Soft Computing

https://doi.org/10.1007/s00500-019-04156-3 (0123456789().,-volV)(0123456789().
,- volV)

FOCUS

Optimization of fuzzy controller design using a Differential Evolution

algorithm with dynamic parameter adaptation based on Type-1
and Interval Type-2 fuzzy systems
Patricia Ochoa1 • Oscar Castillo1 • José Soria1

 Springer-Verlag GmbH Germany, part of Springer Nature 2019

Abstract
This paper proposes the use of the Differential Evolution algorithm with fuzzy logic for parameter adaptation in the optimal
design of fuzzy controllers for nonlinear plants. The Differential Evolution algorithm is enhanced using Type-1 and
Interval Type-2 fuzzy systems for achieving dynamic adaptation of the mutation parameter. In this paper, four control
optimization problems in which the Differential Evolution algorithm optimizes the membership functions of the fuzzy
controllers are presented. First, the experiments were performed with the original algorithm, second the experiments were
performed with the Fuzzy Differential Evolution (in this case the mutation parameter is dynamic), and last, experiments
were performed applying noise to the control plant by using Fuzzy Differential Evolution.

Keywords Differential Evolution  Problem control  Fuzzy control  Parameters

1 Introduction have given control systems a certain degree of intelligence

and flexibility.
The design of fuzzy controllers is an important application The main idea of fuzzy set theory was originally pro-
area for metaheuristic algorithms. In this work, we use the posed by Zadeh in 1965 and was first applied to control
Differential Evolution algorithm enhanced with Type-1 theory in 1974, by Mamdani (1974), Zadeh
and Interval Type-2 fuzzy systems for dynamic parameter (1975a, b, 1978) and Dubois et al. (2000). Based on these
adjustment and its application to fuzzy system design for works, the fuzzy controllers have been successfully applied
three control problems. A comparison is made between the in numerous problems by Gao (2017), Precup et al. (2017),
original algorithm and the proposed algorithm, and finally, Bui et al. (2017) and Driankov and Palm (2013). In Mohan
noise was also added to the control system using the pro- and Sinha (2008), an analytical structure and stability
posed algorithm. analysis of a fuzzy PID controller is presented.
The nonlinear characteristics of ill-defined and complex Recently, the DE algorithm has been applied in different
modern plants make classical controllers inadequate for problems like: Juang et al. (2015), Bi et al. (2014), Sun
such systems because they require complicated mathe- et al. (2014), Liu and Lampinen (2005), Tang et al. (2014),
matical models. However, the use of fuzzy sets and fuzzy Wang et al. (2018), Salehpour et al. (2017) and Aalto and
logic principles have enabled researchers to understand Lampinen (2013, 2014) just to mention some recent works.
better and hence control complex systems that are difficult Similarly, the use of fuzzy logic for different applica-
to model. These newly developed fuzzy logic controllers tions has been increasingly relevant, and we can find the
use of fuzzy logic in areas such as medicine, control,
robotics and artificial intelligence, being the control area
our main interest (Sa-ngiamvibool 2017; Cuevas et al.
Communicated by O. Castillo, D. K. Jana. 2017; Caraveo et al. 2016; Noshadi et al. 2016). In our
case, we focus for this paper on the most frequent use that
& Oscar Castillo is combination of fuzzy logic and metaheuristic algorithms,
[email protected]
since it has been demonstrated that both combined
1
Tijuana Institute of Technology, Tijuana, Mexico methodologies improve the performance of the algorithms,

123
 P. Ochoa et al.

but in particular the use of Type-2 fuzzy systems to men- 2 Fuzzy logic
tion some works of interest (Martı́nez-Soto et al. 2014;
Melin et al. 2013; De La et al. 2017; Olivas et al. 2019; A fuzzy logic system (FLS) that is defined entirely in terms
Castillo et al. 2016a, 2019a, b; Ontiveros-Robles et al. of Type-1 fuzzy system is known as a Type-1 fuzzy logic
2018; Sanchez et al. 2015a; Cervantes and Castillo 2015; system (Type-1 FLS) by Mendel (2014), Mendel and Liu
Amador-Angulo and Castillo 2015) by referring to some (2013) and its elements are defined in Fig. 1.
works of the literature. A Type-2 fuzzy set, Ã, is characterized by
The purpose of this work is to ascertain that the use of
A ¼ fðx; uÞ; u A ðx; uÞj8x 2 X; 8u 2 Jx ½0; 1g ð1Þ
the fuzzy logic combined with the Differential Evolution
algorithm applied to control problems is better than the where 0  u A ðx; uÞ  1.
original algorithm and that the use in particular of Type-2 In a general way, we can say that a Type-2 fuzzy set is a
fuzzy systems helps in problems under a higher level of generalization of a fuzzy set that is associated with a sec-
uncertainty, and for this work the control problems are ondary source of uncertainty related to the definition of a
considered and noise is added to increase their level of set A. This additional source of uncertainty is represented
complexity. In most of the articles mentioned above, the as a secondary membership function. Type-2 fuzzy system,
good performance of combining fuzzy logic and meta- whose secondary membership function is an interval, is
heuristics has been demonstrated, and the novelty of this called an Interval Type-2 fuzzy system and is expressed by
paper is that there are no works in the literature that two membership functions, where one represents the
combine Type-2 fuzzy system and the Differential Evolu- degree of belonging in X and the other gives a weighting to
tion algorithm applied to control problems. each of the Type-1 fuzzy systems. Figure 2 shows the
Previously we worked with the Differential Evolution architecture of an Interval Type-2 fuzzy system.
algorithm combined with fuzzy logic, but applied only to The output processor includes a type reducer and a
benchmark mathematical functions, and in these works the defuzzifier; it generates a Type-1 fuzzy set output (from the
efficiency of the proposed method was proved both for type reducer) or a crisp number (from the defuzzifier).
traditional functions and for competition functions like The main difference between the Type-1 and the Inter-
those of the CEC conference, comparing the original val Type-2 fuzzy systems is the footprint of uncertainty
method and the proposed method. The main contribution (FOU) that defines the uncertainty of A as the union of all
and difference in this work with respect to the previous the primary memberships, which is limited by two mem-
ones are that now we have applied the proposed method bership functions: an upper membership function and lower
which we call in previous works Fuzzy Differentia Evo- membership function. In addition, in the Interval Type-2
lution algorithm to more complex control problems with fuzzy system, the defuzzifier block of a Type-1 is replaced
and without noise, with the advantage offered by using an by the processing block output consisting of a type reducer
Interval Type-2 fuzzy system which dynamically moves followed by defuzzifier. It is important to mention that the
the parameters that were originally fixed during the exe- inference operation in the Interval Type-2 fuzzy system is
cution of the algorithm, and then, we made them dynamic much more complicated than in Type-1 fuzzy system and
with a Type-1 fuzzy system in previous works. This paper that a type reducer is necessary for a Type-2 fuzzy system
also highlights the use of the one main parameter (F) of the to convert Type-2 fuzzy sets into Type-1. These differences
algorithm, which is dynamic using an Interval Type-2 are the ones that make the calculations of an Interval Type-
fuzzy system, since previously we worked with each of the 2 fuzzy system more challenging.
parameters separately to give us an idea of the behavior of
each parameter when using a fuzzy system, and with the
knowledge acquired in said experimentation, the combi-
nation of these two parameters was made in a single fuzzy
system.
The structure of the paper is as follows: Sect. 2
describes the theory of Type-1 and the Interval Type-2
fuzzy systems. Section 3 presents a description of the
Differential Evolution algorithm. Section 4 describes the
control problems used for experimentation. Section 5 pre-
sents the obtained results from the experimentation with
each one of the control problems. Section 6 shows a sta-
tistical test of the results obtained, and finally, a discussion
of results is presented in Sect. 8 with the conclusions. Fig. 1 Type-1 fuzzy logic system

