2019 Similarity-Based Non-Singleton General Type-2 Fuzzy Logic Controller With Applications To Mobile Two-Wheeled Robots

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

Journal of Intelligent & Fuzzy Systems xx (20xx) x–xx 1

DOI:10.3233/JIFS-190683
IOS Press

1 Similarity-based non-singleton general


2 type-2 fuzzy logic controller with

f
applications to mobile two-wheeled robots

roo
3

4 Qian Yua , Songyi Dianb , Yong Lic , Jiaxin Liud and Tao Zhaob,∗
a Qian

rP
5 Yu, College of Electrical Engineering, Sichuan University, Chengdu, China
6
b College of Electrical Engineering, Sichuan University, Chengdu, China
7
c State Grid Shandong Electric Power Company, Jinan, China

8
d Electric Power Research Institute of State Grid Liaoning Electric Power Co., Ltd., Shenyang, Liaoning, China

tho
Abstract. Since the non-singleton fuzzy logic controllers (NFLCs) can effectively reflect the uncertainty brought by the
Au
10 inputs, they are used for balance control and position control of the mobile two-wheeled self-balancing robot (MTWSBR)
11 in this paper. The similarity between the inputs and the antecedent fuzzy sets as the firing strength, that is, the similarity-
12 based non-singleton fuzzy logic controller (Sim-NFLC), is proposed to deal with the problem of information loss caused by
13 the standard non-singleton fuzzy logic controller (Sta-NFLC) in terms of the interaction of the inputs and antecedents. A
14 comparative study among singleton fuzzy logic controllers (SFLCs), Sta-NFLCs and Sim-NFLCs, and interval type-2 fuzzy
15 logic controllers (IT2FLCs) and general type-2 fuzzy logic controllers (GT2FLCs) are also shown. The simulation results
d

16 show that the performance of Sim-NFLCs is better than that of SFLCs and Sta-NFLCs. The similarity-based general type-2
17 fuzzy logic controller (Sim-NGT2FLC) gets the best performance in handling the input uncertainty.
cte

18 Keywords: Mobile two-wheeled self-balancing robot, non-singleton general type-2 fuzzy logic controllers, similarity measure,
19 firing strength
rre

20 1. Introduction the ability of deal with uncertainties becomes an 30

indispensable module for the MTWSBR. 31


co

21 Mobile robot system is a typical multi-variable, Fuzzy logic controllers (FLCs) have excellent per- 32

22 nonlinear, naturally unstable system [1–3]. Mobile formance in terms of dealing with uncertainty, and 33

23 two-wheeled self-balance robot (MTWSBR) has is widely used in nonlinear systems [5–8]. Many 34

24 been developed rapidly, because it has been widely FLCs have been applied to control the balance and 35
Un

25 used in our daily life. In real applications, the position of the MTWSBR in previous papers, which 36

26 MTWSBR is affected by environmental disturbances mainly are singleton fuzzy logic controllers (SFLCs) 37

27 and parameter perturbations. Sensor measurements [9–13]. The singleton fuzzifier maps a clear value to a 38

28 may appear numbers of uncertainties, incompleteness fuzzy singleton number [14] (When x = x , μ(x) = 39

29 and possibly inaccurate information [4, 5]. Therefore, 1; when x =/ x , μ(x) = 0). The SFLC is simple and 40

impersonal, but cannot handle the uncertainties in 41

∗ Corresponding author. Tao Zhao, College of Electrical Engi-


inputs well. Therefore, the expected performance can 42

neering, Sichuan University, Chengdu 610065, China. E-mail: not implemented by the SFLC in a closed-loop system 43

zhaotaozhaogang@126.com. with measurement noises. 44

ISSN 1064-1246/19/$35.00 © 2019 – IOS Press and the authors. All rights reserved
2 Q. Yu et al. / Similarity-based non-singleton general type-2

45 Non-singleton fuzzy logic controller (NSFLC) is and antecedent FSs. Sim-NGT2FLC can better han- 97

46 an extension of the SFLC. The input signal is blurred dle uncertainty to improve the position control of 98

47 into a fuzzy set (FS) rather than a clear value (When MTWSBR in an unknown environment. 99

48 x = x , μ(x) = 1; when x = / x , μ(x) =


/ 0). There- The main academic contributions of the paper are 100

49 fore, the performance of NSFLC is superior to the provided as follows: 101

50 SFLC in the face of multiple uncertainties [5, 8, r A similarity-based general type-2 T-S fuzzy 102
51 15]. In recent years, the general type-2 fuzzy logic
logic model is used to approximate the mathe- 103
52 controller (GT2FLC) has obtained some results in
matical model of MTWSBR for enhancing the

f
104
53 theory and application: [16] uses GT2FLC to process
system’s ability to handle input uncertainties.

roo
r
105
54 dynamic uncertainty for synchronization of uncer-
Sim-NGT2FLC is first applied to adjust the bal- 106
55 tain fractional order chaotic systems; GT2FLC has
ance and position of MTWSBR simultaneously 107
56 been applied to control a mobile robot in [17], which
in the presence of uncertainties.
r
108
57 achieved better results than traditional fuzzy control;
A large number of comparative studies between 109
58 [18] studies on forecasting problems based on the
SFLCs, Sta-NFLCs and Sim-NFLCs, and

rP
110
59 data of torque and revolutions per minute (rpm) of
IT2FLCs and GT2FLCs are implemented. Sim- 111
60 permanent magnetic drive by general type-2 fuzzy
ulation results show that the effectiveness of the 112
61 logic systems, etc. The interval type-2 fuzzy logic
proposed Sim-NGT2FLC for the MTWSBR. 113
62 controller (IT2FLC) outperforms the type-1 fuzzy

tho
63 logic controller (T1FLC), especially under differ- The structure of this paper is as follows. Section 114

64 ent kinds of uncertainties. General type-2 fuzzy set II mainly introduces the GT2FS. Section III briefly 115

65 (GT2FS), as an extension of interval type-2 FS, can introduces the structural composition of the GT2FLC 116

66 better handle real systems with uncertainty. There- and the definition of the firing strength; Section IV 117

67 fore, the GT2FLC obtain better performance than presents the dynamic model of MTWSBR and the 118
Au
68 that of the IT2FLC [17, 19–24]. Compared with the balance and position Sim-NGT2FLCs of MTWSBR. 119

