http://dx.doi.org/10.5755/j01.eie.23.6.19688 ELEKTRONIKA IR ELEKTROTECHNIKA, ISSN 1392-1215, VOL. 23, NO.

6, 2017

ABC Algorithm Based PID Controller Design

for Higher Order Oscillatory Systems
Aytekin Bagis, Halit Senberber
Department of Electrical & Electronics Engineering, Faculty of Engineering, Erciyes University,
38039, Kayseri, Turkey
[email protected]

1Abstract—This paper presents the results of an PID parameters, GA and DEA approaches are
investigation based on the use of artificial bee colony (ABC) comparatively used for three different systems namely high
algorithm to determine the optimum PID parameters for order, with time delay and the non-minimum phase. The
higher order oscillatory systems. For this aim, several
study presented in [20] considers the values of mean squared
oscillatory systems with different orders from the literature
were selected. In order to achieve the desired closed loop error (MSE) and the integral of absolute error (IAE) for
system response, the PID parameters were tuned by the ABC performance evaluation of these algorithms.
algorithm using an appropriate objective function. To In the study of Bagis [18], a modified GA based approach
demonstrate the effectiveness of the considered approach, is used for optimal PID parameters. Karaboga and Kalinli
performance of the ABC algorithm was compared with that of [23] employed the TSA to find the PID parameters for a
the conventional PID tuning methods given in the literature
given process. In a study by Bagis and Savascihabes [24],
and two popular heuristic optimization algorithms namely
particle swarm optimization (PSO) algorithm and genetic the PID tuning operation is performed by the DEA for the
algorithm (GA). desired closed loop system response. A PSO based
application for robust PID controller design is given by
Index Terms—Artificial bee colony algorithm; PID Zhao et al. [28]. Chang and Shih [27] presented a PID
controller design; PID tuning, population based metaheuristics. controller design method for nonlinear systems by using an
improved PSO approach based on modified velocity and
I. INTRODUCTION position updating equations. In another study on the PSO
In the design of the control systems, Proportional- algorithm, useful information on algorithm performance was
Integral-Derivative (PID) type controllers have an exclusive obtained in the modelling problem of high order oscillatory
and important position. To achieve a desired system systems [30]. In this conference paper presented by Bagis, in
response, the controller parameters should be carefully order to define the five different high order processes, the
determined by an appropriate tuning approach. For higher parameters of three different model structures are
order and/or oscillatory systems with complex dynamics, determined by a typical PSO algorithm. Karaboga and Akay
this operation can be quite difficult and time-consuming. reported a comparison of the ABC algorithm, harmony
The performance of the designed control system is directly search, and the bees’ algorithms for the PID design problem
affected by various main factors such as the definition of an [29].
acceptable process model, consideration of the time or The main aim of this study is to investigate the
frequency domain for the controller design, selection of a performance of the ABC algorithm to solve a difficult
suitable objective function for control purposes, or controller design problem based on the use of the PID
determination of a controller tuning method. controller. Another key contribution of this paper is that it
There are many classical approaches and methods in the examines the main idea of how the ABC algorithm can be
literature which use the time or frequency domain response used as an efficient alternative and/or supportive approach
of the systems [1]–[17]. On the other hand, by using PID to improve the performance of the system under control. At
tuning approaches based on artificial intelligence this, several oscillatory systems with different orders in the
optimization algorithms such as genetic algorithms (GAs) literature were selected. The performance of the ABC
[18]–[21], the tabu search algorithm (TSA) [22], [23], algorithm was compared with those of the conventional PID
differential evolution algorithm (DEA) [20], [24], ant colony tuning methods given in the literature and two popular
optimization algorithm [25], particle swarm optimization heuristic optimization algorithms, namely the PSO and GA.
(PSO) [26]-[28], and the artificial bee colony (ABC) The definition of the PID control problem for the ABC
algorithm [29], fast and efficient control solutions can be algorithm based controller design is given in Section II. In
obtained. Wang and Kwok presented a GA based PID Section III, the ABC, GA and PSO algorithms and the
tuning study for a pH neutralization process [21]. In the control parameters used in this study are briefly presented.
study reported by Saad et al. [20], to achieve the optimal The comparison of simulation results is given in Section IV.
Discussion and conclusion are given in the last sections.
Manuscript received 2 March, 2017; accepted 15 September, 2017. II. PID DESIGN PROBLEM
This work was supported by Research Fund of the Erciyes University.
Project Number: FBA-2015-6236. The basic principle in PID tuning is to adjust the
3
ELEKTRONIKA IR ELEKTROTECHNIKA, ISSN 1392-1215, VOL. 23, NO. 6, 2017

