=Paper=
{{Paper
|id=None
|storemode=property
|title=Planning Fitness Training Sessions Using the Bat Algorithm
|pdfUrl=https://ceur-ws.org/Vol-1422/121.pdf
|volume=Vol-1422
|dblpUrl=https://dblp.org/rec/conf/itat/FisterRFFF15
}}
==Planning Fitness Training Sessions Using the Bat Algorithm==
https://ceur-ws.org/Vol-1422/121.pdf
J. Yaghob (Ed.): ITAT 2015 pp. 121–126
Charles University in Prague, Prague, 2015
Planning Fitness Training Sessions Using the Bat Algorithm
Iztok Fister Jr.1 , Samo Rauter2 , Karin Ljubič Fister3 , Dušan Fister1 , and Iztok Fister1
1 University of Maribor, Faculty of Electrical Engineering and Computer Science,
Smetanova 17, 2000 Maribor, Slovenia,
iztok.fister1@um.si
2 University of Ljubljana, Faculty of Sport
Gortanova 22, 1000 Ljubljana
3 University of Maribor, Faculty of Medicine,
Taborska 8, 2000 Maribor, Slovenia
Abstract: Over fairly recent years the concept of an ar- • Recreational cycling marathons - This kind of
tificial sport trainer has been proposed in literature. This mass [7, 8] sports events was very popular approxi-
concept is based on computational intelligence algorithms. mately 10 years ago but still represents a challenge
In this paper, we try to extend the artificial sports trainer for numerous participants. The current world eco-
by planning fitness training sessions that are suitable for nomic problems increased the prices of cycles and
athletes, especially during idle seasons when no competi- consequently less participants could participate on
tion takes place (e.g., winter). The bat algorithm was used cycling marathons.
for planning fitness training sessions and results showed
promise for the proposed solution. Future directions for Participants of the mentioned events participate mostly
development are also outlined in the paper. because of two goals. The first goal is to enjoy (in other
words: to have a nice time) and the second is to finish the
1 Introduction trial. Usually, finishers are awarded with medals which
are big stimulants for participants. In this case, every year
Sport becomes highly addictive for many people in the many more participants have also began to take these com-
world. A few decades ago, people around the world spend petitions more seriously, i.e., semi professional. In line
their free time doing different activities, like: short walks with this, they invest much more time in preparations for
through the park, visiting the cinema or galleries, fish- competitions. Unfortunately, there is a long way for good
ing, visiting a thermal spa and also meet friends. A lot preparations for such kinds of competitions. This good
of leisure studies also proved this. However, over recent preparation consists of proper sports training, good eat-
decades, lifestyles have been substantially changed espe- ing and also good resting. To maintain all these factors as
cially because of globalization that has transformed the high as possible is very hard for numerous athletes, since
whole earth into a global village. Due to a lack of time they do not have enough experience. Newbie athletes es-
as well as personal willingness, people do not want to live pecially suffer from the unwanted effects of irregular train-
like they used to. For this reason, different kind of new ing called over-training syndrome [9, 10, 11] which is re-
activities have emerged over recent past years. One of flected in reduced form. One of the possible solutions for
the bigger revitalizations has been sport which has became avoiding this is to hire a personal trainer or join diverse
extremely in popular because of the emergence of differ- training groups. However, these cost a lot of money and
ent mass sports events. For instance, the main mass sport therefore many of them can not afford them.
events are: In order to break this barrier, we began the development
of an artificial sport trainer. An artificial sports trainer
• Triathlon - This discipline consists of three sports:
was presented recently in [12] and is based on computa-
swimming, cycling and running. Additionally, there
tional intelligence [13] algorithms that are able for plan the
are different distances which vary from short via
sports training over both short-term and long-term. This
medium to long distances. One of the more famous
trainer is also able to discover the different habits of ath-
distances is the Ironman triathlon [1, 2, 3], which is
letes, avoid over-training, etc. Data for the artificial sports
also known as the hardest one day sport event. Iron-
trainer are obtained from sports trackers [14] and sports
man consists of 3.8 km of swimming, 180 km of cy-
watches like Garmin.
cling and 42.2 km of running.
