See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/335203436

A Novel Hybrid MPPT Controller Using (P&O)-neural Networks for Variable

Speed Wind Turbine Based on DFIG

Article  in  Modelling, Measurement and Control A · August 2019

DOI: 10.18280/mmc_a.920104

CITATIONS READS

0 116

3 authors:

Said Azzouz Messalti Sabir

Université de M'sila 49 PUBLICATIONS   312 CITATIONS   
7 PUBLICATIONS   4 CITATIONS   
SEE PROFILE
SEE PROFILE

Abdelghani Harrag
Ferhat Abbas University of Setif
46 PUBLICATIONS   284 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Extraction of Maximum Power From Variable Wind Turbine View project

Grid connected PV inverter with high performance control (STM32F4xx) View project

All content following this page was uploaded by Messalti Sabir on 16 August 2019.

The user has requested enhancement of the downloaded file.

 Modelling, Measurement and Control A
Vol. 92, No., 1, March, 2019, pp. 23-29
Journal homepage: http://iieta.org/journals/mmc_a

A Novel Hybrid MPPT Controller Using (P&O)-neural Networks for Variable Speed Wind
Turbine Based on DFIG
Said Azzouz1,2, Sabir Messalti1*, Abdelghani Harrag3
1
Department of Electrical Engineering, Faculty of technology, University of Msila, Msila 28000, Algeria
2
LGE Laboratory, department of electrical engineering, Faculty of Technology, University of Msila, Msila 28000, Algeria
3
CCNS Laboratory, Electronics Department, Faculty of Technology, Setif 19000, Algeria

Corresponding Author Email: [email protected]

https://doi.org/10.18280/mmc_a.920104 ABSTRACT

Received: 14 December 2018 Although the numerous advantages of wind turbine, the nonlinear characteristic (P–Ω)
Accepted: 27 March 2019 providing a unique maximum power point is the most drawbacks. Therefore a maximum
power point tracker is usually adopted, which the Tip-Speed ratio, the perturbation and
Keywords: observation and the optimum torque control methods are widely used. In this article, a
neural network MPPT controller, combined novel neural network (ANN) MPPT controller based on perturbation and observation has
MPPT controller, P&O MPPT, DFIG, been projected and studied. The ANN MPPT controller of wind turbine system is
variable speed wind turbine established in two independents steps: the offline operation mode is mandatory for training
of different neural networks parameters and the Online operation mode where the most
advantageous neural network MPPT controller is implanted in wind turbine system. The
developed MPPT controller is tested on wind turbine based–DFIG generator, which it is
controlled by sliding mode control (SMC). Simulation results are presented and discussed.

1. INTRODUCTION In recent years, neural network technique has demonstrated

new solution in wind turbine systems. Due to it can be trained
Today, the rapid growth of industry, the huge development off-line, and then used in the on-line physical application [15].
of modern lifestyles, the fast consumption of the fossil fuel The major benefit of neural network is that it does not need an
reserves and the pollution caused by the traditional energy exact mathematical model and be capable of identify
sources have encouraged many countries and governments to compound nonlinear relationships between dependent and
develop a new efficient energy sources. The new energy independent variables.
strategy uses new forms of green energy well-known In this document, a new ANN MPPT controller using the
"renewable energy" to address the global challenges of clean famous perturbation and observation (P&O) has been
energy and able to ensure a part of energy production. These projected and looked for. The necessary ANN model training
energies include the solar, rain, wind; tides, geothermal heat data are calculated using the classical (P&O) rules [17-18].
and various forms of biomass are renewable and can be The proposed MPPT controller is tested on wind turbine
physically replenished continuously. Wind energy is an controlled by sliding mode control (SMC). Simulation results
attractive source of energy and it represent a great potential for are presented and discussed.
power production and consequently it has been developed and
implemented in many countries due to its advantages of clean,
safe energy resource and resource abundance [1-5]. 2. MODELING OF WIND TURBINE AND DFIG
Although the numerous advantages of wind turbine, many
drawbacks can be noted such as the random nature of wind 2.1 Modeling of wind turbine
speed leading to the fluctuation of the produced wind turbine
power. In addition, the wind turbine has a nonlinear The mechanical power of wind turbine is written by [3, 10]:
characteristic (P–Ω) which the maximum power point (MPP)
is guaranteed in only one point, depending on wind speed and 1
speed ratio. Therefore, the wind turbine system must be Paer = C p ( ,  )  Svwind
3
(1)
2
calculated to drive the system at their maximum output power
for any conditions. Therefore, many MPPT methods have been where Paer: is the extracted power from the wind turbine;
proposed and implemented [6-16]. In which, The developed
MPPT controllers can be classified in two groups: i) the first Rt
one is the conventional methods , it includes optimum torque = (2)
control (OTC) [8], tip speed ratio (TSR) [9], perturbation and Vwind
observation (P&O) [4-10], and power speed feedback (PSF)
[4], and ii) the second group or intelligent methods uses fuzzy The wind turbine efficiency Cp is given [19]:
logic, neural-network [11-15], and particle swarm
optimization [16].

