Research Article

Optimal PID tuning for controlling the temperature of electric furnace

by genetic algorithm
Md. Manjurul Gani1 · Md. Saiful Islam1 · Muhammad Ahsan Ullah1

© Springer Nature Switzerland AG 2019

Abstract
Temperature is one of the exigent parameters that needs to be controlled in today’s industries. Importantly this tempera-
ture control should be precise and fast. As the conventional controllers are not optimally tuned, the controller used for
controlling the temperature of the electric furnace does not exhibit better performance. Its rise time and settling time
is too large as well as it has a sizable amount of overshoot. This paper presents a Genetic Algorithm based PID controller
to overcome the low precision, long rise time and settling time of the controller. In this algorithm, Integral of Absolute
Error is taken as the object function for minimizing the error. Using this function, the algorithm engenders the optimum
value of the gain parameters ­(Kp, ­Ki, ­Kd) for the PID controller. It shows better control over the conventional controllers.
As the overshoot, settling time, and rise time are substantially improved, it provides sharp and prompt control over the
temperature. This precise and instant control of temperature has a great impact on the food and medicine industries.
As the temperature could be controlled precisely and instantly, we can avoid the change/degradation of the physical
properties of the materials that are under process.

Keywords Precise · Optimum · PID · Genetic algorithm · Tune

1 Introduction Cohen-Coons as well as Ziegler-Nicholes (ZN) method is

the most well known conventional PID tuning method [3].
Electric furnace is one of the most commonly used fur- Due to the non-linear characteristics of the system, tun-
naces in the industries. It converts the electric energy to ing is a laborious task for the researchers. Besides system
mechanical energy (heat) for heating purpose. Tempera- parameters may change with time. In these cases, an adap-
ture is an exigent parameter that requires control in the tive and intelligent controlling method could be a better
industries and in engineering [1]. If the temperature is option for it. Recently Bio-inspired intelligent computing
not properly controlled, the physical properties of the methods (fuzzy logic (FIS), neural network (NN), genetic
processed material may change/degrade. So, it requires algorithm (GA)) are successfully applied to solve complex
precise and fast control as far as possible. problems [4].
There are manifold controlling methods in industries. Traditional PID controllers are linear and require opera-
Such as proportional-integral-derivative (PID) control, tors experience to adjust the coefficient, which is a time-
internal model control (IMC), predictive control, sliding consuming process [5]. So, it requires the development of
mode control (SMC). But 90% of the industries use PID an optimum tuning method for the PID controller.
controller due to its simplicity, clear functionality and Genetic algorithm is an optimization (minimization)
applicability [2]. technique based on the evolution process. It was first pro-
posed by Holland in 1975 [5] [6]. It is could be a powerful

* Md. Saiful Islam, saiful05eee@gmail.com; Md. Manjurul Gani, gani.cuet@gmail.com; Muhammad Ahsan Ullah, ahsan_cuet@yahoo.com |
1
Chittagong University of Engineering and Technology, Chittagong, Bangladesh.

SN Applied Sciences (2019) 1:880 | https://doi.org/10.1007/s42452-019-0929-y

Received: 21 December 2018 / Accepted: 16 July 2019 / Published online: 20 July 2019

Vol.:(0123456789)
Research Article SN Applied Sciences (2019) 1:880 | https://doi.org/10.1007/s42452-019-0929-y

