Applied Mathematical Modelling: Zhe Sun, Ning Wang, Yunrui Bi

Applied Mathematical Modelling xxx (2014) xxxxxx
Contents lists available at ScienceDirect
Applied Mathematical Modelling

journal homepage: www.elsevier.com/locate/apm
Type-1/type-2 fuzzy logic systems optimization with RNA

genetic algorithm for double inverted pendulum
Zhe Sun a, Ning Wang a,, Yunrui Bi b
a
b
National Laboratory of Industrial Control Technology, Institute of Cyber-Systems and Control, Zhejiang University, Hangzhou 310027, PR China
School of Automation, Southeast University, Nanjing 210096, PR China
a r t i c l e
i n f o
Article history:
Received 14 June 2012
Received in revised form 17 December 2013
Accepted 15 April 2014
Available online xxxx
Keywords:
Type-2 fuzzy logic system
RNA genetic algorithm
Control systems
Double inverted pendulum
a b s t r a c t
This paper proposes a hybrid type-2 fuzzy logic system architecture with the aid of RNA
genetic algorithm for a double inverted pendulum system. As an extension of type-1 fuzzy
logic system, type-2 fuzzy logic system can effectively improve the control performance by
uncertainty of membership function especially when we have to confront with corrupted
data or unpredicted external disturbances. In this proposed method, the related parameters of type-1 and type-2 fuzzy logic systems are respectively optimized by using RNA
genetic algorithm. Through computer simulation and comparisons, the better performance
can be achieved by using optimized type-2 fuzzy logic system with RNA genetic algorithm.
2014 Elsevier Inc. All rights reserved.
1. Introduction
With the advantages of treating ambiguous or vague aspects of human perception and judgment, fuzzy logic system (FLS)
which is based on if-then fuzzy rule expert system has been widely used in various applications. However, recent research
has shown the limitations of traditional type-1 FLS (T1FLS) in treating large uncertainty factors [1] and unexpected disturbances. As an extension of type-1 fuzzy set theory, type-2 fuzzy sets theory [2,3] can well compensate these limitations
when there are too many uncertainties to determine exact membership grade. Therefore, type-2 fuzzy logic system
(T2FLS) theory and its application attract researchers attention in the last few years. For example, Mendel et al. particularly
proposed the design interval type-2 FLS method in literatures [4,5]; a stock index forecasting and stock price analysis model
by using type-2 fuzzy theory were proposed in [6,7]; Hagras presented a type-2 fuzzy logic system controller for autonomous mobile robots [8]; the interval type-2 fuzzy sets were adopted to construct a type-2 support vector machines fusion
classiers [9]; the interval type-2 TS fuzzy impulsive controller and indirect adaptive fuzzy controller were respectively
addressed by Li et al. for nonlinear systems [10,11]. A Fuzzy Lyapunov Synthesis method was proposed to design type-1
and type-2 fuzzy controller for nonsmooth mechanical systems [12].
Moreover, the ongoing challenge for T2FLS is how to come up with suitable parameters in controlling complex systems. In
this sense, the use of intelligent methods demonstrates the usefulness in optimizing FLS. On basis of the type-1 FLS optimization approaches [1317], recent literature has focused the type-2 FLS optimization approaches and shows the advantages
of optimized type-2 FLS especially in control performance, corrupted data processing, noise disturbance and knowledge
uncertainty aspect. An interval type-2 neural network was depicted for position control using ultrasonic motor [18]. In
[18], Lin et al. employed supervised gradient descent learning method to optimize interval type-2 fuzzy neural network
Corresponding author.
E-mail addresses: [email protected], [email protected] (N. Wang).
http://dx.doi.org/10.1016/j.apm.2014.04.035
0307-904X/ 2014 Elsevier Inc. All rights reserved.
Please cite this article in press as: Z. Sun et al., Type-1/type-2 fuzzy logic systems optimization with RNA genetic algorithm for double
inverted pendulum, Appl. Math. Modell. (2014), http://dx.doi.org/10.1016/j.apm.2014.04.035
Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx
(IT2FNN) system parameters and has shown the superior control performance and robust ability of IT2FNN control system;
Lin et al. described an adaptive hybrid interval type-2 fuzzy neural network controller for uncertain nonlinear multivariables dynamic systems [19], which was incorporated sliding mode and Lyapunov synthesis approaches to deal with
the corrupted data by noise or knowledge uncertainty and got better results than adaptive hybrid type-1 FNN control
system; Wu and Tan presented an optimized interval type-2 controller with genetic algorithm for coupled-tank liquid-level
control system [20]. Compared with another GA evolved type-1 fuzzy controller, the optimized type-2 FLC can well eliminate
persistent oscillations; In literature [2125], involving optimization of type-1 and type-2 fuzzy controllers were proposed by
Castillo et al. The type-2 fuzzy controller incorporating evolutionary algorithm for autonomous wheeled mobile robot, DC
motor and non-linear plant were respectively proposed in [2123]; In the work [24], Castillo et al. utilize swarm intelligence
to optimize the type-1 and type-2 controller parameters and got better results; Finally, in [25], an overview is presented
summarizing various optimization method related to type-2 fuzzy systems.
Since Adleman rst developed based on DNA biological computing method for solving a computationally hard problem of
the directed Hamiltonian path problem [26], researchers begin to devote to the research and applications of RNA and DNA
genetic algorithm. Tao and Wang proposed a DNA computing based RNA-genetic algorithm for chemical engineering process
parameter estimation problems [27]; Two kinds of novel RNA genetic algorithms were presented respectively by Wang and
Wang for the hydrocracking of heavy oil and the dynamic systems parameter estimation problems [28,29]. The adaptive RNA
genetic algorithm and a modied DNA genetic algorithm were presented respectively by Zhang and Wang for modeling of
proton exchange membrane fuel cells and the 2-chlorophenol oxidation in supercritical water [30,31]. Chen and Wang proposed the two DNA genetic algorithms were for the parameter estimation in the reaction hydrogenation and the gasoline
blending scheduling problem respectively [32,33]; The literature [34] described a evolutionary neural network with DNA
computing for the dual-axis motion control of a multi-variables system. As an evolutionary algorithm, RNA genetic algorithm
can overcome the drawbacks of traditional genetic algorithm, such as weak local search capability and premature convergence [2730]. Owing to advantages of RNA genetic algorithm, in this work, RNAGA is used to optimize T2FLS. Moreover,
the T2FLS with RNA genetic algorithm can well eliminate persistent oscillations in the double inverted pendulum system. So
it demonstrates that the T2FLS with RNA genetic algorithm can show better systems stability and robustness.
This paper is organized as follows: interval type-2 fuzzy logic system design method is introduced in Section 2. RNA
genetic algorithm is described in Section 3. Section 4 addresses the optimization method of type-1 and type-2 FLSs with
RNA genetic algorithm. The experimental simulation results with different FLSs for double inverted pendulum are given
in Section 5. The conclusions are summarized in the nal section.
2. Interval type-2 fuzzy system design