123
Optimization of fuzzy controller design using a Differential Evolution algorithm with dynamic…

3.3 Mutation
 
vi;g ¼ xr0 ;g þ F  xr1 ;g  xr2 ;g ð6Þ

3.4 Crossover
  
vj;i;g if randj ð0; 1Þ  Cr or j ¼ jrand
ui;g ¼ uj;i;g ð7Þ
xj;i;g otherwise

3.5 Selection
    
ui;g if f ui;g  f xi;g
xi;gþ1 ¼ ð8Þ
xi;g otherwise
Figure 3 corresponds to the pseudocode of the original
Differential Evolution algorithm, which summarizes more
clearly the equations mentioned above.
In summary, we can say that the Differential Evolution
algorithm involves defining a population of NP vectors,
these vectors are initialized, and the lower and upper limits
have to be previously defined depending on the problem.
Then, three individuals are selected, mutation and cross-
over operations are applied, and if the resulting value is
better than the one chosen for replacement, then it replaces
it. Otherwise, the main individual is retained.
Fig. 2 Type-2 fuzzy system

3 The Differential Evolution algorithm 4 Benchmark control problems

Differential Evolution is basically composed of four steps For this article, we use four Benchmark control problems,
(Price et al. 2006): initialization, mutation, crossing and which are considered to validate the efficiency of our
selection. proposed algorithm and make a comparison of results
This is the mathematical representation of the Differ- between Type-1 and Interval Type-2 fuzzy systems.
ential Evolution algorithm by Price et al. (2006):
4.1 Water tank controller
3.1 Population structure
  The first case study is of the water tank controller, whose
Px;g ¼ xi;g ; i ¼ 0; 1; . . .; Np  1; g ¼ 0; 1; . . .; gmax ; main objective is controlling the water level in a tank, and
 
xi;g ¼ xj;i;g ; j ¼ 0; 1; . . .; D  1 Fig. 4 graphically represents the way in which the valve
ð2Þ operates and hence the filling process in the tank by Fierro
  and Castillo (2013) and Amador-Angulo and Castillo
Pv;g ¼ vi;g ; i ¼ 0; 1; . . .; Np  1; g ¼ 0; 1; . . .; gmax ; (2017).
 
vi;g ¼ vj;i;g ; j ¼ 0; 1; . . .; D  1 The mathematical model of the water tank controller is
ð3Þ presented as follows:
  d dH pffiffiffiffi
Pu;g ¼ ui;g ; i ¼ 0; 1; . . .; Np  1; g ¼ 0; 1; . . .; gmax ; Vol ¼ A ¼ bV  a H ð9Þ
  ð4Þ dt dt
ui;g ¼ uj;i;g ; j ¼ 0; 1; . . .; D  1
where Vol is the volume of water in the tank, A is the cross-
sectional area of the tank, b is a constant related to the flow
3.2 Initialization rate into the tank and a is a constant related to the flow rate
  out of the tank. The equation describes the height of water
xj;i;0 ¼ randj ð0; 1Þ  bj;U  bj;L þ bj;L ð5Þ
H as a function of time, due to the difference between flow
rates into and out of the tank.

123
 P. Ochoa et al.

Fig. 3 Pseudocode Differential

Evolution algorithm

central one is triangular. The second input is called flow

and is composed of three membership functions which are
called soft which is of trapezoidal type, good which is of
triangular type and hard which is of trapezoidal type.
The combination of rules shown in Fig. 8 simulates the
speed of the water flow at the desired temperature.

4.3 Mobile robot controller

The third control problem is the case of a mobile robot; in

this case, the plant is of a unicycle mobile robot by Sanchez
et al. (2015b), consisting of two driving wheels located on
the same axis and a front free wheel, and Fig. 9 shows a
Fig. 4 Graphical illustration of the water tank controller graphical description of the robot model.
The robot body is symmetrical around the perpendicular
Figure 5 illustrates the structure of the fuzzy system for axis, and the center of mass is at the geometric center of the
this control problem, the way in which the membership body. It has two driving wheels that are fixed to the axis
functions are granulated, the type of functions and how many that passes through the center of mass ‘‘C’’ represented by
membership functions correspond to each input and output. {C, Xm, Ym}, and one passive wheel that prevents the robot
The five fuzzy rules of the controller are presented in Fig. 6. from tipping over as it moves on a plane.
The fuzzy system designed for the water tank controller The mathematical model of the robot is given by the
is built based on the actual filling behavior of a water tank, following equation:
and the set of rules are made in terms of the theory of how M ðqÞv_ þ C ðq; q_ Þv þ Dv ¼ s þ PðtÞ ð10Þ
this process is performed.
where q ¼ ðx; y; hÞT is the vector of the configuration
4.2 Temperature controller coordinates. v ¼ ðv; wÞT is the vector of velocities. s ¼
 
s1; s2 is the vector of torques applied to the wheels of the
The second case study is a temperature controller, whose robot where s1 and s2 denote the torques of the right and
main objective is to control the temperature in the water left wheels, respectively. P 2 R2 is the uniformly bounded
flow. Figure 7 shows the fuzzy system for the temperature disturbance vector. M ðqÞ 2 R2X2 is the positive-definite
controller, the system contains two inputs and two outputs, inertia matrix. C ðq; q_ Þ# is the vector of centripetal and
it is of Mamdani type, and the rules are presented in Fig. 8 Coriolis forces. D 2 R2X2 is a diagonal positive-definite
by Castillo et al. (2016a). damping matrix.
The fuzzy system of this controller contains two inputs: The kinematic system is represented by Eq. (11):
the first one is called temperature and is composed of three 2 3
membership functions, which are called cold, good and cos h 0 h i
v
hot: the two membership functions are trapezoidal and the q_ ¼ 4 sin h 0 5 ð11Þ
w
0 1