23
   (  + 0.1)  The characteristics of the typical captured output power for
C p ( ,  ) = ( 0.35 − 0.0167(  − 2) ) sin   different pitch angles  (2, 4 to 14) are showed in Figure 2. It
 14.34 − 0.3(  − 2)  (3) can be observed that captured output power of wind turbine
− ( 0.00184 (  − 3) (  − 2) ) depends on pitch angles and it is clear that there is one specific
point (CPmax-opt) at which the turbine is maximized. Hence,
Based on the previous equations, a functional block diagram for each wind speed, there is a one turbine speed that provides
model developed under Simulik of wind turbine system is a maximum output power as shown in Figure 3.
presented in Figure 1.
2.2 DFIG model and control

Today, doubly fed induction generators become the most

popular variable speed wind turbine concept [20-22]. These
class of generators are commonly selected as one of the
appropriate wind energy conversion systems since it is able to
operate in wide range of speed variation approximately the
synchronous speed ±30 %, the independent active and reactive
control possibilities and accessibility to the stator and rotor
providing the opportunity to control active and reactive
powers as well as power factor. The Wind turbine with DFIG
system is exposed in Figure 4.

Figure 1. Bloc Simulink of wind turbine

0.4
Beta=2°
Beta=4°
0.35 Beta=6°
Beta=8°
Beta=10°
0.3 Beta=12°
Beta=14°
Figure 4. The Wind turbine based on DFIG
Power coeffcient

0.25

0.2 The DFIG model is expressed as follows [20-22]:

0.15
 d
0.1 Vds = − Rs .I ds + ds - sqs
dt

0.05 V d
= -Rs .I qs + qs + sds
 qs dt
0
0 2 4 6 8 10 12 
V d
Speed ratio(Lamda)
= Rr .I dr + dr - rqr
 dr dt
Figure 2. Power coefficients Cp(λ) of 3MW wind turbine  d (4)
[19] Vqr = Rr .I qr + qr + rdr
 dt
6
x 10

ds = − Ls .I ds + M .I dr
3
13 m/s


12 m/s

qs = − Ls .I qs + M .I qr
11 m/s
2.5
10 m/s
9 m/s

dr = Lr .I dr − M .I ds
8 m/s
2 7 m/s

qr = Lr .I qr − M .I qs (5)
Power(W)

1.5

1
The DFIG torque is expressed by:
0.5

M
Cem = p qs idr − ds iqr  (6)
Ls 
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Speed generator(tr/min)

Figure 3. Power versus rotational speed of generator for To accomplish independent control [19-24], the subsequent
turbine 3 MW [19] equations can be obtained:

24
 ds = s and qs = 0 (7) where the convergence condition is defined by the equation
Lyapunov s( x). s( x)  0
The voltage Vdr and Vqr shown in Figure 5 are expressed .
respectively by:
2.3.1 Active power control
For n = 1 ,the sliding surface representing the error between
  M2 
Vdr = Rr .I dr - gs  Lr −  I qr the measured and reference active power is given by:
  Ls 

  M2  M s   (12)
 s( P) =  Ps − ref − Ps 
Vqr = Rr .I qr + ?g s  Lr −  I dr + gs (8)  
  Ls  Ls
The derivative of the surface is given by

 
s( P) =  P s − ref − P s  (13)
 