General T2FLS are computationally intensive because its inference and type-reduction are very intensive. To simplify,
type-2 fuzzy set can be amended to interval type-2 fuzzy set [4] if the secondary memberships are either zero or one as
shown in Fig. 1. The structure of interval type-2FLS consists of fuzzier, inference engine, rule base, type-reducer and defuzzier, which is shown in Fig. 2.
2.1. Fuzzier
The fuzzier part is used for mapping the crisp inputs x x1 ; . . . ; xn into different kinds of fuzzy sets. It includes singleton, type-1 fuzzy set and interval type-2 fuzzy set and so on. Here, singleton fuzzication method is adopted to map the
input variable.
1
0.9
0.8
0.7
1
0.6
UMF
0.5
0.4
0.3
0.2
Primary MF
LMF
0.1
0
10
Fig. 1. Interval type-2 membership function.
Defuzzifier
Rule base
Crisp
Inputs
Fuzzifier
Crisp
Output
Type-reducer
Inputs
Fuzzy set
Inference engine
Outputs
Fuzzy set
Fig. 2. Structure of IT2FLS.
2.2. Rule base

Rule base is expressed with the form of if-then which represents expert knowledge. By considering MISO interval type2FLS, the two kinds of if-then expressions can be depicted as followed:
~l.
if x1 is F~l1 and x2 is F~l2 ; . . . ; xp is F~lp , then yl is G
if x1 is F~l1 and x2 is F~l2 ; . . . ; xp is F~lp , then yl f x1 ; x2 ; . . . ; xn .
2.3. Fuzzy inference
The inference is the key process in fuzzy system. Considering MISO interval type-2FLS, each rule is set up p antecedents.
The expression of lth rule is listed bellow:
~l.
Rl : if x1 is F~l1 and x2 is F~l2 ; . . . ; xp is F~lp , then yl is G
l
l
~
~
Where F i and G are the fuzzy language partition of the input variable and output variable. In additions, the ith input variable and output variable membership functions (MF) are expressed by the way of lF~l xi and lG~ l y. The inference results [4]
i
of lth red rule depicted as:
lB~l y lG~ l y u tx2X flX~ 1 x1 u lF~l1 x1 u u lX~ p xp u lF~lp xp g
where lX~ i xi is ith membership function of fuzzication.

By using the singleton fuzzication, the lth red rule result can be simplied as:
lB~l y lG~ l y u tx2X flF~l1 x1 u u lF~lp xp g
Because the fuzzy sets are interval type-2 fuzzy sets, the inference of antecedents can be addressed as:
f l lX~ 1 x1 u lF~l x1 u u lX~ p xp u lF~lp xp
f l lX~ 1 x1 u lF~l x1 u u lX~ p xp u lF~lp xp
2.4. Type-reducer and defuzzier

After fuzzy inference, an interval type-2 fuzzy set output can be acquired. Type-reducer is the way to convert interval
type-2 fuzzy set into type-1 fuzzy set. Type-reduction methods have been developed by Karnik and Mendel and the center-of-set type-reduction is one of the most popular methods in type-reduction approaches. The center-of-set type-reduction
method [3] can be expressed as:
Y cos Y 1 ; . . . ; Y M :F 1 ; . . . ; F M
Z
y1
Z
yM
Z
f1
Z
fM
PM
i
i1 f
i
i
i1 f y
PM
where Y cos is interval sets determined by yl and yr ; yi yl ; yr
PM
yl Pi1
M
fli yil
i
i1 fl
PM i i
fr yr
yr Pi1
M
i
i1 fr
In order to compute yl and yr we need to compute ffli ; i 1; 2; . . . ; Mg and ffri ; i 1; 2; . . . ; Mg. According to literature [4],
K-M iteration procedure was shown as followed. Without lost of generality, the yir s are arranged in ascending order, i.e.,
y1r 6 y2r 6 . . . 6 yM
r .
(1) Compute yr in Eq. (6) by initially setting fri f i f i =2 for i 1; 2; . . . ; M, where f i and f i have already compute in
Eq. (3) and (4) and let y0r , yr ;
(2) Find R1 < R < M 1 such that yRr 6 y0r 6 yR1
;
r
(3) Compute yr in Eq. (6) with fri f i for i 6 R and fri f i for i > R and let y00r , yr ;
(4) If y0r y00r then go to step (5). If y0r y00r , then stop and set yr , y00r ;
(5) Set y0r equal to y00r and return to step (2);
By using this procedure yr can be obtained. It is the same way for computing yl . Just replaced yir by yil and, in step (2), nd
L1 6 L 6 M 1 such that yLl 6 y0l 6 yL1
and, in step 3), let fli f i for i 6 L, and let fli f i for i > L.
l
Because Y cos is an interval set, we defuzzify it using the average of yl and yr . Therefore,the the crisp output can be obtained
as:
f x
yl yr
2
3. RNA genetic algorithm

