International Journal of Power Electronics and Drive Systems (IJPEDS)

Vol. 14, No. 3, September 2023, pp. 1722~1730

ISSN: 2088-8694, DOI: 10.11591/ijpeds.v14.i3.pp1722-1730  1722

Performance enhancement of small-scale wind turbine based on

artificial neural network

Mohammad Ahmed Ibrahim, Ali Saleh Alhfidh, Ali Nathem Hamoodi

Department of Power Techniques Engineering, Technical College of Engineering, Northern Technical University, Mosul, Iraq

Article Info ABSTRACT

Article history: Small-scale wind turbine is typically designed to resisted extreme wind; this
work aims to adjust their pitch angle based on simulations that use
Received Dec 10, 2022 standardization codes for wind turbines. Proportional integral derivative
Revised Mar 18, 2023 (PID) and artificial neural network (ANN) controllers are used to control the
Accepted Mar 29, 2023 speed of wind turbines. The ideal action for controlling the blade pitch angle
can be attained by providing the controller with speed information ahead of
time, allowing the controller to provide the best action for blade pitch angle
Keywords: control. The results of this work represent the relationship between the
turbine speed with respect to time at different pitch angle. It has been
ANN controller concluded that the ANN controller produced the best time response as
Cascade feed forward compared with the PID controller.
MATLAB/Simulink
PID controller
Small scale wind turbine This is an open access article under the CC BY-SA license.

Corresponding Author:
Mohammed Ahmed Ibrahim
Department of Power Techniques Engineering, Technical College of Engineering,
Northern Technical University
Mosul, Iraq
Email: [email protected]

1. INTRODUCTION
The adjustment of pitch angle can be handled using conventional methods, such as creating a
controller after linearizing equipment working locations. However, wind turbines are non-linear of
equipment. This type of method needs careful calculation and a lot of design work. To simplify and make the
control system realistic, the system can be trained to use an offline-learnable mathematical model [1]. Wind
energy is now the most competitive kind of renewable energy due to its rapid growth over the previous few
decades. Modern wind energy systems depend heavily on control. By controlling the wind turbine, the load
that it produces due to aerodynamics and mechanicals may be reduced and the turbine's capacity can be used
more effectively [2]. There are two main operating zones for variable-speed wind turbines: below-rated
power and above-rated power. The turbine runs at a varied speed of rotation to harness the most amount of
wind energy when power output is below the rated power [3], [4]. In the region of higher power, the major
objective is to maintain the power output consistent. This is often achieved by reducing the quantity of wind
energy captured, which is done by varying the blades pitch angle. This work suggests a new method for
controlling the pitch angle of the blade that makes use of neural networks. Nonlinear control systems have
been designed using neural networks. In the region of above-rated power, this neural network pitch controller
is employed to maintain a consistent power [5].
The manufacturers of wind turbines produce this turbine without control of its power. Due to the
over speeding of the turbine rotor, small wind turbines cannot withstand strong winds velocity. The tiny WT
system becomes more difficult and costs more as a result of the addition of pitch control [6]. Figure 1 shows
depicts the wind turbines (WT's) construction. The main parts of the WT are the main controller, nacelle,

Journal homepage: http://ijpeds.iaescore.com

Int J Pow Elec & Dri Syst ISSN: 2088-8694  1723

tower, and blades [7]. A gearbox, generator, transformer, and power converter are all inside the nacelle. The
turbine and generator shafts connected by a gear box because the wind rotates the blades [8]. A transformer
transmits the generator's power to the grid [9]. The mechanical stress and load on the drive system of a wind
turbine are affected by loads that do not involve torque, which are mainly applied to the blades and generator.
These non-torque loads act as input and output on both sides of the WT, and can impact its overall
mechanical load and stress [10]. Additionally, the mechanical load is impacted by the rotating load, which
stresses the WTs. Unexpected loads are also produced by wind on the input shaft, including rapid changes in
wind speed and direction, unequal blade loading, and wind disturbance [11].
From the previous studies, the comparison between small-scale and large-scale wind turbines can be
led to the conclusion that the small-scale wind turbine is designed to spin at a high speed compared to slow
wind turbines, which may give it many more revolutions over a period of time compared to large wind
turbines. This will continue to produce electrical energy within a narrow band of wind speed.

