Sustainable Energy Technologies and Assessments: Abolfazl Pourrajabian, Maziar Dehghan, Saeed Rahgozar
Sustainable Energy Technologies and Assessments: Abolfazl Pourrajabian, Maziar Dehghan, Saeed Rahgozar
Original article
Genetic algorithms for the design and optimization of horizontal axis wind
turbine (HAWT) blades: A continuous approach or a binary one?
Abolfazl Pourrajabian *, Maziar Dehghan , Saeed Rahgozar
Department of Energy, Materials and Energy Research Center (MERC), 14155-4777 Tehran, Iran
A R T I C L E I N F O A B S T R A C T
Keywords: The continuous and the binary techniques are the main types of genetic algorithms that could be employed for an
Wind turbine optimization problem. The study investigates the robustness and accuracy of each technique for the wind turbine
Blade blades design and optimization problem. For that purpose, the geometry of the blade was designed for the
Optimization
maximization of the output power, the desirable goal of the blade design. The design variables consist of the
Continuous genetic algorithm
distribution of the chord and twist along the blade. Results indicate that the continuous genetic algorithm
Binary genetic algorithm
outperforms the binary one from the standpoint of the accuracy and also the computational time. However, the
use of uniform crossover could improve the convergence rate of the binary genetic algorithm. Moreover, the
sensitivity analysis of the applied genetic algorithms with respect to the population size and the mutation rate
was performed to find the appropriate values for those parameters. The outcomes emphasized that adopting a
small number of population size together with a large number of generations could speed up the convergence
rate of the problem. Also approved in the study was the efficiency of the so-called “Superblade” operator which
improved the convergence rate of the algorithm and resulted in the powerful blades.
* Corresponding author.
E-mail address: a.pourrajabian@merc.ac.ir (A. Pourrajabian).
https://doi.org/10.1016/j.seta.2021.101022
Received 17 September 2020; Received in revised form 16 December 2020; Accepted 11 January 2021
Available online 3 February 2021
2213-1388/© 2021 Elsevier Ltd. All rights reserved.
A. Pourrajabian et al. Sustainable Energy Technologies and Assessments 44 (2021) 101022
the convergence rate and also the computational time. Moreover, the
effect of GA input parameters values, the population size and the mu
tation rate, on the convergence rate of both algorithms will be fully
investigated. This analysis would be of paramount importance as an
efficient and robust GA could reduce the number of objective function
evaluations, making it a less computationally expensive algorithm.
The study begins with the concepts of binary and real-coded GAs.
Thereafter, the way of the objective function calculation is introduced
and then followed by results, discussion and conclusions.
2
A. Pourrajabian et al. Sustainable Energy Technologies and Assessments 44 (2021) 101022
Fig. 2. The continuous/binary GA procedure (The dashed blocks are exclusively used in the BGA).
blade element equations as follows [11] Number of blades (N) = 3 U = 10 m/s λp = 5.71
Airfoil type: SG6043 R = 1.21 m Ω = 450 rpm
3
A. Pourrajabian et al. Sustainable Energy Technologies and Assessments 44 (2021) 101022
mutation rate (μ) and also the number of generations are summarized in Table 3
Table 2. As will be outlined in the next section, the optimization pro The allocated range for design variables [40].
cedure was performed within the combination of these values for both Parameter and value Parameter and value
BGA and CGA. More specifically, 24 optimization cases will be per
Minimum c/R = 0.01 Maximum c/R = 0.2
formed. Allocating lower values to the mutation rate is recommended by Minimum θ = -5o Maximum θ = 25o
the researchers as higher values would slow the GA convergence [30].
For the population size, however, a broad range was considered; Table 2.
A combination of the small value for the population size and a large one
Table 4
for the generation would guarantee a sufficient exploration of the so The characteristics of Andersons’s turbine [11,41,42].
lution space. The reverse case holds also true and will be analyzed.
