A Model Reference Adaptive Controller for a Wind Energy

Conversion System Based on a Permanent Magnet Synchronous

Generator Fed by a Matrix Converter
Majid Pahlevaninezhad, Suzan Eren, Alireza Bakhshai and, Praveen Jain
Queen’s University, Kingston Ontario
[email protected], [email protected] [email protected], [email protected]

Abstract- This paper presents a Model Reference Adaptive link capacitor and provides a direct AC/AC conversion.
Control (MRAC) approach for a Wind Energy Conversion Thus, it is a good candidate for WECS applications. The
System (WECS) to track the maximum power point. This matrix converter can control the magnitude, frequency and
WECS includes a permanent magnet synchronous generator phase angle of the output voltage as well as the input power
(PMSG) fed by a matrix converter. Since the mechanical factor. Despite the attractive features of the matrix
power generated by the wind turbine is a function of its shaft
speed at a given wind velocity, the proposed controller
converter, the matrix converter suffers from some problems
provides the desired voltage at the output of the matrix such as low voltage gain, complicated control, bi-directional
converter so as to control the generator speed. This controller switches and lack of ride-through capability [6-7]. However,
is based on the model reference adaptive control approach to there are many papers that are reporting solutions to
track the optimum speed. In addition, the controller is robust mitigate these difficulties.
against system uncertainties and high frequency disturbances. Due to the erratic nature of wind-based energy systems,
Theoretical analysis verifies the global stability and there are so many uncertainties in WECSs. Thus, the control
performance of the proposed approach. Simulation results are system of the WECS should accommodate for the effects of
included which show that the proposed approach has a uncertainties and keep the system stable against large
superior performance compared to the conventional PI
controller.
parameter variations. The conventional PI-based controllers
can not fully satisfy stability and performance requirements.
Index term- Wind Energy, PMSG, Matrix Converter, On the other hand the system is highly nonlinear and has a
Maximum Power Point Tracking, Model Reference Adaptive wide range of operating points; therefore linearization can
Controller. not be used to design the controller. Adaptive control
methods can be employed to effectively solve this issue [11-
14]. In this paper, a model reference adaptive control
I-Introduction approach is used to design the controller.
This paper is organized as follows. First, the proposed wind
Nowadays, renewable energies are becoming increasingly energy system is presented in section II. In section III, the
important as alternative energy sources. Many factors such wind turbine aerodynamic model is given. Maximum power
as diminishing fossil-fuel resources, energy security point tracking is explained in section IV. In section V, the
concerns, and global warming increase the need for proposed controller is introduced. In section VI, the stability
renewable energies. Their main advantages are the analysis of the proposed controller is investigated. The
elimination of harmful emissions and inexhaustibility while matrix converter structure used to connect the generator to
the main drawbacks are the cost and uncontrollability [1-4]. the grid is given in section VII. Finally, the simulation
Wind is one of the most abundant energy sources, which can results are presented in section VIII.
be harnessed by wind turbines and converted to electricity
by generators. Thus, a wind energy conversion system
(WECS) consists of a wind turbine, an electric generator II-Proposed Wind Energy Conversion System
and a power converter [5].
Different configurations of WECSs have been reported in Fig.1 illustrates the WECS block diagram. The mechanical
the literature [8]. The most commonly used generators are energy is converted to electricity by a generator and a
Doubly-Fed Induction Generators (DFIG), Squirrel Cage matrix converter is the interface between the generator and
Induction Generators (SCIG), and Permanent Magnet the utility grid. The adaptive controller provides proper
Synchronous Generators (PMSG). switching signals for the matrix converter to extract the
The conventional back-to-back voltage source converters maximum power from the wind turbine. It also
are widely used to connect the generator to the grid. In this accommodates for the system uncertainties through an
structure, DC-link capacitors are applied to decouple the adaptive algorithm.
generators and grid. However, they are bulky and have a
limited life time. The matrix converter does not use the DC-