Figure 1. Main parts of WT

2. LITERATURE REVIEW
Salem et al. [6] research focused on how to provide a good pitch angle controlling action for wind
turbines in order to achieve the desired speed curve. It has been concluded the neural net (NN) fitting
provided an accordant pitch angle control exertion, which made the response of the wind turbine follow the
desired speed and gave an unavoidable change between pitch angle and speed. Najd et al. [12] studied the
control strategy in wind turbine due to the effect of pitch angle using neural networks, then focused on
increasing wind turbine speed and power by controlling the blades pitch angle. Simulation results showed
that the suggested neural network controller was very operant for correcting the pitch angle. It has been
concluded that changes in system parameters that may occur over time have no effect on the operation of the
control system.
A study was conducted by Yan et al. [13] on the potential of small wind turbines for power
generation based on rotor speed. The study found that installing small wind turbines on platforms is a good
alternative to solar panels because they take up less space. However, it was determined that small wind
turbines were not feasible for harnessing wind power in certain areas due to low wind speeds throughout the
year. The minimum cut-in speed for the selected models of small wind turbines was 2.5 m/s.

3. WIND TURBINE BASED ON PID CONTROLLER

In a digitally controlled system, analog input channels are used to convert analog quantities into
digital quantities after first sampling the process parameters. When dealing with these analog values in
accordance with certain control algorithms, the results of the calculations are being appeared through an
output channel and control the objects using a mover [14]. The most fundamental and popular industrial
control technique is PID control, which is also commonly employed on wind turbines [15]. In Laplace
notation, a PID controller is:

𝑘 𝑘𝐷
𝑦 = ( 𝐼 + 𝑘𝑝 + )𝑥 (1)
𝑠 1+𝑠𝜏

K I , K P and K D represent the proportional, integral and differential gains, while, x is used to correct
the signal and y is a signal that give an indication about control behavior, τ represents the time constant. It
can be modelled the whole wind turbine generator by a one-order inertia link, and the objects controlled by

Performance enhancement of small-scale wind turbine based on artificial … (Mohammed Ahmed Ibrahim)
1724  ISSN: 2088-8694

out-put have been kept constant under outboard external circumstances using a PID organizer. The transfer
function of the regulator's is given as:

𝑈(𝑠)
𝐷(𝑠) =
𝐸(𝑠)
1+𝐾 𝑆 +𝑇𝐷 𝐾 𝑆
𝐼 𝐼
𝐷(𝑠) = 𝐾𝑃 (2)
𝐾𝐼 𝑆

while, 𝐾𝑃 is a system of proportional gain, 𝑇𝐼 is time integral constant, 𝑇𝐷 is the differential time constant.
The form of increment used to represent PID control algorithms:
𝑇 𝑇𝐷
𝑢(𝑛) = 𝑢(n-1) + 𝐾𝑃 [𝑒(𝑛) − 𝑒(𝑛 − 1)] + e(n)+ [𝑒(𝑛) − 2𝑒(𝑛 − 1) + 𝑒(𝑛 − 2)] (3)
𝑇𝐼 𝑇

while, 𝑢(𝑛) is control action, 𝑢(n-1) is last control action, 𝑒(𝑛)is deviation, 𝑒(𝑛 − 1)is last deviation and 𝑇 is
sampling period.

4. NEURAL NETWORK
The cascade feed forward approach, which is a three-layers feed forward network with one single
layer hidden was used [16]. Figure 2 shows the internal structure of the proposed artificial neural network
(ANN). The input layer node as shown in Figure 3, cascade feed forward transfers input signals to the hidden
layer. The hidden nodes are constructed from radiation effect functions such as the Gaussian function, but the
output nodes are constructed from basic linear functions. The primary function of the hidden layers responds
to incoming signals in a local area. That is, when an input signal approaches the basic function’s center
scope, hidden nodes increase the output [17], [18].

