BULLETIN

BULLETIN OFOF THE POLISH ACADEMY

THE POLISH ACADEMY OF OF SCIENCES
SCIENCES CONTROL AND INFORMATICS
TECHNICAL
TECHNICAL SCIENCES, Vol. 69(3), 2021, Article
SCIENCES, Vol. 69(3), 2021, Article number:
number: e137348
e137348
DOI:
DOI: 10.24425/bpasts.2021.137348
10.24425/bpasts.2021.137348

Nonlinear PID controller parameter optimization

optimization
using
using modified hybrid artificial
modified hybrid artificial bee
bee colony
colony algorithm
algorithm
for continuous
for continuous stirred
stirred tank
tank reactor
reactor
Nedumal Pugazhenthi P 1*, 1S. SELVAPERUMAL 1 , 1and K. VIJAYAKUMAR 2
Nedumal P. PUGAZHENTHI , S. SELVAPERUMAL and K. VIJAYAKUMAR2∗
1
Department of EEE, Syed Ammal Engineering College, Ramanathapuram, Tamilnadu, India
1 Department of EEE, Syed Ammal Engineering College, Ramanathapuram, Tamilnadu, India
2
Department
2 Department of electronicsand
of electronics instrumentation, Dr. Mahalingam College of Engineering and Technology, Pollachi, Tamilnadu, India
and instrumentation, Dr.Mahalingam College of Engineering and Technology, Pollachi, Tamilnadu, India

Abstract. The artificial bee colony (ABC) algorithm is well known and widely used optimization method based on swarm intelligence, and it
Abstract.
is inspired byThethe
artificial
behavior beeofcolony (ABC)
honeybees algorithm
searching forisa high
well amount
known and widelyfrom
of nectar usedtheoptimization methodthis
flower. However, based on swarm
algorithm has intelligence, and it
not been exploited
is inspired by the behavior of honeybees searching for a high amount of nectar from the flower. However, this algorithm
sufficiently. This research paper proposes a novel method to analyze the exploration and exploitation of ABC. In ABC, the scout bee searches has not been exploited
sufficiently.
for a source This research
of random paper
food for proposes a novel
exploitation. method
Along with torandom
analyzesearch,
the exploration
the scout and
bee exploitation
is guided byofa ABC. In ABC,
modified thealgorithm
genetic scout beeapproach
searches
for a source of random food for exploitation. Along with random search, the scout bee is guided by a modified genetic algorithm
to locate a food source with a high nectar value. The proposed algorithm is applied for the design of a nonlinear controller for a continuously approach to
locate atank
stirred foodreactor
source(CSTR).
with a high nectar value.
The statistical The of
analysis proposed algorithm
the results confirmsis applied
that the for the design
proposed of a nonlinear
modified controller
hybrid artificial bee for a continuously
colony (HMABC)
stirred tank
achieves reactor (CSTR).
consistently The statistical
better performance analysis
than of the results
the traditional ABC confirms
algorithm. that
Thetheresults
proposed modified hybrid
are compared artificial beeABC
with conventional colony
and(HMABC)
nonlinear
PID (NLPID) to show the superiority of the proposed algorithm. The performance of the HMABC algorithm-based controller isand
achieves consistently better performance than the traditional ABC algorithm. The results are compared with conventional ABC nonlinear
competitive
PID (NLPID) to show the superiority of the proposed algorithm. The performance
with other state-of-the-art meta-heuristic algorithm-based controllers in the literature. of the HMABC algorithm-based controller is competitive
with other state–of-the–art meta-heuristic algorithm-based controllers in the literature.
Key words: artificial bee colony; stirred tank reactor; genetic algorithm; nonlinear PID; controller performance measures.
Key words: artificial bee colony, stirred tank reactor, genetic algorithm, nonlinear PID, controller performance measures.

1. Introduction To design a nonlinear controller for the CSTR process, linear

models at different operating regions are identified and the con-
The proportional-integral-derivative controller is the most com- troller parameters for the identified local linear model are found
monly employed practical controller since it has simple imple- using conventional tuning methods such as the ZN method,
mentation architecture and robustness [1]. The PID controller IMC method, and these linear controllers are combined to form
has three controller parameters, namely proportional gain (K p ), a nonlinear controller [8]. The local linear controller designed
integral time (Tr ), and derivative (Td ). The value of these gains for the local linear region will not afford satisfactory set-point
determines the servo and regulatory performance of a con- tracking and disturbance rejection operation for the shifted op-
troller. These value gains depend upon the process character- erating region. Therefore, in these shifted operating regions, the
istics [2]. In linear systems, determining these parameters us- controller has to change its parameters to adapt to the new re-
ing conventional tuning methods such as Ziegler Nichols (ZN) gion or the controller parameters must be optimized to handle
or classical forced oscillation (CFO) may establish better servo the shifted operating point. To design the optimal controller,
and regulatory performance. These conventional methods will meta-heuristic approaches are used [9]. Some bio-inspired op-
not provide better performance for nonlinear systems. Most of timization techniques such as genetic algorithm (GA) and its
the processes found in industries exhibit a nonlinear relation- variants [10] differential evolution (DE), ant colony optimiza-
ship between input and output [3]. Hence, the controller design tion (ACO), particle swarm optimization (PSO) are also used.
for such systems using a conventional tuning method will not Among all other heuristic approaches, the ABC algorithm is
provide good servo and regulatory performance [4]. tuning of widely adapted to find the optimal parameters [11].
internal model control (IMC) controller for the nonlinear sys- Dervis Karaboga proposed a swarm-based meta-heuristic al-
tem is presented in [5]. The other design procedures for the gorithm ABC, which is an optimization that imitates honeybee
nonlinear controller design available in the literature such as swarms foraging for food [12]. Currently, this ABC algorithm
nonlinear model predictive controller design [6], Fuzzy logic is gaining popularity among researchers since it yields better
controller design [7], and neural network-based model predic- optimization performance, is simple and easily implemented.
tive controller are very complex and involve heavy mathemati- A multi-colony artificial bee colony algorithm was proposed
cal computations for determining the control action. in [13]. The number of algorithm parameters in this technique
is smaller than any heuristic optimization technique. The ABC
∗ e-mail:
algorithm is also capable of optimizing nonconvex optimization
[email protected]
*e-mail: [email protected] problems. For this reason, ABC is employed in all engineering
Manuscript submitted
Manuscript submitted2020-07-30, revisedinitially
20XX-XX-XX, 2021-03-24, initiallyfor
accepted accepted
publication optimization problems [14]. A high nectar food source is iden-
for publication 2021-04-26,
20XX-XX-XX, published published in June 2021
in June 2021. tified by certain bees of ABC collectively. Normally, in a non-