123
Optimization of fuzzy controller design using a Differential Evolution algorithm with dynamic…

Fig. 5 Fuzzy control system for the water tank

static state-feedback controller exists that can stabilize the

closed-loop system around the equilibrium point.

4.4 Inverted pendulum control problem

This control problem, unlike the aforementioned ones, is of

Fig. 6 Rules of the water tank controller
the Takagi–Sugeno type, which makes the level of com-
plexity increase. In addition, complexity originates from the
where (x, y) is the position in the X–Y (world) reference nonlinear nature of the problem. The objective of this
frame, h is the angle between the heading direction and the problem is to keep the pendulum in balance without falling
x-axis and v and w are the linear and angular velocities, off the car. This plant consists of a straight-line rail, a cart, a
respectively. pendulum and a driving unit. The cart can move left or right
Equation (12) represents the non-holonomic constraint, on the rail freely. The pendulum is hinged on the center of the
which this system has, which corresponds to a non-slip top surface of the cart and can rotate around the pivot in the
wheel condition preventing the robot from moving same vertical plane with the rail. Given that no friction exists
sideways. in the system, Eqs. (13, 14) represent the mathematical
y_ cos h  x_ sin h ¼ 0 ð12Þ model of the inverted pendulum by Caraveo et al. (2016).
   
Figure 10 illustrates the fuzzy system for the robot /¼ mc þ mp gsigh  F þ mp lp x2 sin h
n   o ð13Þ
controller, and this controller is composed of two inputs cos h 4=3 mc þ mp  mp ðcos hÞ2 lp ;
and two outputs. The first input is ev (error in the linear  
velocity), and the second is ew (error in the angular a ¼ 4=3 F þ mp lp x2nsin h o
velocity), which have three membership functions with  
 mp g sin h cos h 4=3 mc þ mp  mp ðcos hÞ2
linguistic values of N, Z and P. The first output is t1 (torque
1) and the second output is t2 (torque 2) with three mem- ð14Þ
bership functions with the same linguistic values, it is of The parameters mc and mp are, respectively, the mass of
Mamdani type, and Fig. 11 shows the fuzzy rules. the cart and the mass of the pendulum in the unit (kg), and
The system fails to meet Brockett’s necessary condition g = 9.8 m/s2 is the gravity acceleration. The parameter lp is
for feedback stabilization, which implies that no continuous the length from the center of the pendulum to the pivot in

123
 P. Ochoa et al.

Fig. 7 Fuzzy system for the temperature controller

respectively. The variables x, v and a denote the position of

the cart from the rail origin, its velocity and its accelera-
tion, and right direction is positive.
Figure 12 shows the main idea of the inverted pendulum
controller.
The fuzzy system structure of the controller is shown in
Fig. 13, which contains four inputs which are pendulum
angle, angular velocity, cart position and cart velocity, in
Fig. 8 Fuzzy rules of the temperature controller which all membership functions are Gaussian type, and one
output that contains 16 linear functions.
The rule set for this controller is shown in Fig. 14,
which simulates the vehicle’s behavior on the cart to
maintain balance.
The combination of the rules in Fig. 14 is given by the
following meanings:

in1 = Pendulum angle in1mf1 = Low

in2 = Angular velocity in1mf2 = High
in3 = Car position in2mf1 = Low
in4 = Car velocity in2mf2 = High
Fig. 9 Mobile robot model in3mf1 = Low
in3mf2 = High
the unit (m) and equals to the half-length of the pendulum. in4mf1 = Low
The variable F represents the driving force in the unit in4mf2 = High
(N) applied horizontally to the cart. The variables h, x and
a represent, respectively, the angle of the pendulum from
upright position, its angular velocity and its angular
acceleration, and the clockwise direction is positive,

123
Optimization of fuzzy controller design using a Differential Evolution algorithm with dynamic…

Fig. 10 Fuzzy system for the robot controller

outmf 2 : 40:41in1 þ 10:05in2 þ 3:162in3 þ 4:288in4

þ 0:2068
outmf 3 : 41:37in1 þ 10:03in2 þ 3:162in3 þ 4:288in4
þ 0:3386
outmf 4 : 41:41in1 þ 10:05in2 þ 3:162in3 þ 4:288in4
þ 0:2068

Fig. 11 Rules of the robot controller outmf 5 : 38:56in1 þ 10:18in2 þ 3:162in3 þ 4:288in4
 0:04893
outmf 6 : 37:6in1 þ 10:15in2 þ 3:162in3 þ 4:288in4
They are represented the 16 Sugeno coefficients in the  0:1807
outputs corresponding to the combination of the rules outmf 7 : 38:56in1 þ 10:18in2 þ 3:162in3 þ 4:288in4
shown in Fig. 14:  0:04893
outmf 1 : 41:37in1 þ 10:03in2 þ 3:162in3 þ 4:288in4
outmf 8 : 37:6in1 þ 10:15in2 þ 3:162in3 þ 4:288in4
þ 0:3386
 0:1807

Fig. 12 Inverted pendulum

controller

123
 P. Ochoa et al.

Fig. 13 Fuzzy system for the pendulum control problem

Fig. 14 Rules of the pendulum

control problem

outmf 9 : 37:6in1 þ 10:15in2 þ 3:162in3 þ 4:288in4 outmf 11 : 37:6in1 þ 10:15in2 þ 3:162in3 þ 4:288in4
þ 0:1807 þ 0:1807
outmf 10 : 38:56in1 þ 10:18in2 þ 3:162in3 þ 4:288in4 outmf 12 : 38:56in1 þ 10:18in2 þ 3:162in3 þ 4:288in4
þ 0:04891 þ 0:04892