By replacing active power equation in the equation of the
switching surface, the expression of the surface becomes:

 VM 
s( P) =  P s − ref + s I rq  (14)
 Ls 

From the equation (11) (During the sliding mode and in

permanent regime), we can write

•
. (Vrq − Rr I rq )
1
Figure 5. Field oriented control of DFIG I rq = (15)
Lr
2.3 Sliding mode control of DFIG
The control law design Vrq is given by:
Recently, the sliding mode control (SMC) has been
considered as one of most used techniques for DFIG drive Vrq = Vrqeq + Vrqn (16)
control among nonlinear control strategies, due to its
simplicity, easy implementation and robustness [25-27]. The
sliding mode control is based on three steps: the first is the Replacing equation (16) in equation (14), the equation we
choice of control surface, the second is convergence condition, have
and the third is control law design.
 
SMC algorithm consists to calculate the equivalent and
discontinuous components of control variable from an

VM
s( P) =  P s − ref + s
Ls Lr 
((V eq
rq )
+ Vrqn ) − Rr I rq  (17)

adequate surface of sliding mode chosen. In this case we chose
the error as being the sliding surface.
During the sliding mode and in permanent regime, we have:
The control algorithm is defined by following expression:

u = ueq + un (9) s ( P) = 0, s( P) = 0, Vrqn = 0.

where Replacing in previous equation, the equivalent control Vrqeq

u is the control vector, u eq is the equivalent control vector,
and correction factor are given by:
u n is the switching part of the control (the correction factor)
u eq can be obtained by considering the condition for the  eq Ls Lr 
sliding regime, s = 0 Vrq = P s − ref V M + Rr I rq 
The control law is defined as follows:  s  (18)
V n = K sign ( S ( P ) ) 
 rq V rq 
𝑢𝑛 = 𝑢𝑚𝑎𝑥(𝑠(𝑋)/𝜃)
{ 𝑠𝑖𝑔𝑛(𝑠) 𝑖𝑓 |𝑠| > 𝜃 (10)
𝑠𝑎𝑡(𝑠(𝑋)/𝜃) { where: KV rq positive constant [25].
𝑠/𝜃 𝑖𝑓 |𝑠| < 𝜃
2.3.2 Reactive power control
The sliding surface is calculated by the general equation J.J.
For n=1, the reactive power control surface becomes is
Slotine:
given by
s() = ( s − ref − s )
n −1
d  (19)
s(X ) =  +   e (11)
 dt 
The derivative of the surface is given by

25
   The obtained simulation results of the DFIG using S.M.C is
s() =   s − ref −  s  (20) shown in Figure 6, in which the produced active and reactive
 
power follow perfectly their references.
From the equation (11) the current expression is given by

•
3. PROPOSED ANN MPPT CONTROLLER BASED ON
1
I rd = (Vrd − Rr I rd ) (21) P&O METHOD
Lr
3.1 Perturb & observe (P&0) method
The control law design Vrd is given by:
The perturb & observe (P&O) MPPT controller is widely
Vrd = Vrdeq + Vrdn (22) applied, it can be used by small step size disturbing the control
variable and analyzing the resulting raising or decreasing in
generated power (the controller must move it right or left of
From equation (24), (21) and (22), the derivative of the
the MPP) as shown in Figure 7 presenting the output power of
surface can be expressed by:
3MW wind turbine since a function of generator speed for its
nominal wind speed (13 m/s). If it results in the increase in
 

VM
s() =   s − ref + s
Ls Lr
((V eq
rd
) )
+ Vrdn − Rr I rd 

(23) power, then the similar perturbation is useful for the next
control occurrence. Otherwise the sign of the perturbation is
inverted to track in the way of increasing power. Once the
During the sliding mode and in steady state stable state is achieved the process oscillates about the MPP.

s() = 0, s() = 0, Vrdn = 0.

Therefore, the equivalent control Vrdeq and the switching

term is given by can be expressed by:

 eq Ls Lr 
Vrd = −  s − ref V M + Rr I rd 
 s  (24)
V n = K sign ( S (  ) ) 
 rd V rd 

where: KV rd is positive constant [25].