controller parameters fast and accurately as possible for ABC algorithm having an efficient objective function for
aims of the control design. These parameters are entitled by controlling the higher order systems. The similar structures
proportional tuning constant, Kp, integral tuning constant, to this function type were used in our previous studies, and
Ki, and derivative tuning constant, Kd. For a typical PID impressive results were obtained [18], [19], [22], [24]. Thus,
controller, the controller output signal u(t) can be written as it is preferred to use such a weighted objective function that
the following in time and s domain [18]: is formed different performance specifications. In this study,
because minimization of the ITAE value is primarily aimed
t de(t ) in the step response of the closed loop system under control
u  t   Kp  e  t   Ki  0 e(t )dt  Kd , (1)
dt action, the coefficient of this performance index was
 Ki  selected as 10.
U  s    Kp   Kd  s   E  s  , (2)
 s  III. ALGORITHMS USED IN STUDY
where e(t) is the error signal, and calculated by difference A. ABC Algorithm
between the reference input, r(t), and system output, y(t); s One of the quite successful optimization algorithms is the
is a complex variable which has a real part and an imaginary artificial bee colony (ABC) algorithm presented by
part. The variables of s and (1/s) in (2) can be considered as Karaboga [29], [31]–[33]. This population based approach
differential and integral operators in time domain, presents efficient solutions for numerical optimization
respectively. It is a well-known fact that parameter tuning problems especially. In main procedure of the algorithm,
approach directly affects behaviour of the PID controller foraging behaviours of bee colonies are simulated by three
and satisfactoriness of the control system designed for the groups of the bees. These are employed, onlooker, and scout
control aims. Therefore, in order to meet the desired closed bees. Locations of food sources are evaluated as possible
loop system specifications in time and/or frequency domain, solutions for the problem. The possible and new solutions
PID controller parameters must be carefully determined. For are determined by the employed bees. Nectar amounts
this purpose, one of the efficient approaches is to use the (fitness values) in the food sources define the qualities of the
specifications of the system’s step response. Thus, a solutions. The new possible food sources are selected by the
controller design strategy that provides the aims of the onlooker bees according to information of nectar amounts.
control system can be proposed by analysing the step Selection procedure of the scout bees is controlled by an
response of the controlled system. Definition of an important control parameter called “limit” which is the
appropriate objective function based on the use of important value of predetermined number of cycles.
performance indicators such as the overshoot, settling time, To improve the quality of the solution, generally, a
and the integral time absolute error (ITAE) value can mathematical description is used by the algorithm as the
provide the remarkable solutions for the controller tuning. following [32], [33]
In this study, an ABC algorithm based design mechanism
for PID control of higher order oscillatory systems is Vij  X ij  rand (-1,1)( X ij - X ). (5)
proposed as shown in Fig. 1. kj

This description denotes the location of the new possible

sources with more quality according to a nectar source in the
position i. In here, k (k{1,2,…..,pop_size}) a randomly
chosen index that has to be different from i and j are
randomly selected indice. In here, Vij, Xij and Xkj are new,
current and neighbor solutions, respectively.
There are also two important control parameters that
directly impress the performance of the algorithm: colony
size and the number of the maximum cycles. In this paper,
Fig. 1. PID tuning by using ABC algorithm. for the limit value, description of [(colony size÷2) × number
of the parameters] is used.
The main purpose of the algorithm is to tune the
parameters of the PID controller by using an objective B. PSO Algorithm
function (OF) given in (3). This equation comprise of a PSO presented by Kennedy and Eberhart [34], [35] is a
combination of the ITAE, percent maximum overshoot population based stochastic optimization algorithm [26]–
(OS), and 5 percent settling time (ts) [28], [30]. In the algorithm, there is possible solution
elements called as particles in the population. During the
OF  10  ITAE  5  ts  3  OS .
  (3) search procedure, solution quality is improved by altering
the positions and velocities of the particles. Best values for
In this function, ITAE performance index is described as
the particles and swarm are kept by the algorithm for use
T
when needed. PSO approach has a simple and flexible
 ITAE  0 t  e(t ) dt. (4) structure.
New position (x) and velocity (v) values for the ith particle
As mentioned earlier, the main contribution and are generally obtained by using the descriptions in (6)–(7)
motivation of this work is in exploring the results of the [26]–[28], [30]. In these equations, k is the currently