Blade length (m) Design speed Blade profile Number of blades
Fifteen blade elements were selected along the blade and therefore,
the problem consists of 30 design variables. Table 3 lists the range of the 1.5 10 m/s NACA4412 2
twist angle (θ) and also the non-dimensional chord (c/R) which are
forced by the manufacturing limitations and difficulties [40]. In the
binary GA, five bits were dedicated to each design variable and therefore airfoil occurs at α = 5◦ and the corresponding Cl is 1.22 [43].
each allocated range listed in Table 3 was divided into 32 pieces (25) Table 6 summarizes the values of Cp for 24 cases: 12 for BGA and 12
from which the optimal ones should be determined in fifteen stations for CGA. Each case was evolved through a large number of generations.
along the blade. In the continuous GA, however, any floating-point As an example, Fig. 7 shows the variations in Cp during the GA cycle for
number could be selected from the allocated range. one of the CGA cases. Also shown is the Betz-Joukowsky limit (16/27 ~
0.6). As can be seen, the convergence rate of CGA is faster than the BGA.
Results Results highlight that values of Cp for CGA are slightly larger than
BGA, implying that the blades designed by CGA could harness more
This section presents the outcomes of the study. At first, the valida kinetic energy from the wind than those designed by the BGA. The
tion of the BEM analysis is investigated in subsection 3.1. Then, the flexibility of the CGA for choosing any arbitrary real number for the
optimization results are provided. Also will be investigated are the effect design variables (chord and twist angle) is the main reason for that while
of employing the uniform crossover and the “Superblade” operator. this is not the case for the BGA where the values of the chord and twist
Finally, the computational cost of the BGA and CGA is analyzed. angle should necessarily be selected from the allocated range deter
mined by the number of bits.
Validation of the numerical procedure Irrespective of adopting BGA or CGA, results emphasize that the use
of lower values for the mutation rate results in higher Cp. Particularly for
An in-house code written by MATLAB was used for both optimization BGA, applying the larger value for the mutation rate (0.15) led to the
procedure and also BEM analysis. Prior to the optimization study, it is noticeable lower Cp listed in Table 6.
necessary to evaluate the accuracy of the BEM analysis as the fitness Although the use of BGA led to the lower values for Cp, the results
values (objective function values) are calculated through that analysis. summarized in Table 6 indicate that choosing a small number of popu
In this regard, wind tunnel data measured by Anderson et al. [41] was lation size for the BGA, could increase the power coefficient. The same is
selected to assess the programmed code of the present study. Table 4 also true for the CGA, albeit by the smaller degree (Table 6). These re
summarizes the technical parameters of Andersons’s turbine including sults highlight that applying the small values for both population size
the blade length, the design speed, the blade profile and the number of and also the mutation rate would result in higher Cp for both BGA and
blades [41]. Moreover, the distributions of the chord and twist for this CGA. For the 30 design variables along the blade, the outcomes of the
two-bladed turbine are given in Table 5. The lift and the drag co optimization study suggest the use of 0.05 for the mutation rate and 50
efficients of NACA4412 profile are shown in Fig. 4 for a broad range of for the population size.
angle of attack (α) and also Reynolds number (Re). These values were
tabulated and called during the BEM analysis. Optimal variables
Power coefficient (Cp) values for the present study are compared
with the wind tunnel data in Fig. 5 where the variations in Cp is depicted Figs. 8 and 9 show the optimal distributions of the chord and twist for
over a broad range of the tip speed ratio (λ). As can be seen, the nu both BGA and CGA. Note that the successive plots were displaced up
merical results are close to the experimental data, ensuring that the wards. Also shown are the “ideal” distributions of the chord (c) and twist
numerical procedure is on the right track. (θ) which were derived by Burton et al. [34]:
4
A. Pourrajabian et al. Sustainable Energy Technologies and Assessments 44 (2021) 101022
Table 5
The geometric characteristics of Andersons’s turbine blade [11,41,42].
Radius (cm) 19.5 28.5 37.5 46.5 55.5 64.5 73.5 82.5 91.5 100.5 109.5 118.5 127.5 136.5 145.5
Chord (cm) 25.3 20.5 16.9 14.0 11.9 10.3 9.1 8.2 7.5 6.9 6.3 5.8 5.4 5.0 4.7
Twist (deg.) 24.7 16.8 11.3 7.6 5.2 3.7 2.7 1.9 1.2 0.6 0.2 − 0.3 − 0.7 − 1.1 − 1.5
Fig. 5. Variations in the power coefficient with respect to the tip speed ratio for Anderson’s turbine [41].