Bull. Pol. Acad.

Bull. Pol. Acad. Sci.
Sci. Tech.
Tech. Sci.
Sci. 69(3)
69(3) 2021,
2021,e137348
e137348 11

© 2021 The Author(s). This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
 Nedumal
NedumalPugazhenthi P, S. S.
P. Pugazhenthi, Selvaperumal, and
Selvaperumal, K.K. Vijayakumar
Vijayakumar

linear optimization-based problem, ABC may usually get stuck Step 2. Employed bee phase
in a local minimum as it is deprived of local searching ability. These bees explore the neighboring nectar sources, Ei, j ex-
To overcome these issues, a search scheme–modified GA is de- pressed by (2). The identified nectar replaces the randomly se-
signed to improve the local searching ability. lected initial food source.
In the proposed HMABC, the suitable choice of nectar source
by a scout bee is performed by a modified genetic algorithm. A Ei, j = Fi, j + rand[−1, 1] × (Fi, j − Fk, j ), i = k, (2)
scout bee is assumed to be present at an arbitrary nectar source
from where it is guided to travel to a new food source by us- where
ing GA reproduction operators. This method results in a high k = {1, 2 . . . , Fn },
probability of identifying a nectar source of high quality by the j = {1, 2, 3}, (number of parameters to be optimized),
scout bee. The proposed novel algorithm is applied to design the Fk, j , denotes random food source.
nonlinear PID for the CSTR system and a comprehensive study Step 3. Onlooker bee phase
is made with the basic tuning IMC methods, hybrid heuristic The onlooker bee selects a nectar source from the updated
tuning techniques GA-ABC. The developed algorithm perfor- fitness function done during the employed bee phase. Eq. (3)
mance is verified based on the values of integral square error expresses high nectar value selection by onlooker bee.
(ISE), integral absolute error (IAE), and integral time absolute The onlooker bee searches for a better nectar source near the
error (ITAE). Only the ISE-based results are presented in this available selected nectar source as in Eq. (2). For unimproved
paper. nectar sources, the trail value of the bees is changed.
The present research is structured as follows: Section 2 dis-
cusses the salient feature of the ABC algorithm; Section 3 fitness(i)
presents the HMABC optimization method; Section 4 describes pi = NF
. (3)
the design of the HMABC-based NLPID controller; and finally, ∑ fitness( j)
Section 5 discusses the results obtained from the proposed con- j=1
trol strategies. The conclusion of the research work is briefed in
the last part. Step 4. Scout bee phase
The employed bee which fails to identify better nectar
sources is (trail greater than the limit) transformed into a scout
2. Artificial bee colony algorithm bee and the total search colony is accessed randomly by the
scout bee looking for a new food source based on Eq. (1).
Dervis Karaboga proposed the ABC algorithm based on honey One scout bee per cycle is engaged to eliminate the multiple
searching behavior for nonlinear optimization problems. The random searches. The proper trail value determines the number
optimization phases of ABC-employed bee, onlooker bee, and of scout bees. Once the scout bee phase ends, the search cycle
scout bee are generally used. The bee starts a food search from is incremented by one.
a random food source; once it finds the food, the employed bee Step 5. Stopping condition
waves to and fro. On seeing the waggle dance, the on-looker In this algorithm further searching for a new food source is
bees would select the nectar source and look for fresh varia- stopped as the maximum cycle is reached and the best food
tions of nectar sources. The variation of the existing food source source obtained as far as the optimized value of parameters is
may consist of good nectar or could be worse than the previous concerned, or else Step 2 is repeated.
food source. This exhausted food source which has a low nec- The ABC search algorithm has been modified by many re-
tar value is replaced by a random search conducted by the scout searchers to enhance its performance. The modification is done
bee. This cycle is repeated until a better food source is found. by either hybridization of other known search algorithms or by
The colony size (Cs ) and maximum cycle count (Cmax ) are ini- updating the equations. The hybridization of ABC with other
tialized, and the following stages are conducted to mimic the algorithms is a well-known approach to enhancement [15].
foraging behavior of the bee. ABC hybridization is performed with differential evolution
Step 1. Initialization Phase (DE) [16], where the number of scout bees is changed by DE.
The initialization of ABC parameters Tv , Fn , and Mc is done The ABC and the combination of midway disposal trip selec-
first, where Tv – trail value (max); Fn – food source and Mc – tion are discussed in [17]. ACO and harmonic search hybridiza-
number of the cycle. tion with ABC are discussed in [18], which outperforms the
The initial food source is randomly selected as shown in gradient search, PSO, ACO, and ABC, but takes more time for
Eq. (1) completing one cycle. ABC with tabu search [19, 20] and ABC
F1, j = lb1, j + rand[0, 1] × (upi, j − lbi, j ), (1) with PCO in [21] are used for optimizing the problem in het-
erogeneous wireless networks. Hill climbing optimizer algo-
where: i refers to the number of food source and j refers to the rithm [22] and crossover operator [23] are executed for better
dimension of the search (set as 3); lb1, j , denotes lower limit searching and data clustering. However, the ABC suffers from
of i-th food source; upi, j , denotes the upper limit of i-th food ineffective exploitation [24].
source; rand[0, 1], represents a random value between 0 and 1. To improve the exploitation of ABC, the genetic algorithm
These food sources are treated as controller parameters. hybridization is proposed [25], where GA improves the per-