3.1. RNA encoding
In the view of modern biology, RNA is one of the genetic materials for life and contains plentiful genetic information. A
RNA chromosome is composed of four kinds of nucleotides namely adenine (A), uracil (U), guanine (G) and cytosine (C).
Based on the biological RNA structures, the RNA encoding method is designed for practical optimization problems.
minf x1 ; x2 ; . . . ; xn
xmin 6 xi 6 xmax ;
i 1; . . . ; n
Considering the optimization problem Eq. (8), f x1 ; x2 ; . . . ; xn is the objective function and xi is the variable with the
bounds xmin ; xmax . Every variable xi is represented by the combination of four nucleotide bases. For better processing in simulation, integer 0, 1, 2 and 3 are adopted to encode the RNA nucleotide bases with the mapping principle AUGC=0123, that
states A accords with 0, U with 1, G with 2 and C with 3.
3.2. Selection strategy
The purpose of selection operation is to decide which individuals are kept to next generation and how many individuals
are replicated to offspring. Generally, those individuals with higher tness value have a greater opportunity to survival or
reproduce. Among the various selection methods like roulette wheel selection, tournament selection, ranking selection
and Boltzmann selection, roulette wheel selection is more popular and efcient. But the roulette wheel selection method
may cause the higher tness values individuals to dominate all the offspring individuals, which may lead to premature convergence easily. Therefore, a roulette wheel selection with balance way is used in RNA genetic algorithm. In this method a
group of deleterious individuals are allowed to be picked out for the genetic operation and the elitism is adopted to guarantee the best individual to be reserved for the next generation. Further details of this selection operation is described in
Section 3.6.
3.3. Crossover operators
Crossover operation simulates the process of reproduction where the offspring individuals inherit information from their
respective parents. In GAs, crossover operator is the most important operator, which can exchange information from different chromosomes. However, the traditional crossover operators in GAs, such as single-point, multi-point and arithmetic
crossover operators, are based on simple abstraction of biological concepts, which weakens the searching performance. In
order to improve the searching ability and exploit more useful information of individuals, three main operators namely
translocation operator, transformation operator and permutation operator [27] are adopted to replace traditional crossover
operator.
(1) Translocation operator: Make subsequence RNA sequence transfer to the new location. For example, in the original
RNA sequence X X 5 X 4 X 3 X 2 X 1 , X i i 1; . . . ; 5 is the subsequence of the RNA sequence (X), and the new chromosome
after translocation operation becomes X X 5 X 2 X 4 X 3 X 1 . The mathematical expression of the translocation operation is
shown as follows:
X i k
X i n1 k 1;
k 6 np n1 1
X i n1 np 1 k; np n1 1 < k 6 np
(2) Transformation operator: Let the segments of RNA sequence to exchange their locations. For example, the sequence X
becomes X X 5 X 2 X 3 X 4 X 1 after their transformation operation of X 4 and X 2 . The formula of transformation operator is
described in Eq. (10).

X i k
X i n2 n1 k; n1 6 k 6 n2
X i k;
otherwise
10
(3) Permutation operator: One subsequence of RNA sequence is permutated by the other subsequence. For example,
when X 02 subsequence from the same or other RNA sequence is selected to replace X 2 , the new sequence is
X X 5 X 4 X 3 X 02 X 1 . The new sequence is formed by using the formula given in Eq. (11):
X i k
X j k; k 2 n1 ; n2
11
X i k; otherwise
where, X i and X j are the ith and jth RNA sequence, k 1; 2; 3; . . . ; np is the position of RNA sequence, n1 and n2
(n1 < n2 6 np ) are the random insert points and np is the size of population.
3.4. Self-adaption mutation operator
Mutation operator plays a signicant role when the algorithm traps into local optimum. It can guide the algorithm to
come out of the local optimum when most individuals in the population become similar. According to [35], there exists
hot spots and cold spots in the DNA sequence model and the nucleotide bases in the cold spots mutate more slowly
than those in the hot spots, which accords with the fact that the spots in different bit positions have different effects on the
solutions to the problem. Hence, the shift probability in different evolution stages is employed. For example, at the beginning
of the evolution stage we employ large probability in the high bit position to obtain large searching space and at the end of
evolution stage the large probability in the low bit position is necessary for accurate results. Thus, the two kinds of mutation
probability P mh and P ml , are described as follows:
Pmh a1
b1
1 expaag g 0
12
Pml a1
b1
1 expaag g 0
13
where a1 is denoted as the initial mutation probability, b1 is the range of mutation probability, g is the evolution generation,
g 0 is denoted as the generation where great change of mutation probability occurs and aa is the speed of change.
3.5. Reconstruction operator
In order to improve the diversity of population, inspired by DNA double helix complementary principle, reconstruction
operator is introduced for producing new complementary individuals. The reconstruction operation follows the principle
where A complements with U and C complements with G. An example of reconstruction operator is shown in Fig. 3. Each
sequence represents one of the candidate solutions, and then the corresponding complementary sequence is formed automatically after the mutation operation.
3.6. Procedure of RNAGA
Based on the RNA encoding method and the RNA genetic operations mentioned above, the procedure of RNAGA is summarized below and shown in Fig. 4.
Step 1: Initialize a population with (np N) individuals;
Step 2: Calculate the tness value of each individual;
Step 3: Judge population size np > 5N, if afrmative, goto step 5, otherwise implement step 4;
Step 4: Choose the np =2 better individuals and np =4 deleterious individuals and goto step 6;
Step 5: Choose the np =3 better individuals and np =6 deleterious individuals and carry out next step;
Step 6: Gather these individuals into individual pool and implement the permutation operation;
Step 7: Set the probability pc 0:5, If the random number (q 2 0; 1) q > pc then implement the translocation operation,
otherwise perform transformation operation;
Step 8: Carry out the mutation operation with adaptive mutation probability;
The former RNA Chromosome:
A U C U A...U C A G U
0 1 3 1 01 3 0 2 1
Complementary RNA Chromosome: U A G A U...A G U C A
1 0 2 0 10 2 1 3 0
Fig. 3. Example of reconstruction operator.
Reconstruction operator
Start
RNA encoding and initialize population
np
pc
No
Yes
Calculate individual fitness value
No
np
5N
Translocation operator
Transformation operator
Yes
Mutation operator
Choose n p / 3 better individuals

