Adaptive Model Predictive Control of DFIG-based

Wind Farm: A Model-Free Control Approach

Zahra Rafiee Rasool Heydari Mansour Rafiee
Faculty of Electrical Engineering Electrical Engineering Department Faculty of Electrical Engineering
Shahid Beheshti University University of Southern Denmark Shahid Beheshti University
z rafiee@sbu.ac.ir rah@sdu.dk m rafiee@sbu.ac.ir

Mohammad Reza Aghamohammadi Jose Rodriguez Frede Blaabjerg

Faculty of Electrical Engineering Department of Engineering Science Department of Energy Technology
Shahid Beheshti University Universidad Andres Bello Aalborg University, Denmark
m aghamohammadi@sbu.ac.ir jose.rodrigues@unab.cl fbl@et.aau.dk

Abstract—In this paper a new control strategy of doubly- control reduces the use of machine parameters and reduces the
fed induction generator based wind farms (DFIG-based WFs) is complexity of the vector control algorithm. The DTC method
proposed. Since DFIG has an inherent nonlinear behaviour along directly controls the flux and torque of the DFIG by selecting
with condition variant parameters, the system model cannot
easily be extracted. Therefore, the conventional model predictive voltage vectors from the switching table using torque and
control (MPC) of DFIG-based WFs cannot perform accurately. stator flux information. One of the problems with the DTC
In this paper a novel model-free adaptive MPC structure is method is that its performance deteriorates during start-up
presented to adaptively update the system model by utilizing and in low-speed operations [6]. To address this problem,
model identification and auto-regressive moving average (AR- a modified switching table can be used to select voltage
MAX) model, for each sampling time. Simulation results verify
the performance of the proposed control structure of DFIG-based vectors in the appropriate sequence or predictive techniques.
WFs compared to the conventional control strategies. Conventionally, linear controllers (e.g. proportional-integral)
Index Terms—Adaptive model predictive control, Doubly fed are widely employed to control the rotor current [7]. However,
Induction Generator (DFIG), Auto-regressive–moving-average linearized control structures are designed for a particular
model (ARMAX). operation point [8]. Due to the nonlinear nature of inverters,
the limitation of switching modes, and the nature of changes
I. I NTRODUCTION in machine parameters during operation, nonlinear control
Doubly-fed induction generator (DFIG) as a variable speed techniques such as direct power control (DPC) have been
generator has been widely utilized in the wind farms. DFIG proposed in recent decades [9], [10]. The basic principles of
allows wind system to operate in various wind ranges and these techniques are based on the selection of a voltage vector
increases the system efficiency, reduces converter rating, and and an average control signal for instantaneous control of
diminishes acoustic noise [1]. The rotor of DFIG is connected electromagnetic torque or active power and flux size or reactive
to the grid through a back-to-back converter while its stator power. The main advantages of direct power control are high
is connected to the grid directly. The back-to back converter dynamics, strong performance and ease of implementation.
controls the active and reactive power between DFIG and the The conventional proportional-integral (PI) controllers are
main grid [2]. well applied in the engineering field because of their reliability
Therefore, by using rotor current controllers which regulates and simplicity. However, with the approximated linearized
the decoupled rotor currents through the rotor side converter DFIG model, the parameters of PI controllers are usually
(RSC) and the grid side converter, the reactive and active tuned. In this situation, there must be a compromise be-
power of DFIG can be controlled [3]. Different control meth- tween the robustness and control performance. Furthermore,
ods use for controlling the grid side converter (GSC) and the control performance will be degraded significantly under
rotor side converter (RSC) of the DFIG by creating switching harmonic grid voltage situation [11], [12].
pulses for them in such a way that the desired conditions are Hence, PI controllers don’t have an acceptable performance
created. Control of DFIG-based wind farms is traditionally in nonlinear systems. Motivated by the aforementioned prob-
based on stator flux vector control. Active and reactive power lem, various advance control strategies for DFIG-based wind
control is obtained by controlling the rotor current, where the turbine have been proposed in the literature, e.g., multi-mode
active power is proportional to the orthogonal component of control [13], robust control [14], fuzzy control [15], inner
the rotor current and the reactive power is proportional to the control [16], neural network control [17], and sliding mode
direct component of the rotor current [4]. Direct torque control controller [18].
(DTC) [5] is another method of vector control. Direct torque All these control techniques show a good performance in