22 Bull.
Bull.Pol.
Pol.Acad.
Acad.Sci.
Sci.Tech.
Tech.Sci.
Sci. 69(3)
69(3) 2021,
2021, e137348
e137348
 Nonlinear PID controller parameter optimization
Nonlinearusing
PIDmodified hybrid
controller artificialoptimization
parameter bee colony algorithm
... for continuous stirred tank reactor

formance of an onlooker bee. This hybridization is achieved Step 1. Initialization of ABC parameters.
by incorporating ABC with an artificial fish swarm (AFS) al- Step 2. Employed bee phase.
gorithm [26]. Various hybridization algorithms are discussed
in [27] and an improved dolphin swarm optimization algorithm Step 3. Onlooker bee phase.
in [28], A nonlinear PID using a neural networks-like approach Step 4. Scout bee phase – the onlooker bee determines the ini-
is discussed in [29]. ABC with ACO for CSTR is applied in tial population.
[30]. Differential evaluation (H–ABCDE) algorithm [31] and Step 5. The food source is selected using a roulette wheel se-
artificial immune system (AIS) [32] are combined with ABC to lection based on nectar values.
enhance their performances. The enhancement of the scout bee Step 6. The selected food source is considered as parent chro-
phase has not been dealt with in any of the earlier approaches. mosomes p1 and p2.
To improve the scout bee performance, a novel hybrid ap-
proach is proposed, which is discussed in a detailed manner in Step 7. The mutation is performed, and new food source loca-
the next section. tions are identified.
Step 8. The crossover is conducted by a two-point crossover
method.
3. Modified hybrid artificial bee colony Assign two points x_pt1, x_pt2.
algorithm The crossover kids c1 and c2 are derived as per the equations
below:
 
The ABC convergence rate is improved by replacing unim- c1 = p1(1 : x_pt1)p2(x_pt1 + 1 : x_pt2)p1(x_pt2 + 1 : CL ],
 
proved food sources with a high nectar food source in the scout c2 = p2(1 : x_pt1)p1(x_pt1 + 1 : x_pt2)p2(x_pt2 + 1 : CL) ,
bee phase. The ABC with a better convergence rate is used in
where CL is the chromosome length.
the optimization of controller parameters in many engineering
applications. In the proposed HMABC algorithm, genetic op- Step 9. The elite food source is identified.
erators are used in the scout bee phase to improve its intel- Step 10. Group the mutation, crossover, and elite offspring into
ligence. In the HMABC, the scout identifies the best nectar the next generation population.
source through onlooker bees. If all onlooker bees share the
Step 11. The new food source locations are evaluated for the
same location with the scout bee, along with this location the
nectar values. And the best nectar value food source is as-
scout bee generates other random locations. It starts applying
signed as a fluctuated food source. If the maximum generation
the crossover and mutation operators on the positions of the
is reached, then perform Step 12; otherwise repeat Step 5.
food source to identify the new improved food source. This
search method improves the identification of the quality of the Step 12. Terminating the condition evaluation.
food source identified by scout bees and makes the proposed If the upper limit of iteration is reached, then perform
HMABC converge faster than a conventional ABC. Step 13; otherwise, repeat Step 2.
The proposed HMABC algorithm involves the following Step 13. Terminate the algorithm and the best nectar source is
steps: considered as the optimum result.
/*Step 1 to Step 3 are similar to the conventional ABC opti- The flow chart of the proposed hybrid modified ABC is de-
mization method*/ picted in Fig. 1.

Fig. 1. Flow chart of hybrid modified ABC

Bull. Pol. Acad.