This paper, extends the artificial sport trainer with plan-
• Road marathons - A road marathon [4] consists of ning fitness training sessions. These kinds of training ses-
42.2 km pure running and is a challenge for myriad sions are very important for athletes especially during idle
of people. Big city marathons especially are the most seasons. In Europe, the idle season is usually during win-
popular and attract large numbers of runners. Some ter months when the athletes prepare their form for the
marathons can accommodate more than 40,000 run- whole season. Planning fitness sessions were performed
ners [5, 6]. using a bat algorithm [15, 16] which is a member of the
122 I. Fister Jr., S. Rauter, K. L. Fister, D. Fister, I. Fister
computational intelligence family. The planning of the
fitness sessions was defined as a constraint satisfaction
problem, where the bat algorithm searches for feasible so-
lutions arising when the number of constraint violations
achieved the value of zero.
Organization of the remainder of this workshop paper
is as follows: in section 2 we discuss about characterists
of fitness training, while section 3 presents swarm intelli-
gence algorithms and bat algorithm. Experiments and re-
sults are presented in section 4, while section 5 concludes
the paper.
Figure 2: Squats exercise
2 Characteristics of Fitness Training
This study, focused on fitness training in regard to cycling.
Incorporation of strength training in cyclists preparatory
periods has received more attention over the last two
decades. Most of the serious and competitive cyclists also
include strength training in their training programs. It is
also evident in some previous research that adding strength
training to an endurance training program can increase en-
durance performance [17, 18].
A combination of endurance and strength training (con-
current training) might therefore be a potential training
strategy for promoting muscle oxidative capacity. It might
be related to an improved cycling economy, as observed Figure 3: Lunge exercise
after adding strength training to the ongoing endurance
training, namely because a stronger muscles at a certain For the smart planning of sports training, quantifica-
intensity operates longer with a lower percentage of max- tions, regulating the intensity of a workout is the key for
imum capacity. It is well-known that adding strength success as indicated as basic knowledge in sports training
training to endurance training can increase the maximal literature. This fact also holds for fitness training. As an
strengths and rate of force developments in cyclists. In estimate of the intensity of a fitness workout, two main
theory, this may improve pedaling characteristics by in- measures are employed like a:
creasing peak torque in the pedal stroke, reducing time
to peak torque and reducing the pedaling torque relative • the number of repetitions per set of exercises (NR),
to maximal strength, which in turn may allow for higher
• the maximum amount of weight that can be generated
power output and/or increased blood flow.
in one maximum contraction (1RM).
The most important thing for developing a cycling
strength program is to know, which muscle groups are The logic behind the first measure is as follows. The heav-
the most active during the pedal stroke. Some previous ier the weight, the higher the intensity and the fewer rep-
studies have detected a strong correlation between cycling etitions (also reps) an athlete will be able to lift it for. On
performance and some strength exercises, like leg presses, the other hand, the 1RM determines the desired load for
squats, and deadlifts [17, 18, 19, 20]. Some of the more an exercise (typically as a percentage of the 1RM). Let us
useful exercises for fitness training are presented in Figs. 1 notice that a coach determines the measure of 1RM for
to 3. a definite athlete using tests at the beginning of the fitness
training and then calculates the number of repeats (NR) in
regard to this characteristic value.
3 Swarm Intelligence Based Algorithms
Swarm intelligence (SI) is a paradigm that belongs to com-
putational intelligence (CI). According to the [21], SI con-
cerns the collective, emerging behavior of multiple, inter-
acting agents that are capable of performing simple ac-
Figure 1: Deadlift exercise tions. While each agent may be considered as unintel-
ligent, the whole system of multiple agents shows some
Planning Fitness Training Sessions Using the Bat Algorithm 123
self-organizational behavior and thus can behave like some Algorithm 2 Bat algorithm
sort of collective intelligence. The basic pseudo-code of Input: Bat population xi = (xi1 , . . . , xiD )T for i = 1 . . . N p,
SI-based algorithms is presented in Algorithm 1. Nowa- MAX_FE.
days, the bat algorithm is one of the promising members Output: The best solution xbest and its corresponding
of the SI family. It is very easy to implement and shows ef- value fmin = min( f (x)).