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on March 10,2023 at 14:54:58 UTC from IEEE Xplore. Restrictions apply.
 DFIG Wind Turbine
120kV 25kV 575V
25kV
Y/Delta 30 Km Delta/Yg
GRID DFIG
HV T2 Line LV T1
~120kV/25kV PCC 25kV/575V Gearbox

Filter
Grounding
Transformer AC DC Crowbar
C resistance
C
DC AC GND1
GSC DC link RSC
Chopper

Fig. 1. System under study.

terms of disturbance rejection and robustness. However, these In the expressed state space, the voltage input signal applied
control structures rely on the linearized system model and to the rotor and the output signal is its current. Assuming that
generator parameters. Therefore, if the operation point varies, the system is at time t, the MPC determines the rotor voltage u
the controller cannot react fast and accurate. Recently model (t) such that the rotor current y (t + 1) is as close as possible
predictive control (MPC) is widely employed in power elec- to its reference value. In MPC design, an integrator is first
tronic applications [19]–[24]. However, the MPC relies on the added to the system to eliminate steady state error. Therefore,
accurate model of the system. Therefore, any model mismatch the following new state space are defined:
or parameter changes degrade the control performances.
" # " #" #
In this paper, a model-free adaptive MPC (A-MPC) is  ∆xm (k + 1) Am OmT
∆xm (k)
proposed to control DFIG-based wind turbine through the

 =
y(k + 1) Cm Am 1 y(k)



RSC. In this method, transfer functions among idr and vdr ,



| {z } | {z }| {z }
" A
 x(k+1) x(k)
iqr and vqr are adaptively updated at each sampling time. 

 #
Bm

Then, based on the updated transfer functions, the A-MPC


 + ∆u(k)
generates the controller signals. Therefore, with controller can Cm Bm (2)

be operated with no need to any knowledge about the system

 | {z }
B


 " #
model. 
y(k) = [O 1] ∆xm (k)


m
The paper is organized as follows. Section II explains



 | {z } y(k)
the system under study. The proposed conrtol structure has



 C | {z }
x(k)
been illustrated in Section III. The simulation results are
discussed and presented in Section IV, while the conclusions
are followed finally. In the next step, based on the Augmented Model, vari-
ables and states predictions are obtained along the prediction
II. S YSTEM D ESCRIPTION horizon. Np is prediction horizon and Nc is control horizon
Fig. 1 depicts an exemplary wind farm (WF) connected with.Therefor, the prediction of state variables is computed
to the grid. It comprises the six 1.5 MW DFIG-based Wind over the prediction horizon, which is performed as a recursive
turbine with 575 V and 60 Hz generator-turbines. In this placement:
system, WF is connected to 120 kV transmission line through
transformers T1 and T2 and a 30 km line. The test system x(ki + 1|ki ) = Ax(ki ) + B∆u(ki )
parameters are based on the system model presented in [25]. x(ki + 2|ki ) = Ax(ki + 1|ki ) + B∆u(ki + 1)
= A2 x(ki ) + AB∆u(ki ) + B∆u(ki + 1)
III. P ROPOSED CONTROL STRUCTURE ..
.
A. Model-Free A-MPC Strategy x(ki + NP |ki ) = ANP x(ki ) + ANP −1 B∆u(ki )
An accurate model of the system is required to design the +ANP −2 B∆u(ki + 1) + ... + ANP −NC B∆u(ki + NC − 1).
(3)
MPC. The discrete state space model is as follows:
( x(ki + 1|ki ) and x(ki + Np |ki ) means a one and Np step
xm (t + 1) = Am (t)xm (t) + Bm (t)u(t) forward prediction of state variables based on system infor-
(1)
y(t) = Cm (t)xm (t) mation until time ki , respectively. So, the outputs prediction

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on March 10,2023 at 14:54:58 UTC from IEEE Xplore. Restrictions apply.
 Average Model

Slip RSC
Rings C.T
C.T +
Elec. DFIG
Bus
Udc_ Link
-
P.T
usabc irabc
isabc
Photoelectric
Encoder abc/ αβ
ωr PWM
abc/ αβ abc/ αβ iαβr
PLL
uαβs iαβs   dt
r
ϴr
e  j (e r ) ϴe
 j e  j e
e αβ/abc
ωe ϴe
e iqr idr
udqs ϴr uαβr
ϴe ϴe e j (e  r )
ϴr