Bull. Pol. Acad. Sci.
Sci. Tech.
Tech. Sci.
Sci. 69(3)
69(3) 2021,
2021,e137348
e137348 33
 Nedumal
NedumalPugazhenthi P, S. S.
P. Pugazhenthi, Selvaperumal, and
Selvaperumal, K.K. Vijayakumar
Vijayakumar

Table 1
Benchmark functions and their best values tested on ABC and HMABC

Test function Range Result (ABC) (HMABC)

Best value 0.1936 0.1037
Bukin N 6 x1 ∈ [−15, −5], x2 ∈ [−3, 3]
Worst value 0.1723 0.1129
Best value 5.380e–19 8.5792e–20
Levy xi ∈ [−10, 10]
Worst value 3.98e–18 4.34e–19
Best value 1.8103e–19 1.5282e–19
Levy N 13 xi ∈ [−10, 10]
Worst value 1.323e–18 1.234e–18
Best value 2.074e–07 1.1999e–07
Perm Function 0, d, β xi ∈ [−10, 10]
Worst value 1.923e–06 2.476e–06
Best value 6.516e–19 4.7396e–19
Rotated Hyper-Ellipsoid Function xi ∈ [−65.536, 65.536]
Worst value 2.987e–9 8.348e–18
Best value 5.916e–19 5.5296e–19
Sphere xi ∈ [−5.12, 5.12]
Worst value 4.345e–17 3.478e–18
Best value 0.2925 0.292578
Schaffer N 4 xi ∈ [−100, 100]
Worst value 0.9846 0.6473
Best value 8.88e–16 8.881 e–16
Ackley xi ∈ [−32.768, 32.768]
Worst value 6.347e–15 7.896e–15

The efficiency of the conventional ABC and proposed 4. HMABC-based optimum controller for CSTR
HMABC algorithm is tested in a few standard benchmark func-
tions [33–35] and the results are listed in Table 1. The results To evaluate the performance of the proposed HMABC, it is
clearly show that the proposed HMABC algorithm produces evaluated as shown in Fig. 2 to design an optimum controller
better results than the conventional ABC. Different values are for the nonlinear CSTR process. CSTR is one of the bench-
chosen and tested and after multiple runs are done, the best val- mark problems considered by researchers in testing the con-
ues of parameters used in the algorithms for this test are pro- troller performance as is evident in [6, 29, 36].
vided in Table 2.

Table 2
The best value used for ABC and HMABC

Parameters ABC HMABC

Number of food sources 20 20
Dimension of the population 15 15
Maximum cycle 200 200
Trail limit 20 20
Population size of GA – 40
Crossover rate – 0.8 Fig. 2. CSTR process
Number of crossover sites – 2
Mutation rate – 0.2
The CSTR process exhibits a high nonlinear relationship be-
Elite Count – 2 tween the inlet coolant flow rate (qc (t)) and the concentration
Number of generations – 20 C(t). The objective of the controller is to manipulate the inlet
coolant flow rate to maintain the concentration C(t) of the out-
Chromosome length – 15
let [29].

44 Bull.
Bull.Pol.
Pol.Acad.
Acad.Sci.
Sci.Tech.
Tech.Sci.
Sci. 69(3)
69(3) 2021,
2021, e137348
e137348
 Nonlinear PID controller parameter optimization
Nonlinearusing
PIDmodified hybrid
controller artificialoptimization
parameter bee colony algorithm
... for continuous stirred tank reactor

The process models described in (4) and (5) are nonlinear in the local model structure is needed to give the global model
equations of the state variables T (t) and C(t) structure, which is expressed in (8)
  N
dT qi E
= (Ti − T (t)) + K1C(t) exp − ẋ = ∑ fi (x, u, v, θi ) µi (ϕ),
dt V RT (t) i=1
   (8)
K3   N
+ K2 qc (t) 1 − exp − Tc f − T (t) , (4)
qc (t) y = ∑ gi (y, w, θi )µi (ϕ),
  i=1
dC qi E
= (Ci −C(t)) − K0C(t) exp − . (5) where µi is the interpolation function expressed in (9)
dt V RT (t)

The initial parameters of the CSTR are provided in Table 3 [29]. ρi (ϕ)
µi (ϕ) = N
), (9)

Table 3
∑ ρ j (ϕ
j=1
Parameters of CSTR
where ρi is the model validity function and it is designed such
Input flow, qi – 100 l/min Input temperature Ti – 350 K that the value closes to 1 for a good model structure otherwise
Input concentration, Inlet coolant, equal to zero. A fuzzy nonlinear PID (F-NLPID) controller is
Ci – 1 mol/l Tc f – 350 K the interpolation of the local PID controllers to form the global
model structure.
Tank volume, V – 100 l Energy (E/R), 104 K
The grade of the membership function should be µi : ϕ →
K1 = 1.44 × 1013 Kl/min/mol K2 = 0.01/l [0, 1] is a normalization of the model validity function ρi and
N
K3 = 700 l/min K0 = 7.2 × 1010 min−1
has the property ∑ µi (ϕ) = 1 for all ϕ ∈ Φ.
i=1
The NLPID controller output is expressed by (10)
The researchers reported several tuning techniques for the
PID controller design. A ZN-based PID controller for the CSTR 
Ts
process is presented in [29]. IMC-based tuning for a PID con- ui (k) = K p,i e(k) − e(k − 1) + e(k)
Tr,i
troller is used in [6]. The IMC tuning technique is considered to 
be better than other classical methods due to its robust behav- Td,i
+ (e(k) − 2 × e(k − 1) + e(k − 2)) + ui (k − 1), (10)
ior, i.e., it provides fewer oscillations and overshoot. The IMC Ts
tuning formulae for the PID settings is given in (6)
where Ts is the sampling time; K p,i , Tr,i , and Td,i values of
2ζi 2ζi 1 NLPID are derived from classical tuning approaches. The lo-
K p,i = , Tr,i = , Td,i = , (6) cal PID of the CSTR process derived through the IMC method
ωn,i ki λ ωn,i 2ζi ωn,i
for various stable operating regions is listed in Table 4.
where Kp,i – proportional gain, Tr,i – integral time, Td,i – deriva-
tive time, ξ – damping factor, ωn,i – natural frequency of oscil- Table 4
lation and ki – steady state gain. To determine the ξ , ωn,I and CSTR operating regions and local PID settings
ki , the non-linear process is modeled using the Takagi–Sugeno
Operating point K p,i Tr,i Tr,i
(T–S) fuzzy multiple model design approaches [6].
The possible local linear models are identified over its com- At qc1 = 97; C1 = 0.077; T1 = 443.46 119.43/lamda 0.337 0.192
plete operating regions and the multi-model is derived from At qc2 = 100; C2 = 0.089; T2 = 441.15 92.69/lamda 0.298 0.254
the local model through interpolation, using T–S fuzzy [6].
At qc3 = 103; C3 = 0.098; T3 = 438.77 67.42/lamda 0.25 0.360
The NLPID is derived for the multi-model with fuzzy fusion
through the local model PID settings tuned from the IMC tun- At qc4 = 106; C4 = 0.111; T4 = 436.31 43.28/lamda 0.188 0.580
ing technique. The local model is expressed in (7) At qc5 = 109; C5 = 0.126; T5 = 433.70 19.18/lamda 0.103 1.312