69 type-1 FS and interval type-2 FS, the degrees of free- In section V, the corresponding simulation results and 120

70 dom of GT2FS have been expanded. Therefore, it discussions are presented. Section VI draws conclu- 121

71 has potential advantages in dealing with uncertain sions and provides the direction of future work. 122

72 complex system. Compared with the non-singleton


T1FLC and the non-singleton IT2FLC, the non-
d

73

74 singleton GT2FLC (NGT2FLC) may provide better 2. General type-2 fuzzy set 123
cte

75 performance in the presence of input uncertainties,


76 but few works have been implemented in the design Figure 1 shows the three-dimensional and top view 124

77 of NGT2FLC. Therefore, designing NGT2FLC is of the GT2FS. GT2FSs are 3-D MF whose secondary 125

78 still major significant and challenging for practical MFs are type-1 FSs. The secondary membership 126

79 systems. degree is μA


(x, u) ∈ [0, 1] (the IT2FS is the special 127
rre

80 Besides the fuzzifier, the NSFLC has the same case of the GT2FS, and its secondary membership 128

81 structure as the SFLC. In the inferencing of fuzzy degrees are 1). LMF (A)  and UMF (A)  are defined as 129

82 logic systems, the firing strengths are obtained lower MF and upper MF, respectively. The footprint 130

83 by interacting between the inputs and antecedent of uncertainty (FOU) is composed of the main MFs 131
co

84 membership functions (MFs). Standard NFLC (Sta- between the upper and lower membership functions. 132

85 NFLC) is most commonly used. The firing strength of In order to reduce the computational complexity of 133

86 Sta-NFLC is determined by the maximum member- the GT2FLC, the secondary MFs are cut into multi- 134

87 ship grade of the intersection between the inputs and level horizontal slices (α -cuts) [17, 26]. Each slice is 135
Un

88 antecedent MFs. However, using the maximum point combined in some way to get the 3-D GT2FS. Com- 136

89 of the intersection to determine the firing strength monly used slicing methods are vertical slice and 137

90 results in a large amount of information loss between horizontal slice. 138

91 the inputs and the antecedent MFs [5, 8, 15, 25].


92 Aiming at these problems, this paper proposes 2.1. Vertical slice representation of a GT2FS 139

93 similarity-based NGT2FLC (Sim-NGT2FLC) to


94 adjust the balance and position of the MTWSBR. The vertical slice of a GT2FS is shown in Fig. 2.
95 In this approach, the firing strength of each rule The corresponding of the vertical slice to each x is
96 is calculated by the similarity between the inputs described as:
Q. Yu et al. / Similarity-based non-singleton general type-2 3

f
roo
rP
Fig. 3. Horizontal slice representation of a GT2FS.


The α-cut (α ∈ [0, 1]) of the secondary MF A(x)
is defined as follows [14]:

tho
 α (x) = {u|μ (u) ≥ α} = [aα (x), bα (x)]
A (3)

A(x)

In the formula (3), the interval [aα (x), bα (x)] is


located on the y axis. We can also combine each slice
to get the 3-D general type-2 MF (GT2MF) (A). 
Au
 
 
∀α∈[0,1] [α/Aα (x)]
A= . (4)
∀x∈X x
Fig. 1. General type-2 fuzzy set. (a) Three-dimensional represen-
d

tation; (b) top view.


2.2. Horizontal slice representation of a GT2FS 140
cte

With formula (4), we can get the horizontal slice


representation of a GT2FS.

 [α/A  α (x)]

A=
rre

∀x∈X x
∀α∈[0,1]
 (5)
 α
=

[  α (x)/x]
A
∀α∈[0,1] ∀x∈X
co

The horizontal slice can be regarded as an IT2FS 141

whose secondary membership degree is equal to α 142

(instead of 1 as standard IT2FS). Thus it not only 143


Un

can simplify the computational but can use all the 144
Fig. 2. Vertical slice representation of a GT2FS.
calculations of IT2FSs to deal with GT2FSs [28–30]. 145
  Therefore, the horizontal slice is used in this paper. 146
= A(x)
A . (1)
x∈X x
2.3. Secondary MF of a GT2FS 147
where

μ (u) Common shapes of the secondary MFs are triangle

A(x) = A
. (2)
148

∀u∈[0,1] u and symmetrical trapezoid [27, 28, 31]. This paper 149

uses triangle as the secondary MF. 150


4 Q. Yu et al. / Similarity-based non-singleton general type-2

f
Fig. 5. Structure diagram describing a GT2FLC.

roo
GT2MF is a type-1 MF, and the FOU of each 171

α-plane is an interval type-2 MF. The FOU of 172

Fig. 4. Horizontal slice representation of a GT2FS.


a standard GT2MF is a type-1 MF when α = 1 173

[14].

rP
174

As shown in Fig. 6, according to the type of blur- 175


It can be observed from Fig. 4 that the two triangle ring, four fuzzifiers of the GT2FLC are built: 176
base angle points of the secondary MF are respec- singleton fuzzifier, type-1 non-singleton fuzzi- 177
tively located in the UMF (μAx (x)) and the LMF fier, interval type-2 non-singleton fuzzifier and 178

tho
(μA (x)) of the GT2FS, and the triangle vertex is general type-2 non-singleton fuzzifier.
r
x 179

at α = 1. α-cuts of the triangles are as follows:(w ∈ Fuzzy rule: IF-THEN rules are commonly used.
[0, 1]) Suppose there is p inputs for x = [x1 , ..., xp ] ∈
⎧ X, an output of y ∈ Y , and M rules, where the

⎨Aα (x) = [aα (x), bα (x)] lth rule is as follows [16]:
Au
aα (x) = μA (x) + w[μA (x) − μA (x)]α (6)

⎩  l , ..., xm is A
 l , ..., xp is A
l
bα (x) = μA (x) − (1 − w)[μA (x) − μA (x)]α Rl : If x1 is A 1 m p

151 When w = 0, the secondary MF is a right-angled l


Then y is G l = 1, ..., M
152 triangle whose right angled is perpendicular; when
Where xm is the input of a non-singleton num-
w = 1/2, the secondary MF is an isosceles triangle;
180
d

153
 l is the antecedent of the
ber for the lth rule, A
when w = 1, the secondary MF is a left-angled trian-
181
m
 is the consequent of the lth rule.
154
l
α r
lth rule, G
cte