As can be seen, the optimal distributions for both BGA and CGA are the very small contribution of this element in production of total aero
identical to these ideal relations, implying that the optimization tool of dynamic torque (Q). In fact, the main part of Q is generated by the el
the present study performs well. However, a close look at Figs. 8 and 9 ements near the blade tip rather than the root ones. This forces GA to
reveals that the chord/twist distribution is not completely smooth evolve the design variables, i.e. chord and twist, near the tip part
especially for the first element in the root part of the blade. This is due to compared to the root one, leading to the smooth distribution from the
5
A. Pourrajabian et al. Sustainable Energy Technologies and Assessments 44 (2021) 101022
Table 6
Values of Cp with respect to the generation, population size and the mutation rate.
GA type Mutation rate Population Generation Population Generation Population Generation Population Generation
mid part of the blade toward the tip. To investigate this issue in detail, Effect of employing uniform crossover
Fig. 10 depicts the aerodynamic torque along the blade. As shown, it
gradually increases towards the tip. The small drop at the blade tip is due The crossover operator performed here was the single-point one
to the inclusion of the tip loss in the BEM analysis mentioned earlier. As shown in Fig. 3. It is argued that this technique could not appropriately
labeled in Fig. 10, the contribution of the fifteenth element (corre cover the space of design variables [30]. In fact, the children produced
sponding to 0.044) is far larger than the first one (0.004). Fig. 11 also by this operator are very similar to their parents except for one design
shows the variations in the aerodynamic torque of these elements during variable (See Fig. 3). This reduces the convergence rate of the GA and
the CGA cycle. It is seen that the local torque for the first element adds to the computational time.
converged to the value that is not the highest one which is not the case The uniform crossover, instead, is deemed to be more effective
for the fifteenth element for which the optimization algorithm [30,46]. Fig. 12 shows how the uniform crossover operates in the BGA.
converged to the highest one (See Fig. 11). Following the randomly generation of the so-called “Mask” vector, the
corresponding bits with zero values are selected for the crossover.
Similar to the single point crossover, the uniform crossover is also
6
A. Pourrajabian et al. Sustainable Energy Technologies and Assessments 44 (2021) 101022
with respect to each other in the BEM analysis. In fact, the BEM analysis
is carried out for all the blades elements to give the local thrust and the
local aerodynamic torque. Since the BEM computations for each element
is independent of the others, choosing the local design variables (i.e., the
chord and the twist) based on the maximum value of the local aero
dynamic torque leads to a new blade called the “Superblade” which is
not already in the population matrix. This idea is very beneficial as the
corresponding chord/twist of the maximum aerodynamic torque could
be selected for each element in each generation. The elements of the blade
generated by this creative operator produce maximum aerodynamic
torque and hence output power. By unchangeably moving this specific
blade from the current generation to the next one, the algorithm is
guided toward a more powerful blade, the aim of the optimization.
Fig. 7. Variations in CP through the CGA (Ps = 50, μ = 0.05). To find the efficiency of this operator, the optimization procedure
was repeated by including that in the GA cycle. Fig. 14 shows the var
performed in the CGA as follows [30]: iations in Cp for one of the BGA cases. As indicated, the convergence rate
As can be seen, the uniform crossover is a robust operator which of the algorithm expedited significantly thanks to the use of “Super
forces the algorithm to efficiently explore the solution space. Fig. 13 blade” operator. Moreover, the optimal value of Cp has slightly increased
shows the fitness function of this study for both the single-point cross (Fig. 14).
over and the uniform one. It is seen that the adoption of the uniform
crossover led to a more powerful blade (Fig. 13).
Fig. 8. Variations in the non-dimensional chord along the blade (successive plots were displaced upwards by 0.2).
Fig. 9. Variations in the twist angle along the blade (successive plots were displaced upwards by 25◦ ).