System
Identification

udqs
udr
idr-ref
Ps-ref Vector
Control iqr-ref MPC
uqr
Theory
Qs-ref
iqr

Fig. 2. The proposed model-free A-MPC method structure.

along the prediction horizon is obtained as follows: B. Identification algorithm

y(ki + 1|ki ) = CAx(ki ) + CB∆u(ki ) The ARMAX mathematical model has been used to identify
y(ki + 2|ki ) = CA2 x(ki ) + CAB∆u(ki ) + CB∆u(ki + 1) the system model. The mathematical model of the ARMAX
.. can be defined as follows:
.
y(ki + NP |ki ) = CANP x(ki ) + CANP −1 B∆u(ki ) A(q −1 )y(t) = B(q −1 )u(t) + C(q −1 )e(t) (8)
+ · · · + CANP −NC B∆u(ki + NC − 1).
(4) where y(t) is [idr iqr ]T and u(t) is [udr uqr ]T . A, B, and C
So the input and output vectors have the following array: are Matrix Polynomials proportional to q −1 and have orders
nA , nB , and nC , respectively:
Y = [y(ki + 1|ki ) y(ki + 2|ki ) · · · y(ki + NP |ki )]T
(5) A(q −1 ) = I + A1 q −1 + A2 q −2 + · · · + AnA q −nA
∆U = [∆u(ki ) ∆u(ki + 1) · · · ∆u(ki + NC − 1)]T
B(q −1 ) = B1 q −1 + B2 q −2 + · · · + BnB q −nB (9)
Therefore, (4) is summarized as follows: C(q −1 ) = I + C1 q −1 + C2 q −2 + · · · + CnC q −nC
Y = F x(ki ) + Φ∆U T
T θ = [A1 , · · · , AnA , B1 , · · · , BnB , C1 , · · · , CnC ] is pa-
F = CA, CA2 , · · · , CANP

rameters vector which is updated at each sampling time. By

CB 0 ··· 0

(6) using identification system, At each control interval, the A-
CAB CB 0 0 MPC controller updates the plant model and nominal condi-
Φ=
 

tions. Once updated, the model and conditions remain constant
CANP −1 B CANP −2 B 0 CANP −NC B over the prediction horizon. Fig. 2 shows the proposed method
Finally,objective function for finding the control signal is structure.
expressed as follow: IV. S IMULATION R ESULTS
T
J = W1 (RS − Y ) (RS − Y ) + ∆U W2 ∆U + W3 T
(7) In order to validate the proposed model-free A-MPC control
structure, a simulated model in MatLab/SimPowerSystems is

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on March 10,2023 at 14:54:58 UTC from IEEE Xplore. Restrictions apply.
 (a) (b)

Fig. 3. (a) Active power and (b) reactive power of DFIG-based WF under active power reference changing.