978-1-4244-4649-0/09/$25.00 ©2009 IEEE 65

 According to Fig. 2, at a fixed wind velocity the maximum
achievable power occurs at a specific shaft speed. Thus, the
optimal control of a WECS means it must track the optimal
value of the shaft rotational speed [9]
v.λopt
ωT (opt ) = (5)
R
Due to the erratic nature of wind, nonlinear variant behavior
of the WECS, poor reliability of the measured information
Fig.1 WECS Block Diagram and unknown parameters, the control method used to track
the optimal speed should be robust.
III- Wind Turbine Aerodynamic Model
P ∂ ∂P V
The wind turbine model provides an expression for the ω ∂ω
ref

∂t
torque as follows: [9]
T T

Γw = Γw ( β , v, ωT ) (1)
where β is the rotor blade pitch angle, v is the wind Fig.3 MPPT block diagram
velocity and ωT is the shaft angular speed. The torque Fig. 3 shows the proposed maximum power point tracker for
model is given by [24] the WECS. The approach is based on the calculation of the
P 1 power gradient. Since the power curve is convex, there is no
Γw = w = .π .ρ .v 2 .R 3 .CΓ (λ ) (2)
ωT 2 local maximum and thus, the algorithm can find the
maximum point based on the power gradient. The search
where CΓ = C p / λ is the torque coefficient, R is the blade
algorithm provides the reference value (for the shaft
length of the wind turbine, ρ is the air density and rotational speed) which is the command signal for the
λ = RωT / v is the tip speed ratio of the wind turbine. C p is adaptive model reference controller. This controller
produces the desired voltage vector for the matrix converter.
a function of tip speed ratio and blade pith angle and is
usually found by curve fitting [9] IV- Model Reference Adaptive Controller
− c5
c2 λi
C p (λ , θ ) = c1 ( − c3 β − c4 )e + c6λ (3) Fig. 4 shows the proposed controller block diagram. In this
λi
diagram, the reference value for the speed is provided by the
1 1 0.035 search algorithm and is applied to the nonlinear controller.
= − 3 (4) The nonlinear controller produces the reference voltage
λi λ + 0.08β β + 1
vector for the matrix converter so as to get the optimum
where c1=0.5176, c2=116, c3=0.4, c4=5, c5=21 and shaft speed. The coefficients of this controller are adaptively
c6=0.50068. changed based on the error between the output of the
Typical curves of wind output power and torque versus the reference model and the output of the WECS. Thus, it can
shaft angular speed is illustrated in Fig.1. accommodate for the effects of parameter uncertainties and
0.25 is robust against parameter variations.
0.2

0.15

0.1
P

0.05

-0.05

-0.1
1 2 3 4 5 6 7
â
WT

Fig.2 Wind Turbine Power versus Shaft Angular Speed Fig.4 Model Reference Adaptive Controller
III- Maximum Power Point Tracking (MPPT)
A- Nonlinear Controller Design for WECS