6
x 10

6 Ps-mes Figure 7. Maximal power curve P(Ω) of 3MW wind turbine

Ps-ref
4
Active power[W]

-2

-4

0 5 10 15 20
Time[S]
6
x 10

6 Qs-mes
Qs-ref
4
Reactive power[W]

-2

-4 Figure 8. Diagram of the P&O method [2]

0 5 10 15 20
Time[S] Figure 8 explains the flowchart of the P&O strategy. The
output power is maximum at a rotor speed well known most
Figure 6. Active and reactive power of the DFIG using favorable rotor speed Ωopt, corresponding to optimum tip speed
S.M.C ratio opt. So, to produce maximum possible power, the turbine

26
must be driven at Ωopt. This is achievable by scheming the The developed ANN system operates in two epochs:
rotational speed of the turbine so that it always turns at the 1) The offline epoch necessary to trying different set of
finest speed rotation. neural network parameters to find the most favorable neural
network controller (activation function, organization, and
3.2 Proposed ANN MPPT training algorithm), when the training and testing performed
in off-line, a neural network bloc model is obtained, and then
Technological developments in recent years has allowed it is introduced on-line in MPPT [8].
scientists to develop and refine methods for different 2) The online epoch that employs the best ANN MPPT
applications, especially in the renewable energies areas, which regulator to track the MPP point. The fundamental diagram of
they require the use of chic techniques for an perfect wind turbine system is demonstrated in Figure 9.
estimation of the accessible energy prospective and any Figure 10 shows the ANN MPPT regulator developped via
successful control of systems operation [8]. Simulink.
Over the last few decades, many control strategies have
been developped. Artificial neural networks techniques have
been selected as one of the best solution for computational
system since they offer many benefits compared to the
conventional computational systems [8, 17-18].
The proposed artificial neural network (ANN) MPPT
controller is established on the similar law of Perturbation and Figure 9. Block diagram of the wind turbine system control
Observation (P&O) method decreasing or increasing the
generator speed regarding the sign of (dP/dΩ), the
fundamental rule of ANN MPPT controller is recapitulated in
Table 1.

Table 1. Fundamental rule of artificial neural network MPPT

controller

dP dΩ dP/ dΩ Increase /deacrease of Ω

+ + + Ω(s) = Ω(s-1) + ∆Ω
+ - - Ω(s) = Ω(s-1) - ∆Ω Ω
- + - Ω(s) = Ω(s-1) - ∆Ω Ω
- - + Ω(s) = Ω(s-1) + ∆Ω Ω
Figure 10. ANN MPPT regulator based on P&O rule

Figure 11. Developed ANN MPPT regulator using P&O rule

A simulation study was tested for the wind speed estimation wind turbine connected to DFIG is employed.
and tracking control of the most favorable maximum power The simulation results have been establised using Simulink
using a feed-forward neural networks having three layers. The environment using the wind speed profile depicted in Figure
inputs variables of the projected ANN MPPT controller are: 12.
the output power (P) and generator speed (dΩ). However, the Figures 13, 14 and 15 show the power coefficient, the output
output ANN parameter is the consequent rising or lowering of power and the optimal turbine rotational speeds, respectively,
generator speed +/_∆Ω. determined based on the optimum operating points of the
system.
From figures, its clear that the wind energy- DFIG system
4. SIMULATION AND RESULTS operates at its maximum power which it can be verified for
nominal wind speed (13m/s). The power coefficient and output
In this manuscript segment, the competence of the projected power correspond exactly at it nominal value 0.35 and 3MW,
artificial neural network MPPT regulator using P&O rule is respectively.
presented. A wind energy conversion system based on 3MW

27
 13.5
Modeling and control strategies of the generally system have
13 been developed. Simulation results have carried out using
12.5
Simulink under randomly and fast changeable wind rapidity.
Obtained results exhibit the elevated performance of proposed
Wind spped(m:s)