optimization tool for designing the controllers [3]. It can be three (03) basic gain parameters (­ Kp, ­Ki, ­Kd) Schematic of
used to optimize different error functions to obtain con- PID controller is shown in Fig. 1. If the proportional gain
venient PID parameters. rises, the rise time and the steady-state error decrease. If
Vunlop and Wudhichai in [7] applied Nelder Mead the ­Ki increases, the rise time ­(trt) decreases but the settling
method in search of the suitable value of PID using IAE time ­(tst) and the overshoot increase which can be over-
as the cost function. And they show that their proposed come by the increase of the derivative gain. Here the input
method is more effective than Cohen-Coon method, ZN to the PID controller is e(t) and output is u(t). The relation
method, and Direct Synthesis method. In [8] Wei Jiang between the input and output is shown in Eq. 1 [14].
and Xuchu Jiang implement a Fuzzy tuned PID for a first
∫
de(t)
order system (temperature controlling) and compared its u(t) = kp e(t) + ki e(t)+kd (1)
dt
response with the conventional PID controller. In paper [5]
they apply the genetic algorithm based PID controller for Taking Laplace transformation:
a gas mixture system and found sharp and prompt control
over the conventional PID controllers. The authors in paper E(s)
u(s) = kp E(s) + ki + skd E(s) (2)
[9] present a comparative study between GA PID controller s
and fuzzy self tuned PID controller for DC motor position
controllers. In paper [10] they design a self learning fuzzy u(s) k
= kp + i + skd (3)
algorithm to optimize the PID controller parameters and E(s) s
found better performance. In [11] the authors design PID
controller, GA based PID controller and GA based IMC con- ki
C(s) = kp + + sk d (4)
troller to control the temperature of the heat exchanger s
and they found that GA based IMC controller shows better
performance. 2.2 Genetic algorithm
As the temperature is an important parameter in today’s
industries, it needs to be controlled precisely and quickly. Genetic Algorithms (GAs) are an optimization process that
This paper presents a Genetic Algorithm (GA) based PID follows the process of natural evolution [15]. It is inspired
controller to overcome the low precision, long rise time by the process of selection and the mechanics of genet-
and settling time of the controller. The Genetic Algorithm ics [16]. These are a part of evolutionary computing i.e.
(GA) is used to tune the PID controller. In this algorithm artificial intelligence. The important part of GA is the selec-
integral of absolute error (IAE) is considered as the object tion of proper objective functions. We have to find out the
function. It shows better control over the conventional optimum value of this function. Genetic algorithms search
controllers. As the overshoot, settling time, and rise time for the optimum solution and start blindly in search of best
are substantially improved, it provides sharp and prompt solution from a collection of a potential solution. These
control over the temperature. This precise and instant con- collections of the solutions are known as population and
trol of temperature has a great impact on the food and the each solution of the population is called chromosome.
medicine industries. As the temperature could be con- These chromosomes are compared among themselves
trolled precisely and instantly, we can avoid the change/ based on their fitness value. Then the fittest member is
degradation of the physical properties of the materials selected. GA algorithm composed of 03 basic opera-
that are under process. tions, mutation, crossover, and reproduction [17]. If this
The subsequent sections of this paper describe the fol- member does not meet the stopping criteria, the genetic
lowing: PID controller, Genetic algorithms, temperature
control system, proposed control approach and simula-
tion result and analysis.

2 Literature review

2.1 PID controller

PID controller employs a fundamental control method

with strong control property [12]. PID controllers are
most commonly (around 90% of the controllers) adopted
controller in the industries [13]. PID controller consists of Fig. 1  PID controller

Vol:.(1234567890)
SN Applied Sciences (2019) 1:880 | https://doi.org/10.1007/s42452-019-0929-y Research Article

3. Mutation After crossover, mutation takes place. Muta-

tion means the flipping of the bits of the individuals
i.e. flipping it 0 to 1 or 1 to 0. In the mutation process, it
flips the bits randomly based on small mutation prob-
ability.

Then again it goes to step 02 to evaluate the fitness.

This process goes on until the stopping criteria are met.

2.3 Temperature control system

The electric furnace comprises of a furnace, controller,

temperature, power regulator, sensor, heating coil etc.
Here, the temperature is sensed by the sensor and pro-
duce a voltage, which is negative feedback to the input
(set temperature). Based on the error voltage, the control-
ler regulates the input power to the electric furnace.
The mathematical model of the above furnace can be
described as [7]:
0.15
G(s) = e−1.5s (5)
s2 + 1.1s + 0.2
Fig. 2  Genetic Algorithm (GA)
Here the delay e−1.5s is approximated as:
1.5
algorithm applies its processes to create new population. 1− 2
s 1 − 0.75s
e−1.5s ≈ = (6)
This process goes on until the stopping criterion is met. 1+ 1.5
s 1 + 0.75s
The flow chart of the GA process is presented in Fig. 2 and 2