ẋ = fi (x, u, v, θi ) ,
(7) The (qci , Ci , Ti ) represents the linearization point of the ith
y = gi (y, w, θi ), local model. The stable region of CSTR is expressed by the
phase plane analysis shown in Fig. 3. From the response, it is
where x, u, and v represent the state vector, input vector, and observed that CSTR exhibits nonlinearity behaviour when con-
disturbance vector, respectively; y and w represent the output centration c(t) rises more than 0.13. Figure 3 shows the phase
and noise vector, respectively; and θ is the parameterized vector plane analysis of a stable operating point. If the concentration is
used to describe the system dynamics. If the system operates raised more than 0.13, the phase plant plot becomes the saddle
a wider range, then the interpolation of each operating point point.

Bull. Pol. Acad.

Bull. Pol. Acad. Sci.
Sci. Tech.
Tech. Sci.
Sci. 69(3)
69(3) 2021,
2021,e137348
e137348 55
 Nedumal
NedumalPugazhenthi P, S. S.
P. Pugazhenthi, Selvaperumal, and
Selvaperumal, K.K. Vijayakumar
Vijayakumar

5. Simulation results

The algorithm parameters of ABC and HMABC are given in

Table 5 and HMABC results are related to ABC.

Table 5
Algorithm parameters

Parameters ABC HMABC

Number of food sources (i) 20 20
Dimension of the population ( j) 15 15
Maximum cycle 200 200
Trail limit 20 20
Fig. 3. Phase plane analysis – CSTR stable operating point
Population size of GA – 40
Crossover rate – 0.8
In the literature, it is derived that the classical tuning ap-
Number of crossover sites – 2
proaches are not sufficient to give satisfactory performance for
the complex nonlinear system. The recent optimization tech- Mutation rate – 0.2
nique ABC is applied to calculate the best possible combina- Elite count – 2
tions of the NLPID parameters. The various integral perfor- Number of generations – 20
mance criteria as shown in (11), (12), (13), and (14) can be
taken as cost functions. These Cost functions were minimized Upper limit of Kp,i 200 200
by ABC and HMABC. Upper limit of Tr,i 0.3 0.3
Integral absolute error Upper limit of Td,i 0.4 0.4
Chromosome length – 15
∞
Cost function ISE ISE
IAE = |e(t)| dt. (11)
0
The convergence characteristics of the ABC and HMABC
for minimizing the objective function value are shown in Fig. 4.
Integral time absolute error
In the HMABC, the objective function value is reduced in ev-
ery cycle and displays a faster response than the conventional
∞
ABC. In the HMABC, 183 cycles to reach the objective func-
ITAE = t |e(t)| dt. (12) tion amount to be 2.46, whereas the ABC uses 426 cycles.
0

Integral squared error

∞
ISE = e2 (t) dt. (13)
0

Integral time squared error

∞
ITSE = te2 (t) dt. (14) Fig. 4. Convergence comparison of ABC and proposed HMABC
0
The designed nonlinear controller is applied to CSTR and
ISE can be considered as an objective function when the er- various performance characteristics are evaluated. The obtained
ror value has large positive and negative values. IAE can be results are explained briefly in the following sections. Table 6
considered when the error value has smaller positive and nega- shows the numerical evaluation of the results such as standard
tive values. Time–weighted ISE and IAE can be considered as deviation, mean of convergence curve, and a detailed compar-
objective when we need a lower steady-state error. In this paper, ison of the scout bee fitness value of the HMABC with ABC.
controller performance based on ISE as an objective function is From the values, it is found that the HMABC scout bee is better
discussed. than a conventional scout bee.