and n p / 6 deleterious individuals
Meet end
condition
Choose n p / 2 better individuals

and n p / 4 deleterious individuals
No
Yes
End
Permutation operator
Fig. 4. The ow chart of the RNA genetic algorithm procedure.
Step 9: Perform the reconstruction operation to produce the corresponding complementary individuals and combine
them with the former individuals;
Step 10: Judge whether the optimum in current generation meets the termination criteria. If met then stop, else return to
step 2.
3.7. Numeric experiments
To demonstrate the searching ability of RNA genetic algorithm, ve nonlinear constrained benchmark functions are taken.
Table 1 compiles a list of commonly used test functions, which represent a group of landscape classes with various characteristics: large search space, numerous local minima and fraudulence. For comparison, standard GA is adopted to optimize
these functions. The corresponding parameters setting of GA and RNA genetic algorithm are shown in Table 2.
For every test function, the optimization process will end when the inequality jOb O j 6 e 107 is satised, where Ob
denotes the best optimized objective function value, O is the real optimal value, and e is the precision requirement. Each
algorithm runs R 50 times for every test function. Here, Gav e ; Gmax and Gmin are denoted the average evaluation generation
number, the maximum and minimum evaluation generation for measuring the convergence speed of the algorithm. Obviously, the less Gav e ; Gmax and Gmin are, the faster the algorithm is. The corresponding results are shown in Table 3. The global
search ability is measured by, F av e ; F max and F min , which represent the average, maximum and minimum optimal value of the
best objective function over R runs respectively. The testing results are shown in Table 4. From these results, we can see that
Table 1
Test Functions.
Test functions
minf1 x 100x2
x1 ; x2 2 5:12; 5:12
2
x21
1 x1
x2 100
minf2 x 1 x1 1004000
Optimal value
Value
(1,1)
(100,100)
(0, . . .,0)
(0, . . .,0)
(0, . . .,0)
p
cosx1 100 cosx2 100
2
x1 ; x2 2 200; 200
P
minf3 x ni0 x2i
xi 2 5:12; 5:12; n 10
P
minf4 x ni0 x2i 10 cos2pxi 10
xi 2 5; 5; n 10
Pn 2 Q n
xi
1
p
minf5 x 4000
i0 xi
i0 cos i 1
xi 2 10; 10; n 10

Table 2
Parameters setting of algorithms.
GA
RNAGA
Generation = 1000
Population size = 20
Binary encoding method
Roulette wheel selection
Elitist mechanism
Single point crossover operation
with pc 0:8
Adaptive mutation operation
Generation = 1000
Population size = 20
Quaternary encoding method
Roulette wheel selection with balance way
Elitist mechanism
Permutation operation
Transformation operation
Translocation operation
Adaptive mutation operation
Reconstruction operation
Table 3
Comparison of efciency.
Functions
RNAGA
f1
f2
f3
f4
f5
GA
G av e
Gmax
Gmin
G av e
Gmax
Gmin
69.92
99.88
778.08
1000
1000
687
365
1000
1000
1000
8
11
238
1000
1000
276.2
167.72
1000
1000
1000
1000
1000
1000
1000
1000
4
12
1000
1000
1000
Table 4
Comparison of accuracy ability.
Functions
f1
f2
f3
f4
f5
RNAGA
GA
F av e
F max
F min
F av e
F max
F min
0
0
3.742E07
0.7921
0.0525
0
0
3.818E06
4.0363
0.1109
0
0
6.275E08
1.483E05
0.0148
0.0345
0.0177
4.22E04
1.1542
0.1637
1.0303
0.1105
5.0E4
9.1815
0.2772
0
0
4.0E04
0.0949
0.0839
RNA genetic algorithm shows better performance in convergence speed, searching ability and especially works well for optimization problems with high dimensional space.
4. Type-1 and type-2 FLSs optimization
In this section, we propose type-1 and type-2 FLSs optimization method by using the RNA genetic algorithm to optimize
the associated membership function parameters. The fuzzy system optimized process is depicted in Fig. 5.
4.1. Type-1 and type-2 MFs encoding and objective function formulation
Here, two input variables and one output variable Mamdani fuzzy inference system with seven fuzzy language partitions
is adopted, and the corresponding rule base is shown in Table 5. According to the type-1 and interval type-2 MFs formula in
Eqs. (14) and (15), these two types MFs parameters are encoded respectively by using RNA encoding, and the relevant encoding process for one of these variables is shown in Figs. 6 and 7.
lA x exp
"
#
x m2
2r2
"
#
x m2
leA x exp
2r2
14
r 2 r1 ; r2
15
RNA genetic
algorithm
Learning
Process
Knowledge Base
RuleBase
Input
Fuzzification
Defuzzification
Inference Engine
Output
Type-Reducer
Interval Type-2 Fuzzy Logic System

Fig. 5. RNA genetic algorithm design and fuzzy optimized process.
Table 5
Rule base of FLSs for double inverted pendulum system.
EC
NB
NM
NS
ZO
PS
PM
PB
NB
NB
NM
NM
NM
NS
NS
NB
NM
NM
NS
NS
ZO
PS
NM
NM
NS
NS
ZO
PS
PM
NM
NM
NS
ZO
PS
PM
PM
NM
NS
ZO
PS
PS
PM
PM
NS
ZO
PS
PS
PM
PM
PB
NS
PS
PM
PM
PM
PB
PB
E
NB
NM
NS
ZO
PS
PM
PB
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-6
mNB
NB
NB
mNM
NM
NS
-4
NM
(* * * * )(* * * * )
ZO
-2
mNS
NS
PS
mZO
2
ZO
mPS
PB
PM
4
PS
mPM
PM
mPB
PB
(* * * * )(* * * * )
Fig. 6. RNA encoding method for T1FLS MFs parameters.
In order to carry out the optimization process, the objective function need to be conrmed. According to the double
inverted pendulum system description which will be presented in Section 5, the uctuation of h2 ; h1 and x (upper pole angle,
lower pole angle and cart position) should be small during the control process stage. Based on the characteristics of double
inverted pendulum, the upper pole angle should have more weight than the lower pole angle and cart position in control
P