it is described step by step below: Putting the value from Eq. (6) to Eq. (5) we get the over-
Step 01: Initial Population all system function is as follows.
The First step of the genetic algorithm is to declare the
initial population which contains the possible solutions of −0.1125s + 0.15
G(s) = (7)
the given problem. 0.75s3 + 1.825s2 + 1.25s + 0.2
Step 02: Evaluate the fitness
In this step, fitness of each individual is evaluated. The
individuals i.e. the possible solutions are ranked based on 3 Proposed temperature control approach
the fitness value.
Step 03: Genetic Operation Figure 3 shows the proposed GA based control method
The genetic operations follow the natural evaluation to tune the PID for the electric furnace. Here the plant
process. It includes three basic operations to create new G(s) indicates the transfer function of the temperature
population to find the optimum solution: control system of Fig. 4. The output temperature y(t) is
feedback to the input and compared with the set tem-
1. Selection In this operation, the individuals (chromo- perature x(t). Thus the generated error signal e(t) which
somes) are selected from the initial population for
further processing. Now, the problem is how to select
the chromosomes. According to Darwin’s theory, the
best individuals will survive and breeds new individu-
als (solutions). Generally, individuals with high fitness
value are selected as the parent for crossover.
2. Crossover In crossover operation, it combines two
selected parents to produce offspring (new solutions).
It is a way of creating new solutions from the exist-
ing solutions. The crossover takes place according to
a defined crossover probability. Fig. 3  Proposed GA based PID controller

Vol.:(0123456789)
Research Article SN Applied Sciences (2019) 1:880 | https://doi.org/10.1007/s42452-019-0929-y

Fig. 4  Temperature control system

Fig. 6  Variation of characteristics parameters with the change of ­Ki

Fig. 5  Variation of characteristics parameters with the change of K

­p

Fig. 7  Variation of characteristics parameters with the change of ­Kd

is the difference between the x(t) and y(t), governs the

PID controller. The PID controller acts in such a way that Table 1  Variation of characteristics parameters with the change of
it tries to minimize the error. On the other hand, the GA ­Kp
forms the object function (Integral of Absolute Error Kp Rise time (s) Settling time (s) Percentage
(IAE)) using the error signal to tune the PID controller. It overshoot
is the function which is optimized by genetic algorithm. (%)
The proposed object function is mentioned in Eq. 8. The 1 4.04 12.7 4.69
overall methodology is described step by step below: 2 2.53 15.2 20.8
Step 01: Selection of the gain parameters 3 1.90 18.4 38.9
The PID controller is used in cascade with the elec-
tric furnace. Here the three basic gain parameters of the
PID type controller are ­KP, ­K i, and ­Kd. If ­KP rises, the rise Table 2  Variation of characteristics parameters with the change of
time and the steady-state error decrease. Again, if K ­i ­Ki
increases, the rise time decreases but the overshoot and Ki Rise time (s) Settling time (s) Percentage
the settling time increase, which could be overcome by overshoot
increasing the ­Kd. Following the Figs. 5, 6 and 7 and the (%)
Tables 1, 2 and 3 represents the variations of the char- 0.5 1.79 10.2 0
acteristics parameters of the system with the change of 1.25 1.56 17.5 27.7
gain parameters. From these tables it is apparent that 2 1.43 45.2 61.6
the variations of the characteristics parameters are not
uniform rather these are sometimes increasing, some-
time decreasing, and sometimes fluctuating. is applied to find the suitable (optimum) values of these
As the control characteristics (overshoot, rise time, gain parameters.
and settling time) change with the variation of the gain Step 02: Choosing the object function
parameters, one needs to choose the optimum value of The object function is an indispensable part of the
the gain parameters (­ KP, ­Ki, ­Kd). So, the genetic algorithm genetic algorithm, which affects the functionality of the

Vol:.(1234567890)
SN Applied Sciences (2019) 1:880 | https://doi.org/10.1007/s42452-019-0929-y Research Article

Table 3  Variation of characteristics parameters with the change of

­Kd
Kd Rise time (s) Settling time (s) Percentage
overshoot
(%)

