Optimal control of variable speed wind turbines

Andreea Pintea, Nicolai Christov, Pierre Borne, Dumitru Popescu, A. Badea

To cite this version:

Andreea Pintea, Nicolai Christov, Pierre Borne, Dumitru Popescu, A. Badea. Optimal con-
trol of variable speed wind turbines. The 19th Mediterranean Conference on Control and
Automation June, Aquis Corfu Holiday Palace, Corfu, Greece., Jun 2011, Corfu, Greece. <hal-
00719468>

HAL Id: hal-00719468

https://hal.archives-ouvertes.fr/hal-00719468
Submitted on 19 Jul 2012

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 Optimal control of variable speed wind turbines
Andreea Pintea, Nicolai Christov, Pierre Borne, Dumitru Popescu, Adrian Badea

Abstract—this paper proposes a MIMO linear quadratic on ensuring an optimal operation, but also on load reduction
regulator (LQR) controller designed for a horizontal variable and grid integration. Another important challenge is to
speed wind turbine with focus on the operating range referring provide good quality energy delivery from a profoundly
to the above rated wind speeds. The operating conditions of irregular primary source, the wind.
wind turbines make them subject to fluctuating loads that
create fatigue and lead to damage. Alleviating these loads would
The characteristics of the wind energy source are
reduce the needed materials, and increase the lifespan and the important in different aspects regarding wind energy
quality of the produced energy. The optimality of the entire exploitation. The energy available in the wind varies with the
system is defined in relation with the trade-off between the cube of the wind speed. The wind is variable both in space
wind energy conversion maximization and the minimization of and in time [2].
the fatigue in the mechanical structure. The solution of a Based on the value of the wind speed, there were two
control using an LQR regulator is presented. The performances
of the optimal control are assessed and discussed by means of a
essential functioning regimes identified for the wind
set of simulations. turbines. The first one corresponds to low wind operation,
and here the main control goal is to maximize the energy
I. INTRODUCTION capture.
Classical control system design is generally a trial and This region ends when the wind’s speed reaches the “rated
error process in which various methods of analysis are value”, above which, the turbine enters the second regime.
iteratively used to determine the design parameters of a This value is usually around 14m/s.
system. Acceptable performance is generally defined in In the above rated region, the pitch angle and the
terms of time and frequency domain criteria such as rise electromagnetic torque are the control variables that are used
time, settling time, overshoot, gain and phase margin and to reduce the structural loads and to maintain the output
bandwidth. power around a constant nominal value, also called the rated
Radically different performance criteria must be satisfied, power of the turbine (Fig. 1).
however, by the complex, multiple inputs, and multiple
outputs systems required to meet the demands of modern
technology.
The objective of optimal control theory is to determine the
control signals that will cause a process to satisfy the
physical constraints and at the same time to minimize or
maximize some performance criterion. [1]
The wind industry offers many challenges in designing
effective wind turbines that will harness wind energy and
will transform it into electricity. Wind turbines are large,
complex dynamically flexible structures that operate in
Fig.1 Tipical wind turbine power curve
turbulent and unpredictable environmental conditions where
efficiency and reliability are highly dependent upon a well Therefore, in this regime, the system is multivariable and
designed control strategy. multi-objective. Many applications used classical controls to
From a control point of view, the importance lies not only address more than one control objective, by adding multiple
control loops.
Manuscript received January 30, 2011. These added complexity to the control design and
A. P. Author is with the University “Politehnica” Of Bucharest, system’s behavior but, nevertheless, it was difficult to
Romania (corresponding author, e-mail: andreea.pintea@ gmail.com). properly address control-structure interaction issues because
N. C. Author is with Universite Lille 1, Lille, France, LAGIS
Department (e-mail: Nicolai.Christov@univ-lille1.fr).
the controller used only a single measured turbine output as
P. B Author is with Ecole Centrale de Lille, France, LAGIS Department the basis of its control and did not have direct knowledge of
(e-mail: Pierre.borne@ec-lille.fr ) the dynamics of the turbine. Modern control designs using
D. P. Author is with the University “Politehnica” of Bucharest,
Romania, (e-mail: popescu_upb@yahoo.com)
state space methods, can handle these issues in a better way,
A. B. Author is with the University “Politehnica” of Bucharest, since the controllers in these cases use a model to determine
Romania, (e-mail: badea@energ.pub.ro) the system’s states. Controllers can be designed not only to
maximize power or to regulate the turbine’s speed, but also
to add damping to its flexible modes, through state feedback
[3]. In the same context, the LQR regulator, proved to be a
good solution due to the fact that it facilitates multivariable
and multi-objective control design.
The paper is organized as follows: after a short
introduction and the presentation of the context in which the
LQR controller was chosen, one continues with Section II in
which the mathematical model of the turbine is presented in
detail. Section III provides a description of the LQR control
method and Section IV presents the analysis of the results
and the concluding remarks of this study. In the end of the
paper, an APPENDIX with the numerical values of the wind
turbine’s parameters used is provided.