182
155 gle whose left angled is perpendicular. The α-cut A
156 in formula (3) can be obtained by calculating formula Inference engine: it combines fuzzified inputs 183

157 (6). with IF-THEN rules using a t-norm to get the 184

firing strength .
r
185

Type-reduction: the inference output GT2FS by 186


rre

calculating the center of gravity is converted 187


158 3. General Type-2 fuzzy logic controller
into a type-1 FS.
r
188

Defuzzifier: defuzzification combines all acti- 189


159 This section will briefly introduce the background
vated output sets in a way to get a clear output 190
160 material, including the structural composition of a
value.
co

191
161 GT2FLC and two methods of obtaining the fir-
162 ing strength (standard method and similarity-based
163 method). 3.2. Definition of firing strength 192
Un

164 3.1. Structural composition of a GT2FLC In this section, two approaches for determining the 193

firing strength are briefly introduced, namely the stan- 194

165 Figure 5 shows the structure diagram of a GT2FLC, dard method and the similarity-based method. In most 195

166 which includes five modules as follows: practical applications, input data is usually corrupted 196

by noise. Therefore, capturing noise (uncertainty)


r Fuzzifier: The general type-2 non-singleton
197

167 becomes especially important. 198

168 fuzzifier maps the input x = x into a general Suppose there are two given FSs, A (transform the
169 type-2 fuzzy number, which is a GT2MF asso- input data x into a non-singleton MF) and B (for the
170 ciated with the input. The secondary MF of the antecedent MF). (α ∈ [0, 1])
Q. Yu et al. / Similarity-based non-singleton general type-2 5

f
roo
rP
tho
Au
Fig. 6. Different types of fuzzifier. (a)singleton fuzzifier; (b) type-1 non-singleton fuzzifier; (c) interval type-2 non-singleton fuzzifier; (d)
General type-2 non-singleton fuzzifier.

 α (x) = {u|μ (u) ≥ α} ≡ [aα (x), bα (x)]


A (7)

d

A(x)

 α (x) = {u|μ (u) ≥ α} ≡ [cα (x), dα (x)]


cte

B 
B(x)
(8)

199 3.2.1. Standard firing strength definition


The standard reasoning method is most common
rre

in the design of FLC. As can be observed in Fig. 7,


the firing strength of Sta-NFLC is determined by the
maximum membership degree grade of the intersec-
tion between the inputs and antecedent MFs [8, 14]. Fig. 7. The intersection of two different input fuzzy sets X1 and
When x = x , the firing strengths of lth rule with the
co

X2 with a single antecedent fuzzy set A.


secondary membership degree α are as follows:

Fαl (x ) ≡ α/[f lα (x ), f α (x )]
l



⎪ respectively. In discrete systems, including most 202

Un


⎪ l  fuzzy logic system applications,  represent the
⎪ f α (x ) = sup(μ
⎪ 203
(x)  μ (x))
⎨ A B t-norm function, and the product and minimum oper- 204

= max(aα (x)  cα (x)) (9) ations are often used. This paper uses the minimum


205


⎪ operation. sup denotes the supremum for all x, and

⎪ f α (x ) = sup(μ
l
(x)  μ (x))
206


⎪ A A the maximum operation is generally adopted.


207

= max(bα (x)  dα (x)) Recent literature indicates that the standard method 208

not only results in substantial information loss in 209


200 Among them, The upper and lower firing strengths terms of the interaction of the inputs and antecedent 210
l
201 of the lth rule under α level are f lα (x ) and f α (x ), MFs, but is unable to separate of input uncertainty and 211
6 Q. Yu et al. / Similarity-based non-singleton general type-2

212 linguistic antecedent uncertainty [5, 15, 25]. Aiming between inputs and antecedent FSs with the highest 235

213 at these problems, this paper shows two examples in fidelity possible [15]. In order to improve this prob- 236

214 Figs. 7 and 8. From the two figures, we can see that lem, the similarity-based inference method is used in 237

215 two different degrees of fuzzy input X1 (dark red) and this paper. In this approach, the firing strength of each 238

216 X2 (light red) interact with the same antecedent MF A rule is calculated by the similarity between the inputs 239

217 (yellow), and different firing strengths are expected. and antecedent FSs instead of being calculated by the 240

218 However, the result shows two firing strengths are standard approach.
241
219 equal with the standard method to determine the

f
l

⎪ Fαl (x ) ≡ α/[S lα (x ), S α (x )]

roo

⎪ 



⎨ l  min(aα (x),cα (x))dx min(bα (x),dα (x))dx
S α (x ) = min
x∈X
,
x∈X
max(aα (x),cα (x))dx max(bα (x),dα (x))dx (10) 242

⎪ 
x∈X

x∈X 



⎩S lα (x ) = max
x∈X
min(aα (x),cα (x))dx
,
x∈X
min(bα (x),dα (x))dx
max(aα (x),cα (x))dx max(bα (x),dα (x))dx

rP
x∈X x∈X
243

⎧ l

⎪Fαl (x ) ≡ α/[S lα (x ), S α (x )]

⎪  n n 

⎨ l  min(aα (x),cα (x))dx min(bα (x),dα (x))dx
S α (x ) = min  i=1
n 
, ni=1

tho
max(aα (x),cα (x))dx max(bα (x),dα (x))dx (11) 244

⎪   i=1
i=1 

⎪ n n

⎩S lα (x ) = max ni=1
min(aα (x),cα (x))dx

, n i=1
min(bα (x),dα (x))dx

i=1
max(aα (x),cα (x))dx i=1
max(bα (x),dα (x))dx
220 firing strengths. Even if large difference exists among
221 the input and the antecedent MF in Fig. 8, the fir-
Au
222 ing strengths of standard method are 1. Through the 3.2.2. Similarity-based firing strength definition 245

223 above two cases, the magnitude of the firing strength Similarity is often used on FSs to measure the sim- 246

224 ignores the uncertainty brought by the input in some ilarity degree between two FSs [15, 43]. Similarity 247

225 special cases. In other words, the input uncertainties function s: A × B −→ [0, 1], where A and B are the 248

226 are not reflected in the firing strenth and act on the same type of FS. It must satisfy the following four 249
d

227 system output. attributes [15, 32, 33]: 250

228 Based on the above discussions, using the maxi-  B)


 = 1 ⇐⇒ A  = B.