ficient results especially when solving small dimensional 1: init_bat();
problems. 2: eval = evaluate_the_new_population;
3: f min = find_the_best_solution(xbest ); {initialization}
Algorithm 1 Swarm Intelligence 4: while termination_condition_not_meet do
1: initialize_population_with_random_candidate_particles; 5: for i = 1 to N p do
2: eval = evaluate_each_particle; 6: y = generate_new_solution(xi );
3: while termination_condition_not_meet do 7: if rand(0, 1) > ri then
4: move_particles_towards_the_best_individual; 8: y = improve_the_best_solution(xbest )
5: eval += evaluate_each_particle;
9: end if{ local search step }
6: select_the_best_individuals_for_the_next_generation;
10: fnew = evaluate_the_new_solution(y);
7: end while
11: eval = eval + 1;
12: if fnew ≤ fi and N(0, 1) < Ai then
Next subsection describes the mentioned algorithm in 13: xi = y; fi = fnew ;
detail. 14: end if{ save the best solution conditionally }
15: fmin =find_the_best_solution(xbest );
3.1 Bat Algorithm 16: end for
17: end while
The bat algorithm was developed by Yang in 2010. The
main purposes of this algorithm were to be: simple, effi-
cient and applicable to varios problem domains. The inspi- • evaluate_the_new_solution (line 10): evaluating the
ration for the bat algorithm came from the phenomenon of new solution,
the echolocation characteristics of some types of micro-
bats. Developer used a three simplified rules describing • save_the_best_solution_conditionaly (lines 12-14):
the bat behavior, as follows [22]: saving the new best solution under some probability
Ai ,
• All bats use echolocation to sense distance to target
objects. • find_the_best_solution (line 15): finding the current
best solution.
• Bats fly randomly with the velocity vi at position xi ,
the frequency Qi ∈ [Qmin , Qmax ] (also the wavelength Generating the new solution is governed by the follow-
λi ), the rate of pulse emission ri ∈ [0, 1], and the loud- ing equation:
ness Ai ∈ [A0 , Amin ]. The frequency (and wavelength)
can be adjusted depending on the proximities of their (t)
Qi = Qmin + (Qmax − Qmin )N(0, 1),
targets.
= vti + (xti − best)Qi ,
(t+1) (t)
vi (1)
• The loudness varies from a large (positive) A0 to a (t+1) (t) (t+1)
minimum constant value Amin . xi = xi + vi ,
The algorithm’s pseudo-code is presented in Algorithm 2. where N(0, 1) is a random number drawn from a Gaussian
The main bat algorithm components [23] are summa- distribution with zero mean and a standard deviation of
rized as follows: one. A RWDE heuristic implemented in the function im-
prove_the_best_solution modifies the current best solution
• initialization (lines 1-3): initializing the algorithm according to the equation:
parameters, generating the initial population, evalu-
ating this, and finally, determining the best solution (t)
x(t) = best + εAi N(0, 1), (2)
xbest in the population,
• generate_the_new_solution (line 6): moving the vir- where N(0, 1) denotes the random number drawn from
tual bats in the search space according to the physical a Gaussian distribution with zero mean and a standard de-
(t)
rules of bat echolocation, viation of one, ε being the scaling factor, and Ai the loud-
ness.
• local_search_step (lines 7-9): improving the best Contemporary work on bat algorithms captures many
solution using random walk direct exploitation variants and application domains. Some recent works are
(RWDE) heuristic, presented in papers [24, 25, 26, 27, 28]
124 I. Fister Jr., S. Rauter, K. L. Fister, D. Fister, I. Fister
3.2 Bat Algorithm for Planning Fitness Sessions Defining constraints The purpose of the fitness training
plan is to prescribe sufficient numbers of exercises for each
Based on the original bat algorithm, we have developed of the prescribed muscle groups, their number of repeats
a modified bat algorithm for planning fitness sessions. De- (NR) and the proper intensities (%1RM) such that an ath-
velopment of this algorithm demanded the following four lete simultaneously develops all the muscle groups needed
steps: for building the cyclist’s basic form. Therefore, trainers
• determining the fitness exercises, determine the proper amount of a specific exercise in the
plan in regarding to the others. In order to regulate the re-
• defining constraints, lations between exercises in the fitness training plan, the
following constraints are defined:
• modifying the original bat algorithm,
• at least four exercises must have the number of re-
• representing the results and their visualizations.
peats over 25 times (i.e., NR>25),
In the remainder of this paper, all these steps are described
in detail. • each training plan should have at least two exercises
of high intensity,
Selecting Fitness Exercises. We need to determine spe- • each muscle group repeating in Table 1 more than
cific exercises for different muscle groups before the fit- once does not have the same exercise,
ness training can start. Although sports medicine recog-
• if the last exercise in the fitness training plan was
nizes more than 15 muscle groups that must be included
of higher intensity, the next exercise should be of
within fitness training, we focus on four groups (i.e., legs,
medium or high intensity.
core, arms and back) in this preliminary study. Further-
more, some of these groups can be repeated during the In the remainder of this paper, these constraints were
training. Each muscle group is associated with three pre- captured within the algorithmic structure of the original
scribed exercises as presented in Table 1. bat algorithm for planning the fitness training plan.