II. MATHEMATICAL MODEL

Fig. 2 The mechanical structure of the wind turbine

At present, there are several variable speed wind turbine
configurations that are being widely used. For this study, a
M ⋅ q + C ⋅ q + K ⋅ q = Q (q, q, t , u ) (1)
horizontal variable speed wind turbine was chosen. The
variable speed wind turbine is currently the most used where M, C and K, are the mass, damping and the stiffness
technology and it has proven its advantages over the years matrices, Q is the vector of forces acting on the system, and
[4]. The major advantage is that by allowing the rotor to qi is the generalized coordinate. For our model, the
operate at various speeds, one can obtain a more efficient generalized coordinates are: q = (ωT , ω G , ζ 1 , ζ 2 , yT ) , where
capture of the wind energy with less stress in the turbine ωT is the angular speed of the rotor, ωG stands for the angular
drive train during wind gusts. The reader can find different speed of the generator, ζ1 and ζ2 are the flaps of the blades,
wind turbine modeling techniques in [2] and also detailed while yT represents the horizontal movement of the tower
explanations regarding the use of each type of model. (Fig. 2).
Generally, a model for an entire wind energy conversion
system can be structured as several interconnected subsystem
models: an aerodynamic, a mechanical, electrical and
actuator subsystems. But since the dominant dynamics lie in
the mechanical subsystem, special attention will be paid in
this direction. The mechanical structure that we chose to
study is seen as being arranged into several rigid bodies
linked by flexible joints. The amount of these joints or Fig. 3 The two mass model representation of the drive train
degrees of freedom, determines the order of the model.
In [5] [6] and [3] one can observe the way in which the Since the thrust forces acting on the blades are equal, it is
number of degrees of freedom of the system can increase the naturally to consider ζ1 = ζ2= ζ and Faero1 = Faero2 = Faero,
order of the non linear models of the turbine. Therefore, it is which transforms q into q = (ωT , ω G , ζ , yT ) . In the same
important to consider on the model just those degrees of time, one can find Q as being:
freedom that are directly coupled to the control [4]. Q = (C aero ,−C em , Faero ,2 ⋅ Faero ) (2)
By this reason, the model presented here, will include just The considered forces that are acting on the system are:
the first mode of the drive train, the first mode of tower Caero, the aerodynamic torque, Cem, the electromagnetic
bending dynamics, and the first mode of the flapping of the torque, and Faero, representing the thrust. The aerodynamic
blades. These degrees of freedom will suffice for the torque and the force acting on the entire rotor are expressed
controller design that will be presented (Fig. 2). The drive in terms of non-dimensional power coefficient CP and thrust
train is modeled as a two rigid bodies linked by a flexible coefficient CT respectively, as follows
shaft (Fig. 3). Also it was supposed that the two blades move
1 v3
in unison and support the same forces. C aero = ⋅ ρ ⋅ π ⋅ R 2 ⋅ C P (λ , β ) ⋅
In order to compute the model, we have started from a 2 ωT
theory that states that a mechanical system of arbitrary 1
complexity can be described by the equation of motion: Faero = ⋅ ρ ⋅ π ⋅ R 2 ⋅ C T (λ , β ) ⋅ v 2 (3)
2
where ρ represents the air density, R is the blade radius, and
v is the average speed of the wind. the model, the pitch controller. This was modeled here as a
The power coefficient is one of the most important first degree order system [7]:
parameters of the wind turbine because it offers information β 1
upon the efficiency of the turbine, it helps defining the = , where βref is the desired pitch angle
control objectives in the below rated regime and also it
β ref 1 + Tβ ⋅ s
characterizes the aerodynamic torque that moves the and β is the actual pitch angle of the blades.
turbine’s rotor. The power and the thrust coefficients can be We have taken into consideration the fact that the pitch
expressed in a polynomial form, and depend on two servomotor has some physical limitations, and we have
parameters which are the tip speed ratio λ and the pitch angle modeled them by including into our model one saturation in
β of the blades. the position and one in the speed. For this study we have
In order to derive the mathematical model, one has used supposed that the saturation values in position are -45˚ and
the Lagrange equation that offers a systematic procedure to 45˚, and that the servomotor does not exceed the speed of
calculate such models 10˚ /s. In Fig. 4 one can observe the way the pitch
d δE c δE δE δE servomotor’s dynamics were modeled.
( )− c + d + P =Q (4)
dt δqi δq i δqi δq i
Here, Ec, Ed, and Ep denote the kinetic, dissipated and
potential energies. After a few calculations, applied for our
system, one obtains
Jt J M Fig. 4 The pitch servomotor dynamics modeling
Ec = ⋅ ωT2 + G ⋅ ωG2 + T ⋅ yT2 + MP ⋅ ( yT + rP ⋅ ζ )2 After combining all these equations, one can put (4) into
2 2 2
dA d the into the classical state-space representation
ED = ⋅ (ωT − ωG )2 + dP ⋅ (rP ⋅ ζ )2 + T ⋅ yT2 (5) x(t ) = A ⋅ x(t ) + B ⋅ u (t ) + E ⋅ m v (8)
2 2
k k y (t ) = C ⋅ x(t ) + D ⋅ u (t )
EP = A ⋅ (θT − θG )2 + kP ⋅ (rP ⋅ ζ )2 + T ⋅ yT2
2 2 in which mv represents a perturbation acting on the system,
These energies were calculated under the supposition that and from a physical point of view it models the eventual
the generalized force that acts on the rotor is applied on a wind gusts that appear.
point situated at the distance rP on each blade from the hub  0 0 0 1 −1 0 0 0 0 
 