compared with the conventional linear control structure. Fig. selects appropriate rotor voltages based on the active and
3 shows active power tracking under a constant wind speed reactive power reference and rotor currents errors. In this
condition using PI controller and proposed A-MPC approach. method, using the ARMAX model and online identification
The q-axis current reference of the grid side converter (GSC) system, transfer functions between input and output variables
regulator is set to zero, so that the output reactive and active (udr , idr : uqr , iqr ) are adaptively identified. Therefore, the
power of the WF is controlled by the RSC. The active power difficulties associated with finding accurate state space model
references changed from 9.5 to 1.625, 6.5 and 10 MW at of the DFIG-based WF are eradicated.The A-MPC can address
t=1.2 s, t=1.5 s,and t=1.8 s, respectively. As it can be seen nonlinear plant, degradation of LTI prediction, and variation
from Fig. 3, the dynamic responses of the proposed control of DFIG parameters by adapting the prediction model for
structure are much superior compared to the linear control changing operating conditions. The simulation results confirm
strategies. Moreover, as the proposed control strategy is model- the effectiveness and robustness of the proposed A-MPC
free, parameter changes cannot affect the performance of the strategy during various active power changes.
system.
Fig 4 shows the other variable of the DFIG under active ACKNOWLEDGMENT
power reference step. Fig. 4(a) presents the rotor current, The authors would like to acknowledge the support of ANID
Fig. 4(b) shows the d-axes rotor current, the generator speed through projects FB0008, ACT192013 and 1170167.
has been indicates in Fig. 4(c), Fig. 4(d) presents the q-
axes rotor current, the PCC voltage has been shown in Fig. R EFERENCES
4(e), and Fig. 4(f) presents the DC-link voltage of the DFIG
[1] Y. M. Alsmadi, L. Xu, F. Blaabjerg, A. J. P. Ortega, A. Y. Abdelaziz,
under active power reference changing. Slow change of rotor A. Wang, and Z. Albataineh, “Detailed investigation and performance
speed is duo to the large inertia of the wind turbine (Fig. improvement of the dynamic behavior of grid-connected DFIG-based
4(c)). Corresponding to theses mentioned curves, there is low wind turbines under lvrt conditions,” IEEE Transactions on Industry
Applications, vol. 54, no. 5, pp. 4795–4812, 2018.
over shoot of rotor currents or the output active power. With [2] X. Liu and X. Kong, “Nonlinear model predictive control for DFIG-
regard to the result, the DC-link voltage, the PCC voltage, based wind power generation,” IEEE Transactions on Automation Sci-
and generator speed are controlled as well using the A-MPC ence and Engineering, vol. 11, no. 4, pp. 1046–1055, 2013.
[3] P. Kou, D. Liang, J. Li, L. Gao, and Q. Ze, “Finite-control-set model
strategy under various active power steps. Fig. 5(a) and Fig. predictive control for DFIG wind turbines,” IEEE Transactions on
5(b) shows the q-axes and d-axes error between actual rotor Automation Science and Engineering, vol. 15, no. 3, pp. 1004–1013,
current and it’s reference. As can be seen, by applying the 2017.
[4] S. Shao, E. Abdi, F. Barati, and R. McMahon, “Stator-flux-oriented
proposed A-MPC strategy the rotor currents also follow the vector control for brushless doubly fed induction generator,” IEEE
reference more accurately. Moreover, as the proposed control Transactions on Industrial Electronics, vol. 56, no. 10, pp. 4220–4228,
strategy is model-free, parameter changes cannot affect the 2009.
[5] M. A. Kashkooli, S. M. Madani, and T. A. Lipo, “Improved direct torque
performance of the system. control for a DFIG under symmetrical voltage dip with transient flux
damping,” IEEE Transactions on Industrial Electronics, vol. 67, no. 1,
V. C ONCLUSION pp. 28–37, 2019.
[6] H. Jadhav and R. Roy, “A comprehensive review on the grid integration
A new model-free adaptive MPC (A-MPC) strategy for a of doubly fed induction generator,” International Journal of Electrical
DFIG-based WF has been presented in this paper. The strategy Power & Energy Systems, vol. 49, pp. 8–18, 2013.

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on March 10,2023 at 14:54:58 UTC from IEEE Xplore. Restrictions apply.
 (a) (d)

(b) (e)

(c) (f)

Fig. 4. (a) Rotor current, (b) d-axes rotor current, (c) generator speed, (d) q-axes rotor current, (e) the PCC voltage and (f) Dc-link voltage
of the DFIG under active power reference changing.

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on March 10,2023 at 14:54:58 UTC from IEEE Xplore. Restrictions apply.
 (a) (b)

Fig. 5. (a) Error between q-axes and (b) d-axes actual rotor current and rotor current reference of the DFIG under active power reference
changing .