Muscle groups Exercise
LEGS LEG PRESS, SQUATS, LUNGE Modifying the original bat algorithm Each solution in
CORE LEG SCISSORS, PLANK, LEG LIFTS the modified bat (MBA) algorithm consists of 24 floating-
ARMS PULLDOWN, PUSH UPS, point elements representing the fitness training plans for
UNDERARM ISOMETRIC EXERCISE some athlete. The elements of the solution are divided into
LEGS LEG PRESS, SQUATS, LUNGE three groups of elements. In other words, the solution is
BACK BACK EXTENSION, DEADLIFT, expressed as
BAR ROWS
CORE LEG SCISSORS, PLANK, LEG LIFTS xi = (xi1 , . . . , xi8 , xi9 , . . . , xi16 , xi17 , . . . , xi24 )T , (3)
LEGS LEG PRESS, SQUATS, LUNGE
ARMS PULLDOWN, PUSH UPS, where elements xi1 , . . . , xi8 denote exercises from Table 1,
UNDERARM ISOMETRIC EXERCISE xi9 , . . . , xi16 are the number of repeats NR and xi17 , . . . , xi24
the corresponding intensity, respectively. This means,
Table 1: Muscle groups and associated exercises each fitness training plan consists of eight exercises with
an assigned number of repeats and corresponding intensi-
On the other hand, the intensities of the exercises must ties. While the number of repeats is selected from inter-
be determined in a fitness training plan. This intensity is val NR ∈ [1, 40], parameters exercises and intensities are
associated with a measure 1RM measured for a specific drawn from the interval [0, 1], and their proper values are
athlete. Here, three levels of intensity are supported in our encoded as indices into a discrete set of features according
study, where each level is mapped according to the 1RM, to the following equations
as can be seen in Table 2.
ex(xi, j ) = d3.0 · xi, j e, for j = 1, . . . , 8, (4)
Intensity 1RM measure
int(xi, j ) = d3.0 · xi, j e, for j = 17, . . . , 24, (5)
HIGH 1RM > 80%
MEDIUM 60% < 1RM ≤ 80% where ex(xi, j ) and int(xi, j ) determine the element in the
LOW 1RM ≤ 60% feature sets as represented in Tables 1 and 2. For instance,
the function intensity can obtain the following values from
the feature set
Table 2: Intensity mapping
HIGH, if 0 ≤ xi, j < 31 ,
Note that the data in Tables 1 and 2 were specified ac- int(xi, j ) = MEDIUM, if 31 ≤ xi, j < 23 ,
cording to the suggestions of fitness trainers. LOW, if 23 ≤ xi, j < 1,
Planning Fitness Training Sessions Using the Bat Algorithm 125
respectively. The planning of the fitness training sessions Exercise Repeats Intensity
is defined as a constraint satisfaction problem that is for- SQUATS 34 LOW
mally defined as LEG SCISSORS 22 MEDIUM
UNDERARM ISOMETRIC 40 HIGH
k<4
Minimize f (xi ) = ∑ χk (xi ), LEG PRESS 15 LOW
k=1 BAR ROWS 36 HIGH
15 LEG LIFTS 27 LOW
subject to ∑ y j ≥ 4, LUNGE 15 MEDIUM
j=8 PULLDOWN 22 LOW
24
∑ z j ≥ 2, Table 4: Second set of fitness workouts
j=16
xi,1 6= xi,4 6= xi,7 ∧ xi,2 6= xi,6 ∧ xi,3 6= xi,8 , Exercise Repeats Intensity
int(xi, j ) ≡ HIGH ⇒ int(xi, j+1 ) 6= HIGH, LEG PRESS 21 HIGH
LEG LIFTS 29 MEDIUM
where UNDERARM ISOMETRIC 28 MEDIUM
+1 if xi, j ≥ 25,
yj = LUNGE 39 MEDIUM
+0 otherwise,
BACK EXTENSION 15 LOW
and PLANK 35 MEDIUM
+1 if int(xi, j ) ≡ HIGH, SQUATS 25 HIGH
zj =
+0 otherwise,
PUSH UPS 38 MEDIUM
The proper solution to the problem is found, when the
f (x) = 0. Table 5: Third set of fitness workouts
Representation of Results. Although the results could be the other hand, we would also like to present some prob-