of the rotor (Fig. 2). In the above equations, JT and JG  0 0 0 0 0 1 0 0 0 
 0 0 0 0 0 0 1 0 0 
represent the rotor and the generator moments of inertia, MT  
 − 51.4 0 0 − 0.69 0.28 0 0 −2 0.04 
and MP are the masses of the tower and of the blade, dP, dA 
A =  268.29 0 0 1.46 − 1.46 0 0 0 0 

and dT represent the damping coefficients for the blade, drive  0 − 390.47 30.35 0.15 0 − 3.9 0.17 − 0.13 0.04 
 
shaft and tower. Similarly, kP, kA and kT stand for the spring  0 457.14 − 242.85 − 0.21 0 4.57 − 1.42 − 0.21 0.07 
 0 0 0 0 0 0 0 − 5.55 0 
coefficients of the blade, drive shaft and tower. ΘT and ΘG  
 0 0 0 0 0 0 0 0 − 0.14 
are the angular positions of the rotor and generator.
The interconnection of the models of different plant 0 0 0 0 0 0 0 1 T
0
subsystems, leads to a global highly non linear system,  
B =  0 0 0 0 − 2.44 ⋅ 10 − 5 0 0 0 0
mainly because of the expressions of the aerodynamic torque 0 0 0 0 0 0 0 5.55 0 

