-
depend on the point where the control signal tuations from large offshore wind farms, in Proc. Nordic Wind Power
Conf., Espoo, Finland, May 2223, 2006.
is introduced in the turbine controller. [3] F. Boettcher, C. Renner, H.-P. Wald, and J. Peinke, On the statistic of
The use of pitch angle control may increase possibility of wind gusts, Boundary-Layer Meteorol., vol. 108, pp. 163173, 2003.
fatigue of the blade and blade-hub material. This factor [4] S. T. Frandsen, Turbulence and turbulence-generated structural loading
in wind turbine clusters, Riso National Laboratory, Roskilde, Denmark,
should be taken into account when designing a control Rep. Riso-R-1188(EN), Nov. 2005.
system. [5] Z. Lubosny, Wind Turbine Operation in Electric Power Systems. Ad-
Practical implementation of either of the schemes would need vanced Modeling. New York: Springer-Verlag, 2003.
[6] E. Muljadi, C. P. Butterfield, J. Chacon, and H. Romanowitz, Power
to be preceded by a thorough cost-benefit analysis. Using ex- quality aspects in a wind power plant, in Proc. IEEE Power Eng. Soc.
ternal energy storage would mean additional capital investment General Meeting, Jun. 1822, 2006.
and maintenance costs, while using power reserve would mean [7] M. P. Palsson, T. Toftevaag, K. Uhlen, and J. O. G. Tande, Con-
trol Concepts to Enable Increased Wind Power Penetration, in Proc.
a loss of revenues for the owner and increased possibility of IEEE Power Eng. Soc. General Meeting, Jul. 1317, 2003, vol. 3, pp.
fatigue of the blade and blade-hub material due to additional 19841990.
pitch angle control. Our simulations showed that the loss of rev- [8] N. W. Miller, E. V. Larsen, and J. M. MacDowell, Advanced control
of wind turbine-generators to improve power system dynamic perfor-
enues amounted to only 0.65% for the assumed wind conditions. mance, in Proc. 11th Int. Conf. Harmonics and Quality of Power, Sep.
With more turbulent winds, the amount of power kept in reserve 1215, 2004, pp. 330332.
would obviously have to be higher. [9] I. Erlich and U. Bachmann, Grid code requirements concerning con-
nection and operation of wind turbines in Germany, in Proc. IEEE
The crucial element in the proposed control structures is the Power Eng. Soc. General Meeting, Jun.e 1216, 2005, pp. 22302234.
capacity of energy storage or power reserve. The capacity de- [10] I. Erlich, W. Winter, and A. Dittrich, Advanced grid requirements for
termines the tasks that can be efficiently executed by the pro- the integration of wind turbines into the German transmission system,
in Proc. IEEE Power Eng. Soc. General Meeting, Jun. 2024, 2006.
posed supervisory controller, i.e., power quality improvement, [11] Advanced Grid Option 2, Technical Description, Item No. 951421.RO,
reduction of output power variation, power system stability im- 2003-15-05, Vestas Wind Systems A/S.
provement, and frequency control. The last two tasks require a [12] NORDEX N90, Technical Description, N90-1-techn-descrip-
tion-en.doc, 2005-04-19, NORDEX Energy GmbH.
higher capacity of storage or reserve. For all the control tasks, [13] A. Dittrich and A. Stoev, Comparison of fault ride-through strate-
the access to energy in storage or reserve has to be fast. gies for wind turbines with DFiM generators, in Proc. Eur. Conf.
The improvement of power quality, which requires a rela- Power Electronics and Applications (EPE2005), Dresden, Germany,
Sep. 1114, 2005.
tively small capacity of storage or reserve, can be executed on [14] J. R. Saenz, A. Tapia, G. Tapia, F. Jurado, X. Ostolaza, and I. Zubia,
each turbine operating in the wind farm. Elimination of larger Reactive power control of a wind farm through different control al-
power variations, which require a higher capacity of energy gorithms, in Proc. 4th IEEE Int. Conf. Power Electronics and Drive
Systems, Oct. 2225, 2001, vol. 1, pp. 203207.
storage, can be done more effectively using a central (single) [15] P. Sorensen, A. D. Hansen, K. Thomsen, T. Buhl, P. E. Morthorst, L.
energy storage. This is because the variation in the overall farm H. Nielsen, F. Iov, F. Blaabjerg, H. A. Nielsen, H. Madsen, and M. H.
output power is smaller than the algebraic sum of variations in Donovan, Operation and Control of Large Wind Turbines and Wind
FarmsFinal Report, Riso National Laboratory, Roskilde, Denmark,
each turbine output. That leads to a smaller capacity of the cen- Riso-R-1532(EN), Sep. 2005.
tral energy storage or power reserve than the sum of capacities [16] A. D. Hansen, P. Sorensen, F. Iov, and F. Blaabjerg, Centralised con-
for each wind turbine. It also simplifies the control algorithm trol of wind farm with doubly-fed induction generators, in Proc. 26th
Int. Conf. Power Electronics, Intelligent Motion, Power Quality (PCIM
and increases effectiveness of the system. 2005), Nuremberg, Germany, Jun. 79, 2005.