123
Optimization of fuzzy controller design using a Differential Evolution algorithm with dynamic…

Fig. 15 Original algorithm with

a dynamic F parameter

outmf 13 : 40:41in1 þ 10:05in2 þ 3:162in3 þ 4:288in4 to dynamically move the F parameter, first with Type-1
 0:2068 fuzzy logic and then with Interval Type-2 fuzzy system,
which is in charge of dynamically moving the F parameter.
outmf 14 : 41:37in1 þ 10:03in2 þ 3:162in3 þ 4:288in4 Figure 15 shows the structure of the original algorithm
 0:3386 and the fuzzy systems used for experimentation.
outmf 15 : 40:41in1 þ 10:05in2 þ 3:162in3 þ 4:288in4 Equation (15) expresses the way in which the input of
 0:2068 the fuzzy system is calculated with which the F parameter
is obtained as output, and in this way it can be used in the
outmf 16 : 41:37in1 þ 10:03in2 þ 3:162in3 þ 4:288in4 algorithm.
 0:3386
Current generations
Generations ¼ ð15Þ
Maximum of generatios
Figure 16 represents the pseudocode in which the fuzzy
5 Experimentation with control problems system is used to dynamically adapt F parameter.
The Type-1 fuzzy system is explained in more detail in
For the experiments, we use the proposed method, which
previous works (Castillo et al. 2016b); mathematical
we call FDE, which has as main dynamic adjustment the
Eq. (16) of the membership functions and their parameters
F (mutation) parameter of the original algorithm, which is
is included. In the same way, our Interval Type-2 fuzzy
recommended to be used in a range of [0, 1]. The proposed
system contains one input and one output, mathematical
method makes the F parameter to be moved dynamically
Eq. (17) represents the knowledge of the membership
during the execution of the algorithm within the range [0,
function, and Tables 1 and 2 contain the parameters of the
1]. We have previous works where we have used our
functions of Type-1 and Interval Type-2 fuzzy system.
proposed FDE method for Benchmark problems, and we 8
obtained good results by Ochoa et al. (2016a, b, 2017 and >
> 0; xa
>
> xa
Castillo et al. (2016b). < ; axb
a
trimf ðx; a; b; cÞ ¼ bc  ð16Þ
For this work, we use our FDE method to change the > x
>
> ; bxc
parameter values of the membership functions for each of >
:c  b
0; cx
the fuzzy controllers. A comparison is made where we use

123
 P. Ochoa et al.

Fig. 16 Pseudocode for the

Fuzzy Differential Evolution

Table 1 Parameters of the membership functions for the Type-1 Table 2 Parameters of the membership functions of the Interval
fuzzy system Type-2 fuzzy system
8
Low > 0; x   0:5 Low l1 ð xÞ ¼ max min x0:5 0:4x
>
> 0:08þ0:5 ; 0:4þ0:08 ;0
>
> x þ 0:5
< ; 0:5  x  0
trimf ðx; a; b; cÞ ¼ 0 þ 0:5
xþ0:4 0:5x
l2 ð xÞ ¼ max min 0:08þ0:4 ; 0:50:08 ;0
>
> 0:5  x
>
> ; 0  x  0:5
>
: 0:5  0 lð xÞ ¼ maxðl1 ð xÞ; l2 ð xÞÞ8x 62 ð0:08; 0:08Þ
0; 0:5  x
8 lð xÞ ¼ 18x 2 ð0:08; 0:08Þ
Medium > 0; x0
>
> x0 lð xÞ ¼ minðl1 ð xÞ; l2 ð xÞÞ
>
< ; 0  x  0:5
trimf ðx; a; b; cÞ ¼ 0:5 0 Medium xþ0:83
l1 ð xÞ ¼ max min 0:4þ0:83 0:92x
; 0:920:4 ;0
>
> 1x
>
> ; 0:5  x  1   x0:08 1:07x  
: 1  0:5 l2 ð xÞ ¼ max min 0:50:08 ; 1:070:5 ; 0
0; 1x
8 lð xÞ ¼ maxðl1 ð xÞ; l2 ð xÞÞ8x 62 ð0:4; 0:5Þ
High > 0; x  0:5
>
> x  0:5 lð xÞ ¼ 18x 2 ð0:4; 0:5Þ
>
< ; 0  x  0:5
 0:5
trimf ðx; a; b; cÞ ¼ 11:5 lð xÞ ¼ minðl1 ð xÞ; l2 ð xÞÞ
>
>  x   x0:4  
>
> ; 1  x  1:5 High l1 ð xÞ ¼ max min 0:920:4 1:4x
; 1:40:92 ;0
: 1:5  1
0; 1:5  x   x0:5 1:5x
 
l2 ð xÞ ¼ max min 1:070:5 ; 1:51:07 ; 0
lð xÞ ¼ maxðl1 ð xÞ; l2 ð xÞÞ8x 62 ð0:92; 1:07Þ
lð xÞ ¼ 18x 2 ð0:92; 1:07Þ
lð xÞ ¼ minðl1 ð xÞ; l2 ð xÞÞ
h i
l~ð xÞ ¼ lð xÞ; lð xÞ ¼ itristype2ðx; ½a1 ; b1 ; c1 ; a2 ; b2 ; c2 Þ
ð17Þ
in low, medium and high, at the input and output, and
where a1 \a2 ; b1 \b2 ; c1 \c2
  Fig. 17 illustrates the Interval Type-2 fuzzy system.
x  a1 c 1  x The rules of the fuzzy system and the surface repre-
l1 ð xÞ ¼ max min ; ;0
b1  a1: c1  b1 senting the fuzzy model are presented in Figs. 18 and 19,
  and the rules are defined so that the F variable changes in a
x  a2 c 2  x
l2 ð xÞ ¼ max min ; ;0 decreasing form.
b2  a2: c2  b2
The method used to perform the experiments is as fol-
lð xÞ ¼ maxðl1 ð xÞ; l2 ð xÞÞ8x 62 ðb1 ; b2 Þ lows: the Differential Evolution algorithm initializes all
parameters, and for the case where the F parameter is
lð xÞ ¼ 18x 2 ðb1 ; b2 Þ
dynamic the fuzzy system performs that task (Type-1 or
lð xÞ ¼ minðl1 ð xÞ; l2 ð xÞÞ Interval Type-2 as appropriate). Subsequently, the algo-
rithm searches for the best structure for the membership
Tables 1 and 2 contain the parameters of the functions
functions of the controller, and finally, the controller is
of the Type-1 and Interval Type-2 fuzzy system, both fuzzy
simulated with the structure proposed by the algorithm, and
systems have one input and one output, they are granulated