visualized, the numerical results in the tables are presented lems and bottlenecks which we encountered during devel-
only in this preliminary version of the modified bat algo- opment. Firstly, it seems that it will be good to test our idea
rithm. with evolutionary algorithms in the future. Experiments
showed that the success of the bat algorithm in satisfying
all constraints was about 25% of runs only. The problem
4 Experiments and Results is that the bat algorithm is highly dependent on the best
solution. From this reason, our algorithm went into local
The results of our experiments are illustrated in Tables 3 optimum a lot of times. We believe that advanced mech-
to Table 5, where the tables represent the three sets of ex- anisms e.g. arithmetic crossover would behave much bet-
ercises. An athlete has some free time for resting after ter. Moreover, using adaptive and self-adaptive bat vari-
finishing each set. We run algorithm 25 times and after the ants could also be suggested since we spent a lot of time
run we selected three generated training sessions which tuning parameters. On the other hand, many more con-
were successfully found by bat algorithm. straints should be defined in order to have very precise
solutions which should be very similar to those solutions
Exercise Reps[NR] Intensity[1RM]
created by the human sport trainer.
LUNGE 26 HIGH
LEG SCISSORS 23 LOW
PULLDOWN 18 MEDIUM
SQUATS 27 HIGH 5 Conclusions
BAR ROWS 40 MEDIUM
LEG LIFTS 22 HIGH
LEG PRESS 35 MEDIUM In this workshop paper, we presented a simple, yet effi-
PUSH UPS 39 LOW cient solution for planning fitness training sessions auto-
matically. The bat algorithm was employed in order to
Table 3: First set of fitness workouts tackle this problem. This algorithm successfully generated
training sessions which were evaluated and confirmed by a
The obtained results were evaluated by human trainer human trainer who had more than 20 years of experience.
who evaluated and approved it. The obtained results con- In the future, there are many tasks to do in this direction
firm that the idea of automatic fitness training sessions was like for example testing with other nature-inspired algo-
worth investigation and the promising results also show rithms, employing arithmetic crossover and taking more
the potentials of the solution when used in practice. On constraints and exercices into account.
126 I. Fister Jr., S. Rauter, K. L. Fister, D. Fister, I. Fister
Acknowledgement [17] Aagaard, P., Simonsen, E. B., Andersen, J. L., Magnus-
son, P., Dyhre-Poulsen, P.: Increased rate of force develop-
The research reported in this paper has been partially sup- ment and neural drive of human skeletal muscle following
ported by the Czech Science Foundation grant 13-17187S. resistance training. Journal of Applied Physiology 93(4)
(2002), 1318–1326
[18] Rønnestad, B. R., Hansen, E. A., Raastad, T.: Strength
training improves 5-min all-out performance following
185 min of cycling. Scandinavian Journal of Medicine &
References Science in Sports 21(2) (2011), 250–259
[19] Hausswirth, C., Bigard, A. X., Berthelot, M.,
[1] Petschnig, S.: 10 Jahre Ironman Triathlon Austria. Meyer Thomaidis, M., Guezennec, C. Y.: Variability in energy
& Meyer, 2007 cost of running at the end of a triathlon and a marathon.