N
2
2
2
stage. Then the objective function can be denoted as F obj min
.
i1 x 50h1 250h2
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
mNB
NB1
NB
-6
NB 2
NM
NS
ZO
PS
PM
-4
-2
mNM
NM 1
...
NM 2
mPM
PM 1
(* * **)(* * **)(* * **)
PB
PM 2
mPB
PB1
PB 2
(* * **)(* * **)(* * **)
Fig. 7. RNA encoding method for IT2FLS MFs parameters.
Table 6
Optimized MFs parameters of T1FLS and T2FLS with RNA genetic algorithm
NB
NM
NS
ZO
PS
PM
PB
T1FLS
m:
r:
6.000
0.9379
4.024
1. 098
2.001
0.8587
0.1960
0.5274
1.630
0.8069
4.244
1.215
6.000
1.393
T2FLS
m:
r1 :
r2 :
6.000
0.5978
1.197
4.167
0.5997
0.6574
1.860
0.9037
1.272
0.4853
1.1080
1.229
1.779
1.206
1.230
3.561
0.9184
1.302
6.000
0.5841
0.9643
NS
PS
390
1
NB
NM
ZO
PM
PB
GAT1
RNAGAT1
380
0.8
Fitness value
370
0.6
0.4
360
350
336.4266
340
330
0.2
319.5155
320
0
6
(a)
310
10
15
20
25
30
Generation
35
40
45
50
(b)
Fig. 8. T1 MFs with RNA genetic algorithm and evolutionary tendency.
4.2. Optimization of type-1 and type-2 FLSs

According to double inverted pendulum system requirements, the objective function need to be minimized for getting the
best parameters. Here, taking type-2 FLS for example, the procedure of type-2 FLS with RNA optimization is addressed as
follows.
Step 1: Set the type-2 variable ranges and encode the variables with RNA encoding method in a chromosome;
Step 2: Initialize the population (np 20) and evolutionary generation (G 50), then implement type-2 FLS and calculate
tness value according to the objective function;
Step 3: Carry out the selection operation with elitism strategy in the population to select the high value individuals and
reproduce the similar individuals;
Step 4: Implement the permutation operation and set the probability pc 0:5, If the random number (q 2 0; 1) q > pc
then implement translocation operation, otherwise perform transformation operation;
10
Step 5: Set high bit position probability and low bit position probability and implement mutation operation; then execute
the reconstruction operation to produce the corresponding complementary individuals;
Step 6: If the termination condition is met or iteration is completed then stop, else return to step 2.
The procedure is similar for type-1 FLS optimization. For further comparison, standard genetic algorithm is adopted to
optimize two types of FLSs. The generation of GA is denoted G 50, and others parameters are stated in Table 2. Finally
the optimization results and evolved tendency are shown in Table 6, Figs. 8 and 9.
350
1
NB
NM
NS
ZO
PS
PM
GAT2
RNAGAT2
PB
0.8
300
Fitness value
0.6
0.4
240.2925
250
0.2
0
6
213.1699
200
10
15
20
25
30
35
40
45
50
Generation
(a)
(b)
Fig. 9. IT2 MFs with RNA genetic algorithm and evolutionary tendency.
m2 g
m1 g
Cart
x
Fig. 10. The double inverted pendulum system.
Table 7
Parameters and values of the double inverted pendulum
Item
Symbol
Value
Unit
Mass of cart
Mass of lower pole
Mass of upper pole
Inertia of lower pole around the lower joint
Inertia of upper pole around the upper joint
Length of lower pole
Length of upper pole
Length from lower joint to the center of mass of lower pole
Length from upper joint to the center of mass of upper pole
Friction of cart
Friction of lower pole
Friction of upper pole
Gravitational acceleration
m0
m1
m2
J1
J2
L1
L2
l1
l2
f0
f1
f2
g
1.32
0.215
0.132
0.008398
0.01825
0.16
0.40
0.13
0.22
23.6782
0.003425
0.003425
9.8
kg
kg
kg
kg m2
kg m2
m
m
m
m
N/s/m
N/s/m
N/s/m
m/s2
11
5. Simulation results of double inverted pendulum

5.1. Double inverted pendulum system
The double inverted pendulum system is a complicated MISO system, which is a typical high-speed multi-variable, nonlinear and dynamically unstable system. As shown in Fig. 10, it composes of a cart and two rigid poles with the length L1 and
L2 . The end of the lower pole is mounted on the car. The top end of the lower pole is connected with the end of the upper
pole. The control goal is to balance the poles by pushing the cart back and forth on a track of limited length.
Table 8
Double inverted pendulum initial values
Variable
x_
h1
h_1
h2
h_2
Con1
Con2
Con3
Con4
0.3
0
0.1
0.2
0.2
0.2
0.1
0.1
0.1
0.25
0.15
0.15
0.5
0.8
0.1
0.5
0.1
0.1
0.05
0.1
0.4
0.5
0.8
0.6
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
0.3
Angle1
0.2
0.1
0
0.1
0.2
0.4
0
0.5
1.5
2.5
3.5
4.5
6
0
0.5
1.5
2.5
3.5
4.5
time(s)
0.05
0
0.05
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
Angle2 rate
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
0.1
Angle2
time(s)
0.15
0.5
0
0.5
0.1
0.15
1
0
0.5
1.5
2.5
3.5
4.5
time(s)
0.35
0.2
0.15
0.1
1.5
2.5
3.5
4.5
time(s)
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
Cart rate
0.25
0.5
1.5
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
0.3
cart position
0.3
0.2
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
Angle1 rate
0.4
0.5
0
0.5
0.05
1
0
0.05
0.5
1.5
2.5
time(s)
3.5
4.5
1.5
0.5
1.5
2.5
3.5
4.5
time(s)
(a)
(b)
Fig. 11. The comparison of different FLSs with (0.3 0.2 0.1 0.5 0.1 0.4) initial value.
12
_ h1 ; h_1 ; h2 ; h_2 ; F), denoted as the