Figure 2. Internal structure of the proposed ANN Figure 3. Structure of cascade feed forward
neural network

Cascade feed forwarded way has been chosen from ANN, this is comparable to a weight link from
the input layer succeeding layers, which are analogous to feed-forward networks. Feed-forward networks
with further layers may learn complex relationships further quickly, even if two-layer feed-forward networks
have the potential to learn almost any input-output relationship [19]. Networks with cascade-forward are built
using the newly created function as an illustration, a three-layers network has engagements between layers 1
and 2, layers 2 and 3, and layers 1 and 3. Additionally, the input is connected to all three layers of the three-
layer network. It’s possible that the extra connections will hasten the network’s learning of the intended
relationship [20]. CF artificial intelligence model is similar to the basic characteristic of the feedforward
backpropagation neural is which each neuron in the current layer is linked to every neuron in the previous
layer. The backpropagation method is used by this network to update its weights [10]. To achieve the optimal
state, all three types of transfer functions log-sigmoid, tan-sigmoid were used, as well as pure linear threshold
functions [21].

𝑄𝑒𝑥𝑝 −𝑄𝑐𝑎𝑙 2
𝑀𝑆𝐸 = [∑𝑣1 ( ) ] (4)
𝑛

2
1 𝑄𝑒𝑥𝑝 −𝑄𝑐𝑎𝑙
𝑅𝑀𝑆𝐸 = √ [∑𝑣1 ( ) ] (5)
𝑛 𝑄𝑒𝑥𝑝

Int J Pow Elec & Dri Syst, Vol. 14, No. 3, September 2023: 1722-1730
Int J Pow Elec & Dri Syst ISSN: 2088-8694  1725

2
𝑄𝑒𝑥𝑝 −𝑄𝑐𝑎𝑙
𝑅2 = 1 − [∑𝑣1 ( ) ] (6)
𝑄𝑒𝑥𝑝

Where, MSE mean squire error Qexp = value as observed; Qcal = value predicted; Qexp = mean predicted
value; n = the dataset’s number of observations. RMSE =root mean square and R2 = normalized version,
were utilized to constant the prediction performance of the advanced models. The best score for R2 outcome
is 1 and for latest measures is zero. Figure 4 show the trained ANN’s performance and state plot. The
regression performance plot is depicted in Figure 5.

Figure 4. The effectiveness plot of the trained ANN

Figure 5. Regression performance plot

Performance enhancement of small-scale wind turbine based on artificial … (Mohammed Ahmed Ibrahim)
1726  ISSN: 2088-8694

5. NEURAL NETWORK BASED PITCH CONTROLLER ALGORITHM

The flowchart of the NN pitch controller algorithm is shown in Figure 6. The initial phase of startup
is split into three situations based on whether the present generator speed is more, equal, or less than the rated
generator speed [22]. Furthermore, the varying rate of generator speed is taken into account when deciding
whether to raise or reduce pitch angle β [23]. When the current rotor speed exceeds the rated speed, the
changing rate of the rotor speed is taken into account once more to increase the pitch angle. If the rate of
change of the rotor speed increases, the pitch angle is decided to increase. Otherwise, no changes are made to
the pitch angle [24], [25]. When the current rotor speed is lower than the ordained speed, the same process is
used to decrease pitch angle. When the present rotor speed equals the ordained speed, the pitch angle changes
proportionately to the rate at which the rotor speed changes. When the changing rate of the rotor speed is
positive, the pitch angle increases [1], [26].

Figure 6. Flowchart of the neural pitch controller

6. MODELLING OF WIND TURBINE BASED ON PID CONTROLLER

The modeling of a 1 kW self-excitation induction generator (SEIG) wind turbine is utilized for
simulation. The speed control of the shaft is adjusted by the pitch angle of the blade. Figure 7 represents the
block diagram model of this wind turbine.