[2] Knechtle, B., Wirth, A., Baumann, B., Knechtle, P., Rose- International Journal of Sports Medicine 17(8) (1996),
mann, T.: Personal best time, percent body fat, and train- 572–579
ing are differently associated with race time for male and [20] Psilander, N., Frank, P., Flockhart, M., Sahlin, K.: Adding
female ironman triathletes. Research quarterly for exercise strength to endurance training does not enhance aerobic ca-
and sport 81(1) (2010) 62–68 pacity in cyclists. Scandinavian Journal of Medicine & Sci-
[3] McCarville, R.: From a fall in the mall to a run in the sun: ence in Sports, 2015
One journey to ironman triathlon. Leisure Sciences 29(2) [21] Fister Jr, I., Yang, X.-S., Fister, I., Brest, J., Fister, D.:
(2014), 159–173 A brief review of nature-inspired algorithms for optimiza-
[4] Lehto, N.: Effects of age onmarathon finishing timeamong tion. Elektrotehniški vestnik 80(3) (2013), 116–122
male amateur runners in stockholm marathon 1979–2014. [22] Fister Jr., I., Fister, I., Yang, X.-S., Fong, S., Zhuang, Y.:
Journal of Sport and Health Science, 2015 Bat algorithm: recent advances. In: Computational Intel-
[5] Rauter, S.: Mass sport events as a way of life: differencess ligence and Informatics (CINTI), 2014 IEEE 15th Interna-
between participants in a cycling and running event. Kine- tional Symposium on, 163–167, IEEE, 2014
siologica Slovenica 20(1) (2014), 5–15 [23] Fister Jr., I.: A comprehensive review of bat algorithms and
[6] Shipway, R., Jones, I.: The great suburban everest: An “in- their hybridization. Master’s Thesis, University of Maribor,
siders” perspective on experiences at the 2007 Flora Lon- Slovenia, 2013
don marathon. Journal of Sport and Tourism 13(1) (2008), [24] Hassan, E. A., Hafez, A. I., Hassanien, A. E., Fahmy, A. A.:
61–77 A discrete bat algorithm for the community detection prob-
[7] Rauter, S., Topic, M. D.: Differences in travel behav- lem. In: Hybrid Artificial Intelligent Systems, 188–199,
iors of small and large cycling events participants. AP- Springer, 2015
STRACT: Applied Studies in Agribusiness and Commerce [25] Soto, R., Crawford, B., Olivares, R., Johnson, F., Pare-
7(1), (2013) des, F.: Online control of enumeration strategies via bat-
[8] Rauter, S.: Socialni profil športnih turistov - udeležencev inspired optimization. In: Bioinspired Computation in Ar-
množičnih športnih prireditev v Sloveniji. PhD thesis, Uni- tificial Systems, 1–10, Springer, 2015
versity of Ljubljana, Slovenia, 2012 [26] Cai, X., Wang, L., Kang, Q., Wu, Q.: Adaptive bat al-
[9] Budgett, R.: Fatigue and underperformance in athletes: the gorithm for coverage of wireless sensor network. Inter-
overtraining syndrome. British Journal of Sports Medicine national Journal of Wireless and Mobile Computing 8(3)
32(2) (1998), 107–110 (2015), 271–276
[10] Cardoos, N.: Overtraining syndrome. Current sports [27] Zhao, D., He, Y.: Chaotic binary bat algorithm for analog
medicine reports 14(3) (2015), 157–158 test point selection. Analog Integrated Circuits and Signal
Processing, 1–14, 2015
[11] Lehmann, M., Foster, C., Keul, J.: Overtraining in en-
durance athletes: a brief review. Medicine & Science in [28] Premkumar, K., Manikandan, B. V.: Speed control of
Sports & Exercise 25 (1993), 854–862 brushless dc motor using bat algorithm optimized adaptive
neuro-fuzzy inference system. Applied Soft Computing 32
[12] Fister Jr., I., Ljubič, K., Suganthan, P. N., Perc, M., Fis-
(2015), 403–419
ter, I.: Computational intelligence in sports: challenges
and opportunities within a new research domain. Applied
Mathematics and Computation 262 (2015), 178–186
[13] Engelbrecht, A. P.: Computational intelligence: an intro-
duction. John Wiley & Sons, 2007
[14] Fister, I., Rauter, S., Yang, X -S., Ljubič, K., Fister Jr., I.:
Planning the sports training sessions with the bat algorithm.
Neurocomputing 149 (2015), 993–1002
[15] Yang, X. -S.: A new metaheuristic bat-inspired algorithm.
In: Nature Inspired Cooperative Strategies for Optimiza-
tion (NICSO 2010), 65–74, Springer, 2010
[16] Fister Jr, I., Fister, D., Yang, X. -S.: A hybrid bat algorithm.
Elektrotehniški vestnik 80(1-2) (2013), 1–7