cte

mum point of the intersection to determine the firing 1) Reflexivity: s(A, 251
229
   
2) Symmetry: s(A, B) = s(B, A). 252
230 strength results in a large amount of information  ≤B  then s(A,
 ≤ C,  B)
 ≥
loss between the inputs and the antecedent MFs and 3) Overlapping: if A 253
231
 
s(A, C). 254
232 imprecise firing strength is likely to appear. The  ∩B=  B)
 > 0;
standard inference method may not necessarily be 4) Transitivity: if A / ?, then s(A, 255

rre

233

otherwise s(A, B) = 0. 256
234 the best choice in terms of capturing the interaction
This paper adopts the ZXLD similarity measure 257

function for GT2FSs [33]. This similarity measure 258

function first defined the similarity between two


co

259

GT2FSs as a Type-1 FS [34]. When x = x , in the 260

lth rule, the similarity is shown in formula (10). In 261

the discrete domain, formula (11) can be obtained, 262


Un

where n is the number of discretization levels (we set 263

n = 100). 264

4. Main results 265

For further analysis the performance of the pro- 266


Fig. 8. The intersection of two different input fuzzy sets X1 and X2
with a single antecedent fuzzy set A. Note that the firing strengths
posed Sim-NGT2FLC, it is used to adjust the balance 267

for the standard NSFLS approach are equal to one. and position of the MTWSBR. 268
Q. Yu et al. / Similarity-based non-singleton general type-2 7

269 4.1. Mathematical model of the MTWSBR and


270 similarity-based GT2 T-S fuzzy modeling

271 Let the state variable be x = [x1 , x2 , x3 , x4 ]T =


272 [θ, θ̇, ψ, ψ̇]T , where θ ∈ [− π6 , π6 ]. The basic physical
273 parameters are as follows [35, 36]: θ is the inclination
274 angle of the MTWSBR; ψ is the rotation angle of two
275 wheels; The mass of the main body is mb = 2.58kg;

f
276 The mass of each wheel is mw = 0.14kg; The radius

roo
277 of each wheel is r = 0.04m; The half of the distance
278 between two wheels is b = 0.15m;
279 Rule 1 : If θ is X1 , then ẋ
⎡ ⎤ ⎡ ⎤
0 1 0 0 0

rP
⎢ ⎥ ⎢ ⎥
⎢ 93.31 −0.044 0 0.296 ⎥ ⎢ −22.11 ⎥
=⎢⎢
⎥ x +⎢
⎢ 0 ⎥u,

0 1 ⎥
280
⎣ 0 0 ⎦ ⎣ ⎦
−123.36 0.47 0 −4.88 233.37

tho
281

282 Rule 2 : If θ is X2 , then ẋ


⎡ ⎤ ⎡ ⎤
0 1 0 0 0
⎢ ⎥ ⎢ ⎥
⎢ 79.26 −0.038 0 0.22 ⎥ ⎢ −20.04 ⎥
=⎢ ⎥ x +⎢ ⎥
Au
283
⎢ 0 0 0 1 ⎥ ⎢ 0 ⎥u.
⎣ ⎦ ⎣ ⎦
−90.74 0.45 0 −4.78 227.23

The distance from the wheel axle to the center of


gravity is l = 0.0622m; The moment of inertia of
the body about the Y axis is nyb = 0.00128kg · m2 ;
d

The moment of inertia of each wheel about its axis is Fig. 9. The antecedents of the general type-2 T-S fuzzy model. (a)
Horizontal slice representation; (b) Three-dimensional represen-
nwa = 0.00014kg · m2 ; τi is the combination of mea-
cte

tation.
surement uncertainties and external disturbances of
the system (i = 1, 2, 3). Referring to [36], we can get
the mathematical model of the MTWSBR, as shown
in (12).

rre

cosine functions. The nonlinearity can be elimi- 289


⎪x˙1 = x2

⎪ nated by locally approximating (let x1 reaching

290

⎪x˙2 = 1 (m212 x22 cos x1 sin x1 − 2m12 cos x1 dw x4 0 and ± π6 ), and then, the T-S fuzzy model is


291



⎪ −m11 gb sin x1 + 2m12 cos x1 db (x2 − x4 ) obtained [37–39]. Two general type-2 fuzzy rules 292


⎪ Rule1 and Rule2 are used to describe the MTWSBR
co


293
+2m11 db (x2 − x4 )) + (m11 + m12 cos x1 )u2
mathematical model by generalizing the existing 294


⎪x˙3 = x4 type-1 and interval type-2 fuzzy modeling methods

295


⎪x˙4 = 1 (−m22 m12 x22 sin x1 + 2m22 dw x4

[36, 40]. 296

Un


⎪ Figure 9 shows the 3-D GT2MFs of the antecedent 297

⎪ −2m22 db (x2 − x4 ) + m12 gb cos x1 sin x1

⎪ Xi . The mathematical expression of the secondary 298

⎩ MF is shown in formula (6). It can be seen that the
−2m12 db (x2 − x4 )cos x1 )+(−m22 − m12 cos x1 )u4 299

two upper and lower membership functions of the 300


(12)
antecedents are complementary to reduce the com- 301
284 where u is the input of the MTWSBR system, and putational complexity and facilitate analysis. The 302
285  = m212 cos2 θ − m11 m22 , m11 = (mb + 2mw )r 2 + standard firing interval and similarity-based firing 303
286 2nwa , m12 = mb lr, m22 = mb l2 + nyb , gb = mb gl. interval can be obtained according to the formula (9) 304

287 It can be seen from formula (12) that the non- and formulas (10)-(11), respectively. The expressions 305

288 linearity of the system is caused by sine and of the antecedents of the general type-2 T-S fuzzy 306
8 Q. Yu et al. / Similarity-based non-singleton general type-2

307 model are as follows:


⎧ ⎧ θ+π/7

⎪ ⎪ + w[ θ+π/6 − θ+π/7 ], θ<0
⎪ ⎨ π/7 π/6 π/7



⎪aα = 1 ,
1
θ=0

⎪ ⎪
⎩ π/7−θ

⎪ + w[ π/6 − π/7 ] , θ > 0
π/6−θ π/7−θ




π/7
⎧ θ+π/6

⎪ − (1 − w)[ θ+π/6 − θ+π/7

⎪ ⎪
⎨ π/6 π/6 π/7
], θ<0



⎪bα = 1 ,
1
θ=0

⎪ ⎪

f

⎨ ⎩ π/6−θ
− (1 − w)[ π/6 − π/7 ] , θ > 0
π/6−θ π/7−θ

roo
π/6
308 ⎧ −θ (13)