Figure 7. Speed control of wind turbine based on PID controller

Int J Pow Elec & Dri Syst, Vol. 14, No. 3, September 2023: 1722-1730
Int J Pow Elec & Dri Syst ISSN: 2088-8694  1727

7. MODELLING OF WIND TURBINE BASED ON ANN CONTROLLER

The use of an ANN controller was employed to test the efficiency of a wind turbine. MATLAB was
chosen for its vast collection of algorithms capable of solving various mathematical problems such as
differential equations, matrix algebra, and linear and nonlinear systems. The findings suggest that small wind
turbines mounted on platforms could be a feasible alternative to solar panels due to their smaller size.
However, certain areas with low wind speeds made it unfeasible to harness wind power using small turbines
with a minimum cut-in speed of 2.5 m/s. MATLAB provides a fascinating environment with numerous built-
in algorithms that are efficient and dependable. These functions can assist in solving a wide range of
mathematical problems, including linear systems, nonlinear systems, differential equations, matrix algebra,
and many other technical solutions. Figure 8 gives the modeling of wind turbine based on NN controller.

Figure 8. Speed control of wind turbine based on ANN controller at reference pitch angle 15°

8. SIMULATION RESULTS
For PID controller case, at reference pitch angle, the behaviour of speed response of wind turbine is
given in Figure 9. At reference pitch angle the behaviour of peed response curve of wind turbine with respect
to time is given in Figure 10. For ANN controller case, the behaviour of speed response of wind turbine with
respect to time is shown in Figure 11. At reference pitch angle, the behaviour of speed response of wind
turbine with respect to time is given in Figure 12. The transient response of rotor speed for wind turbine-
based PSD controller with respect to time are listed in Table 1.

Figure 9. Speed response of wind turbine at pitch angle 15°

Table 1. Time response parameters of 1 kW wind turbine

Controller Pitch Angle Peak over shoot % Peak time tp (sec) Rise time tr (sec) Setting time (sec)
PID-controller 15° 15.15 0.749 0.238 1.727
35 ° 16.145 0.72 0.246 1.718
ANN-controller 15° 0.1113 0.782 0.2565 0.4
35 ° 0.12 0.76 0.26 0.42

Performance enhancement of small-scale wind turbine based on artificial … (Mohammed Ahmed Ibrahim)
1728  ISSN: 2088-8694

Figure 10. Speed response of wind turbine at pitch angle 35°

Figure 11. Speed response of wind turbine at pitch angle 15°

Figure 12. Speed response of wind turbine at pitch angle 35°

Int J Pow Elec & Dri Syst, Vol. 14, No. 3, September 2023: 1722-1730
Int J Pow Elec & Dri Syst ISSN: 2088-8694  1729

9. CONCLUSION
The performance of pitch angle behavior with small wind turbine could provide quick action and an
effective wind turbine blades The ANN technique provides a good and suitable control action to control pitch
angle. At a pitch angle of 15° ANN, the controller made an enhancement in percentage peak overshoot and
setting time by reducing these parameters by 15.038% and 1.327 sec, respectively as compared by the PID
controller, at a pitch angle of 35°, the ANN controller made an enhancement in percentage peak overshoot
and setting time by reducing these parameters by 16.025% and 1.258 sec, respectively as compared to the
PID controller. The ANN controller at pitch angle 35° enhancement the peak over shoot by 0.987 % and the
settling time by 0.065 sec as compared with pitch angle 15°. Finally, the ANN controller provided a
continuous relationship between Pitch angle and variations in wind speed.

ACKNOWLEDGEMENTS
The researchers like to thank Northern Technical University/Technical College of Engineering,
Mosul for supplying us with a simulation tool that enabled us to complete our research.