978-1-4244-4649-0/09/$25.00 ©2009 IEEE 66

 ⎛ e&1 ⎞ ⎛ e2 ⎞
To design the controller, a mathematical model for each ⎜ ⎟ ⎜ ⎟
⎜ e&2 ⎟ = ⎜ − a5 (a6 − a5ω + a4iq ) + a4 (− a1iq + pω id − a3ω g + a2uq ) ⎟
block in Fig. 1 must be derived. A model for the drive train ⎜ e& ⎟ ⎜⎜ ⎟⎟
system, the interface between the wind turbine and the ⎝ 3⎠ ⎝ − a1id + pω iq + a2u d ⎠
generator, is derived in [9] as: ⎛ 0 ⎞
ω g = ngωT ⎜ ⎟
+ ⎜ a2uq + k r1 x1r + k r 2 x2 r − k r1ωref ⎟
dω g η ⎜⎜ a2u q + k r 3 x3r − k r1idref ⎟⎟
Jh = ⋅ TW − Tg (6) ⎝ ⎠
dt ng (11)
η This system includes the uncertain parameters and these
where J h = (J1 + JW ) ⋅ + J 2 + J g and J1 and J 2 are the parameters will be estimated so as to get a robust
ng
performance against parameter variations. Therefore,
inertias of the multiplier gearings. equation (11) is rewritten as
The matrix converter does not have any energy storage
⎛ e&1 ⎞ ⎛ e2 ⎞
element and thus has no dynamics. Thus, the WECS is ⎜ ⎟ ⎜ ⎟
modeled as a third order system defined by: e&
⎜ 2⎟ = ⎜ − a (
5 6a − a 5ω + a i
4 q ) + a 4 ( − a i
1q + pω i d − a ω
3 g + a 2 q ⎟
u )
⎜ e& ⎟ ⎜⎜ − a1id + pω iq + a2u d ⎟⎟
d id ⎝ 3⎠ ⎝ ⎠
= −a1id + pω iq + a2u d
dt ⎛ Δf1 ⎞ ⎛ 0 ⎞
d iq ⎜ ⎟ ⎜ ⎟
= −a1iq + pω id − a3ω g + a2u q (7) ⎜ Δf 2 ⎟ + ⎜ a2 u q + k r1 x1r + k r 2 x2 r − k r1ω ref ⎟
dt ⎜ Δf ⎟ ⎜⎜ a2u q + k r 3 x3r − k r1idref
⎟⎟
⎝ 3⎠ ⎝ ⎠
d ωg
= a6 − a5ω + a4iq (12)
dt where Δf i represents the effects of uncertainties on the error
R 1 Pφ 3 Pφ model. The uncertain parameter errors are defined as
Where a1 = , a2 = , a3 = , a4 = ,
L L L 2Jh a~i = ai − aˆ i (13)
B nΓ If we look at the first row of equation (12), e2 can be
a5 = and, a6 = .
Jh J h ng considered as a virtual controller and thus its reference is
The control objective is to have the generator rotational given by