12 ANN MPPT regulator. The simulation results are close to the

11.5 theoretically expected results. The advantages of this proposed
method include: good accuracy, does not require information
11
of power coefficient Cpmax and the tip speed ratio opt and the
10.5 characteristic curve of wind turbine. In addition, a variable
10 step size can be used in the proposed neural Network MPPT
0 50 100 150 200 250 300 350 400 450 500
Time(s) controller for achieving better performances. This last feature
will be considered for our future work.
Figure 12. Wind speed profile
0.4 REFERENCES
0.35

0.3
[1] Suresh, K., Babu, A.R.V., Venkatesh, P.M. (2018).
Power coefficient Cp

Experimental investigations on grid integrated wind

0.25
energy storage system using neuro fuzzy controller.
0.2
Modelling, Measurement and Control A, 91(3): 123-130.
0.15 https://doi.org/10.18280/mmc_a.910304
0.1 [2] Suresh, K., Babu, A.R.V., Venkatesh, P.M. (2018).
0.05
Experimental investigations on grid integrated wind
energy storage system using neuro fuzzy controller.
0
0 50 100 150 200 250 300 350 400 450 500 Modelling, Measurement and Control A, 91(3): 123-130.
Time(s)
[3] Magrini, A., Lazzari, S., Marenco, L., Guazzi, G. (2018).
Cost optimal analysis of energy refurbishment actions
Figure 13. Power coefficient Cp
depending on the local climate and its variations.
6
Mathematical Modelling of Engineering Problems, 5(3):
x 10
3.5 268-274. https://doi.org/10.18280/mmep.050321
3
[4] Manukonda, D., Gorantla, S.R. (2018). A fuzzy logic
controller based vortex wind turbine for commercial
2.5
applications. Modelling, Measurement and Control A,
Output power(W)

2 91(2): 54-58. https://doi.org/10.18280/mmc_a.910204

[5] Messaoud, M., Abdessamed, R. (2011). Modeling and
1.5
optimization of wind turbine driving permanent magnet
1 synchronous generator. Jordan Journal of Mechanical
and Industrial Engineering, 5(6): 489-494.
0.5
[6] Tanaka, T., Toumiya, T., Suzuki, T. (1997). Output
0
0 50 100 150 200 250 300 350 400 450 500
control by hill-climbing method for a small scale wind
Time(s) power generating system. Renewable Energy, 12(4):
387-400. https://doi.org/10.1016/S0960-1481(97)00055-
Figure 14. Tracking MPPT power ANN-P&O 4.
[7] Hosseini, S.E., Abdul Wahid, M. (2014). The role of
250
renewable and sustainable energy in the energy mix of
Malaysia: A review. International Journal of Energy
200
Research, 38(14): 1769-1792.
turbine rotational speed

https://doi.org/10.1002/er.3190
150
[8] Musunuri, S., Ginn, H.L. (2011). Comprehensive review
100
of wind energy maximum power extraction algorithms.
In 2011 IEEE power and energy society general meeting,
50
pp. 1-8. https://doi.org/10.1109/PES.2011.6039023
[9] Saad, N.H., El-Sattar, A.A., Metally, M.E. (2018).
0
Artificial neural controller for torque ripple control and
0 50 100 150 200 250
Time(s)
300 350 400 450 500
maximum power extraction for wind system driven by
switched reluctance generator. Ain Shams Engineering
Figure 15. Turbine rotational speed Journal, 9(4): 2255-2264.
https://doi.org/10.1016/j.asej.2017.03.005
[10] Mahvash, H., Taher, S.A., Rahimi, M. (2017). A new
5. CONCLUSION approach for power quality improvement of DFIG based
wind farms connected to weak utility grid. Ain Shams
In this article, a novel ANN MPPT regulator using (P&O) Engineering Journal, 8(3): 415-430.
rules has been presented considering changeable speed wind https://doi.org/10.1016/j.asej.2015.09.001
turbine moving doubly-fed induction generator (DFIG). [11] Goh, H.H., Anwar, M.S., Chua, Q.S., Ling, C.W., Goh,