3 1.43 9.82 24.3

5 1.01 7.38 24.6
7 0.694 8 38.4

Table 4  GA parameters

Parameters Values
Fig. 8  Response (step) of the uncontrolled system
Population type Double vector
Fitness scaling (scaling function) Rank Table 6  Obtain parameters Parameters Values
Creation function Constraint dependent without controller
Crossover Single point Rise time 10.2 s
Selection Roulette Overshoot (%) 0
Mutation Uniform Settling time 19.8 s
Population size 20

performance. The optimized value of these gain param-

Table 5  Range of K
­ p, ­Ki and ­Kd Parameters Mini- Maxi- eters are shown in Table 9.
mum mum
value value

Kp 1 4
4 Simulation and analysis
Ki 0 2
The simulation of the proposed GA optimized PID control-
Kd 3 7
ler is accomplished in MATLAB. The step response of the
system is observed under various conditions such as with/
without PID controller. Also, the obtained results using dif-
GA [5]. The genetic algorithm searches for the optimum ferent methods are compared among themselves. Here
(minimum) value of this object function. The genetic the step response means the output of the system for a
algorithm uses the error signal e(t), which is the dif- step input. The step input indicates the input temperature.
ference between the set input and output to form the The output is also in temperature.
object function. It is the function which is optimized by
genetic algorithm. Here the integral of absolute error 4.1 Step response of conventional controller
(IAE) is considered as the object function for optimiza-
tion which is mathematically represented by Eq. (8). Here Fig. 8 shows the step response of the uncontrolled
∞ system. It is apparent from the response that, it is not com-
∫0
IAE = |e(t)|dt (8) petent. The rise time and setting time is too high. Besides,
it has a large offset error which is difference between the
Step 03: Finding the optimum value using GA reference input and output. Here the reference is step
In this step genetic algorithm is applied to tune the input that is one and the output is approximately 0.75.
PID controller i.e. to find the optimum value of ­Kd, ­Ki, and Table 6 presents the values of settling time, rise time and
­K p. Parameters used for the genetic algorithm is given percentage of overshoot that are found for the uncon-
in Table 4. And Table 5 shows the rages of gain param- trolled system.
eters that are used to find the optimum value of them. Figure 9 shows the response of the system that is tuned
There is no specific rule for selecting the population size. by different methods of [7]. And Table 7 represents the
Generally it is selected randomly. Here the population characteristics parameters of the respective response.
size and rages of gain parameters are determined based Here, the ZN method has the largest percentage of over-
on trial and error for which the algorithm shows better shoot. On the other hand, the Nelder-Mead method and
direct synthesis have least amount of overshoot. Though

Vol.:(0123456789)
Research Article SN Applied Sciences (2019) 1:880 | https://doi.org/10.1007/s42452-019-0929-y

The response of the electric furnace with proposed

GA-PID controller is shown in Fig. 10. Here, the perfor-
mance of the proposed method is evaluated based on
the percentage of overshoot, settling time, and rise time.
From Fig. 10 it is found that transient response (settling
time and rise time) is significantly improved and the set-
tling time, percentage of overshoot, and rise time have
the least value compared to the methods mentioned
in [7] and steady-state error is nil. The values of these
characteristics parameters of the proposed method are
shown in Table 10.
Fig. 9  Response (step) of the system with conventional PID control-
ler

Table 7  Characteristics Parameters ZN method [7] Cohen-Coon Direct synthesis Nelder-

parameters of different method [7] method [7] Mead
methods method method [7]

Rise time (s) 1.2927 1.8049 3.0855 1.31

Overshoot (%) 37.3952 17.5964 3.6878 7.0007
Settling time (s) 9.9689 20.8248 9.211 7.65

Table 8  Dependance of the Increasing the Rise time Overshoot Settling time Steady state error
charateristics parameters on parameter
the gain parameters
Kp Decreases Increases Small change Decreases
Ki Decreases Increases Increases Highly reduced
Kd Small change Decreases Decreases Small change

direct synthesis has a minimum amount of overshoot, Table 9  Optimized value of K

­ p, Parameters Optimized values
its rise and settling time is the largest. Comparing the ­Ki and ­Kd
above mention methods Nelder-Mead method shows a Kp 3.4066
better result than the other methods. But its percentage Ki 0.6221
of overshoot is greater than the direct synthesis method. Kd 6.7512
Our proposed GA based PID controller reduces the above
limitations which is presented in the next section.