123
Optimization of fuzzy controller design using a Differential Evolution algorithm with dynamic…

Fig. 17 Interval Type-2 fuzzy system

original algorithm, then we used Type-1 fuzzy system to

modify the F variable (mutation), and finally, we used
Interval Type-2 fuzzy system to modify the F variable, and
to validate the efficiency of the proposed method we add
noise to each of the control problems and experiments are
Fig. 18 Rules for the Interval Type-2 fuzzy system
also performed with Type-1 and Interval Type-2 fuzzy
systems.
Table 3 shows the parameters for each of the variants of
the algorithms where D is the number of dimensions, NP is
the number of elements in the population, F is the mutation
which is dynamic for the Type-1 and the Interval Type-2
fuzzy systems, CR is the crossover parameter, G is the
number of generations and noise represents the level the
noise in the controller plant, and this noise is a uniform
random number value in decibels.
In this paper, we use the following abbreviations for
each one of the variants with which the experiments were
performed:
• DE: Differential Evolution
• DE ? T1FS: Differential Evolution with Type-1 fuzzy
system
Fig. 19 Surface of the fuzzy system • DE ? T1FS ? N: Differential Evolution with Type-1
fuzzy system plus noise
Fig. 20 illustrates the general idea of how the complete • DE ? T2FS: Differential Evolution with Interval Type-
process is working. 2 fuzzy system
For each of the control problems, the objective function • DE ? T2FS ? N: Differential Evolution with Interval
is defined by the calculation of the mean squared error Type-2 fuzzy system plus noise
(RMSE), which is expressed in Eq. (18).
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Tables 4, 5, 6 and 7 represent the results for each of the
u N
u1 X control problems described above, and in each table the
RMSE ¼ t ðxt  x^t Þ2 ð18Þ best result, the worst results, the averages and the standard
N t¼1
deviations are presented.
The presented results are the average of 30 experiments Figure 21 shows the plots for each of the best results
performed for each of the variants. The methodology of our obtained for the water tank controller, where the blue line
work is as follows: we performed experiments with the represents the trajectory to follow and the pink line is the

123
123
Fig. 20 Fuzzy controller experimentation
P. Ochoa et al.
Optimization of fuzzy controller design using a Differential Evolution algorithm with dynamic…

Table 3 Parameters of the algorithm Figure 23 shows the plots for each of the best results
D NP F CR GEN Noise
obtained for the temperature controller, where the blue line
represents the trajectory to follow and the pink line is the
DE 30 50 0.5 0.5 50 – result obtained using the algorithm with each of its
DE ? T1FS 30 50 Dynamic 0.5 50 – variants.
DE ? T1FS ? N 30 50 Dynamic 0.5 50 0.50 For the case of the temperature controller, it is observed
DE ? IT2FS 30 50 Dynamic 0.5 50 – that Interval Type-2 fuzzy system with noise is better, in
DE ?IT2FS ? N 30 50 Dynamic 0.5 50 0.50 comparison with the other variations, although the differ-
ence between all the results is minimal and the plot in
result obtained using the algorithm with each of its Fig. 24 illustrates all the best results.
variants. Table 6 represents the results obtained for the experi-
Figure 22 shows the behavior of the best result of the ments carried out with the mobile robot controller, and
different experiments performed with the water tank con- Figs. 25 and 26 illustrate the simulation of the best results
troller. We can observe that the Interval Type-2 fuzzy obtained for each of the variations made; then, a plot of the
systems without noise and with noise are better than the best results obtained by each one of the variations is pre-
original DE algorithm and the algorithm using Type-1 sented, respectively.
fuzzy system. Table 7 represents the results obtained for the experi-
ments carried out with the inverted pendulum controller,

Table 4 Results for the water

DE DE ? T1FS DE ? T1FS ? N DE ? IT2FS DE ? IT2FS ? N
tank controller
RMSE
Best 1.38E-02 2.45E-04 5.66E-03 5.40E-02 1.23E-03
Worst 2.45E-01 1.21E-01 6.37E-02 6.50E-02 1.46E-01
Mean 1.12E-01 3.40E-02 6.00E-02 6.02E-02 3.22E-02
SD 5.98E-02 3.00E-02 2.46E-03 2.45E-03 3.63E-02

Table 5 Results for the

DE DE ? T1FS DE ? T1FS ? N DE ? IT2FS DE ? IT2FS ? N
temperature controller
RMSE
Best 5.42E-02 5.40E-02 5.40E-03 5.40E-02 5.66E-03
Worst 6.49E-02 6.49E-02 6.48E-02 6.50E-02 6.37E-02
Mean 6.11E-02 6.18E-02 6.14E-02 6.02E-02 6.00E-02
SD 2.09E-03 2.51E-03 2.68E-03 2.45E-03 2.46E-03

Table 6 Results for the mobile

DE DE ? T1FS DE ? T1FS ? N DE ? IT2FS DE ?IT2FS ? N
robot controller
RMSE
Best 6.37E?00 6.37E-01 1.15E-02 6.37E-03 6.37E-04
Worst 1.40E?00 1.40E?00 7.92E-01 1.40E?00 1.40E?00
Mean 2.12E?01 2.06E-01 1.91E-01 3.62E-02 2.12E-03
SD 2.81E?01 2.01E-01 1.73E-01 3.18E-02 2.91E-03

Table 7 Results for the inverted

DE DE ? T1FS DE ? T1FS ? N DE ? IT2FS DE ?IT2FS ? N
pendulum control
RMSE

Best 5.84E-01 1.02E-01 4.33E-02 1.83E-02 1.40E-02

Worst 2.46E?00 1.40E?00 1.26E?00 1.40E?00 4.33E-01
Mean 1.49E?00 4.30E-01 3.30E-01 2.15E-01 2.41E-01
SD 5.20E-01 3.62E-01 2.69E-01 2.75E-01 1.65E-01

123
123
Fig. 21 Simulation of results for the water tank
P. Ochoa et al.
Optimization of fuzzy controller design using a Differential Evolution algorithm with dynamic…