4
ELEKTRONIKA IR ELEKTROTECHNIKA, ISSN 1392-1215, VOL. 23, NO. 6, 2017

iteration number, and w is the inertia weight factor. The best the minimized objective functions were obtained.
position values for the particles and swarm are represented
by P and G symbols. Cognitive and social parameters are TABLE I. CONTROL PARAMETER VALUES.
Control Parameter PSO GA ABC
specified with the constants of c1 and c2, respectively.
Population (or colony) size 30
Random variables of r1 and r2 are in the interval of [0-1].
Cognitive and social parameters c1 = c2 = 2 - -
The variables of v and x are (population size  optimized
Limit - - 45
parameters number)-dimensional vectors. In this study, the 0.9 /
size of these vectors is (30  3): Crossover and Mutation Rate - -
0.3
Number of Generation (maxit) 100
vi  k  1  wi  vi  c1  r1  w = (maxit-
Inertia weight factor - -
iteration)/maxit
  Pi - xi  k    c2  r2   G - xi  k   , (6) Parameter Search
[0-100]
Interval for PID
xi  k  1  xi  k   vi  k  1 . (7)

The PID tuning procedure based on the ABC, PSO and

C. Genetic Algorithm (GA)
GA was repeated as 30 times for 100 iterations in each
Population based GAs are one of the general-purpose system. To present the efficiency of the ABC approach,
search algorithms [18]–[21], [31]. The population consists search intervals for the PID parameters were preferred in the
of chromosomes which represent the possible solutions for intervals of [0–100]. The simulation times for the G1 and
the problem. To obtain fitness values or qualities of the G3 processes were used in the interval of [0–50] s. On the
chromosomes, an appropriate objective or cost function other hand, this period for process G2 was taken as [0–
called as fitness function is used. The fitness function is an 100] s. In the simulation, the Matlab programming package
important and critical performance index to describe the [39], [40] and a Mobile Intel Core i7, 2400 MHz computer
qualities of each individual in the population. The qualities were used.
and diversities of the solutions in the population are The optimum PID parameters obtained by the algorithms
improved by using some critical procedures such as for the considered processes are given in Table II. In this
reproduction, crossover and mutation. The number of table, the step response specifications of the PID controlled
transfer to the next generation of the most fits chromosomes systems are also given. As shown in Table II, these results
in the populations is controlled by reproduction operation. are compared with different methods in the literature [36]–
In solving a specific optimization problem, some critical [38]. For this purpose, the time response values of the
decisions such as genetic representation of the solutions, systems under PID control, based on the use of PID
selection mechanism, description of the fitness function, and parameters in the related references, are calculated by using
determination of the control parameters are performed by the Matlab program. The time responses of the systems
the GA. In this study, the real coded GA with roulette wheel controlled by different methods are also plotted in Fig. 2.
selection operator is used the single point crossover It is clearly shown in Table II that the values of the PID
operation. The fitness values for the chromosomes are controllers optimized by the algorithms are very close to
obtained by using the definition of “100/(1+objective each other. Thus, the step responses of the algorithm based
function value)”. systems are inherently very similar. Therefore, the drawing
of the ABC algorithm based system response alone is shown
IV. EXPERIMENTAL RESULTS in Fig. 2. From Table II, it can be clearly seen that the
For the PID tuning problem, various higher order optimum PID controller parameters obtained by the
oscillatory systems from the literature were considered as algorithms are almost identical to each other for all the
the following [36]–[38]: processes. From Table II again, the performance of the
ABC, PSO and GAs is quite impressive. The desired values
1 of the step response specifications are strikingly achieved by
 G1  s   e-0.1s , (8)
 2

s  s  1  s  2
2 the ABC, PSO and GA based PID controllers. In particular,
this case can be distinguished from the ITAE and settling
1 time values obtained by the algorithm based systems.
G2  s  
  , (9)
 s  1 6
 s2  2 In the design of a PID control system, some important
preferences such as fast control with a large overshoot or
1 slow control with no overshoot can be made inevitably.
G3  s  
  e-0.3s . (10)
  Fortunately, using the presented ABC, PSO and GA
2 3
s  2s  3  s  3 approaches, settling time and overshoot can be concurrently
decreased as compared to other methods. It is possible that
The G1, G2, and G3 systems given in (8)–(10) are the significant improvements on the controlled system
investigated in [37], [38], [37], and [36], [38], respectively. response can be achieved by performing the required
In this study, the control parameters for the algorithms are modifications in the objective function.
selected and are given in Table I. In order to obtain the most It can be seen from Table II and Fig. II that the ABC,
efficient parameters, many trials were made by using the PSO and GA based systems show a considerable
trial-and-error method and our previous experiences of improvement in the values of ITAE, settling time, and
heuristic algorithms. Thus, the values in Table I that provide maximum overshoots for all the processes.