4.2 Step response of the proposed GA‑PID

controller

As the characteristics of the step response mainly depend

on the value of the gain parameters, these values should
be selected carefully. The dependence of the character-
istics parameters on the gain parameters are shown in
Table 8 [18]. In order that all the characteristics parameters
are in acceptable range, one needs to use the optimum
value of the gain parameters ­(KP, ­Ki, ­Kd). After applying the
genetic algorithm, the obtained optimized value of the
gain parameters are shown in Table 9. Fig. 10  Response of the system (GA-PID controller)

Vol:.(1234567890)
SN Applied Sciences (2019) 1:880 | https://doi.org/10.1007/s42452-019-0929-y Research Article

Table 10  Obtain parameters Parameters Values

with proposed method
Rise time (s) 1.14
Overshoot (%) 3.27
Settling time (s) 7.21

Table 11  Comparison of different methods

Parameters Cohen- Direct Nelder- Pro-
Coon synthesis Mead posed
method [7] method [7] method [7] method

Rise time (s) 1.8049 3.0855 1.31 1.14

Overshoot 17.5964 3.6878 7.0007 3.27
(%) Fig. 12  Response of the system with proposed GA-PID controller
Settling time 20.8248 9.211 7.65 7.21
(s)
5 Conclusion and future scope

This research proposed an IAE based genetically opti-

mized PID controller to control the temperature of the
electric furnace. The IAE is considered as the object func-
tion for the optimization of error. From this optimization
we acquire the optimum value of ­Kp = 3.4066, ­Ki = 0.6221,
and ­Kd = 6.7512. For these values of the gain parameters,
the step response of the system exhibits better result than
conventional methods. Besides using the Routh-Hurwitz
criterion for stability it is found that the system is stable.
This sharp and prompt control over the temperature has a
great impact on the food and medicine industries. As the
temperature could be controlled precisely and instantly,
we can avoid the change/degradation of the physical
Fig. 11  Step response of the system in different methods
properties of the materials that are under process. Due to
the long term operation of a process, its parameters may
change with time. This will affect the precision as well as
4.3 Performance comparison fast response in transient period. This needs further inves-
tigation to overcome the above limitation.
Here Table 11 shows the comparative analysis of the
proposed method with other methods such as Cohen-
Coon method, Direct synthesis method, Nelder Mead Compliance with ethical standards
method that are shown in [7]. The performance of the
PID controller is appraised based on settling time, per- Conflict of interest The authors declare they have no conflict of in-
centage of overshoot, and rise time. From the Table 11, terest.
it is apparent that the proposed method has minimum
rise time 1.14 s, settling time 7.21 s and percentage of
overshoot 3.27% i.e. the proposed GA-PID controller
shows more refined performance than other methods References
(Fig. 11).
1. Li J, Yan C, Liu J (2012) Design of temperature control system
Figure 12 represents the values of the gain param- based on fuzzy PID, advanced materials research, vol 418–420.
eters ­(KP, ­Ki, ­Kd) that are used in different methods of [7]. Trans Tech Publications, Switzerland, pp 1756–1759
From the above simulation and data, it is apparent 2. Jayachitra A, Vinodha R (2014) Genetic algorithm based PID
that the proposed genetic algorithm based optimiza- controller tuning approach for continuous stirred tank reactor,
Hindawi Publishing Corporation. Adv Artif Intell 2014, 791230
tion technique exhibits better performance than other 3. Sabura Banu U, Uma G (2007) ANFIS gain scheduled CSTR with
methods. genetic algorithm based PID minimizing integral square error,

Vol.:(0123456789)
Research Article SN Applied Sciences (2019) 1:880 | https://doi.org/10.1007/s42452-019-0929-y