speed track the reference value derived by the MPPT e2 ref = − k1e1 − Δf1 (14)
algorithm. Then the new variables are defined as
To transform the system to the appropriate form for e~1 = e1 ; e~2 = e2 − e2 ref ;~
e3 = e3 (15)
designing the controller, new variables are defined as
Based on the new variables, the system dynamics is given
x1 = ω
by
x2 = ω& ⎛~ &⎞ ⎛ ⎞
⎜ e1 ⎟ ⎜ e2
⎟
x3 = id ~&
⎜e2 ⎟ =⎜− a5 (a6 − a5ω + a4iq ) + a4 (−a1iq + pωid − a3ωg + a2uq )⎟
Therefore the system dynamics is given by ⎜⎜~& ⎟⎟ ⎜⎜ ⎟⎟
e − a1id + pωiq + a2ud
⎝ 3⎠ ⎝ ⎠
⎛ x&1 ⎞ ⎛ a6 − a5ω + a4iq ⎞
⎜ ⎟ ⎜ ⎟ ⎛ e2ref ⎞ ⎛ Δf1 ⎞ ⎛ 0 ⎞
⎜ x&2 ⎟ = ⎜ − a5 (a6 − a5ω + a4iq ) + a4 (−a1iq + pω id − a3ωg + a2uq ) ⎟ (8) ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ x& ⎟ ⎜⎜ ⎟⎟ &
+ ⎜− k1(e2 + e2ref + Δf1) − Δf1 ⎟ + ⎜Δf2 ⎟ +⎜a2uq + kr1x1r + kr2x2r − kr1ωref ⎟
⎝ 3⎠ ⎝ − a1id + pω iq + a2ud ⎠ ⎜ 0 ⎟ ⎜Δf ⎟ ⎜⎜ a u + k x − k i ⎟⎟
The reference model is considered as ⎝ ⎠ ⎝ 3⎠ ⎝ 2 q r3 3r r1 dref ⎠
⎛ x&1r ⎞ ⎛ 0 1 0 ⎞⎛ x1r ⎞ ⎛ 0 0 ⎞ (16)
⎜ ⎟ ⎜ ⎟⎜ ⎟ ⎜ ⎟⎛ ω ref ⎞ Finally, the Lyapunov function is defined as
⎜ x& 2 r ⎟ = ⎜ − k r1 − k r 2 0 ⎟⎜ x2 r ⎟ + ⎜ k r1 0 ⎟⎜⎜ ⎟ (9)
idref ⎟⎠ 1 1 1 1
6 ~2
ai
⎜ x& ⎟ ⎜ 0
⎝ 3r ⎠ ⎝ 0 ⎟⎜ ⎟ ⎜ ⎟
− k r 3 ⎠⎝ x3r ⎠ ⎝ 0 k r 3 ⎠⎝ V = e~12 + e~22 + e~32 +
2 2 2 2 i =1 μi ∑ (17)
Therefore, the difference between the state variables and the
reference variables is defined as the error This is the Control Lyapunov Function (CLF) and the
control laws and adaptation laws are obtained so as the
⎛ e1 ⎞ ⎛ x1 − x1r ⎞
⎜ ⎟ ⎜ ⎟ derivative of this Lyapunov function is negative definite or
⎜ e2 ⎟ = ⎜ x 2 − x 2 r ⎟ (10) at least negative semi-definite. Thus, the time derivative of
⎜e ⎟ ⎜ x − x ⎟ the Lyapunov function is given by
⎝ 3⎠ ⎝ 3 3r ⎠
6 ~~
Thus, error dynamics is given by ai a&i
V& = e~1~
e&1 + ~
e2 e~&2 + ~
e3e~&3 + ∑i =1
μi
(18)