5
ELEKTRONIKA IR ELEKTROTECHNIKA, ISSN 1392-1215, VOL. 23, NO. 6, 2017

TABLE II. PID CONTROLLER PARAMETERS [36]–[38] AND STEP studies as compared with other approaches. Similarly, all the
RESPONSE SPECIFICATIONS.
processes exhibit the best behavior in terms of ITAE. As
PID seen in Table II, in the PID control of G2, the best
Step response specifications
Parameters
(minimum) overshoot values are obtained. When the ABC
Pr Method
Kp, algorithm is used, this value is 5.0021 % for G2.
Ki, ITAE ts OS % OF
Kd
1.3039
Malwatkar
1.3104 6.1999 6.2875 0.3143 94.3796
et al. [37]
1.3351
1.5030
Wang et al.
1.3660 6.1833 6.6230 0.6808 96.9898
[38]
1.7150
2.1470
Ho et al.
1.4840 7.2006 9.4044 3.5306 129.620
[38]
0.7770
4.2331
G1 ZN 1.6277 8.0067 11.6359 2.4391 145.564
2.7522
2.7106
PSO 2.2147 2.6440 3.1886 1.8618 47.9686
2.8134
(a)
2.7131
GA 2.2389 2.6312 3.1698 1.9619 48.0472
2.8597
2.7121
ABC 2.2136 2.6452 3.1883 1.8582 47.9683
2.8118
0.78
Malwatkar
0.3708 69.269 20.782 16.257 845.37
et al. [37]
0.8964
0.5996
Wang et al.
0.6299 216.52 53.313 48.14 2576.16
[37]
2.281
1.959
Ho et al.
0.5727 48.045 23.778 30.106 689.66
[37]
1.676
2.3702
G2 ZN 0.4356 29.329 21.716 14.466 445.27 (b)
3.2245
1.9742
PSO 0.4919 23.575 13.058 4.9997 316.04
3.033
1.9814
GA 0.4943 23.554 13.060 5.3497 316.89
3.0328
1.9738
ABC 0.4923 23.576 13.055 5.0021 316.04
3.0306
Kappa- 4.5 1
Tau 0.9756 38.0459 16.6627 0 463.7730
[36] 0.1485
4.93
Karimi et
15.6013 15.6880 8.6716 0.6982 202.3331
al. [36]
0.6162 (c)
17.562 Fig. 2. Step responses of different methods for process: (a) G1(s),
Wang et al.
22.485 10.0070 9.1222 5.1766 161.2102 (b) G2(s), (c) G3(s).
[38]
14.13
43.1985 In addition to other step response specifications, these
G3 ZN 14.37 35.1072 22.8617 0.6884 467.446 striking reductions in the overshoots are clear indicators of
32.465
the algorithm’s power. On the other hand, although the
29.177
PSO 25.729 6.0179 4.6798 1.3326 87.576
maximum overshoots of the ABC method are not the best
20.632 values for all the processes, these are quite satisfactory
29.2213 values when considering the long settling times of the other
GA 25.745 6.0178 4.6758 1.3585 87.633 methods.
20.7187
In this study, the time domain specifications of the PID
29.148
ABC 25.706 6.0205 4.6845 1.3162 87.576 controlled systems are considered for PID tuning based
20.573 control system design. On the other hand, in frequency
response based approaches, some important performance
In particular, again for all processes, the best values for indicators such as sensitivity, gain and phase margin (GM,
settling time are obtained by ABC, PSO and GA based PM) values are often used [11]–[13], [16], [28], [40], [41].
6
ELEKTRONIKA IR ELEKTROTECHNIKA, ISSN 1392-1215, VOL. 23, NO. 6, 2017

In terms of the stability of the systems, two main sensitivity show that the computation time of the ABC algorithm is
values are important. These are the maximum and maximum dramatically longer than those for the PSO and genetic
complementary sensitivities (Ms and Mp). Although these algorithms. Admittedly this situation is a disadvantage of the
indicators do not exist in the objective function, the values ABC algorithm.
of these robustness characteristics for PID controlled Furthermore, in order to demonstrate the convergence
systems are also presented in Table III. speed of the ABC, PSO and genetic algorithms, the
iteration-objection function (OF) variation curves of the PID
TABLE III. ROBUSTNESS CHARACTERISTICS OF PID controlled systems obtained by the algorithms for each
CONTROLLED PROCESSES.
process are given in Fig. 3.
Characteristic of Robustness
Pr Method
Ms Mp GM PM (deg)
Malwatkar et al. [37] 1.2843 0.9999 8.9310 70.0145
Wang et al. [38] 1.2534 0.9999 7.8182 71.9707
Ho et al. [38] 1.7883 0.9998 2.7680 70.6715
G1
PSO 1.5288 1.0000 4.3452 66.1671
GA 1.5265 1.0000 4.3428 65.8825
ABC 1.5294 1.0000 4.3426 66.1943
Malwatkar et al. [37] 1.6332 1.1247 4.1004 52.7666
Wang et al. [37] 2.7116 2.2370 4.8544 27.132
Ho et al. [37] 2.2397 1.4879 1.9991 45.358
G2
PSO 1.9358 1.0010 2.1266 64.3008
GA 1.9417 1.0076 2.1200 64.1526 (a)
ABC 1.9363 1.0014 2.126 64.328
Kappa-Tau [36] 1.3449 1.0000 4.2933 72.5729
Karimi et al. [36] 1.5763 0.9999 3.0140 63.9997
Wang et al. [38] 1.5989 1.0020 3.1451 60.6215
G3
PSO 1.7226 1.0000 2.4134 62.4362
GA 1.7236 1.0000 2.4104 62.4415
ABC 1.7228 0.9999 2.413 62.446