In: IET-UK international conference on information and com- computation of power, energy, information and communication
munication technology in electrical sciences (ICTES 2007), Dr. (ICCPEIC), Chennai, pp 489–495
M.G.R. University, Chennai, Tamil Nadu, India. Dec. 20–22, pp 12. Han Y, Jinling J, Guangjian C, Xizhen C (2011) Temperature
57–62 control of electric furnace based on fuzzy PID. In: International
4. Rahmat MF, Yazdani AM, Movahed MA, Mahmoudzadeh S (2011) conference on electronics and optoelectronics (ICEOE 201 1),
Temperature control of a continuous stirred tank reactor by 2011, Proceedings of 2011 international conference on electron-
means of two different intelligent strategies. Int J Smart Sens ics and optoelectronics, pp 41–44
Intell Syst 4(2):244–267 13. Vahedpour M, Ranjbar Noei A, Agheh Kholerdi H (2015) Compar-
5. Chunchen W, Feng C, Guang Z, Ming Y, Liangzhe L, Taihu W ison between performance of conventional, fuzzy and fractional
(2017) Design of genetic algorithm optimized PID controller for order PID controllers in practical speed control of induction
gas mixture system. In: 2017 IEEE 13th international conference motors. In: 2nd international conference on knowledge-base
on electronic measurement & instruments, 13th IEEE interna- engineering and innovation (KBEI) November 5–6, pp 912–916
tional conference on electronic measurement & instruments 14. Jaen-Cuellar AY, de Romero-Troncoso RJ, Morales-Velazquez L,
(ICEMI), 20–22 Oct. 2017, pp 6–9 Osornio-Rios RA (2013) PID-controller tuning optimization with
6. Ayam M, Liqaa SM, Raoof TH (2016) Speed control for separately genetic algorithms in servo systems. Int J Adv Robot Syst 11:324
excited DC motor with PID controller, GA, And PSO. Int J Recent 15. Elsrogy WM, Fkirin MA, Moustafa Hassan MA (2013) Speed con-
Sci Res 7(7):12673–12678 trol of DC motor using PID controller based on artificial intel-
7. Sinlapakun V, Assawinchaichote W (2015) Optimized PID con- ligence techniques. In: International conference on control,
troller design for electric furnace temperature systems with decision and information technologies (CoDIT), pp 196–201
Nelder Mead Algorithm. In: 12th international conference on 16. Meena DC, Devanshu A (2017) Genetic algorithm tuned PID con-
electrical engineering/electronics, computer, telecommunica- troller for process control. In: International conference on inven-
tions and information technology (ECTI-CON), 24–27 June 2015, tive systems and control, international conference on inventive
pp 1–4 systems and control (ICISC), pp 1–6
8. Jiang W, Jiang X (2012) Design of an intelligent temperature con- 17. Mustafa Kamal Md, Mathew L, Chatterji S (2014) Speed control
trol system based on the fuzzy self-tuning PID. In: International of brushless DC motor using intelligent controllers. In: Students
symposium on safety science and engineering in China, 2012 conference on engineering and systems. pp 1–5
(ISSSE-2012), Procedia Engineering 43, pp 307–311 18. Sharma K, Palwalia DK (2017) A modified PID control with adap-
9. Flores-Morán E, Yánez-Pazmiño W, Barzola-Monteses J (2018) tive fuzzy controller applied to DC motor. In: Proceedings of
Genetic algorithm and fuzzy self-tuning PID for DC motor posi- the international conference on information, communication,
tion controllers. In: 2018 19th international carpathian control instrumentation and control (ICICIC), Indore, India, 17–19 Aug.
conference (ICCC), Szilvasvarad, pp 162–168 2017, pp 1–6
10. Deng C, Wang Z (2018) Self-learning fuzzy algorithm opti-
mized temperature control and efficiency monitoring of heat Publisher’s Note Springer Nature remains neutral with regard to
exchanger. In: 2018 37th Chinese control conference (CCC), jurisdictional claims in published maps and institutional affiliations.
Wuhan, pp 3394–3399
11. Valarmathi R, Theerthagiri PR, Rakeshkumar S, Gomathi V (2018)
Design of genetic algorithm based internal model control-
ler for a heat exchanger. In: 2018 international conference on

Vol:.(1234567890)