978-1-4244-4649-0/09/$25.00 ©2009 IEEE 67

Therefore, the following control laws and adaptation laws Which is bounded, hence; V& is uniformly continuous.
make the Lyapunov function derivative negative. Invoking Barbaret’s lemma, the errors converge to zero.
ud =
1
aˆ 2
( )
aˆ1id − pω iq + k 2 ed + k r1 (ω ref − x1r ) − k r 2 x2 r (19)
lim E (t ) = 0 (26)
t →∞
⎛ k3eq − aˆ 4 e + aˆ1iq + pω id + aˆ3ω + aˆ5iq ⎞
⎜ ⎟ which means that the objective of tracking is achieved under
1 ⎜ aˆ5 k1 ⎟
2
aˆ5 2 k1 parameter uncertainties and disturbances. Through this
uq = ⎜− ω+ aˆ6 − k1iq + aˆ5ω − aˆ6 ⎟ (20) analysis, global stability is proven.
aˆ 2 ⎜ aˆ 4 aˆ 4 aˆ 4 aˆ 4 ⎟
⎜ + k (i − x3r ) ⎟
⎝ r 3 dref ⎠
1 1 ( d d q q)
&aˆ = μ e i + e i

a&ˆ 2 = μ 2 (− ed ud − equq )
VIII- Matrix Converter
aˆ&3 = μ3 (ω eq )
The matrix converter is an array of semiconductors which
⎛ aˆ k ⎞
a&ˆ4 = μ 4 ⎜⎜ − eiq + 5 eqiq − 1 eqiq ⎟⎟ interfaces between two multi-phase systems with different
⎝ aˆ 4 aˆ 4 ⎠ frequencies. Fig. 5 shows a three-phase-to-three-phase
matrix converter with 9 bi-directional switches. The
⎛ aˆ k ⎞
a&ˆ5 = μ5 ⎜⎜ eω − 5 ω eq + 1 ω eq ⎟⎟ operating principles of the matrix converter and its
⎝ aˆ 4 aˆ 4 ⎠ modulation schemes are comprehensively explained in [4-
⎛ aˆ k ⎞ 5]. According to Fig. 5, the output voltage reference
a&ˆ6 = μ6 ⎜⎜ e − 5 eq + 1 eq ⎟⎟ (22) vector and input current reference vector are applied by the
⎝ aˆ4 aˆ4 ⎠ MPPT to the matrix converter controller. The matrix
Substituting (19) and (20) into (18), the Lyapanov function converter produces a set of voltages to control the PMSG
derivative is given by angular speed.
V& = − k e~ 2 − k e~ 2 − k e~ 2
1 1 2 2 3 3 (23)

VII-Stability Analysis

According to (23), the Lyapunov function derivative is

negative semi-definite (not definite) so the stability theorem
only guarantees the global boundedness and regulation of
e~1 , e~2 , e~3 . Whereas, it only guarantees the boundedness
~ [10]. Therefore, the estimated
but not the regulation of a i
V
parameters do not necessarily converge to the actual values. o , ref

This can be proven by Barbalet’s lemma [10].

I in , ref

∫ V& dt = −∫ ( )
t t Fig.5 Matrix Converter
k e~12 + k 2 e~22 + k3e~32 dt
0 0 114 4424443
E (T )

= V (t ) − V (0) The matrix converter is controlled by selecting an

t appropriate set of switching functions. There are two
∫ E (t ) dt = V (0) − V (t )
0
(24) general approaches to derive the switching functions,
namely the direct and indirect modulation approaches [4-5].
∞
∫ E (τ ) dτ
In this paper, the direct method has been applied. In this
Therefore, lim exists and is finite. It is only
t →∞ 0 method, the input current vector and output voltage vector
required to show that E (t ) is uniformly continuous. The are given as the reference vectors to the modulator and the
switching functions are determined so as to get the desired
time derivate of V& (t ) is output voltage and input current. Fig. 6 illustrates the
V&& = −2k1e~1e~&1 −2k 2 e~2 ~
e&2 − 2k3~
e3~
e&3 (25) simulated direct space vector modulation in PSIM.

978-1-4244-4649-0/09/$25.00 ©2009 IEEE 68

 Fig.8-1 Line and Phase Voltages

Fig.6 SVM for Matrix Converter

IX- Simulation Results

Fig. 7 shows the simulated WECS. This system is simulated Fig.8-2 Line Voltages
in PSIM.

S ij VAB
ω Vref
IABC

Fig. 10 Shaft speed response with the proposed controller

Fig.7 WECS Block Diagram

The PMSG parameters are given in Table I .

Table I: PMSG parameters

Rs=1.93Ω Ld=8mH Lq=8mH
P=2 J=2m B=0.0002
Po=1KW

Fig. 8-1 and 8-2, illustrate the output line and phase voltages
and Fig. 9 shows the line voltage frequency spectrum. Fig. 9
and Fig. 10 illustrate the shaft speed tracking for a step
change in the wind speed with the proposed controller and Fig. 11 Shaft speed response with PI controller
the conventional PI controller respectively. The proposed
controller is much faster than the conventional PI one. The
MPPT is shown in Fig. 12 and Fig. 13 for step and
sinusoidal changes in the wind speed profile respectively.