[7] Z. Rafiee, M. Rafiee, M. Aghamohammadi, et al., “An investigation of [20] X.-m. Li, X. Zhang, Z.-w. Lin, Y.-g. Niu, M.-y. Li, and N. V. Nathan,
the influences of the voltage sag on the doubly fed induction generator “Adaptive multiple MPC for a wind farm with DFIG: a decentralized-
using tuned pi controllers,” Tehnički vjesnik, vol. 27, no. 1, pp. 149–156, coordinated approach,” Journal of Electrical Engineering & Technology,
2020. vol. 11, no. 5, pp. 1116–1127, 2016.
[8] J. J. Justo, F. Mwasilu, and J.-W. Jung, “Enhanced crowbarless FRT [21] R. Heyderi, M. Alhasheem, T. Dragicevic, and F. Blaabjerg, “Model pre-
strategy for DFIG based wind turbines under three-phase voltage dip,” dictive control approach for distributed hierarchical control of vsc-based
Electric Power Systems Research, vol. 142, pp. 215–226, 2017. microgrids,” in 2018 20th European Conference on Power Electronics
[9] P. Xiong and D. Sun, “Backstepping-based DPC strategy of a wind and Applications (EPE’18 ECCE Europe), pp. P–1, IEEE, 2018.
turbine-driven DFIG under normal and harmonic grid voltage,” IEEE [22] S. Huang, Q. Wu, Y. Guo, and F. Rong, “Optimal active power control
Transactions on Power Electronics, vol. 31, no. 6, pp. 4216–4225, 2015. based on MPC for DFIG-based wind farm equipped with distributed
[10] D. Zhi and L. Xu, “Direct power control of DFIG with constant switch- energy storage systems,” International Journal of Electrical Power &
ing frequency and improved transient performance,” IEEE Transactions Energy Systems, vol. 113, pp. 154–163, 2019.
on Energy Conversion, vol. 22, no. 1, pp. 110–118, 2007. [23] R. Heydari, M. Gheisarnejad, M. H. Khooban, T. Dragicevic, and
[11] J. Hu, H. Xu, and Y. He, “Coordinated control of DFIG’s RSC and GSC F. Blaabjerg, “Robust and fast voltage-source-converter (vsc) control
under generalized unbalanced and distorted grid voltage conditions,” for naval shipboard microgrids,” IEEE Trans. Power Electron., 2019.
IEEE Transactions on Industrial Electronics, vol. 60, no. 7, pp. 2808– [24] R. Heydari, Y. Khayat, A. Amiri, T. Dragicevic, Q. Shafiee, P. Popovski,
2819, 2012. and F. Blaabjerg, “Robust high-rate secondary control of microgrids
[12] Y. Khayat, Q. Shafiee, R. Heydari, M. Naderi, T. Dragičević, J. W. with mitigation of communication impairments,” IEEE Transactions on
Simpson-Porco, F. Dörfler, M. Fathi, F. Blaabjerg, J. M. Guerrero, and Power Electronics, vol. 35, no. 11, pp. 12486–12496, 2020.
H. Bevrani, “On the secondary control architectures of ac microgrids: [25] Z. Rafiee, S. S. Najafi, M. Rafiee, M. R. Aghamohammadi, and
An overview,” IEEE Transactions on Power Electronics, vol. 35, no. 6, M. Pourgholi, “Optimized control of coordinated series resistive limiter
pp. 6482–6500, 2020. and SMES for improving LVRT using TVC in DFIG-base wind farm,”
[13] F. A. Bhuiyan and A. Yazdani, “Multimode control of a DFIG-based Physica C: Superconductivity and its Applications, p. 1353607, 2020.
wind-power unit for remote applications,” IEEE Transactions on Power
Delivery, vol. 24, no. 4, pp. 2079–2089, 2009.
[14] T. Surinkaew and I. Ngamroo, “Coordinated robust control of DFIG
wind turbine and PSS for stabilization of power oscillations considering
system uncertainties,” IEEE Transactions on Sustainable Energy, vol. 5,
no. 3, pp. 823–833, 2014.
[15] S. A. Taher, Z. D. Arani, M. Rahimi, and M. Shahidehpour, “Model
predictive fuzzy control for enhancing frt capability of DFIG-based wt
in real-time simulation environment,” Energy Systems, vol. 9, no. 4,
pp. 899–919, 2018.
[16] G. Marques and M. F. Iacchetti, “Inner control method and frequency
regulation of a DFIG connected to a dc link,” IEEE Transactions on
Energy Conversion, vol. 29, no. 2, pp. 435–444, 2014.
[17] R. A. de Marchi, P. S. Dainez, F. J. Von Zuben, and E. Bim, “A
multilayer perceptron controller applied to the direct power control of
a doubly fed induction generator,” IEEE transactions on sustainable
energy, vol. 5, no. 2, pp. 498–506, 2014.
[18] L. Xiong, P. Li, F. Wu, M. Ma, M. W. Khan, and J. Wang, “A
coordinated high-order sliding mode control of DFIG wind turbine for
power optimization and grid synchronization,” International Journal of
Electrical Power & Energy Systems, vol. 105, pp. 679–689, 2019.
[19] R. Heydari, T. Dragicevic, and F. Blaabjerg, “High-bandwidth secondary
voltage and frequency control of vsc-based ac microgrid,” IEEE Trans-
actions on Power Electronics, 2019.

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on March 10,2023 at 14:54:58 UTC from IEEE Xplore. Restrictions apply.