Fig. 22 Plot of the best result

and Figs. 27 and 28 illustrate the simulation of the best • The alternative hypothesis states that the DE ? T1FS
results obtained for each of the variations made; then, a algorithm average is lower than the average of the DE.
plot of the best results obtained by each one of the varia-
In the second statistical test, we compare the Type-1
tions is presented, respectively.
Fuzzy Differential Evolution algorithm with the Interval
We can clearly notice in both plots that the results
Type-2 Fuzzy Differential Evolution algorithm where
obtained for this controller are good since the optimization
of the algorithm using fuzzy logic is better than the original • The null hypothesis states that the average of the
algorithm. DE ? T2FS is greater than or equal to the average of
Figure 29 illustrates the mean for each of the con- the DE ? T1FS.
trollers, and we can observe the performance of the original • The alternative hypothesis states that the DE ? T2FS
algorithm and the variants proposed for the algorithm with algorithm average is lower than the average of the
Type-1 and Interval Type-2 fuzzy systems. DE ? T1FS.
In the third statistical test, we compare the Fuzzy Dif-
ferential Evolution algorithm plus noise with Interval
6 Statistical tests Type-2 Fuzzy Differential Evolution algorithm plus noise
where
To verify the results of both methods and to provide a
statement of which method is better, a statistical test was • The null hypothesis states that the average of the
performed. The Z statistical test of two samples was used to DE ? T2FS ? N is greater than or equal to the average
make a comparison between the original Differential of the DE ? T1FS ? N.
Evolution algorithm and the Fuzzy Differential Evolution • The alternative hypothesis states that the DE ?
with Type-1 and Interval Type-2 fuzzy systems. T2FS ? N algorithm average is lower than the average
The statistical test used for comparison is the z-test, of the DE ? T1FS ? N.
whose parameters are defined in Table 8.
The region of rejection is defined for all values
In this case, 12 statistical tests were made, three for each
below- 1.645.
controller, and a brief explanation of these tests is pre-
The equation for the tests is expressed as follows:
sented as follows:
In the first statistical test, we compare the Differential ðX1  X2 Þ  ðl1  l2 Þ
Z¼ ð19Þ
Evolution algorithm with the Fuzzy Differential Evolution rX1 X2
algorithm where
The data from the values of the mean and standard
• The null hypothesis states that the average of the deviation for the original method and the proposed method
DE ? T1FS is greater than or equal to the average of are obtained from Tables 4, 5, 6 and 7, respectively.
the DE.

123
123
Fig. 23 Simulation of the temperature controller
P. Ochoa et al.
Optimization of fuzzy controller design using a Differential Evolution algorithm with dynamic…

Fig. 24 Plot of the best results

Fig. 25 Simulation results for the mobile robot controller

123
 P. Ochoa et al.

Fig. 26 Plot of the best error for the mobile robot controller

Table 9 contains the results for each of the z-values and are listed in Table 8. Table 13 shows the statistical test for
indicates whether the statistical test is significant or not for the inverted pendulum controller.
each controller. Two different types of statistical tests were performed:
Table 9 clearly shows that the obtained results are the first where the controller has no noise and the second
favorable compared to the original algorithm, except for where the controller has noise for each of the plants, and of
two of the nine tests performed, where we did not obtain a these statistical tests we can conclude that for the more
favorable result, but in general, it can be can concluded that complex problems that in this case are the mobile robot
the use of Interval Type-2 fuzzy logic is better in all four controller and inverted pendulum controller our method-
controllers used. ology obtains better results.
To complement our study, a statistical test is carried out
with other methods, and thus, to know the competitiveness
we have compared with other algorithms that combine 7 Conclusions
fuzzy logic to make dynamic some or several parameters;
for the case of the water tank, temperature and mobile The use of fuzzy logic in DE is the main contribution of
robot controllers we make a statistical comparison with the this paper, and a study was performed by using Type-1 and
harmony search algorithm by Peraza et al. (2017). Interval Type-2 fuzzy systems to be able to make a com-
Tables 10, 11 and 12 show the statistical tests between parative study of the performance of each metaheuristic. In
both methods using an Interval Type-2 fuzzy system to this case, the different Differential Evolution variants were
make dynamic a parameter of each algorithm; the param- applied to control problems, and as a conclusion of our
eters for the statistical tests are those contained in Table 8. work we can state that the use of fuzzy logic to optimize
In the case of inverted pendulum controller, the com- the parameters of some metaheuristics improves perfor-
parison is made with another reference, but with the same mance, and for this article we can rely on the statistical test
harmony search algorithm by Castillo et al. (2016c), sta- two-state two main conclusions:
tistical tests are performed using Type-1 fuzzy system for The first conclusion is that the use of Type-1 or Interval
comparison and the parameters of the statistical test used Type-2 fuzzy systems in DE is better than the original
algorithm although it will depend a lot on the particular

123
Optimization of fuzzy controller design using a Differential Evolution algorithm with dynamic…

Fig. 27 Simulation results for the inverted pendulum

123
 P. Ochoa et al.

problem used to measure the performance since as we can

note for the case of the water tank controller, the use of a
Type-1 fuzzy system was better than the Interval Type-2
fuzzy system.
The second conclusion drawn from the study is that for
problems with a higher level of uncertainty the use of
Interval Type-2 fuzzy system is better since in this work in
the four case studies when we considered noise for the
controller the use of Interval Type-2 fuzzy system obtained
significant evidence of better results with respect to a Type-
1 fuzzy system.
Fig. 28 Plot of the best error for the inverted pendulum control In a general way, we can conclude that the proposal to
use fuzzy logic combined with the Differential Evolution
algorithm is a good one since statistically it is verified that
the use of an Interval Type-2 fuzzy system to dynamically
change some parameter is better than the original algo-
rithm. In the same way, statistically it is verified that when
compared with another fuzzy algorithm it obtains better
results, as it can be seen that in the statistical tests in
Tables 10, 11, 12 and 13, in most cases enough evidence is
obtained that our fuzzy algorithm is better on average than
the reference algorithm; of eight statistical tests carried out,
success is obtained in five of them with which the afore-
mentioned is affirmed.
The future work to be done is to make a fuzzy system
that contains the F (mutation) and CR (crossing)
dynamically parameters applied to control problems,
where we can vary the noise level in the controller in
Fig. 29 Plot for the means for the three controllers
such a way that we can verify that the higher the level of
Table 8 Parameters for the statistical testing uncertainty, the fuzzy system has better results. Another
work that we have in mind is to use a generalized fuzzy
Parameter Value
system with some more complex problem in order to see
Level of significance 95% the difference between using a using Type-1 fuzzy sys-
Alpha 5% tem, Interval Type-2 fuzzy system and Generalized Type-
H0 l1 C l2 2 fuzzy system.
Ha l1 \ l2(claim)
Critical value - 1.645