and of the thrust force, both given in (3).
0 0 0 1 0 0 0 0 0 0
 0 10 − 5 
For control design purposes, we linearized the model   
0 0 0 0 0.25 0 0 0 0  0 0 0 
around an operating point Sop C= , D= 
0 0 1 0 0 0 0 0 0
C aero = D cω ⋅ ω T + Dcβ ⋅ β + Dcv ⋅ v 
0
 0 0 0 
(7)  1 0 0 0 0 0 0 0  0
 0 0 
Faero = D fω ⋅ ω T + D fβ ⋅ β + D fv ⋅ v E = (0 0 0 0 0 0 0 0 1)
Here The system is multivariable; there have been identified
 ∂C   ∂C   ∂C  three inputs and four outputs (Fig. 5). As shown in this
Dcω =  aero  , Dcβ =  aero  , Dcv =  aero  control scheme, the input variables of the system are
 ∂ω T  Sop  ∂β  Sop  ∂v  Sop
considered: vm the average value of the wind speed, and the
 ∂F   ∂F  two control variables: the pitch angle, β, and the
,and D fω =  aero  , D fβ =  aero  ,
 ∂ω T  Sop  ∂β  Sop electromagnetic torque Cem.
Here, we have considered the state vector
 ∂F  x T = (θ T − θ G , ζ , y T , ω T , ω G , ζ , y T , β , v ) T , the output of
D fv =  aero  respectively.
 ∂v  Sop the system y = (P_el, ωT, ζ, yT), and the command signal u =
Besides these equations, in order to interconnect the (β, Cem). The first component of the output vector represents
models of the individual subsystems, one must include into the electrical power generated by the turbine. It can be
computed as P_el = ω G ⋅ C em but in this paper, its cost is given by:
normalized value was used. u = −K ⋅ x + Kr ⋅ r ,
where K is given by K = R −1 ⋅ B T ⋅ Pc , Pc is given by the
solution to the equation:
Pc ⋅ A + AT ⋅ Pc − Pc ⋅ B ⋅ R −1 ⋅ B T ⋅ Pc + Q = 0 (10)
while Kr is being defined by:
K r = − ( R1 ) −1 ⋅ B T ⋅ (( A − B ⋅ K ) T ) −1 ⋅ C T ⋅ Q (11)
Fig.5 The block scheme of the controlled system
The other output variables that we are interested in are ωT This matrix ensures the reference input is scaled in order
because the goal is to try to maintain it constant to its to become equal to the feedback signal provided by the LQR
nominal value, no matter the changes that appear in the regulator. This algorithm guaranties that no matter, any two
environment, the flap mode of the blades ζ and of the tower symmetric and positive definite matrixes Q and R that we
yT respectively, because, it is desired that these variables be chose in order to minimize the quadratic criteria, there is
as much as possible. always a matrix Pc, also symmetric and positive definite, that
The two available control variables are the pitch angle and represents the solution of the Ricatti equation (10).
the electromagnetic torque. The numeric values of the wind Through this criterion, by replacing the variables y and u
turbine’s parameters can be found in the APPENDIX, at the by the corresponding vectors presented in Section 2, one
end of the paper. tries to minimize the flap mode of the blades and the tower
oscillation respectively, maintain the electrical power level
III. GENERAL PROCEDURE OF THE LINEAR and the angular speed of the rotor at the desired levels while
QUADRATIC CONTROLLER DESIGN computing the appropriate command.
The typical rule for choosing the weighting matrixes R
As previously said, there is a large variety of control and Q is the Bryson’s rule, which states that these matrixes
techniques that were applied to wind turbines in a permanent should be selected as diagonal with the non-zero elements
attempt to improve their functioning and to benefit as much scaled so that the variables that appear in the optimization
as possible from the energy that they can produce. In criterion have a maximum value of one [10] [11] [12].
literature, one can find proposed solutions for mono-variable This is important especially for the situations when the
systems as well as for multi-variable ones. units used for the different components of the command and
In [8], for instance, one can find a compared study made state vectors are numerically very different from each other.
upon the simulation results obtained with three controllers: a This is also our case, in the command vector, for instance,
classical PID regulator, a full state feedback and a fuzzy the pitch angle and the electromagnetic torque have different
controller. The author’s conclusion is that the PID controller order of degree units.
ensures good performances with power regulation but not Although Bryson’s rule gives good results, often it is just a
with reducing the structure’s mechanical loads. In the same starting point of a trial and error procedure of choosing these
time, the full-state feedback controller manages to reduce matrixes, in order to obtain the desirable properties for the
these loads even under turbulent conditions. closed loop system. Weights reflect the relative importance
The idea of conveniently sizing a trade-off between energy given to the state with respect to the control effort.
efficiency and increasing the lifetime of the wind turbines by Therefore, for our system, if one chooses large values for
alleviating fatigue loads is continuously being paid special Q compared to the values in R, one gives a higher
attention, even when employing controllers like PI or PID. importance on the minimization of the mechanical weights
However, these approaches do not allow a rigorous control and a lower importance to the command effort [13] [14].
design in order to perform a fine tuning of the trade-off
between the energy performance and the reliability demands IV. RESULTS
[9]. The simulations were done using MATLAB/SIMULINK
These aspects, together with its design simplicity and the software and the results proved good performances. The
advantages it could bring, lead us to the idea of choosing a chosen operating point for the linearization of the system
state feedback linear quadratic controller (LQR) for this corresponds to the average value of the wind speed of 18m/s.
study. In Fig. 6 one can see the scheme that was used for the
For its design, one imposes a quadratic cost function simulation.
defined as The two reference variables, for the normalized electrical
∞

∫ (y )
power P_el_ref and for the angular rotor speed ωT_ref
J= T
⋅ Q ⋅ y + u T ⋅ R ⋅ u ⋅ dt (9) respectively, were chosen as constants with the appropriate
0 values because the goal is to minimize the variations of the
The feedback control law that minimizes the value of this electrical power extracted around the nominal value of the
generator and we also want to keep the rotor speed constant.