[17] M. P. Palsson, T. Toftevaag, K. Uhlen, and J. O. G. Tande, Large-
APPENDIX scale wind power integration and voltage stability limits in regional
networks, in Proc. IEEE Power Eng. Soc. General Meeting, 2002, pp.
Data of a GE 3.6-MW wind turbine model 362369.
994 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO. 3, AUGUST 2007

[18] M. P. Palsson, T. Toftevaag, K. Uhlen, and J. O. G. Tande, Control [28] Manual of PSLF (Positive Sequence Load Flow) Program, ver. 14.2,
concept to enable increased wind power penetration, in Proc. IEEE GE.
Power Eng. Soc. General Meeting, Jul. 1317, 2003, pp. 19841990. [29] Large Scale Integration of Wind Energy in the European Power Supply:
[19] F. M. Hughes, O. Anaya-Lara, N. Jenkins, and G. Strbac, Control of Analysis, Issues and Recommendations EWEA, Dec. 2005.
DFIG-based wind generation for power network support, IEEE Trans.
Power Syst., vol. 20, no. 4, pp. 19581966, Nov. 2005.
[20] J. L. Rodrguez-Amenedo, S. Arnalte, and J. C. Burgos, Automatic
generation control of a wind farm with variable speed wind turbines,
IEEE Trans. Energy Convers., vol. 17, no. 2, pp. 279284, Jun. 2002.
[21] A. Joseph and M. Shahidehpour, Battery storage systems in electric
power systems, in Proc. IEEE Power Eng. Soc. General Meeting, Jun. Zbigniew Lubosny (M06) received the M.Sc.,
2024, 2006. Ph.D., and D.Sc. degrees from Gdansk University
[22] G. S. Payne, A. Kiprakis, M. Ehsan, W. H. S. Rampen, J. P. Chick, and of Technology, Gdansk-Wrzeszcz, Poland, in 1985,
R. Wallace, On the efficiency and dynamic performance of digital dis- 1991, and 1999, respectively.
placement hydraulic transmission in tidal current energy converters, He is currently a full Professor (since 2004) at the
Proc. Inst. Mech. Eng., Part A, J. Power Energy (Special Issue on the Gdansk University of Technology and a Research
Status of Tidal-Energy-Based Power Generation), accepted for publi- Fellow at the University of Edinburgh, Edinburgh
cation. U.K. His fields of interests include mathematical
[23] L. Ran, J. Bumby, and P. Tavner, Use of turbine inertia for power modeling, power system stability and control, artifi-
smoothing of wind turbines with a DFIG, in Proc. 11th Int. Conf. Har- cial intelligence utilization to power system control,
monics and Quality of Power, Sep. 1215, 2004, pp. 106111. wind turbines modeling, and control.
[24] B. G. Rawn, P. W. Lehn, and M. Maggiore, Toward controlled wind
farm output: Adjustable power filtering, in Proc. IEEE Power Eng.
Soc. General Meeting, Montreal, QC, Canada, Jun. 1822, 2006.
[25] L. M. Fernndez, C. A. Garca, F. Jurado, and J. R. Saenz, Control Janusz W. Bialek (M95) received the M.Eng. and
system of doubly fed induction generators based wind turbines with Ph.D. degrees in electrical engineering from Warsaw
production limits, in Proc. IEEE Conf. Electric Machines and Drives, University of Technology, Warsaw, Poland, in 1977
May 1518, 2005, pp. 19361941. and 1981, respectively.
[26] C. Luo, H. G. Far, H. Banakar, P.-K. Keung, and B.-T. Ooi, Estimation He was with Warsaw University of Technology
of wind penetration as limited by frequency deviation, in Proc. IEEE from 1981 to 1989. From 1989 to 2002, he was with
Power Eng. Soc. General Meeting, Montreal, QC, Canada, Jun. 1822, University of Durham, Durham, U.K., and since
2006, pp. 18. 2003, he has held the Bert Whittington Chair of Elec-
[27] N. W. Miller, J. J. Sanchez-Gasca, W. Price, and R. W. Delmerico, trical Engineering at the University of Edinburgh,
Dynamic modeling of GE 1.5 and 3.6 MW wind turbine-generators for Edinburgh, U.K. His research interests are in power
stability simulations, in Proc. IEEE Power Eng. Soc. General Meeting, system security, liberalization of the electricity
Jul. 1317, 2003, pp. 19771987. power industry, grid integration of renewables, and in power system dynamics.