REFERENCES
[1] M. J. Kang, and H. C. Kim, “Neural network based pitch controller,” Proceedings of the World Congress on Electrical
Engineering and Computer Systems and Science (EECSS 2015), Barcelona, Spain – July 13 - 14, 2015.
[2] P. Rao, M. L. Crow, and Z. Yang, “STATCOM control for power system voltage control applications,” IEEE Trans. power Deliv.,
vol. 15, no. 4, pp. 1311–1317, 2000, doi: 10.1109/61.891520.
[3] A. H. Norouzi and A. M. Sharaf, “Two control schemes to enhance the dynamic performance of the STATCOM and SSSC,” IEEE
Trans. Power Deliv., vol. 20, no. 1, pp. 435–442, 2005, doi: 10.1109/TPWRD.2004.839725.
[4] A. S. Saleh, R. K. Antar, and A. J. Ali, “Design and implementation of single-phase PV Power system,” Ann. Rom. Soc. Cell Biol.,
vol. 25, no. 6, pp. 12078–12087, 2021.
[5] X. Yao, X. Su, and L. Tian, “Pitch angle control of variable pitch wind turbines based on neural network PID,” in 2009 4th IEEE
Conference on Industrial Electronics and Applications, 2009, pp. 3235–3239, doi: 10.1109/ICIEA.2009.5138799.
[6] M. E. M. Salem, H. M. El-Batsh, A. A. El-Betar, and A. M. A. Attia, “Application of neural network fitting for pitch angle control
of small wind turbines,” IFAC-PapersOnLine, vol. 54, no. 14, pp. 185–190, 2021, doi: 10.1016/j.ifacol.2021.10.350.
[7] A. Crespo, J. Hernandez, and S. Frandsen, “Survey of modelling methods for wind turbine wakes and wind farms,” Wind Energy
An Int. J. Prog. Appl. Wind Power Convers. Technol., vol. 2, no. 1, pp. 1–24, 1999, doi: 10.1002/(SICI)1099-
1824(199901/03)2:1<1::AID-WE16>3.0.CO;2-7.
[8] J. G. Leishman, “Challenges in modelling the unsteady aerodynamics of wind turbines,” Wind Energy An Int. J. Prog. Appl. Wind
Power Convers. Technol., vol. 5, no. 2‐3, pp. 85–132, 2002, doi: 10.1002/we.62.
[9] L. J. Vermeer, J. N. Sørensen, and A. Crespo, “Wind turbine wake aerodynamics,” Prog. Aerosp. Sci., vol. 39, no. 6–7, pp. 467–
510, 2003, doi: 10.1016/S0376-0421(03)00078-2.
[10] J. Antonelli, “Reduced order modeling of wind turbines in MatLab for grid integration and control studies,” M.S. thesis, Master of
Science Programme in Wind Power Project Management, Department of Wind Energy, Gotland University, Swede 2012.
[11] A. G. Ameen, I. Hamdan, S. Mohamed, and G. T. Abdel-Gaber, “Dynamic performance improvement of pitch angle control using
PID and PIDA controllers for wind generation systems,” SVU-International J. Eng. Sci. Appl., vol. 1, no. 1, pp. 32–36, 2020, doi:
10.21608/svusrc.2020.53401.1005.
[12] A. H. Najd, G. Goksu, and H. F. Hammood, “Pitch angle control using neural network in wind turbines,” in IOP Conference
Series: Materials Science and Engineering, 2020, vol. 928, no. 2, doi: 10.1088/1757-899X/928/2/022118.
[13] C. S. Yan, S. C. Lim, and K. Y. Chan, “Potential of small-wind turbine for power generation on offshore oil and gas platforms in
Malaysia,” Int. J. Energy Econ. Policy, vol. 12, no. 6, 2022, doi: 10.32479/ijeep.13433.
[14] R. Jawad, M. Ahmed, H. Salih, and Y. Mahmood, “Variable speed controller of wind generation system using model predictive
control and NARMA controller,” Iraqi J. Electr. Electron. Eng., vol. 18, no. 2, pp. 43–52, 2022, doi: 10.37917/ijeee.18.2.6.
[15] S. A. Hamoodi, F. I. Hameed, and A. N. Hamoodi, “Pitch angle control of wind turbine using adaptive Fuzzy-PID controller,” EAI
Endorsed Trans. Energy Web, vol. 7, no. 28, 2020, doi: 10.4108/eai.13-7-2018.165500.
[16] M. Sahib and T. Nayl, “A new pitch angle control method of wind turbine generators based on feedforward wind speed
information,” E3S Web Conf., vol. 122, 2019, doi: 10.1051/e3sconf/201912204001.
[17] R. Chedid, F. Mrad, and M. Basma, “Intelligent control for wind energy conversion systems,” Wind Eng., vol 22, no. 1, pp. 1–16,
1998.
[18] F. D. Kanellos and N. D. Hatziargyriou, “A new control scheme for variable speed wind turbines using neural networks,” in 2002
IEEE Power Engineering Society Winter Meeting. Conference Proceedings (Cat. No. 02CH37309), 2002, vol. 1, pp. 360–365, doi:
10.1109/PESW.2002.985020.
[19] S. Baburajan and A. Ismail, “Design and control of the pitch of wind turbine through PID,” Int. Res. J. Eng. Technol., vol. 4, no. 9,
pp. 657–661, 2017.
[20] M. H. Beale, M. T. Hagan, and H. B. Demuth, “Neural network toolbox,” User’s Guid. MathWorks, vol. 2, pp. 77–81, 2010.
[21] F. S. Abdullah, A. N. Hamoodi, and R. A. Mohammed, “Performance improvement in steam turbine in thermal power plants using
artificial neural network,” J. Eur. des Systèmes Autom., vol. 54, no. 6, pp. 891–895, 2021, doi: 10.18280/jesa.540611.
[22] A. Asgharnia, A. Jamali, R. Shahnazi, and A. Maheri, “Load mitigation of a class of 5-MW wind turbine with RBF neural network
based fractional-order PID controller,” ISA Trans., vol. 96, pp. 272–286, 2020, doi: 10.1016/j.isatra.2019.07.006.
[23] R. Sitharthan, C. K. Sundarabalan, K. R. Devabalaji, T. Yuvaraj, and A. Mohamed Imran, “Automated power management
strategy for wind power generation system using pitch angle controller,” Meas. Control, vol. 52, no. 3–4, pp. 169–182, 2019, doi:
10.1177/0020294019827330.
[24] H. Erol, “Stability analysis of pitch angle control of large wind turbines with fractional order PID controller,” Sustain. Energy,
Grids Networks, vol. 26, p. 100430, 2021, doi: 10.1016/j.segan.2021.100430.