28
 K.C. (2017). Analysis between perturb & observe Renewable and Sustainable Energy Reviews, 68(1): 221-
controller and fuzzy logic controller for a photovoltaic 233
system with CUK and SEPIC converter. Journal of [20] Gaillard, A. (2010). Système éolien basésur une MADA:
Telecommunication, Electronic and Computer Contribution àl'étude de la qualitéde l'énergie électrique
Engineering (JTEC), 9(3-7): 81-88. et de la continuité de service. Doctoral dissertation,
[12] Wang, Q., Chang, L. (2004). An intelligent maximum UniversitéHenri Poincaré-Nancy 1.
power extraction algorithm for inverter-based variable [21] Zemmit, A., Messalti, S., Harrag, A. (2017). A new
speed wind turbine systems. IEEE TRANSACTIONS on improved DTC of doubly fed induction machine using
Power Electronics, 19(5): 1242-1249. GA-based PI controller. Ain Shams Engineering Journal,
https://doi.org/10.1109/TPEL.2004.833459 9(4): 1877-1885.
[13] Algazar, M.M., El-Halim, H.A., Salem, M.E.E.K. https://doi.org/10.1016/j.asej.2016.10.011
(2012). Maximum power point tracking using fuzzy logic [22] Akhmatov, V. (2003). Variable-speed wind turbines with
control. International Journal of Electrical Power & doubly-fed induction generators Part IV: Uninterrupted
Energy Systems, 39(1): 21-28. operation features at grid faults with converter control
https://doi.org/10.1016/j.ijepes.2011.12.006 coordination. Wind Engineering, 27(6): 519-529.
[14] Mohamed, A.Z., Eskander, M.N., Ghali, F.A. (2001). https://doi.org/10.1260/030952403773617481
Fuzzy logic control based maximum power tracking of a [23] Muller, S., Deicke, M., De Doncker, R.W. (2002).
wind energy system. Renewable Energy, 23(2): 235-245. Doubly fed induction generator systems for wind
https://doi.org/10.1016/S0960-1481(00)00099-9 turbines. IEEE Industry Applications Magazine, 8(3):
[15] Jabr, H.M., Lu, D., Kar, N.C. (2011). Design and 26-33. https://doi.org/10.1109/2943.999610
implementation of neuro-fuzzy vector control for wind- [24] Yamamoto, M., Motoyoshi, O. (1991). Active and
driven doubly-fed induction generator. IEEE reactive power control for doubly-fed wound rotor
Transactions on Sustainable Energy, 2(4): 404-413. induction generator. IEEE Transactions on Power
https://doi.org/10.1109/TSTE.2011.2160374 Electronics, 6(4): 624-629.
[16] Pucci, M., Cirrincione, M. (2010). Neural MPPT control https://doi.org/10.1109/63.97761
of wind generators with induction machines without [25] Adjoudj, M., Abid, M., Aissaoui, A.G., Ramdani, Y.,
speed sensors. IEEE Transactions on Industrial Bounoua, H. (2011). Sliding mode control of a doubly
Electronics, 58(1): 37-47. fed induction generator for wind turbines. Rev. Roum.
https://doi.org/10.1109/TIE.2010.2043043 Sci. Techn.–Électrotechn. etÉnerg, 56(1): 15-24.
[17] Tianpei, Z., Wei, S. (2013). MPPT method of wind [26] Boukezzar, B., M'Saad, M. (2008, June). Robust sliding
power based on improved particle swarm optimization. mode control of a DFIG variable speed wind turbine for
TELKOMNIKA Indonesian Journal of Electrical power production optimization. In 2008 16th
Engineering, 11(6): 3206-3212. Mediterranean Conference on Control and Automation,
[18] Messalti, S., Boudjellal, B., Said, A. (2015). Artificial pp. 795-800.
neural networks controller for power system voltage https://doi.org/10.1109/MED.2008.4602145
improvement. In IREC2015 The Sixth International [27] Liu, X., Han, Y. (2014). Sliding mode control for DFIG-
Renewable Energy Congress, pp. 1-6. based wind energy conversion optimization with
https://doi.org/10.1109/IREC.2015.7110897 switching gain adjustment. In Proceeding of the 11th
[19] Messalti, S., Harrag, A., Loukriz, A. (2017). A new World Congress on Intelligent Control and Automation,
variable step size neural networks MPPT controller: pp. 1213-1218.
Review, simulation and hardware implementation. https://doi.org/10.1109/WCICA.2014.7052892

29

View publication stats