66 Bull.
Bull.Pol.
Pol.Acad.
Acad.Sci.
Sci.Tech.
Tech.Sci.
Sci. 69(3)
69(3) 2021,
2021, e137348
e137348
 Nonlinear PID controller parameter optimization
Nonlinearusing
PIDmodified hybrid
controller artificialoptimization
parameter bee colony algorithm
... for continuous stirred tank reactor

Table 6 Table 7
Statistical evaluation of convergence characteristics curve Statistical values of scout bee fitness

Type of optimization technique Mean Standard deviation Type of optimization technique Mean Standard deviation
ABC 2.705 0.2622 ABC 8.7549 5.890
HMABC 2.537 0.15 HMABC 2.4992 0.026

Table 5 shows that the mean of convergence characteristic Figure 7 shows the comparison of results of HMABC-based
curves plotted in Fig. 4. From this table, it is clear that, in the PID responses for the set-point variations with ABC-based
ABC-based optimization and with the objective function mini- PID. The local PID parameters obtained through the ABC and
mization, the mean is 2.705, whereas, in the HMABC, the mean HMABC optimization methods are given in Tables 8 and 9, re-
is only 2.537. This indicates that the HMABC has a better op- spectively.
timization effort than the ABC. From Fig. 5, it is clear that the
maximum value of the cost function is lower than the value
obtained using a conventional ABC. This reduction in the max-
imum value shows that the HMABC produces a better solution
in each cycle compared to the ABC. Comparing the ABC, the
proposed HMABC has a low standard deviation and this indi-
cates that the HMABC reaches a minimum cost function value
in fewer cycles. From Fig. 5, it is clear that in all other statistical
comparisons, the HMABC is better than the ABC.

Fig. 7. Servo response of CSTR for HMABC PID

Table 8
ABC – PID settings

Operating point Kp Tr,i Td,i

At qc1 = 97; C1 = 0.077; T1 = 443.46 200/lamda 0.300 0.158
At qc2 = 100; C2 = 0.089; T2 = 441.15 200/lamda 0.300 0.193
At qc3 = 103; C3 = 0.098; T3 = 438.77 200/lamda 0.300 0.163
At qc4 = 106; C4 = 0.111; T4 = 436.31 200/lamda 0.300 0.345
Fig. 5. Statistical evaluation of ABC and HMABC
At qc5 = 109; C5 = 0.126; T5 = 433.70 200/lamda 0.291 0.100

The food sources identified by the scout bee to replace the

unimproved food source are monitored in each cycle of the Table 9
HMABC – PID settings
ABC as well as in the HMABC and depicted in Fig. 6, and the
values of the fitness function and the food source are given in
Operating point Kp Tr,i Td,i
Table 7. Based on the results, it is decided that the HMABC per-
forms better as it increases the searching cabbalist of the scout At qc1 = 97; C1 = 0.077; T1 = 443.46 200/lamda 0.300 0.192
bee by three times. At qc2 = 100; C2 = 0.089; T2 = 441.15 200/lamda 0.300 0.178
At qc3 = 103; C3 = 0.098; T3 = 438.77 66.32/lamda 0.300 0.290
At qc4 = 106; C4 = 0.111; T4 = 436.31 178.44/lamda 0.300 0.370
At qc5 = 109; C5 = 0.126; T5 = 433.70 193.26/lamda 0.263 0.119

A lower value of the lamda provides aggressive control per-

formance. The lamda value and its effectiveness are presented
in Fig. 8. The performance of the HMABC is compared with
the ABC method to show its effectiveness.
For a smaller value of lamda, the value of K p will be higher.
Fig. 6. Scout bee objective values of ABC and HMABC Thus, setting a smaller value for lamda provides a more aggres-

Bull. Pol. Acad.

Bull. Pol. Acad. Sci.
Sci. Tech.
Tech. Sci.
Sci. 69(3)
69(3) 2021,
2021,e137348
e137348 77
 Nedumal
NedumalPugazhenthi P, S. S.
P. Pugazhenthi, Selvaperumal, and
Selvaperumal, K.K. Vijayakumar
Vijayakumar

the regulatory operation are presented in Table 11, in which the

ISE values during the sampling time 101 to 151 seconds of the
ABC-based PID is 1.5732e–3, whereas in the HMABC-based
PID controller the ISE value is only about 1.0947e–3. The com-
parison of regulatory action of the HMABC controller is de-
picted in Fig. 9. The feed temperature change is regulated by
the HMABC-based PID controller with a low ISE value com-
pared to the ABC-based PID controller.

Fig. 8. Servo response of CSTR from 400 sampling instant to 1000 for
various controllers

sive control action and more overshoot. The following diagram

shows the servo response of the implemented controllers from
the sampling instants 400 to 1000.
Fig. 9. Regulatory response of CSTR for HMABC PID
The cost-function ISE values are shown in Table 10 obtained
for various controllers and HMABC-PID at various sampling
instants to show the effectiveness of the proposed one.
The results show that the proposed HMABC performs much
better than the other conventional methods for nonlinear pro-
cesses. At sampling instants between 101 and 150, HMABC
– PID exhibits ISE value as 1.2344e–7, whereas ISE value of
conventional ABC is 5.7433e–7. The regulatory performance
of the designed controller is verified by injecting a variation in
the feed temperature of the CSTR. At sampling instant 120, the
feed temperature “T” is raised to 440 K. This abrupt change
introduces a great dynamic in the system. The designed con-
troller can reject that change and brings the system to stabil-
ity with minimum overshoot. The ISE values obtained during Fig. 10. Control action during the regulatory response of CSTR