The state of the double inverted pendulum system can be described by six variables (x; x;
displacement of the cart, cart velocity, lower pole angle and its angular velocity, upper pole angle and its angular velocity and
the force.
The differential equation of double inverted pendulum is described as follows, and the description of parameters [36] are
listed in Table 7.
3
2 3
2 3
x
1
x_
6 7
6 7
6 7
Mh1 ; h2 4 h1 5 Fh1 ; h2 ; h_1 ; h_2 4 h_1 5 Nh1 ; h2 4 0 5
0
h2
h_2
0
m0 m1 m2
m1 l1 m2 L1 cos h1
B
Mh1 ; h2 @ m1 l1 m2 L1 cos h1
J 1 m1 l1 m2 L21
m2 l2 cos h2
0
B
Fh1 ; h2 ; h_1 ; h_2 @ 0
0
0.3
Angle1 rate
0.1
1
0
1
2
0.3
3
1
1.5
2.5
3.5
4.5
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
Angle2 rate
0.02
0
0.02
0.04
0.06
0.08
0
0.5
1.5
1.5
2.5
3.5
4.5
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
Cart rate
0.08
0.06
0.04
0.02
0
0.5
1.5
2.5
2.5
3.5
4.5
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
0.5
1.5
2.5
3.5
4.5
time(s)
0.1
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
0
time(s)
0.12
0.02
0.5
time(s)
0.04
0.1
time(s)
0.06
cart position
0.2
0.5
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
18
0.1
0.4
17
2
m2 l2
0.2
Angle1
J2
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
0.4
Angle2
C
m2 l2 L1 cosh2 h1 A
1
m2 l2 h_2 sin h2
C
m2 l2 L1 h_2 sinh2 h1 f2 A
f2
f1 f2
m2 l2 L1 h_1 sinh2 h1 f2
0.5
m2 l2 cos h2
m2 l2 L1 cosh2 h1
m1 l1 m2 L1 h_1 sin h1
f0
16
3.5
4.5
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
0.5
1.5
2.5
time(s)
time(s)
(a)
(b)
3.5
4.5
Fig. 12. The comparison of different FLSs with (0 0.2 0.25 0.8 0.1 0.5) initial value.
13
B
C
Nh1 ; h2 @ gm1 l1 m2 L1 sin h1 A
19
gm2 l2 sin h2
5.2. The experimental simulation of different FLSs