7
A. Pourrajabian et al. Sustainable Energy Technologies and Assessments 44 (2021) 101022
Table 7 summarizes the computational time of the CGA and BGA for
four considered cases. All were run on a personal computer (PC) with
2.4-GHz Intel Dual Core CPU and 4-GB RAM. It is seen that the
computational time for the binary GA is larger than the continuous one.
The main reason for that is the decoding procedure of the binary space to
the real one and vice versa as shown in Fig. 2. Moreover, the population
size for the BGA is noticeably larger than the CGA as the former consists
of 150 (30 × 5) bits for every candidate blade while the corresponding
figure for the latter is only 30. This also adds to the computational time
during the optimization process.
Conclusions
Fig. 13. Variations of power coefficient during the BGA cycle for two types of crossover (Ps = 50, μ = 0.05).
8
A. Pourrajabian et al. Sustainable Energy Technologies and Assessments 44 (2021) 101022
Fig. 14. Variations of power coefficient during the BGA cycle (Ps = 50, μ = 0.15).
Table 7
Run time (s) for four cases based on the population size and the number of generations (μ = 0.05).
Population Generation Population Generation Population Generation Population Generation
2000 500 50 2000 100 1000 1000 1000
algorithms for finding more powerful blades. This would be of para CRediT authorship contribution statement
mount importance as it could be a step forward towards the improve
ment of wind turbines technology and thus making them more profitable Abolfazl Pourrajabian: Conceptualization, Methodology, Software,
and sustainable. The design variables consist of the distributions of the Validation, Formal analysis, Investigation, Writing - review & editing.
chord and twist angle. The blade-element momentum theory was Maziar Dehghan: Software, Validation, Formal analysis, Writing - re
adopted for the calculation of the objective function. The optimal dis view & editing. Saeed Rahgozar: Software, Validation, Formal analysis,
tribution of the design variables including the chord and twist angle was Writing - review & editing.
assessed with the so-called ideal distribution which approved the ac
curacy of the programmed code. Results highlighted that the continuous Declaration of Competing Interest
type outperforms the binary one. The sensitivity analysis of the genetic
algorithm parameters revealed that the small number of population size The authors declare that they have no known competing financial
and also the mutation rate are suitable for finding the optimal solution interests or personal relationships that could have appeared to influence
provided that a large number of generations is allocated for running the the work reported in this paper.
genetic algorithm cycle. Moreover, results indicated that replacing the
standard single-point crossover with the uniform one could explore the Acknowledgement
solution space more effectively, leading to more powerful blades for a
constant number of generations. Also investigated was the efficiency of This work was supported by the Iran National Science Foundation
the so-called “Superblade” operator which selects the design variables “INSF” (Grant number 97023376).
for each blade element based on the aerodynamic torque. This results in
a blade whose elements produce more aerodynamic torque. For that References
purpose, the design variables, chord and twist, were selected at the end
of each generation and moved unchangeably to the next one. This [1] Dovì VG, Friedler F, Huisingh D, Klemeš JJ. Cleaner energy for sustainable future.
J Cleaner Prod 2009;17(10):889–95.
customized operator was added to the canonical genetic algorithms [2] Kuşkaya S, Bilgili F. The wind energy-greenhouse gas nexus: The wavelet-partial
which not only improved the convergence rate of the algorithm but also wavelet coherence model approach. J Cleaner Prod 2020;245:118872.
led to the powerful blade, the goal of the optimization. To sum up, the [3] Li J, Li S, Wu F. Research on carbon emission reduction benefit of wind power
project based on life cycle assessment theory. Renewable Energy 2020;155:456–68.
simultaneously use of a continuous genetic algorithm and the “Super [4] Pehl M, Arvesen A, Humpenöder F, Popp A, Hertwich EG, Luderer G.
blade” operator requires less computational time and more importantly Understanding future emissions from low-carbon power systems by integration of
results in more powerful blades. Although considering a sophisticated life-cycle assessment and integrated energy modelling. Nat Energy 2017;2(12):
939–45.