In a robust system, as evidence of the system’s relative

stability, the values of the Ms and Mp parameters must be as
low as possible [41]. As known, a large Mp in the frequency
domain corresponds to a large overshoot in the step
response. On the other hand, for a successful disturbance
(b)
rejection behavior, the value of Ms should be low. In
addition to this, in general, large gain and phase margins are
desired system specifications.
In spite of the fact that the robustness specifications such
as the minimum Ms and/or Mp values, and predetermined
margin values are not in the features of the purposed system
behaviors, we can say from Table III that the ABC, PSO and
GA based PID controlled systems provide good robustness
characteristics in general. For all of the processes, the
measurements of the Ms, Mp, and phase margin can be
considered as satisfactory values. However, admittedly,
there clearly needs to be an improvement with regard to gain
margin properties.
(c)
One of the main performance characteristics of the
Fig. 3. Iteration-objection function (OF) variation curves of the systems
algorithm based methods is the computational time for the obtained by the algorithms for each process: (a) G1(s), (b) G2(s), (c) G3(s).
iterations. In this study, the search procedure for each of the
processes was repeated 30 times for an iteration number of Although the algorithm has a large search area given as
100 by the ABC, PSO and genetic algorithms. When the [0–100], it can be seen from Fig. 3 that the ABC, PSO and
ABC algorithm is used, the average calculation times for GA methods are converged rapidly into the desired values of
100 iterations of the G1, G2, and G3 processes are noted as the objective function, requiring only 20–50 iterations. For
500.54 sec, 416.01 sec, and 474.27 sec, respectively. the G2 and G3 processes, the convergence rate of the ABC
Similarly, these values for the same processes in the PSO algorithm is better than that of the PSO algorithm. However,
based studies are obtained as 56.168 sec, 46.751 sec, and the PSO algorithm in the G1 process needs fewer iterations
73.169 sec, respectively. Finally, the average calculation for the best solution as compared with the ABC. On the
times for GA based studies are found as 39.52 sec, other hand, the worst convergence performance for the G1
21.69 sec, and 54.27 sec, respectively. These values clearly process is exhibited by the GA. In the G2 and G3 processes,
7
ELEKTRONIKA IR ELEKTROTECHNIKA, ISSN 1392-1215, VOL. 23, NO. 6, 2017