⎪ ⎪
−θ
+ w[ π/6 −θ
− π/7 ], θ<0

⎪ ⎨ π/7



⎪aα2 = 0 , θ=0

⎪ ⎪
⎩ θ

⎪ + w[ π/6 − π/7

θ θ
⎪ ], θ>0

⎪ ⎧
π/7

rP

⎪ ⎪
−θ −θ
− (1 − w)[ π/6 −θ
− π/7 ], θ<0

⎪ ⎨ π/6

⎪bα = 0 , θ=0


2

⎩ ⎪
⎩ θ
π/6
− (1 − w)[ π/6 −
θ θ
π/7
], θ>0

tho
309

In order to directly get the clear output and reduce


the complexity of the type-reduction calculation, the Fig. 10. Overall control diagram for the MTWSBR.
N-T defuzzification method [36, 41, 42] is used for
the GT2 T-S fuzzy model in this paper.
Au
2
i
[f i (θ)+f α (θ)]/2 Where Xi is the antecedent and u = Ui x is the con-
2 (Ai x + Bi u)
αk k
αk sequent. Then the N-T algorithm is used to get the
 i=1
kmax j
[f j (θ)+f α (θ)]/2
ẋ =
j=1 αk k defuzzified output:
kmax
k=1 k=1 αk
2
i
[f i (θ)+f αs (θ)]/2
2
i
[f i (θ)+f α (θ)] αs 2 αs
Ui x
d


kmax
αk
αk k
(Ai x + Bi u) 
smax i=1 j
[f j (θ)+f αs (θ)]/2
u= smax
i=1 j=1 αs
= 2
kmax
s=1 αs
cte

k=1 k=1 αk s=1

2
i
kmax 
 2

smax [f i (θ)+f αs (θ)]
αs
= mik (θ)(Ai x + Bi u) αs Ui x
= i=1
smax 2
k=1 i=1
s=1 s=1 αs
(14)
rre

310 Compared with the existence results [36, 44–47], 


smax 
2
311 this paper proposes a similarity-based general type-2 = mis (θ)Ui x
312 T-S fuzzy model. With similarity reasoning, the cor- s=1 i=1
313 responding calculation expression of firing strength (15)
co

314 has been improved. It is noted that the firing strengths are obtained 321

by the similarity between the inputs and antecedent 322

FSs. Substituting formula (17) into (14), we can get 323


315 4.2. Balance and position controller design
the formula (16). Ui is the local feedback gains. It 324
Un

can be obtained by Lyapunov stability theory. The 325


316 Figure 10 shows the overall control diagram of the
detailed process to solve Ui is similar to [36, 40]. 326
317 MTWSBR, where the balance controller and position
When Ui is determined, the singleton general type- 327
318 controller use respectively the similarity-based gen-
2 balance controller is extended the similarity-based 328
319 eral type-2 T-S fuzzy model and the similarity-based
general type-2 balance controller. 329
320 general type-2 Mamdani fuzzy model [36].
The Mamdani fuzzy rules for the similarity-based
The fuzzy rules of the similarity-based general
general type-2 position controller are as follows:
type-2 balance controller are as follows:
Ri : If θ is Xi , then u = Ui x. i = 1, 2 Ri : If pe is De and θ̇ is Dθ̇ , then θp is Dp .
Q. Yu et al. / Similarity-based non-singleton general type-2 9

Table 1
Ranges of the input and output variables
Variable Range
pe (m) [−3, 3]
θ̇(rad/s) [−4, 4]
θp (rad) [−π/40, π/40]

Table 2

f
Fuzzy rules of the GT2FLCs for position control

roo
Rules pe
NB NM NS ZO PS PM PB
NB PB PB PB PB PM PS ZO
NM PB PB PB PM PS ZO NS
NS PB PB PM PS ZO NS NM
θ̇ ZO PB PM PS ZO NS NM NB

rP
PS PM PS ZO NS NM NB NB
PM PS ZO NS NM NB NB NB
PB ZO NS NM NB NB NB NB

tho
Au
Fig. 11. FOU of the antecedents and consequents. (a) Three-
dimensional representation; (b) top view.
d

330 where pe is the position error, and θ̇ is the inclination


Fig. 12. The balance control only without random disturbance.
angle rate. De and Dθ̇ are the antecedents, and Dp is
cte

331

332 the consequent. Their MFs are shown in Fig. 11, and The rules in Table 2 have been shown in the lit- 342

333 the secondary MF are defined as triangle in formula erature [36]. Compared with the existing results [36, 343

334 (6). The range of the input and output variables in the 47], the similarity reasoning algorithm is used for 344

335 position controller are shown in Table 1. As can be the balance and position controllers in this paper. 345
rre

336 seen, the area of the antecedents and consequents in [36] has used singleton IT2FLC. [47] has used the 346

337 the position controller are divided into seven linguis- standard firing interval. However, the general type- 347

338 tic variables named as negative big (NB), negative 2 fuzzy logic, non-singleton fuzzifier and similarity 348

339 medium (NM), negative small (NS), zero (ZO), pos- reasoning algorithm have been applied to improve 349

itive small (PS), positive medium (PM) and positive the existing results. It has ability to get better control
co

340 350

341 big (PB). results. 351

2   max 2 
 Ai x + Bi ss=1
Un

kmax α
k i=1 mik (θ) j=1 mjs (θ)Uj x
ẋ = kmax
k=1 k=1 αk
2  smax 2 smax 2 

kmax α
k i=1 m ik (θ) s=1 j=1 m js (θ)A i x + s=1 j=1 m js (θ)B i U j x
= kmax (16)
k=1 k=1 αk


kmax 
smax 
2 
2
= mik (θ)mjs (θ)(Ai x + Bi Uj x)
k=1 s=1 i=1 j=1
10 Q. Yu et al. / Similarity-based non-singleton general type-2