Table 9 Statistical tests for the

Case study l1 l2 z-value Evidence
three case studies
Water tank controller DE ? T1FS DE - 6.3857 Significant
DE ? T2FS DE ? IT1FS 4.7676 Not significant
DE ? T2FS ? N DE ? IT1FS ? N - 4.1851 Significant
Temperature controller DE ? T1FS DE - 0.5031 Not significant
DE ? T2FS DE ? IT1FS - 2.4985 Significant
DE ? T2FS ? N DE ? IT1FS ? N - 2.1079 Significant
Mobile robot controller DE ? T1FS DE - 4.092 Significant
DE ? T2FS DE ? IT1FS - 4.5702 Significant
DE ? T2FS ? N DE ? IT1FS ? N - 5.9791 Significant
Inverted pendulum controller DE ? T1FS DE - 9.1633 Significant
DE ? T2FS DE ? IT1FS - 2.5904 Significant
DE ? T2FS ? N DE ? IT1FS ? N - 1.5447 Not significant

123
Optimization of fuzzy controller design using a Differential Evolution algorithm with dynamic…

Table 10 Statistical tests for the

Method Mean Standard deviation z-value Evidence
water tank controller case study
DE ? IT2FS 6.02E-02 2.45E-03 9.5464 Not significant
FHS2 [45] 2.56E-02 1.97E-02
DE ? IT2FS ? N 3.22E-02 3.63E-02 2.6392 Not significant
FHS2 ? N [45] 1.32E-02 1.254E-02

Table 11 Statistical tests for the

Method Mean Standard deviation z-value Evidence
temperature controller case
study DE ? IT2FS 6.02E-02 2.45E-03 - 4.6692 Significant
FHS2 6.25E-02 1.13E-03
DE ? IT2FS ? N 6.00E-02 2.46E-03 132.0655 Not significant
FHS2 ? N 6.29E-04 1.07E-04

Table 12 Statistical tests for the

Method Mean Standard deviation z-value Evidence
temperature controller case
study DE ? IT2FS 3.62E-02 3.18E-02 - 2.9130 Significant
FHS2 1.11E-01 1.37E-01
DE ? IT2FS ? N 2.12E-03 2.91E-03 - 4.5248 Significant
FHS2 ? N 3.69E-02 4.20E-02

Table 13 Statistical tests for the

Method Mean Standard deviation z-value Evidence
inverted pendulum controller
case study DE ? T1FS 4.30E-01 3.62E-01 - 1.2107 Significant
FHS 7.67E-01 4.81E-01
DE ? T1FS ? N 3.30E-01 2.69E-01 - 4.4369 Significant
FHS ? N 7.54E-01 4.49E-01

Compliance with ethical standards colony optimization in the stabilization of fuzzy controllers. In:
Melin P, Castillo O, Kacprzyk J (eds) Nature-inspired design of
Conflict of interest All the authors in the paper have no conflict of hybrid intelligent systems, vol 667. Springer, Cham, pp 551–571
interest. Bi Y, Srinivasan D, Lu X, Sun Z, Zeng W (2014) Type-2 fuzzy multi-
intersection traffic signal control with differential evolution
Ethical approval This article does not contain any studies with human optimization. Expert Syst Appl 41(16):7338–7349
participants or animals performed by any of the authors. Bui V-B, Tran Q-C, Bui H-L (2017) Multi-objective optimal design
of fuzzy controller for structural vibration control using Hedge-
algebras approach. Artif Intell Rev 50:569
Caraveo C, Valdez F, Castillo O (2016) Optimization of fuzzy
References controller design using a new bee colony algorithm with fuzzy
dynamic parameter adaptation. Appl Soft Comput 43:131–142
Aalto J, Lampinen J (2013) A mutation adaptation mechanism for Castillo O, Amador-Angulo L, Castro JR, Garcia-Valdez M (2016a)
Differential Evolution algorithm. In: 2013 IEEE congress on A comparative study of type-1 fuzzy logic systems, interval
evolutionary computation, Cancun, Mexico, pp 55–62 type-2 fuzzy logic systems and generalized type-2 fuzzy logic
Aalto J, Lampinen J (2014) A mutation and crossover adaptation systems in control problems. Inf Sci 354:257–274
mechanism for differential evolution algorithm. In: 2014 IEEE Castillo O, Cervantes L, Soria J, Sanchez M, Castro JR (2016b) A
congress on evolutionary computation (CEC), Beijing, China, generalized type-2 fuzzy granular approach with applications to
pp 451–458 aerospace. Inf Sci 354:165–177
Amador-Angulo L, Castillo O (2015) Statistical analysis of type-1 Castillo O, Ochoa P, Soria J (2016c) Differential evolution with fuzzy
and interval type-2 fuzzy logic in dynamic parameter adaptation logic for dynamic adaptation of parameters in mathematical
of the BCO. In: 2015 Conference of the International Fuzzy function optimization. In: Angelov P, Sotirov S (eds) Impreci-
Systems Association and the European Society for Fuzzy Logic sion and uncertainty in information representation and process-
and Technology (IFSA-EUSFLAT-15) ing, vol 332. Springer, Cham, pp 361–374
Amador-Angulo L, Castillo O (2017) Comparative analysis of Castillo O, Melin P, Valdez F, Soria J, Ontiveros-Robles E, Peraza C,
designing differents types of membership functions using bee Ochoa P (2019a) Shadowed type-2 fuzzy systems for dynamic