objective function such as a multi-objective one would increase the [5] Welch JB, Venkateswaran A. The dual sustainability of wind energy. Renew Sustain
computational cost, this does not necessarily raise doubts about the Energy Rev 2009;13(5):1121–6.
general outcomes of the study achieved through different combinations [6] Rahgozar S, Pourrajabian A, Kazmi SAA, Kazmi SMR. Performance analysis of a
small horizontal axis wind turbine under the use of linear/nonlinear distributions
of genetic algorithm parameters, the population size and the mutation
for the chord and twist angle. Energy Sustain Develop 2020;58:42–9.
rate, for both continuous approach and the binary one. [7] Lin YT, Chiu PH. Influence of leading-edge protuberances of fx63 airfoil for
horizontal-axis wind turbine on power performance. Sustain Energy Technol Assess
2020;38:100675.
9
A. Pourrajabian et al. Sustainable Energy Technologies and Assessments 44 (2021) 101022
[8] Abdelwaly M, El-Batsh H, Hanna MB. Numerical study for the flow field and power [26] Pourrajabian A, Mirzaei M, Ebrahimi R, Wood D. On the effect of altitude on the
augmentation in a horizontal axis wind turbine. Sustain Energy Technol Assess performance of a small wind turbine blade. ASME Turbo Expo 2014. https://doi.
2019;31:245–53. org/10.1115/GT2014-27335.
[9] Sun X, Zhu J, Hanif A, Li Z, Sun G. Effects of blade shape and its corresponding [27] Pourrajabian A, Dehghan M, Javed A, Wood D. Choosing an appropriate timber for
moment of inertia on self-starting and power extraction performance of the novel a small wind turbine blade: A comparative study. Renew Sustain Energy Rev 2019;
bowl-shaped floating straight-bladed vertical axis wind turbine. Sustain Energy 100:1–8.
Technol Assess 2020;38:100648. [28] Ariyarit A, Kanazaki M, Bureerat S. An approach combining an efficient and global
[10] Didane DH, Rosly N, Zulkafli MF, Shamsudin SS. Numerical investigation of a evolutionary algorithm with a gradient-based method for airfoil design problems.
novel contra-rotating vertical axis wind turbine. Sustain Energy Technol Assess Smart Sci 2020:1–10.
2019;31:43–53. [29] Sessarego M. “A Hybrid Multi-Objective Evolutionary Algorithm for Wind-Turbine
[11] Wood DH. Small wind turbines: analysis, design, and application, green energy and Blade Optimization”.Master’s thesis. Alberta, Canada: University of Calgary; 2013.
technology. London: Springer-Verlag; 2011. [30] Haupt RL, Haupt SE. Practical genetic algorithms. 2nd edition. New York: John
[12] Sessarego M, Wood D. Multi-dimensional optimization of small wind turbine Wiley and Sons; 2004.
blades. Renewables: Wind, Water, and Solar 2015;2(1):1–11. [31] Holland JH. Adaptation in natural and artificial systems. Ann Arbor: University of
[13] Andrew Ning S, Damiani R, Moriarty PJ. Objectives and constraints for wind Michigan Press; 1975.
turbine optimization. J Sol Energy Eng 2014;136(4). [32] Deb K. Multi-objective optimization using evolutionary algorithms. New York:
[14] Amir Nazmi Afshar, P., Gooya, M., Hosseini, S.V. and Pourrajabian, A., 2017. John Wiley and Sons; 2001.
Starting improvement of micro-wind turbines operating in low wind speed regions. [33] Chen Y, Li H, Jin K, Song Q. Wind farm layout optimization using genetic algorithm
International Journal of Green Energy, 14(11), pp.868-877. with different hub height wind turbines. Energy Convers Manage 2013;70:56–65.
[15] Clifton-Smith MJ. Aerodynamic noise reduction for small wind turbine rotors. [34] Burton T, Sharpe D, Jenkins N, Bossanyi E. Wind Energy Handbook. 2nd edition.
Wind Eng 2010;34(4). John Wiley and Sons; 2011.