352 5. Simulation results and discussions

353 When only the balance controller is working, initial


354 state is set to x0 = [0.48, 0, 0, 0]T , and the control
355 performance of SGT2FLC, Sta-NGT2FLC and Sim-
356 NGT2FLC are contrasted.
357 It can be seen from Figs. 12 and 13 that, compared
358 with SGT2FLC and Sta-NGT2FLC, Sim-NGT2FLC

f
359 has high response speed, high stability and good con-

roo
360 trol effect.
361 When the balance and position controllers are
362 tested together, the expected position is selected
Fig. 13. The balance control only with random disturbances in the
363 as Dd = 0.7. To illustrate the effectiveness of the control variable.
364 controller proposed in this paper, a comparative

rP
365 study between SGT2FLC, Sta-NGT2FLC and Sim-
366 NGT2FLC, and Sim-NGT2FLC and Sim-NIT2FLC position in comparison to the SFLCs and the Sta- 377

367 is also shown. NFLCs. Comparing the results of the Sim-NGT2FLC 378

368 In order to visually display the performance of dif- and the Sim-NIT2FLC, it can be seen obviously 379

that the Sim-NGT2FLC is more rapid to reach the 380

tho
369 ferent FLCs, the performance indicators, including
370 ISE,IAE and ITAE, are used in this paper [48–50]. desired position and has better control performance. 381

371 The performance indicators under different cases are In the 6th second, the impulse noise (width is 0.1s 382

372 computed in Tables 3–6. In these tables, b is the incli- and amplitude is 1) is added to the state x1 . For the 383

373 nation angle error of the balance controllers, and p is impulse noise, the result of the SGT2FLC is diver- 384

gent and cannot be stabilized; The adjustment time


Au 385
374 the position error of the position controllers.
375 It can be observed from Figs. 14–18 and Tables under the Sim-NGT2FLC is less than that under the 386

376 3–6 that the Sim-NFLCs are faster to reach the desired Sta-NGT2FLC and the Sim-NT2FLC. Observing the 387

error integration of the position, the Sim-NFLCs have 388


d
cte
rre
co
Un

Fig. 14. Results of the GT2FLCs using different firing strength methods.
Q. Yu et al. / Similarity-based non-singleton general type-2 11

f
roo
rP
tho
Fig. 15. Results of the GT2FLCs using different firing strength methods with impulse noise (width is 0.1s and amplitude is 1).
d Au
cte
rre
co
Un

Fig. 16. Results of the IT2FLCs using different firing strength methods.
12 Q. Yu et al. / Similarity-based non-singleton general type-2

Table 3
Performance index results of the IT2FLCs with different firing strength methods
Performance b p
index SIT 2FLCs Sta − NIT 2FLCs Sim − NIT 2FLCs SIT 2FLCs Sta − NIT 2FLCs Sim − NIT 2FLCs
ISE 0.0220 0.0041 0.0299 0.8710 0.8027 0.5739
IAE 0.2486 0.1439 0.2434 2.9254 3.1196 2.0405
ITAE 0.9402 0.9206 0.6782 13.4601 17.3007 7.3591

Table 4

f
Performance index results of the GT2FLCs with different firing strength methods

roo
Performance b p
index SGT 2FLCs Sta − NGT 2FLCs Sim − NGT 2FLCs SGT 2FLCs Sta − NGT 2FLCs Sim − NGT 2FLCs
ISE 0.0483 0.0101 0.0412 0.5693 0.3835 0.4685
IAE 0.2675 0.1826 0.2765 2.1803 1.4678 1.4239
ITAE 0.7500 0.7188 0.4487 8.7603 3.7789 2.9648

rP
Table 5
Performance index results of the GT2FLCs using different firing strength methods with impulse noise
Performance b p
index SGT 2FLCs Sta − NGT 2FLCs Sim − NGT 2FLCs SGT 2FLCs Sta − NGT 2FLCs Sim − NGT 2FLCs

tho
ISE 190.2728 32.7630 57.9540 8653.1984 84.3955 12.5258
IAE 44.2243 6.1745 9.5101 297.3282 25.5824 6.0494
ITAE 540.3570 43.9953 63.1939 4570.2652 245.5158 36.8715
d Au
cte
rre

Fig. 17. Compare Sim-NGT2FLC with Sim-NIT2FLC for the


position control. Fig. 18. Compare Sim-NGT2FLC with Sim-NIT2FLC for the
position control with impulse noise (width is 0.1s and amplitude
is 1).
co

Table 6
Performance index results of the Sim-GT2FLC and Sim-IT2FLC
for the position control with impulse noise In summary, all the results show that, when there 392

Performance p are input uncertainties, the Sim-NFLC is much more 393


Un

index Sim − NIT 2FLC Sim − NGT 2FLC stable and more rapid to reach the desired posi- 394

ISE 12.2844 12.5258 tion. Meanwhile, the Sim-NGT2FLC proposed in this 395

IAE 6.7622 6.0494 paper is feasible and effective by the above simulation 396
397
ITAE 44.2050 36.8715 results.

6. Conclusions and future work 398

389 the smallest error integral by lateral correlation; the


390 Sim-NGT2FLC has the smallest error integral by lon- In this paper, a Sim-GT2FLC is proposed to adjust 399

391 gitudinal comparison. the balance and position of the MTWSBR. In order 400
Q. Yu et al. / Similarity-based non-singleton general type-2 13

401 to reduce information loss in terms of the interac- [8] D. Pekaslan, J.M. Garibaldi and C. Wagner, Exploring sub- 451

402 tion of the inputs and antecedents as possible, the sethood to determine firing strength in non-singleton fuzzy 452
logic systems[C], 2018 IEEE International Conference on 453
403 firing strength of each rule is calculated by using the Fuzzy Systems (FUZZ-IEEE). IEEE (2018), 1–8. 454
404 similarity between the inputs and the antecedent FSs [9] C.H. Huang, W.J. Wang and C.H. Chiu, Design and 455

405 rather than the maximum of their intersection. The implementation of fuzzy control on a two-wheel inverted 456