123
 P. Ochoa et al.

parameter adaptation in harmony search and differential evolu- Ochoa P, Castillo O, Soria J (2016a) Fuzzy differential evolution
tion algorithms. Algorithms 12(1):17 method with dynamic parameter adaptation using type-2 fuzzy
Castillo O, Valdez F, Soria J, Amador-Angulo L, Ochoa P, Peraza C logic, pp 113–118
(2019b) Comparative study in fuzzy controller optimization Ochoa P, Castillo O, Soria J (2016b) Type-2 fuzzy logic dynamic
using bee colony, differential evolution, and harmony search parameter adaptation in a new Fuzzy Differential Evolution
algorithms. Algorithms 12(1):9 method, pp 1–6
Cervantes L, Castillo O (2015) Type-2 fuzzy logic aggregation of Ochoa P, Castillo O, Soria J (2017) Differential evolution using fuzzy
multiple fuzzy controllers for airplane flight control. Inf Sci logic and a comparative study with other metaheuristics. In:
324:247–256 Melin P, Castillo O, Kacprzyk J (eds) Nature-inspired design of
Cuevas E, Luque A, Zaldı́var D, Pérez-Cisneros M (2017) Evolu- hybrid intelligent systems, vol 667. Springer, Cham, pp 257–268
tionary calibration of fractional fuzzy controllers. Appl Intell Olivas F, Valdez F, Melin P, Sombra A, Castillo O (2019) Interval
47:291 type-2 fuzzy logic for dynamic parameter adaptation in a
De La OD, Castillo O, Soria J (2017) Optimization of reactive control modified gravitational search algorithm. Inf Sci 476:159–175
for mobile robots based on the CRA using type-2 fuzzy logic. In: Ontiveros-Robles E, Melin P, Castillo O (2018) Comparative analysis
Melin P, Castillo O, Kacprzyk J (eds) Nature-inspired design of of noise robustness of type 2 fuzzy logic controllers. Kybernetika
hybrid intelligent systems, vol 667. Springer, Cham, pp 505–515 54(1):175–201
Driankov D, Palm R (2013) Advances in fuzzy control. Physica Peraza C, Valdez F, Melin P (2017) Optimization of intelligent
2013:68 controllers using a type-1 and interval type-2 fuzzy harmony
Dubois D, Hung TN, Henri P (2000) Possibility theory, probability search algorithm. Algorithms 10(3):82
and fuzzy sets misunderstandings, bridges and gaps. In: Dubois Precup R-E, David R-C, Petriu EM, Wolf G (2017) Optimizer
D, Prade H (eds) Fundamentals of fuzzy sets. Springer, Boston, algorithm-based tuning of fuzzy control systems with reduced
pp 343–438 parametric sensitivity. IEEE Trans Ind Electron 64(1):527–534
Fierro R, Castillo O (2013) Design of fuzzy control systems with Price K, Storn RM, Lampinen JA (2006) Differential evolution: a
different PSO variants. In: Castillo O, Melin P, Kacprzyk J (eds) practical approach to global optimization. Springer, Berlin
Recent advances on hybrid intelligent systems, vol 451. Salehpour M, Jamali A, Bagheri A, Nariman-zadeh N (2017) A new
Springer, Berlin, pp 81–88 adaptive differential evolution optimization algorithm based on
Gao Q (2017) Universal fuzzy models and universal fuzzy controllers fuzzy inference system. Eng Sci Technol Int J 20(2):587–597
for stochastic non-affine nonlinear systems. In: Gao Q (ed) Sanchez MA, Castillo O, Castro JR (2015a) Information granule
Universal fuzzy controllers for non-affine nonlinear systems. formation via the concept of uncertainty-based information with
Springer, Singapore, pp 45–70 interval type-2 fuzzy sets representation and Takagi–Sugeno–
Juang C-F, Chen Y-H, Jhan Y-H (2015) Wall-following control of a Kang consequents optimized with Cuckoo search. Appl Soft
hexapod robot using a data-driven fuzzy controller learned Comput 27:602–609
through differential evolution. IEEE Trans Ind Electron Sanchez MA, Castillo O, Castro JR (2015b) Generalized type-2 fuzzy
62(1):611–619 systems for controlling a mobile robot and a performance
Liu J, Lampinen J (2005) A fuzzy adaptive differential evolution comparison with interval type-2 and type-1 fuzzy systems.
algorithm. Soft Comput 9(6):448–462 Expert Syst Appl 42(14):5904–5914
Mamdani EH (1974) Application of fuzzy algorithms for control of Sa-ngiamvibool W (2017) Optimal fuzzy logic proportional integral
simple dynamic plant. Proc Inst Electr Eng 121(12):1585 derivative controller design by Bee algorithm for hydro-thermal
Martı́nez-Soto R, Castillo O, Castro JR (2014) Genetic algorithm system. IEEE Trans Ind Inform 1:1
optimization for type-2 non-singleton fuzzy logic controllers. In: Sun Z, Wang N, Srinivasan D, Bi Y (2014) Optimal tuning of type-2
Castillo O, Melin P, Pedrycz W, Kacprzyk J (eds) Recent fuzzy logic power system stabilizer based on differential
advances on hybrid approaches for designing intelligent systems. evolution algorithm. Int J Electr Power Energy Syst 62:19–28
Springer, Cham, pp 3–18 Tang L, Zhao Y, Liu J (2014) An improved differential evolution
Melin P, Astudillo L, Castillo O, Valdez F, Garcia M (2013) Optimal algorithm for practical dynamic scheduling in steelmaking-
design of type-2 and type-1 fuzzy tracking controllers for continuous casting production. IEEE Trans Evol Comput
autonomous mobile robots under perturbed torques using a new 18(2):209–225
chemical optimization paradigm. Expert Syst Appl Wang Y, Liu Z-Z, Li J, Li H-X, Wang J (2018) On the selection of
40(8):3185–3195 solutions for mutation in differential evolution. Front Comput
Mendel JM (2014) General type-2 fuzzy logic systems made simple: a Sci 12(2):297–315
tutorial. IEEE Trans Fuzzy Syst 22(5):1162–1182 Zadeh L (1975a) The concept of a linguistic variable and its
Mendel JM, Liu X (2013) Simplified interval type-2 fuzzy logic application to approximate reasoning—I. Inf Sci 8(3):199–249
systems. IEEE Trans Fuzzy Syst 21(6):1056–1069 Zadeh L (1975b) The concept of a linguistic variable and its
Mohan BM, Sinha A (2008) Analytical structure and stability analysis application to approximate reasoning—II. Inf Sci 8(4):301–357
of a fuzzy PID controller. Appl Soft Comput 8(1):749–758 Zadeh L (1978) Fuzzy sets as a basis for a theory of possibility. Fuzzy
Noshadi A, Shi J, Lee WS, Shi P, Kalam A (2016) Optimal PID-type Sets Syst 1(1):3–28
fuzzy logic controller for a multi-input multi-output active
magnetic bearing system. Neural Comput Appl 27(7):2031–2046 Publisher’s Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.

123