Performance enhancement of small-scale wind turbine based on artificial … (Mohammed Ahmed Ibrahim)
1730  ISSN: 2088-8694

[25] J. E. Sierra-Garcia and M. Santos, “Combination of neural networks and reinforcement learning for wind turbine pitch control,” in
Hybrid Artificial Intelligent Systems: 17th International Conference, HAIS 2022, Salamanca, Spain, September 5–7, 2022,
Proceedings, 2022, pp. 385–392, doi: 10.1007/978-3-031-15471-3_33.
[26] A. N. Hamoodi, M. A. Ibrahim, and B. M. Salih, “An intelligent differential protection of power transformer based on artificial
neural network,” Bull. Electr. Eng. Informatics, vol. 11, no. 1, pp. 93–102, 2022, doi: 10.11591/beei.v11i1.3547.

BIOGRAPHIES OF AUTHORS

Mohammed Ahmed Ibrahim is an engineer Department of Electrical Power

Technologies Engineering. He received bachelor's degree from the Northern Technical
University 2005 in Electrical Engineering, master's degree from the Northern Technical
University in (2011) in Electrical Power Technologies Engineering. Currently, he is working
as a lecturer in the Electrical Power Engineering specializing, at Northern Technical
University, Technical College of Engineering-Mosul-Iraq. He can be contacted at email:
[email protected].

Ali Saleh Alhfidh obtained his M.Sc. from the Northern Technical University in
(2022) in Electrical Power Technologies Engineering. Currently he is working as a lecturer in
the Electrical Power Engineering specializing, at Northern Technical University-Mosul-Iraq.
He can be contacted at email: [email protected].

Ali Nathem Hamoodi teacher Department of Electrical Power Technologies

Engineering. Master's degree (2000), and Ph.D. (2011) in electrical engineering, from the
University of Aleppo, Syria. Currently, Research published in international journals, Scopus
containers and science Seven, and Five internal and sponsored conferences. His specializing in
the Electrical Power Engineering and he is working as a lecturer in Northern Technical
University, Technical College of Engineering-Mosul-Iraq. He can be contacted at email:
[email protected].

Int J Pow Elec & Dri Syst, Vol. 14, No. 3, September 2023: 1722-1730