406 simulation results show that, in the presence of input pendulum[J], IEEE Transactions on Industrial Electronics 457
58(7) (2011), 2988–3001. 458
407 uncertainties, the Sim-NFLCs can obtain superior [10] Z. Guo, J.X. Xu and T.H. Lee, A gain-scheduling opti- 459
408 control performance compared to the Sta-NFLCs. By mal fuzzy logic controller design for unicycle[C], Ieee/asme 460

f
409 comparing with the control performance of the Sim- International Conference on Advanced Intelligent Mecha- 461

roo
410 NIT2FLC and other controllers, the Sim-NGT2FLC tronics. IEEE (2009), 1423–1428. 462
[11] S. Dian, L. Wei-bo and T. Zhao, Interval Type-2 Fuzzy 463
411 in this paper shows the superiority to control the Logic Control for a Two-Wheeled Mobile Robot Based on 464
412 MTWSBR. Improved QPSO[J], Kongzhi Yu Juece/control and Decision 465

413 In future work, we will focus on reducing the com- (2018), 1–8. 466
[12] L. Cervantes and O. Castillo, Type-2 fuzzy logic aggregation 467
414 plexity of the NGT2FLCs and improving the ability

rP
of multiple fuzzy controllers for airplane flight control[J], 468
415 to deal with higher-level uncertainties. Moreover, it Information Sciences 324 (2015), 247–256. 469
416 is very important to verify the proposed method by [13] M. El-Bardini and A.M. El-Nagar, Interval type-2 fuzzy 470

experiments. PID controller for uncertain nonlinear inverted pendulum 471

417 system[J], ISA transactions 53(3) (2014), 732–743. 472


[14] J.M. Mendel, Uncertain rule-based fuzzy logic system: 473

tho
introduction and new directions[J], (2001). 474
[15] C. Wagner, A. Pourabdollah, J. McCulloch, et al. A 475
418 Acknowledgment similarity-based inference engine for non-singleton fuzzy 476
logic systems[C], 2016 IEEE International Conference on 477

419 This work is supported by the National Key Fuzzy Systems (FUZZ-IEEE). IEEE (2016), 316–323. 478
[16] A. Mohammadzadeh, S. Ghaemi, O. Kaynak, et al. 479
R&D Program of China (2018YFB1307402) and
420
Au
Observer-based method for synchronization of uncertain 480
421 the National Natural Science Foundation of China fractional order chaotic systems by the use of a general 481
(61703291). type-2 fuzzy system[J], Applied Soft Computing 49 (2016), 482
422 544–560. 483
[17] M.A. Sanchez, O. Castillo and J.R. Castro, Generalized 484
type-2 fuzzy systems for controlling a mobile robot and 485
a performance comparison with interval type-2 and type-1 486
d

423 References
fuzzy systems[J], Expert Systems with Applications 42(14) 487
(2015), 5904–5914. 488
cte

424 [1] J. Huang, F. Ding, T. Fukuda, et al. Modeling and Veloc- [18] Y. Chen and D. Wang, Forecasting by general type-2 fuzzy 489
425 ity Control for a Novel Narrow Vehicle Based on Mobile logic systems optimized with QPSO algorithms[J], Inter- 490
426 Wheeled Inverted Pendulum[J], IEEE Transactions on Con- national Journal of Control, Automation and Systems 15(6) 491
427 trol Systems Technology 21(5) (2013), 1607–1617. (2017), 2950–2958. 492
428 [2] S.H. Ri, J. Huang, Y. Wang, et al. Terminal sliding mode [19] J.M. Mendel, General type-2 Fuzzy Logic Controllers made 493
429 control of mobile wheeled inverted pendulum system with simple: a tutorial[J], IEEE Transactions on Fuzzy Systems 494
rre

430 nonlinear disturbance observer[J], Mathematical Problems 22(5) (2014), 1162–1182. 495
431 in Engineering 2014 (2014). [20] C. Wagner and H. Hagras, zSlices based general type-2 FLC 496
432 [3] J. Huang, S. Ri, L. Liu, et al. Nonlinear disturbance observer- for the control of autonomous mobile robots in real world 497
433 based dynamic surface control of mobile wheeled inverted environments[C], Fuzzy Systems, 2009. FUZZ-IEEE 2009. 498
434 pendulum[J], IEEE Transactions on Control Systems Tech- IEEE International Conference on. IEEE (2009), 718–725. 499
co

435 nology 23(6) (2015), 2400–2407. [21] T. Zhao, J. Liu and S. Dian, Finite-time Control for Inter- 500
436 [4] C. Fu, A. Carrio and P. Campoy, Efficient visual odometry val Type-2 Fuzzy Time-delay Systems with Norm-Bounded 501
437 and mapping for unmanned aerial vehicle using ARM-based Uncertainties and Limited Communication Capacity[J], 502
438 stereo vision pre-processing system[C], 2015 International Information Sciences 483 (2019), 153–173. 503
439 Conference on Unmanned Aircraft Systems (ICUAS). IEEE [22] O. Castillo, L. Amador-Angulo, J.R. Castro, et al. A com- 504
Un

440 (2015), 957–962. parative study of type-1 fuzzy logic systems, interval type-2 505
441 [5] C. Fu, A. Sarabakha, E. Kayacan, et al. Similarity-based fuzzy logic systems and generalized type-2 fuzzy logic 506
442 non-singleton fuzzy logic control for improved performance systems in control problems[J], Information Sciences 354 507
443 in UAVs[C], 2017 IEEE International Conference on Fuzzy (2016), 257–274. 508
444 Systems (FUZZ-IEEE). IEEE (2017), 1–6. [23] O. Castillo and L. Amador-Angulo, A generalized type- 509
445 [6] T.A. Runkler, C. Chen and R. John, Type reduction operators 2 fuzzy logic approach for dynamic parameter adaptation 510
446 for interval type C2 defuzzification[J], Information Sciences in bee colony optimization applied to fuzzy controller 511
447 467 (2018), 464–476. design[J], Information Sciences 460-461 (2018), 476–496. 512
448 [7] O. Castillo and P. Melin, Intelligent systems with interval [24] O. Castillo, L. Cervantes, J. Soria, et al. A generalized type- 513
449 type-2 fuzzy logic[J], International Journal of Innovative 2 fuzzy granular approach with applications to aerospace[J], 514
450 Computing, Information and Control 4(4) (2008), 771–783. Information Sciences 354 (2016), 165–177. 515
14 Q. Yu et al. / Similarity-based non-singleton general type-2