Table 10
Sampling time in seconds and ISE values for servo response

Sampling time in seconds & ISE

PID tuning method
0 to 50 51 to 100 101 to 150 151 to 200 201 to 250 251 to 300 301 to 350
IMC 7.3086e–6 7.3933e–3 6.5555e–3 4.3560e–3 4.0491e–3 1.9648e–6 1.7321e–6
ABC 5.8054e–7 8.2912e–4 5.7433e–7 1.3918e–3 4.0855e–9 3.2581e–4 8.004e–10
HMABC 1.6517e–7 6.1247e–4 1.2344e–7 1.4140e–3 1.3630e–9 3.0262e–4 1.5593e–9

Table 11
Sampling time in seconds and ISE values for regulatory operation

Sampling time in seconds & ISE

PID tuning method
0 to 50 51 to 100 101 to 150 151 to 200 201 to 250 251 to 300 301 to 350
ABC 6.7338e–4 4.2315e–4 1.5732e–3 5.9475e–8 7.8730e–18 3.7585e–4 1.1316e–13
HMABC 6.6400e–4 4.1654e–4 1.0947e–3 6.1824e–8 7.5370e–18 3.8105e–4 1.7453e–13

88 Bull.
Bull.Pol.
Pol.Acad.
Acad.Sci.
Sci.Tech.
Tech.Sci.
Sci. 69(3)
69(3) 2021,
2021, e137348
e137348
 Nonlinear PID controller parameter optimization using modified hybrid artificial bee colony algorithm for continuous stirred tank reactor

6. Conclusion [11] S.S. Jadon, R. Tiwari, H. Sharma, and J.C. Bansal, “Hybrid Arti-
ficial Bee Colony algorithm with Differential Evolution”, Appl.
In this manuscript, the authors have presented an effortless and Soft Comput. 58(1), 11‒24 (2017).
uncomplicated strategy that improvises the searching capabil- [12] D. Karaboga, “An Idea Based on Honey Bee Swarm for Nu-
ity of the ABC, which provides the optimum NLPID control merical Optimization”, Technical Report-TR06, Department of
parameters. These NLPID values are tested on a nonlinear Computer Engineering, Engineering Faculty, Erciyes University
(2005).
reactor process to show the effectiveness of its performance.
[13] J. Zhou, X.Yao, F.T.S. Chan, Y. Lin, H. Jin, L. Gao, X. Wang,
The simulation responses obtained, clearly prove the superi-
“An individual dependent multi-colony artificial bee colony al-
ority of the proposed control algorithm by its good servo and gorithm”, Inf. Sci. 485(1), 114‒140 (2019).
regulatory performance capabilities at various operating points. [14] X. Chen, H. Tianfield, and K. Li, “Self-adaptive differential ar-
The proposed algorithm is compared with well-known conven- tificial bee colony algorithm for global optimization problems”,
tional optimization algorithms for the same process. From the Swarm Evol. Comput. 45(1), 70‒91 (2019).
results, it is determined that HMABC – PID helps to reduce the [15] Y. Zhang, S. Cheng, Y. Shi, D.-Wei Gong, and X. Zhao, “Cost-sen-
number of cycles that are needed to obtain optimum controller sitive feature selection using two-archive multi-objective arti-
gain values for good servo and regulatory action. The proposed ficial bee colony algorithm”, Expert Syst. Appl. 137(1), 46‒58
HMABC-based optimization can be considered as a substitute (2019).
to basic old optimization algorithms. [16] R. Szczepanski, T. Tarczewski, and L.M. Grzesiak, “Adaptive
state feedback speed controller for PMSM based on Artificial
Acknowledgements. There is no fund received for this research Bee Colony algorithm”, Appl. Soft Comput. 83(1), 105644
work. We would like to acknowledge our college management (2019).
[17] Q. Wei, Z. Guo, H.C. Lau, and Z. He, “An artificial bee col-
for the moral support provided in completing this research.
ony-based hybrid approach for waste collection problem with
midway disposal pattern”, Appl. Soft Comput. 76(1), 629‒637
References (2019).
[18] T. Sen and H.D. Mathur, “A new approach to solve Economic
[1] M.J. Mahmoodabadi, R.A. Maafi, and M. Taherkhorsandi, “An Dispatch problem using a Hybrid ACO–ABC–HS optimiza-
optimal adaptive robust PID controller subject to fuzzy rules and tion algorithm”, Electr. Power Energy Syst. 78(1), 735–744
sliding modes for MIMO uncertain chaotic systems”, Appl. Soft (2017).
Comput. 52(1), 1191‒1199 (2017). [19] X. Li , Z. Peng , B. Dub, J. Guo, W. Xu, and K. Zhuang, “Hybrid
[2] C. Lorenzini, A.S. Bazanella, L.F. Alves Pereira, and G.R. Gon- artificial bee colony algorithm with a rescheduling strategy for
çalves da Silva, “The generalized forced oscillation method for solving flexible job shop scheduling problems”, Comput. Ind.
tuning PID controllers”, ISA Trans. 87(1), 68‒87 (2019). Eng. 113(1), 10–26 (2017).
[3] S.K. Valluru and M. Singh, “Performance investigations of [20] S. Lua, X. Liua, J. Peia, M.T. Thai, and P.M. Pardalos, “A hybrid
APSO tuned linear and nonlinear PID controllers for a nonlinear ABC-TS algorithm for the unrelated parallel-batchingmachines
dynamical system”, J. Electr. Syst. Inf. Technol. 5(3), 442‒452 scheduling problem with deteriorating jobs and maintenance ac-
(2018). tivity”, Appl. Soft Comput. 66(1), 168–182 (2018).
[4] M. Kumar, D. Prasad, B.S. Giri, and R.S. Singh, “Tempera- [21] S. Goudarzi et.al., “ABC-PSO for vertical handover in hetero-
ture control of fermentation bioreactor for ethanol production geneous wireless networks”, Neurocomputing 256(1), 63–81
using IMC-PID controller”, Biotechnol. Rep. 22(1), e00319 (2017).
(2019). [22] M.A. Awadallah, A.L. Bolaji, and M.A. Al-Betar, “A hybrid ar-
[5] D.B. Santosh Kumar and R. Padma Sree, “Tuning of IMC based tificial bee colony for a nurse rostering problem”, Appl. Soft
PID controllers for integrating systems with time delay”, ISA Comput. 35(1), 726‒739 (2015).
Trans. 63(1), 242‒255 (2016). [23] X. Yan, Y. Zhu, W. Zou, and L. Wang, “A new approach for
[6] J. Prakash and K. Srinivasan, “Design of nonlinear PID control- data clustering using hybrid artificial bee colony algorithm”,
ler and nonlinear model predictive controller for a continuous Neurocomputing, 97(1), 241‒250 (2012).
stirred tank reactor”, ISA Trans. 48(3), 273‒282 (2009). [24] W.-F. Gao and S.-Y. Liu, “A modified artificial bee colony al-
[7] M. Hamdy and I. Hamdan, “Robust fuzzy output feedback con- gorithm”, Eng. Appl. Artif. Intell. 39(1), 3, 687‒697 (2012).
troller for affine nonlinear systems via T–S fuzzy bilinear model: [25] P. Pramanik and M.K. Maiti, “An inventory model for deteriorat-
CSTR benchmark”, ISA Trans. 57(1), 85‒92 (2015). ing items with inflation induced variable demand under two level
[8] V. Ghaffari, S. VahidNaghavi, and A.A. Safavi, “Robust mod- partial trade credit: A hybrid ABC-GA approach”, Biotechnol.
el predictive control of a class of uncertain nonlinear systems Rep. 85(1), 194–207 (2019).
with application to typical CSTR problems”, J. Process Control. [26] V. Hajisalem and S.Babaie, “A hybrid intrusion detection system
23(4), 493‒499 (2013). based on ABC-AFS algorithm for misuse and anomaly detec-
[9] W.-D. Chang, “Nonlinear CSTR control system design using an tion”, Comput. Networks 136(1), 37–50 (2018).
artificial bee colony algorithm”, Simul. Modell. Pract. Theory [27] W. Chmiel, P. Kadłuczka, J. Kwiecień, and B. Filipowicz, “A com-
31(1), 1‒9 (2013) parison of nature inspired algorithms for the quadratic assignment
[10] Y.P. Wang, N.R. Watson, and H.H. Chong, “Modified genetic problem”, Bull. Pol. Acad. Sci. Tech. Sci. 65(4), 513‒522 (2017).
algorithm approach to design of an optimal PID controller for [28] Y. Li and X. Wang, “Improved dolphin swarm optimization algo-
AC–DC transmission systems”, Int. J. Electr. Power Energy Syst. rithm based on information entropy”, Bull. Pol. Acad. Sci. Tech.
24(1), 59‒69 (2002). Sci. 67(4), 679‒685 (2019).