For convenient control of the pendulum in FLSs these six variables are assembled into E and EC. In this simulation process,
we denote the error and rate of error (E; EC) as FLSs input variables and denote the force of control (F) as output variable. The
stability of double inverted pendulum system is much more sensitive at different pole angles. According to the experimental
research, four kinds of initial conditions are set up for testing the control systems i.e., (a) h1 P 0; h2 P 0; (b) h1 0; h2 0; (c)
h1 P 0; h2 0; (d) h1 0; h2 P 0. These four kinds of random initial value are listed in Table 8. Furthermore, we add external
disturbances to this double inverted pendulum system at beginning of stage to demonstrate the superiority of type-2FLS in
treating unpredicted disturbances. The ITAE (integral of time-weighted absolute value of the error) index is used to examine
the control performance of different FLSs.
Based on the optimized results of type-1FLS and type-2FLS, these optimized parameters are combined respectively with
type-1 and type-2 FLSs to keep the double inverted pendulum balanced at different initial status. The experimental simulation results and the comparison with different fuzzy logic systems are shown in Figs. 1114 and Table 9.
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
Angle1
0.2
0.1
0
0.1
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
1.5
Angle1 rate
0.3
1
0.5
0
0.5
1
0.2
0.3
1.5
0
0.5
1.5
2.5
3.5
4.5
0.1
0.08
0.06
0.04
0.02
0
0.02
0.04
0.06
0.08
0.1
0
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
1.5
2.5
3.5
4.5
time(s)
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
0.4
0.2
0
0.2
0.4
0.5
1.5
2.5
3.5
4.5
0.6
0.12
0.08
0.06
0.04
0.5
1.5
2.5
3.5
4.5
time(s)
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
0.6
0.4
Cart rate
0.1
0.8
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
0.14
cart position
0.5
0.6
time(s)
0.2
0
0.2
0.02
0.4
0
0.02
0.8
Angle2 rate
Angle2
time(s)
0.5
1.5
2.5
3.5
4.5
0.6
0.5
1.5
2.5
time(s)
time(s)
(a)
(b)
3.5
4.5
0.2
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
0.15
Angle1
0.1
0.05
0
0.05
2
1
0
1
0.1
0.15
0
0.5
1.5
2.5
3.5
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
0.08
0.06
Angle2
3
0
time(s)
0.1
0.04
0.02
2.5
3.5
4.5
0.3
0.1
1.5
2.5
3.5
4.5
1.5
2.5
3.5
4.5
time(s)
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
0.4
1
4.5
0.2
0
0.5
0.4
0.2
3.5
0.6
0.05
0.05
0.5
0.8
Cart rate
0.15
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
0.8
0
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
0.2
2.5
time(s)
0.2
time(s)
0.25
0.6
1.5
1.5
0.2
0.02
1
0.4
0.4
0.5
0.5
0.6
0.04
0
cart position
4.5
Angle2 rate
0.2
Benchmark
T1
GAT1
RNAGAT1
RNAGAT2
Angle1 rate
14
0.6
0.5
1.5
2.5
time(s)
time(s)
(a)
(b)
3.5
4.5
Table 9
The ITAE index of simulation with different FLSs
Status
Type
x_
h_1
h2
h_2
Sum
T1
GA-T1
RNAGA-T1
RNAGA-T2
T1
16.9435
16.4679
14.9130
13.1954
16.2096
37.5624
34.2746
30.5150
22.7341
33.8640
9.0376
7.5406
7.3769
6.5222
8.1437
61.7064
46.7014
42.4123
40.2563
53.7295
5.1121
5.0725
4.6941
4.1478
4.9397
22.8595
20.9215
19.8659
18.4856
20.0171
153.2215
130.9786
119.7773
105.3413
136.9035
Con2
GA-T1
RNAGA-T1
RNAGA-T2
T1
13.4573
13.2788
10.1193
19.7429
27.5805
26.1803
10.2747
39.1097
5.7144
5.3987
2.9597
9.0953
41.5389
33.4460
23.3137
47.1801
4.1238
3.8972
1.4630
6.4400
16.0934
15.4262
8.0354
22.7006
108.5083
97.6271
56.1659
144.2686
Con3
GA-T1
RNAGA-T1
RNAGA-T2
T1
18.9855
14.0734
11.4509
34.0512
33.7843
29.2253
20.5743
55.0192
7.7565
6.8075
5.4276
10.7468
40.8942
36.3747
29.2688
52.0946
5.8894
4.7399
3.6486
8.0212
19.1682
17.3987
14.7804
24.3365
126.4781
108.6494
85.1506
184.2695
Con4
GA-T1
RNAGA-T1
RNAGA-T2
31.6007
29.4931
24.7384
50.8420
44.6151
35.9629
9.8468
8.7092
7.0171
44.4833
37.8748
31.0495
7.4770
6.8284
5.6376
21.5116
19.5547
14.8201
165.7614
147.0753
119.2256
Con1
h1
15
Comparing the fuzzy system with no optimization technique, GA optimized type-1FLSs and the type-1FLS with RNA optimization, the latter shows better performance under different initial conditions. It demonstrates the RNA genetic algorithm
can t very well for system parameter optimization especially when there are many parameters to be optimized. Moreover,
from Figs. 1114 and Table 9, we can see that the least oscillation and ITAE index can be achieved when we use RNAGA
optimized type-2FLS. So the results indicates that the optimized type-2FLS achieves the aim of system control especially
when we have to confront with more uncertainties and unexpected disturbances.
6. Conclusions
In this paper, we have discussed the control system of the type-1 and type-2 fuzzy systems for the double inverted pendulum. By comparing type-1 fuzzy system with optimized type-2FLS, the latter shows superiority in eliminating the persistent oscillations and vibrations. And comparing with GA, RNA genetic algorithm demonstrates the advantage in multivariable systems optimization problem. Thus, the proposed method in this paper is generic, and can be applied to control
systems that have various parameters and unexpected disturbances.
Acknowledgement
This work was supported partly by the National Natural Science Foundation of China under Grant Nos. 60874072,
61363033, the National Science & Technology Pillar Program of China under Grant No. 2013BAF07B03.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
J.M. Mendel, Uncertain Rule-Based Fuzzy Logic Systems: Introduction and New Directions, Prentice-Hall, Upper-Saddle River, NJ, 2001.
L.A. Zadeh, The concept of a linguistic variable and its application to approximate reasoning-1, Inf. Sci. 8 (1975) 199249.
N.N. Karnik, J.M. Mendel, Centroid of a type-2 fuzzy set, Inform. Sciences 132 (14) (2001) 195220.
Q.L. Liang, J.M. Mendel, Interval type-2 fuzzy logic systems: theory and design, IEEE Trans. Fuzzy Syst. 8 (5) (2000) 535550.
J.M. Mendel, R.I. John, F.L. Liu, Interval type-2 fuzzy logic systems made simple, IEEE Trans. Fuzzy Syst. 14 (6) (2006) 808821.
K.H. Huarng, H.K. Yu, A type 2 fuzzy time series model for stock index forecasting, Physica A 353 (2005) 445462.
M.H. Fazel Zarandi, B. Rezaee, I.B. Turksen, E. Neshat, A type-2 fuzzy rule-based expert system model for stock price analysis, Expert Syst. Appl. 36 (1)
(2009) 139154.
H. Hagras, A type-2 fuzzy logic controller for autonomous mobile robots, in: Proc. IEEE Int. Conf. on Fuzzy Systems. July 2004, pp. 965970.
X.J. Chen, Y. Li, R. Harrison, Y.Q. Zhang, Type-2 fuzzy logic-based classier fusion for support vector machines, Appl. Soft. Comput. 8 (3) (2008) 1222
1231.
Y.M. Li, Y.Y. Sun, Type-2 TS fuzzy impulsive control of nonlinear systems, Appl. Math. Model. 36 (6) (2012) 27102723.
Y.M. Li, Y.J. Du, Indirect adaptive fuzzy observer and controller design based on interval type-2 TS fuzzy model, Appl. Math. Model. 36 (4) (2012)
15581569.
Nohe.R. Cazarez-Castro, Luis T. Aguilar, Oscar Castillo, Designing type-1 and type-2 fuzzy logic controllers via fuzzy Lyapunov synthesis for nonsmooth
mechanical systems, Eng. Appl. Artif. Intell. 25 (5) (2012) 971979.
R.A. Aliev, B.G. Guirimov, B. Fazlollahi, R.R. Aliev, Evolutionary algorithm-based learning of fuzzy neural networks. Part 2: recurrent fuzzy neural
networks, Fuzzy Sets Syst. 160 (17) (2009) 25532566.
M. Otadi, M. Mosleh, Simulation and evaluation of dual fully fuzzy linear systems by fuzzy neural network, Appl. Math. Model. 35 (10) (2011) 5026
5039.
A.M. Tang, C. Quek, G.S. Ng, GATSKfnn: parameters tuning of fuzzy neural network using genetic algorithms, Expert. Syst. Appl. 29 (4) (2005) 769
781.
O. Cordon, F. Gomide, F. Herrera, F. Hoffmann, L. Magdalena, Ten years of genetic fuzzy systems: current framework and new trends, Fuzzy Sets Syst.
141 (1) (2004) 531.
Y.C. Chou, L.W. Lan, Genetic fuzzy logic controller: an iterative evolution algorithm with new encoding method, Fuzzy Sets Syst. 152 (3) (2005) 617
635.
F.J. Lin, S.Y. Chen, P.H. Chou, P.H. Shieh, Interval type-2 neural network control for X-Y-Theta motion control stage using linear ultrasonic motors,
Neurocomputing 72 (46) (2009) 11381151.
T.C. Lin, M.C. Chen, Adaptive hybrid type-2 intelligent sliding mode control for uncertain nonlinear multivariable dynamical systems, Fuzzy Sets Syst.
171 (1) (2011) 4471.
D. Wu, W.W. Tan, Genetic learning and performance evaluation of interval type-2 logic controllers, Eng. Appl. Artif. Intell. 19 (8) (2006) 829841.
R. Martinez, O. Castillo, L.T. Aguilar, Optimization of interval type-2 fuzzy logic controllers for a perturbed autonomous wheeled mobile robot using
genetic algorithms, Inf. Sci. 179 (13) (2009) 21582174.
N.R. Cazarez-Castro, L.T. Aguilar, O. Castillo, Fuzzy logic control with genetic membership function parameters optimization for the output regulation
of a servomechanism with nonlinear backlash, Expert. Syst. Appl. 37 (6) (2010) 43684378.
O. Castillo, P. Melin, A. Alanis, O. Montiel, R. Sepulveda, Optimization of interval type-2 fuzzy logic controllers using evolutionary algorithms, Soft
Comput. 15 (6) (2011) 11451160.
O. Castillo, R. Martinez-Marroquin, P. Melin, F. Valdez, J. Soria, Comparative study of bio-inspired algorithms applied to the optimization of type-1 and
type-2 fuzzy controllers for an autonomous mobile robot, Inf. Sci. 192 (1) (2012) 1938.
O. Castillo, P. Melin, A review on the design and optimization of interval type-2 fuzzy controllers, Appl. Soft Comput. 12 (4) (2012) 12671278.
L.M. Adleman, Molecular computation of solutions to combinatorial problems, Science 266 (5187) (1994) 10211024.
J.L. Tao, N. Wang, DNA computing based RNA genetic algorithm with applications in parameter estimation of chemical engineering processes, Comput.
Chem. Eng. 31 (12) (2007) 16021618.
K.T. Wang, N. Wang, A novel RNA genetic algorithm for parameter estimation of dynamic systems, Chem. Eng. Res. Des. 88 (11) (2010) 14851493.
K.T. Wang, N. Wang, A protein inspired RNA genetic algorithm for parameter estimation in hydrocracking of heavy oil, Chem. Eng. J. 167 (1) (2011)
228239.
L. Zhang, N. Wang, An adaptive RNA genetic algorithm for modeling of proton exchange membrane fuel cells, Int. J. Hydrogen Energy 38 (1) (2013)
219228.
L. Zhang, N. Wang, A modied DNA genetic algorithm for parameter estimation of the 2-chlorophenol oxidation in supercritical water, Appl. Math.
Model. 37 (3) (2013) 11371146.
X. Chen, N. Wang, A DNA based genetic algorithm for parameter estimation in the hydrogenation reaction, Chem. Eng. J. 150 (23) (2009) 527535.
16
[33] X. Chen, N. Wang, Optimization of short-time gasoline blending scheduling problem with a DNA based hybrid genetic algorithm, Chem. Eng. Process.
49 (10) (2010) 10761083.
[34] C.H. Huang, C.L. Lin, Evolutionary neural networks and DNA computing algorithms for dual-axis motion control, Eng. Appl. Artif. Intell. 24 (7) (2011)
12631273.
[35] C. Neuhauser, S.M. Krone, The genealogy of samples in models with selection, Genetics 145 (2) (1997) 519534.
[36] J.H. Xing, J. Chen, Application of fuzzy control in the double inverted pendulum system, J. Anhui Univ. Technol. Sci. 25 (3) (2010) 3135.