516 [25] C. Fu, A. Sarabakha, E. Kayacan, et al. Input uncer- [38] X.P. Xie, D. Yue and C. peng, Multi-instant observer design 565
517 tainty sensitivity enhanced nonsingleton fuzzy logic of discrete-time fuzzy systems: a ranking-based switching 566
518 controllers for long-term navigation of quadrotor UAVs[J], approach[J], IEEE Transactions on Fuzzy Systems 25(5) 567
519 IEEE/ASME Transactions on Mechatronics 23(2) (2018), (2017), 1281–1292. 568
520 725–734. [39] H.Y. Li, S. Yin, Y.N. Pan and H.K. Lam, Model reduc- 569
521 [26] D. Zhai and J.M. Mendel, Uncertainty measures for general tion for interval type-2 Takagi-Sugeno fuzzy systems[J], 570
522 type-2 fuzzy sets[J], Information Sciences 181(3) (2011), Automatica 61 (2015), 308–314. 571
523 503–518. [40] C.H. Huang, W.J. Wang and C.H. Chiu, Design and 572
524 [27] J.M. Mendel, General type-2 Fuzzy Logic Controllers made implementation of fuzzy control on a two-wheel inverted 573
525 simple: a tutorial[J], IEEE Transactions on Fuzzy Systems pendulum[J], IEEE Transactions on Industrial Electronics 574

f
526 22(5) (2014), 1162–1182. 58(7) (2011), 2988–3001. 575

roo
527 [28] J.M. Mendel, F. Liu and D. Zhai, α-Plane Representa- [41] M. Nie and W.W. Tan, Towards an efficient type-reduction 576
528 tion for Type-2 Fuzzy Sets: Theory and Applications[J], method for interval type-2 Fuzzy Logic Controllers[C], 577
529 IEEE Transactions on Fuzzy Systems 17(5) (2009), Fuzzy Systems, 2008. FUZZ-IEEE 2008.(IEEE World 578
530 1189–1207. Congress on Computational Intelligence). IEEE Interna- 579
531 [29] C. Wagner and H. Hagras, zSlicesł towards bridging the gap tional Conference on. IEEE (2008), 1425–1432. 580
532 between interval and general type-2 fuzzy logic[C], Fuzzy [42] D. Wu and W.W. Tan, A simplified architecture for type-2 581

rP
533 Systems, 2008. FUZZ-IEEE 2008. (IEEE World Congress FLSs and its application to nonlinear control[C], Cybernet- 582
534 on Computational Intelligence). IEEE International Con- ics and Intelligent Systems, 2004 IEEE Conference on. IEEE 583
535 ference on. IEEE (2008), 489–497. 1 (2004), 485–490. 584
536 [30] C. Wagner and H. Hagras, zSlices based general type-2 [43] Z. Gao, X. Jian, J. Qiang and Z. Yong, Similarity and inclu- 585
537 fuzzy sets and systems[M], Advances in Type-2 Fuzzy Sets sion measures between IT2FSs[J], Kongzhi Yu Juece/control 586
538 and Systems. Springer, New York, NY, (2013), 65–80. and Decision 26(06) (2011), 861–866. 587

tho
539 [31] L.A. Lucas, T.M. Centeno and M.R. Delgado, General [44] T. Zhao, M. Huang and S. Dian, Stability and stabilization of 588
540 type-2 fuzzy inference systems: analysis, design and com- TS fuzzy systems with two additive time-varying delays[J], 589
541 putational aspects[C], Fuzzy Systems Conference, 2007. Information Sciences 494 (2019), 174–192. 590
542 FUZZ-IEEE 2007. IEEE International. IEEE (2007), 1–6. [45] T. Zhao and S. Dian, State feedback control for interval 591
543 [32] D. Wu and J.M. Mendel, A comparative study of ranking type-2 fuzzy systems with time-varying delay and unreli- 592
544 methods, similarity measures and uncertainty measures for able communication links[J], IEEE Transactions on Fuzzy 593
Au
545 interval type-2 fuzzy sets[J], Information Sciences 179(8) Systems 26(2) (2017), 951–966. 594
546 (2009), 1169–1192. [46] T. Zhao and S. Dian, Delay-dependent stabilization of 595
547 [33] T. Zhao, J. Xiao, Y. Li, et al. A new approach to similarity and discrete-time interval type-2 TCS fuzzy systems with time- 596
548 inclusion measures between general type-2 fuzzy sets[J], varying delay[J], Journal of the Franklin Institute 354(3) 597
549 Soft Computing 18(4) (2014), 809–823. (2017), 1542–1567. 598
550 [34] D. Wu and J.M. Mendel, Similarity Measures for Closed [47] T. Zhao, Q. Yu, S. Dian, et al. Non-singleton General Type-2 599
551 General Type-2 Fuzzy Sets: Overview, Comparisons, and a Fuzzy Control for a Two-Wheeled Self-Balancing Robot[J], 600
d

552 Geometric Approach[J], IEEE Transactions on Fuzzy Sys- International Journal of Fuzzy Systems 1–14. 601
553 tems (2018). [48] T. Kumbasar and H. Hagras, Big BangCBig Crunch 602
cte

554 [35] J. Huang, S. Ri, L. Liu, et al. Nonlinear disturbance observer- optimization based interval type-2 fuzzy PID cascade con- 603
555 based dynamic surface control of mobile wheeled inverted troller design strategy[J], Information Sciences 282 (2014), 604
556 pendulum[J], IEEE Transactions on Control Systems Tech- 277–295. 605
557 nology 23(6) (2015). [49] H.C.W. Lau, T.T. Wong and K.F. Pun, Neural-fuzzy mod- 606
558 [36] J. Huang, M.H. Ri, D. Wu, et al. Interval Type-2 Fuzzy Logic eling of plastic injection molding machine for intelligent 607
559 Modeling and Control of a Mobile Two-Wheeled Inverted control[J], Expert Systems with Applications 17(1) (1999), 608
rre

560 Pendulum[J], IEEE Transactions on Fuzzy Systems 26(4) 33–43. 609


561 (2018), 2030–2038. [50] C.H. Lee, F.Y. Chang and C.M. Lin, An efficient inter- 610
562 [37] X.H. Chang, Robust nonfragile filtering of fuzzy systems val type-2 fuzzy CMAC for chaos time-series prediction 611
563 with linear fractional parametric uncertainties[J], IEEE and synchronization[J], IEEE Transactions on Cybernetics 612
564 Transactions on Fuzzy Systems 20(6) (2012), 1001–1011. 44(3) (2014), 329–341. 613
co
Un

You might also like