the performance ranking in terms of convergence in a low 4. In this study, to define the objective function, three
number of iterations is observed as ABC, GA, and PSO. main performance indicators in the time domain are
Nevertheless, from the simulation results, it can be observed preferred. There is no any robustness specification in the
that, especially in terms of the eventual best outcome objective function. Thus, robustness characteristics
obtained from the PID controlled systems, the ABC, PSO provided perfectly by the controlled systems are not
and GAs have similar ability and comparable convergence expected. However, satisfactory or acceptable results in
rates. terms of the robustness requirements are desired. From
the simulation results, we can say that the robustness
V. DISCUSSION characteristics of the controlled systems are reasonable
In this study, the performance of ABC, PSO and GA but in need of improvement at the same time. More
approaches in the PID based control of the higher order detailed investigations of the robustness considerations
oscillatory systems is investigated. For this research, a are left for subsequent studies.
simple but effective objective function including the basic It is also an undeniable fact that it is important to develop
step response properties was used. The results of the the design considering the different dynamic characteristics
simulations clearly show that all of these algorithms are in the design of a control system and to observe the
extremely successful in identifying the PID parameters to achievement of the results obtained in a more realistic
achieve the desired control objectives. The GA that provides control setup [9], [11], [12], [14], [15], [17]. This work is
the shortest calculation times requires more control one of the first steps of a long-running project. Accordingly,
parameters and procedures to identify the problem. In the main objectives are to investigate the benefits of
addition, more iterations than the other algorithms are different algorithms in the design of control systems, and to
needed by the GA to achieve the minimum objective orient themselves to practical work out of the feasible
function value. In the PSO approach, which requires more outcomes. Hence, more realistic practices will be taken into
parameters to be stored and updated when compared to the account in future studies.
GA, good solutions are achieved with slightly longer
calculation times. The calculation times in PSO are about VI. CONCLUSIONS
1.5 times higher than those in GA. In the PSO algorithm, as The main contribution of this study is to incorporate the
in GA, optimum PID parameters are obtained in high PID control of the higher order oscillatory systems with the
iterations. Unlike other algorithms, the appropriate PID ABC algorithm. In this paper, the results of a PID controller
parameters in the ABC approach can be provided in low design approach based on the use of ABC algorithm are
iterations. Moreover, this algorithm has a simple structure presented for higher order oscillatory systems. In order to
consisting of few control parameters. This provides ease of achieve this aim, the simulation results of different high
application to the user. On the other hand, the most striking order systems existing in the literature are comparatively
disadvantage of this algorithm is the long computation time given.
which is 7–10 times higher than other algorithms. According to simulation results, it can notify that the
According to the simulation results, a general ABC based PID tuning method can be easily and
summarization can be drawn as follows: successfully applied to the higher order oscillatory systems.
1. The main benefit of using the ABC algorithm based Furthermore, the interesting results of this method prove that
method is that the controller tuning can be easily done to they exhibit the competitive solutions to effectively achieve
perform the desired control objectives. Fast and easy the presented control problem. Moreover, to obtain the high
reconfiguration of the algorithm structure or search performance specifications from the PID controlled system,
procedure can be achieved by the users. objective function used by the algorithm can be easily
2. The simulations show that, as compared with the PSO modified for control aims. By considering the time or
and GA, the ABC algorithm cannot rapidly reach optimal frequency domain requirements of the system under control,
controller performance for kinds of higher oscillatory performance parameters and/or robustness measurements
systems considered in this study. Also, to improve the can be inserted into the objective function. Thus, the
quality of the solutions, the ABC algorithm needs long computations needed to meet the goals of control design can
computation times in the iterations. However, the be done without repeated use of the complex or time
algorithm can provide the high improvements on the consuming mathematical operations, which is a significant
solutions for low iteration numbers. Furthermore, the advantage.
ABC algorithm always provides the competitive
outcomes with high quality.
REFERENCES
On the other hand, an important advantage of the ABC
[1] J. G. Ziegler, N. B. Nichols, ‘‘Optimum settings for automatic
algorithm is that it also has fewer control parameters than controllers’’, Transactions of ASME, vol. 64, pp. 759–768, 1942.
the PSO and genetic algorithms. [Online]. Available: http://dx.doi.org/10.1115/1.2899060
3. It is clear that the performance of the controlled [2] G. H. Cohen, G .A. Coon, ‘‘Theoretical consideration of retarded
systems depends directly on the type of objective control’’, Transactions of ASME, vol. 75, pp. 827–834, 1953.
[3] D. W. Pessen, ‘‘A new look at PID-controller tuning’’, Trans. ASME.
function. For non-expert users, this property can provide a Journal of Dynamical Systems Measures and Control, vol. 116,
significant advantage. The advantage of the objective pp. 553–557, 1994. [Online]. Available: http://dx.doi.org/
function used in this study is that the PID parameters can 10.1115/1.2899252
[4] M. Zhuang, D. P. Atherton, ‘‘Automatic tuning of optimum PID
be adjusted effectively by modifying the adjustment controllers’’, IEE Proc. Control and Applications, vol. 140, pp. 216–
coefficients for desired step response specifications. 224, 1993. [Online]. Available: http://dx.doi.org/10.1049/ip-

8
ELEKTRONIKA IR ELEKTROTECHNIKA, ISSN 1392-1215, VOL. 23, NO. 6, 2017

d.1993.0030 [23] D. Karaboga, A. Kalinli, ‘‘Tuning PID controller parameters using