978-1-4244-4649-0/09/$25.00 ©2009 IEEE 69

 Motion Control Conference, 2006. IPEMC '06. CES/IEEE 5th
International, page(s): 1-11 Aug. 2006.
[4] Wheeler, P.W. Rodriguez, J. Clare, J.C. Empringham, L.
Weinstein, A. , “Matrix converters: a technology review ”, IEEE
Transactions on Industrial Electronics, Vol. 49 page. 276-288,
April 2002.
[5] Domenico Casadei, , Giovanni Serra, , Angelo Tani, and Luca
Zarri, “Matrix Converter Modulation Strategies: A New General
Approach Based on Space-Vector Representation of the Switch
State” IEEE Transactions on Industrial Electronics, Vol. 49, No. 2,
April 2002.
[6] Christian Klumpner, Ion Boldea and Frede Blaabjerg,
“Limited Ride-Through Capabilities for Direct Frequency
Fig. 12 Shaft speed response to the wind speed step change Converters” IEEE Transactions on Power Electronics, Vol. 16,
No. 6, November 2001.
[7] Kwak, S. Toliyat, H.A., “An Approach to Fault-Tolerant
Three-Phase Matrix Converter Drives” IEEE Transaction on
Energy Conversion,Vol. 22, Issue 4, Page(s):855 – 863, Dec.
2007.
[8] Klumpner, C. Blaabjerg, F. Thogersen, P. ,” Evaluation of
the converter topologies suited for integrated motor drives” ,
Industry 38th Applications Conference, IAS, page(s): 890- 897,
vol.2 2003.
[9] Munteau, I., Bratcu, A. I. Cutululis, N. A., Ceanga, E.,
“Optimal Control of Wind Energy Systems” Book, Springer-
Verlag London 2008.
[10] Krstic, Kanellakopoulos, Kokotovic, “ Nonlinear & Adaptive
Control Design” Book, John Wiley & Sons, 1995.
[11] Parks, P. ,” Liapunov redesign of model reference adaptive
Fig. 13 Shaft speed response to the wind speed sinusoidal change control systems” IEEE Transactions on Automatic Control,
July 1966, 11, 362-367 .
[12] Kuo-Kai Shyu; Ming-Ji Yang; Yen-Mo Chen; Yi-Fei Lin;,
VIII- Conclusion “Model Reference Adaptive Control Design for a Shunt Active-
Power-Filter System” IEEE Transactions on Industrial Electronics,
This paper presented a model reference adaptive nonlinear Volume 55, Issue 1, Jan. 2008 Page(s):97 – 106.
approach for the maximum power point tracking of a [13] Kim, K.-H.; “Model reference adaptive control-based adaptive
WECS. The proposed approach is also robust against current control scheme of a PM synchronous motor with an
system parameter uncertainties. In addition, the closed loop improved servo performance Power” Electric Applications, IET
stability has been analyzed for the proposed controller Volume 3, Issue 1, January 2009 Page(s):8 – 18.
through a nonlinear analysis and global stability has been [14] Liu Hsu; Costa, R.R.; Lizarralde, F.; “Lyapunov/Passivity-
Based Adaptive Control of Relative Degree Two MIMO Systems
proven. A matrix converter is utilized as an interface With an Application to Visual Servoing” Transactions on
between the wind turbine and the load. Direct space vector Automatic Control, IEEE Volume 52, Issue 2, Feb. 2007
modulation is used to control the matrix converter switches. Page(s):364 – 371.
Theoretical analysis and simulation results are given to
verify the feasibility and performance of the proposed
approach.

References
[1] Blaabjerg, F. Zhe Chen Kjaer, S.B., “Power electronics as
efficient interface in dispersed power generation systems” , IEEE
Transactions on Power Electronics Sept. 2004 Vol. 19, page(s):
1184- 1194.
[2] J. M. Carrasco, L. G. Franquelo, J. T. Bialasiewicz, E. Galvan,
R. C. P. Guisado, M. A. Martin Prats, J. I. Leon, N. M. Alfonso,
“Power electronic systems for grid integration of renewable energy
sources: a survey” IEEE Trans. Industrial Electronics, vol. 53, no.
4, pp. 1002-1016, August. 2006.
[3] F. Blaabjerg Z. Chen R. Teodorescu F. Iov , “Power
Electronics in Wind Turbine Systems” Power Electronics and

978-1-4244-4649-0/09/$25.00 ©2009 IEEE 70