[16] Ram M, Child M, Aghahosseini A, Bogdanov D, Lohrmann A, Breyer C. [35] Song Q, Lubitz WD. BEM simulation and performance analysis of a small wind
A comparative analysis of electricity generation costs from renewable, fossil fuel turbine rotor. Wind Eng 2013;37(4):381–99.
and nuclear sources in G20 countries for the period 2015–2030. J Cleaner Prod [36] Qiyue S, Lubitz WD. Design and testing of a new small wind turbine blade. J Solar
2018;199:687–704. Energy Eng ASME 2014;136:034502.
[17] Kavari G, Tahani M, Mirhosseini M. Wind shear effect on aerodynamic [37] Hansen MO. Aerodynamics of wind turbines. Routledge; 2015.
performance and energy production of horizontal axis wind turbines with [38] Selig MS, McGranahan BD. Wind tunnel aerodynamic tests of six airfoils for use on
developing blade element momentum theory. J Cleaner Prod 2019;219:368–76. small wind turbines. J Solar Energy Eng ASME 2004;126(4):986–1001.
[18] Sessarego M, Dixon KR, Rival DE, Wood DH. A hybrid multi-objective evolutionary [39] Giguere P, Selig MS. New airfoils for small horizontal axis wind turbines. J Sol
algorithm for wind-turbine blade optimization. Eng Optim 2015;47(8):1043–62. Energy Eng 1998;120(2):108–14.
[19] Vorspel L, Stoevesandt B, Peinke J. Optimize rotating wind energy rotor blades [40] Pourrajabian A, Ebrahimi R, Mirzaei M. Applying micro scales of horizontal axis
using the adjoint approach. Appl. Sci. 2018;8(7):1112. wind turbines for operation in low wind speed regions. Energy Convers Manage
[20] Xudong W, Shen WZ, Zhu WJ, Sørensen JN, Jin C. Shape Optimization of Wind 2014;87:119–27.
Turbine Blades. Wind 2009;Energy12 (8):781–803. [41] Anderson MB, Milborrow DJ, Ross NJ. Performance and wake measurements on a 3
[21] Fuglsang P, Madsen HA. Optimization method for wind turbine rotors. J Wind Eng m diameter horizontal axis wind turbine: comparison of theory. wind tunnel and
Ind Aerodyn 1999;80(1–2):191–206. field test data. Department of Energy Technology Support Unit; 1982.
[22] Wang L, Wang T, Luo Y. Improved Non-Dominated Sorting Genetic Algorithm [42] Spera D. Wind turbine technology. New York: ASME Publishing; 1994.
NSGA-II in Multi-Objective Optimization Studies of Wind Turbine Blades. Appl [43] Clifton-Smith MJ, Wood DH. July. Further dual purpose evolutionary optimization
Math Mech 2011;32(6):739–48. of small wind turbine blades. J Phys Conf Ser 2007;Vol. 75, No. 1.
[23] Cai X, Zhu J, Pan P, Gu R. Structural optimization design of horizontal-axis wind [44] Lyon CA, Broeren AP, Giguere P, Gopalarathnam A, Selig MS. Summary of low
turbine blades using a particle swarm optimization algorithm and finite element speed airfoil data, vol 3. Virginia Beach: Soartech; 1997.
method. Energies 2012;5(11):4683–96. [45] Gasch R, Twele J. Blade geometry according to Betz and Schmitz. In: Wind Power
[24] Tahani M, Maeda T, Babayan N, Mehrnia S, Shadmehri M, Li Q, et al. Investigating Plants. Berlin, Heidelberg: Springer; 2012. p. 168–207.
the effect of geometrical parameters of an optimized wind turbine blade in [46] Yi JH, Xing LN, Wang GG, Dong J, Vasilakos AV, Alavi AH, et al. Behavior of
turbulent flow. Energy Convers Manage 2017;153:71–82. crossover operators in NSGA-III for large-scale optimization problems. Inf Sci 2020;
[25] Pourrajabian A, Afshar PAN, Mirzaei M, Ebrahimi R, Wood DH. Hollow blades for 509:470–87.
small wind turbines operating at high altitudes. J Sol Energy Eng 2016;138(6).
10