[5] M. Morari, E. Zafiriou, Robust process control. Englewood Cliffs NJ: tabu search algorithm’’, in Proc. IEEE Int. Conf. on Systems, Man
Prentice-Hall, 1989. and Cybernetics, Adiyaman, Turkey, pp. 134–136, 1996. [Online].
[6] D. E. Rivera, M. Morari, S. Skogestad, ‘‘Internal model control.4. Available: http://dx.doi.org /10.1109/ ICSMC.1996.569754
PID controller design’’, Industrial and Engineering Chemistry [24] A. Bagis, A. Savascihabes, ‘‘PID tuning by using differential
Process Design and Development, vol. 25, pp. 252–265, 1986. evolution algorithm for desired closed loop system response’’, Int.
[Online]. Available: http://dx.doi.org/10.1021/i200032a041 Symposium on Innovations in Intelligent Systems and Applications
[7] W. K. Ho, O. P. Gan, E. B. Tay, E. L. Ang, ‘‘Performance and gain (INISTA 2010), Kayseri, Turkey, pp. 170–174, 2010.
and phase margins of well-known PID tuning formulas’’, IEEE Trans. [25] D. Hai-bin, W. Dao-bo, Y. Xiu-fen, ‘‘Novel approach to nonlinear
Control Systems Technology, vol. 4, pp. 473–477, 1996. [Online]. PID parameter optimization using ant colony optimization
Available: http://dx.doi.org/10.1109/87.508897 algorithm’’, Journal of Bionic Engineering, vol. 3, pp. 073–078,
[8] W. K. Ho, C. C. Hang, L. S. Cao, ‘‘Tuning of PID controllers based 2006. [Online]. Available: http://dx.doi.org/10.1016/S1672-
on gain and phase margin specifications’’, Automatica, vol. 31, 6529(06)60010-3
pp. 497–502, 1995. [Online]. Available: http://dx.doi.org/10.1109/ [26] A. Alfi, H. Modares, “System identification and control using
ISIE.1992. 279722 adaptive particle swarm optimization”, Applied Mathematical
[9] M. Huba, “Performance measures, performance limits and optimal PI Modelling, vol. 35, pp. 1210–1221, 2011. [Online]. Available:
control for the IPDT plant”, Journal of Process Control, vol. 23, http://dx.doi.org/ 10.1016/j.apm.2010.08.008
no. 4, pp. 500–515, 2013. [Online]. Available: http://dx.doi.org/ [27] W. D. Chang, S. P. Shih, ‘‘PID controller design of nonlinear systems
10.1016/ j.jprocont.2013.01.002 using an improved particle swarm optimization approach’’,
[10] I. Mizumoto, D. Ikeda, T. Hirahata, Z. Iwai, “Control Engineering Communications in Nonlinear Science and Numerical Simulation,
Practice Design of discrete time adaptive PID control systems with vol. 15, pp. 3632–3639, 2010. [Online]. Available: http://dx.doi.org/
parallel feedforward compensator”, Control Engineering Practice, 10.1016/j.cnsns.2010.01.005
vol. 18, no. 2, pp. 168–176, 2010. [Online]. Available: [28] J. Zhao, T. Li, J. Qian, ‘‘Application of particle swarm optimization
http://dx.doi.org/10.1016/j.conengprac.2009.09.003 algorithm on robust PID controller tuning’’, Lecture Notes in
[11] J. C. Jeng, W. L. Tseng, M. S. Chiu, “A one-step tuning method for Computer Sciences, vol. 3612, pp. 948–957, 2005. [Online].
PID controllers with robustness specification using plant step- Available: http://dx.doi.org/10.1007/11539902_118
response data”, Chemical Engineering Research and Design, vol. 92, [29] D. Karaboga, B. Akay, ‘‘Proportional-integral-derivative controller
no. 3, pp. 545–558, 2014. [Online]. Available: http://dx.doi.org/ design by using artificial bee colony, harmony search, and the bees
10.1016/j.cherd. 2013.09.012 algorithms’’, in Proc. IMechE Part I: J. Systems and Control
[12] M. N. Anwar, S. Pan, “A frequency response model matching method Engineering, vol. 224, pp. 869–883, 2010. [Online]. Available:
for PID controller design for processes with dead-time”, ISA http://dx.doi.org/10.1243/09596518JSCE954
Transactions, vol. 55, pp. 175–187, 2015. [Online]. Available: [30] A. Bagis, ‘‘Investigation of the performance of particle swarm
http://dx.doi.org/10.1016/j.isatra.2014.08.020 optimization in the modeling of high order oscillatory systems’’,
[13] D. B. S. Kumar, R. P. Sree, “Tuning of IMC based PID controllers for Automatic Control National Conf., Gebze, Kocaeli, Turkey, pp. 168–
integrating systems with time delay”, ISA Transactions, vol. 63, 173, 2010. (in Turkish)
pp. 242–255, 2016. [Online]. Available: http://dx.doi.org/10.1016/ [31] A. Bagis, H. Senberber, ‘‘Modeling of higher order systems using
j.isatra.2016.03.020 artificial bee colony algorithm’’, An International Journal of