Applied Mathematical Modelling: Zhe Sun, Ning Wang, Yunrui Bi

Uploaded by

Copyright:

Available Formats

Applied Mathematical Modelling: Zhe Sun, Ning Wang, Yunrui Bi

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Applied Mathematical Modelling: Zhe Sun, Ning Wang, Yunrui Bi

Uploaded by

Copyright:

Available Formats

Applied Mathematical Modelling xxx (2014) xxxxxx

Contents lists available at ScienceDirect

Applied Mathematical Modelling

Type-1/type-2 fuzzy logic systems optimization with RNA

Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx

2. Interval type-2 fuzzy system design

Fig. 1. Interval type-2 membership function.

Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx

Fig. 2. Structure of IT2FLS.

2.2. Rule base

lB~l y lG~ l y u tx2X flX~ 1 x1 u lF~l1 x1 u u lX~ p xp u lF~lp xp g

where lX~ i xi is ith membership function of fuzzication.

lB~l y lG~ l y u tx2X flF~l1 x1 u u lF~lp xp g

f l lX~ 1 x1 u lF~l x1 u u lX~ p xp u lF~lp xp

f l lX~ 1 x1 u lF~l x1 u u lX~ p xp u lF~lp xp

2.4. Type-reducer and defuzzier

where Y cos is interval sets determined by yl and yr ; yi yl ; yr

Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx

3. RNA genetic algorithm

Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx

The former RNA Chromosome:

Complementary RNA Chromosome: U A G A U...A G U C A

Fig. 3. Example of reconstruction operator.

Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx

Choose n p / 3 better individuals

Choose n p / 2 better individuals

Fig. 4. The ow chart of the RNA genetic algorithm procedure.

Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx

Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx

Interval Type-2 Fuzzy Logic System

Fig. 6. RNA encoding method for T1FLS MFs parameters.

Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx

(* * **)(* * **)(* * **)

(* * **)(* * **)(* * **)

Fig. 7. RNA encoding method for IT2FLS MFs parameters.

Fig. 8. T1 MFs with RNA genetic algorithm and evolutionary tendency.

4.2. Optimization of type-1 and type-2 FLSs

Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx

Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx

5. Simulation results of double inverted pendulum

Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx

_ h1 ; h_1 ; h2 ; h_2 ; F), denoted as the

Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx

5.2. The experimental simulation of different FLSs

Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx

Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx

Z. Sun et al. / Applied Mathematical Modelling xxx (2014) xxxxxx

You might also like