Bull. Pol. Acad. Sci. Tech. Sci. 69(3) 2021, e137348 9

 Nedumal Pugazhenthi P, S. Selvaperumal, and K. Vijayakumar

[29] R. Gao, A. O’dywer, and E. Coyle, “A Nonlinear PID control for [33] S. Surjanovic and D. Bingham, “Virtual Library of Simulation
CSTR using local model networks”, Proceedings of 4th World Experiments: Test Functions and Datasets” [Online]. Available:
Congress on Intelligent Control and Automation, Shanghai, Chi- http://www.sfu.ca/~ssurjano [Accessed: January 21, 2021].
na, 2002. [34] D. T. Pham and M. Castellani, “Benchmarking and comparison
[30] K. Vijayakumar and M. Thathan, “Enhanced ABC Based PID of nature-inspired population-based continuous optimisation al-
Controller for Nonlinear Control Systems”, Indian J. Sci. Tech- gorithms”, Soft Comput. (18), 871–903 (2014).
nol. 8(1), 1‒9 (2015). [35] Y. Zhang, P. Wang, L. Yang, Y. Liu, Y. Lu, and X. Zhu, “Nov-
[31] D. Ustuna and A. Akdagli, “Design of band–notched UWB an- el Swarm Intelligence Algorithm for Global Optimization and
tenna using a hybrid optimization based on ABC and DE algo- Multi-UAVs Cooperative Path Planning: Anas Platyrhynchos
rithms”, Int. J. Electron. Commun. 87(1), 10–21 (2018). Optimizer”, Appl. Sci. 10(14), 4821, 1‒29 (2020).
[32] D. Zhang, R. Dong, Y.-W. Si, F. Ye, Q. Cai, “A hybrid swarm [36] K. Anbarasan and K. Srinivasan, “Design of RTDA controller
algorithm based on ABC and AIS for 2L-HFCVRP”, Appl. Soft for industrial process using SOPDT model with minimum or
Comput. 64(1), 468–479 (2018). non-minimum zero”, ISA Trans. 57, 231–244 (2015).

10 Bull. Pol. Acad. Sci. Tech. Sci. 69(3) 2021, e137348