[14] D. Fister, I. Fister, R. Safaric, “Parameter tuning of PID controller Optimization and Control: Theories & Applications, vol. 6, no. 2,
with reactive nature-inspired algorithms”, Robotics and Autonomous pp. 129–139, 2016. [Online]. Available: http://dx.doi.org/10.11121
Systems, vol. 84, pp. 64–75, 2016. [Online]. Available: /ijocta.01.2016.00298
http://dx.doi.org/10. 1016/j.robot.2016.07.005 [32] D. Karaboga, B. Akay, ‘‘A powerful and efficient algorithm for
[15] J. C. Jeng, G. P. Ge, “Disturbance-rejection-based tuning of numerical function optimization: artificial bee colony (ABC)
proportional–integral–derivative controllers by exploiting closed-loop algorithm’’, Journal of Global Optimization, vol. 39, pp. 459–471,
plant data”, ISA Transactions, vol. 62, pp. 312–324, 2016. [Online]. 2007. [Online]. Available: http://dx.doi.org/10.1007/s10898-007-
Available: http://dx.doi.org/10.1016/j.isatra.2016.02.011 9149-x
[16] S. Srivastava, V. S. Pandit, “A PI/PID controller for time delay [33] D. Karaboga, B. Akay, ‘‘A comparative study of artificial bee colony
systems with desired closed loop time response and guaranteed gain algorithm’’, Applied Mathematics and Computation, vol. 214,
and phase margins”, Journal of Process Control, vol. 37, pp. 70–77, pp. 108–132, 2009. [Online]. Available: http://dx.doi.org/10.1016/
2016. [Online]. Available: http://dx.doi.org/10.1016/j.jprocont. j.amc.2009. 03.090
2015.11.001 [34] R. C. Eberhart, J. Kennedy, ‘‘A new optimizer using particle swarm
[17] Q. H. Seer, J. Nandong, “Stabilization and PID tuning algorithms for theory’’, in Proc. Sixth Int. Symposium on Micromachine and Human
second-order unstable processes with time-delays”, ISA Transactions, Science, Nagoya, Japan, 1995, pp. 39–43. [Online]. Available:
vol. 67, pp. 233–245, 2017. [Online]. Available: http://dx.doi.org/ https://doi.org/10.1109/MHS.1995.494215
10.1016/j.isatra.2017.01.017 [35] J. Kennedy, R. C. Eberhart, ‘‘Particle swarm optimization’’, in Proc.
[18] A. Bagis, ‘‘Determination of the PID controller parameters by IEEE Int. Joint Conf. on Neural Networks; Perth, West Australia,
modified genetic algorithm for improved performance’’, Journal of pp. 1942–1948, 1995. [Online]. Available: https://doi.org/10.1109/
Information Science and Engineering, vol. 23, pp. 1469–1480, 2007. ICNN.1995.488968
[19] A. Bagis, A. Savascihabes, ‘‘Determination of the PID controller [36] A. Karimi, D. Garcia, R. Longchamp, “PID controller tuning using
parameters by using binary and real coded genetic algorithm’’, Bode’s integrals”, IEEE Trans. Control Systems Technology, vol. 11,
International Symposium on Innovations in Intelligent Systems and pp. 812–821, 2003. [Online]. Available: http://dx.doi.org/
Applications (ASYU 2008), Isparta, Turkey, pp. 198–202, 2008. (in 10.1109/TCST.2003.815541
Turkish). [37] G. M. Malwatkar, S. H. Sonawane, L. M. Waghmare, ‘‘Tuning PID
[20] M. S. Saad, H. Jamaluddin, I. Z. M. Darus, ‘‘Implementation of PID controllers for higher-order oscillatory systems with improved
controller tuning using differential evolution and genetic algorithms’’, performance’’, ISA Trans., vol. 48, pp. 347–353, 2009. [Online].
Int. Journal of Innovative Computing, Information and Control, Available: http://dx.doi.org/10.1016/j.isatra.2009.04.005
vol. 8, pp. 7761–7779, 2012. [38] Q. G. Wang, T. H. Lee, H. W. Fung, Q. Bi, Y. Zhang, ‘‘PID tuning
[21] P. Wang, D. P. Kwok, ‘‘Optimal design of PID process controllers for improved performance’’, IEEE Trans. Control Systems
based on genetic algorithms’’, Control Engineering Practice, vol. 2, Technology, vol. 7, pp. 457–465, 1999. [Online]. Available:
pp. 641–648, 1994. [Online]. Available: http://dx.doi.org/10.1016/ http://dx.doi.org/10.1109/87.772161
0967-0661(94)90008-6 [39] MathWorks, Control system toolbox user's guide. R2014a. The
[22] A. Bagis, ‘‘Tabu search algorithm based PID controller tuning for MathWorks, Inc., 2014.
desired system specifications’’, Journal of the Franklin Institute, [40] R. C. Dorf, R. H, Bishop, Modern control systems-8th edt. Addison-
vol. 348, pp. 2795–2812, 2011. [Online]. Available: http://dx.doi.org Wesley Longman, Inc., 1998.
/10.1016/j.jfranklin.2011.09.001 [41] W. S. Levine, The control handbook. CRC Press LLC, USA, 1996.