Fig. 8 The variation of the angular speed of the rotor

Fig. 6 The simulation of the system with LQR regulator

The weighting matrices mentioned in (9) and used for
these simulations are
 0.3 0 0 0 
 
 0 400 0 0 
R = I, Q =  .
0 0 1 0
 
 0 0 0 1 

These values were chosen using the methods mentioned
above and also based on the fact that they provided very
good performance of the system in terms of achieving good
responses and not very strong control actions.
Fig. 9 The tower bending movement in the direction of the nacelle
The cost function was written in the following form:
∞
J = ∫ (x ⋅ Q1 ⋅ x + u T ⋅ R1 ⋅ u + 2 ⋅ x T ⋅ S ⋅ u ) , where
T

Q1 = C1T ⋅ Q ⋅ C1 , R1 = R + D1T ⋅ Q ⋅ D1 , S = C1T ⋅ Q ⋅ D1 ,

and the matrices C1 and D1 being the truncated blocks from
the system matrices C and D. These matrices contain the
lines and columns from C and D corresponding to the control
variables Cem and β.
The system is controllable and it does not contain
unobservable modes. One important property of LQ Fig. 10 The variation of the first flap mode of the blades
regulators is that provided these conditions, they guarantee It can be observed that the electrical output power and the
nominally stable closed loop systems. angular speed of the rotor manage to follow the reference
In Fig. 7-10, one can see the results obtained in and to maintain their nominal imposed values.
simulation. In the same time, the variables that were meant to be
minimized, namely the first flap mode of the blades and the
bending of the tower, have extremely small values. The
blades have a deviation of about 5mm while the tower has an
insignificant movement on the horizontal direction.

APPENDIX

THE NUMERICAL VALUES OF THE WIND TURBINE

PARAMETERS

Symbol Physical measure Value

Fig. 7 The normalized electrical output power of the turbine

Jt Turbine inertia 214 000 Kg
* m2
 Jg Generator inertia 41 Kg * m2
MT Tower and nacelle mass 35000 kg
Mp Blade mass 3000 kg
kP Blade Stiffness Coefficient 1000 Kg *
m2/s2
kT Tower Stiffness Coefficient 8500 Kg *
m/s2
kA Drive Shaft Stiffness 11000 Kg *
Coefficient m2/s2
dP Blade Damping coefficient 10 000 Kg *
m2/s
dT Tower Damping coefficient 50 000 Kg *
m/s
dA Drive shaft damping 60 000 Kg *
coefficient m2/s
rP Distance from the rotor hub 8m
N Number of blades 2
D The rotor diameter 34 m
Pn Nominal Power 400 kW
nom Nominal rotor speed 4 rad/s
h Tower height 47 m

REFERENCES
[1] D. O. Kirk, “Optimal control theory-An Introduction”, Dover
Publications, 2004, pp. 3-5.
[2] N.A Cutululis, I. Munteanu, E. Ceanga, “Optimal control structure for
variable speed wind power system”, The annals of “Dunarea de Jos”
University of Galati, Fascicle III, pp 95-102, 2002.
[3] A. D. Wright, “Modern control design for flexible wind turbines –
Technical Report”, NREL/TP-500-35816, July 2004, pp.23-26.
[4] D. F. Bianchi, H. Battista, “Wind turbine control systems –Principles,
Modeling and gain scheduling design”, Springer-Verlag, London,
2002
[5] L. Lupu, B. Boukhezzar, “Pitch and torque control strategy for
variable speed wind turbines”, Proceedings EWEC, Athens, 2006.
[6] F. A. Vanegas, M. Zamacona, “Robust control solution of a wind
turbine - A simulation study”, International Master’s Thesis in
Information Technologies, Halmstadt University, February, 2008.
[7] F. Lescher, P.Borne, “Robust gain scheduling controller for pitch
regulated variable speed wind turbine”, Studies in Informatics and
Control, vol 14, No.4, pp 299-315, 2005.
[8] M. Jelavic, I. Petrovic, “Design of a wind turbine pitch controller for
loads and fatigue reduction”, Electrical Engineering Institute,
Proceedings of the European Wind Energy Conference & Exhibition –
EWEC, Milan, Italy, 2007.
[9] I. Munteanu, E. Ceanga, “Optimal control of wind energy systems –
Towards a global approach”, Springer, 2007.
[10] B. D. O. Anderson, J. B. Moore, “Optimal control: Linear quadratic
methods”, Prentice Hall, Englewood Cliffs, NJ, 1990.
[11] F. C. Callier, J. L. Willems, “Criterion for convergence of the solution
of the Riccati differential equations”, IEEE Trans. Automat. Contr.
AC-26, pp. 1232-1242, 1981.
[12] R. Sivan, H. Kwakernaak, “Linear optimal control systems”, Wiley-
Interscience, 1st edition, October 1972.
[13] A. E. Bryson, “Applied linear optimal control – Examples and
Algorithms”, Cambridge University Press, 2002.
[14] J. L. Hellerstein, Y. Diao, S. Parekh, D. M Tilbury, ”Feedback control
of computing systems“, John Wiley & Sons, Inc., 2004.