Ye Infrmaton Sciences xx (2016) ence ‘Contents lists availabe at ScienceDirect Information Sciences Ea Journal homepage: www.elseviercom/locatelins = A comparative study of type-1 fuzzy logic systems, interval type-2 fuzzy logic systems and generalized type-2 fuzzy logic systems in control problems Castillo Oscar’, Amador-Angulo Leticia, Juan R. Castro, Mario Garcia-Valdez iuoa vate of etnelngy 22079 Tana Mee ARTICLE INFO ABSTRACT ‘race none “This paper presents a comparative study of eype-2 fuzzy Tog systems with respect to Received 25 March 208, terval type-2 and type-t fy logic systems to show the efceney and performance of _Accepend 14 ar 2035, ‘esable one xe -eneralized type-? fuzzy lpi controller (GT2RLO) We used diferent types of fuzy logic {ystems for designing the fuzzy contoles of complex non-linear plants. The theory ofa pha planes is used for approximating generalized type-2 fury loge in fuzzy controllers. Fora Inthe detizttistion process, the Karmieand Mendel Algorithm i used, simulation re- Ae ne representation sults with 2 type-t fuzzy loge contcller(THFLO) an interval type-2 fuzzy logic controller Faay controler (QF2ALC) and with a generalized type-2 fuzzy logic controler (GT2FLC) fer beachmark Generale type-2 ty lee Plants are presented. The advantage of using generalized type-2 fuzzy logic in furzy con- Footprint uncertain let is venied with four Benchmark problems. We considered diferent levels of nose, number of alpha planes and four types of membership functions in the simulations for comparison and to analyze the approach of generalized type-2 fuzy logic systems whe, applied in fuzy conto (© 2016 Published by Elsevier tne Introduction. Fuzzy control systems combine information of human experts (natural language) with measurements and mathematical models. Fuzzy systems transform the knowledge base into a mathematical formulation that has proven to be very eficient in many applications [145,6.1445.48} ‘Type-2 fuzzy models have emerged as an interesting generalization of type-1 fuzzy models based upon fuzzy sets. In fact, these models are also called generalized type-2 fuzzy models. There have been a number of claims put forward as to the relevance of type-2 fuzzy sets being regarded as generic building constructs of fuzzy models Fuzzy controllers have the advantage that they can be adaptive when disturbances in the plant are present. The main advantage of using generalized type-2 fuzzy control (GT2FLC) is that the plants show a higher degree of stability in the simulations. When we consider noise presence forthe GT2FLC, the results show that the Generalized Type-2 Fuzzy Logic System has better stability characteristics ‘This paper considers several experiments inthe simulation of four benchmark control problems with a type-1 fuzzy logic controller (TIFLC) and with an interval rype-2 fuzzy logic controller (TT2FLC), where the authors conclude that the IT2FLC is Tareponding ator Tel: =595645935016 ‘mal edie: acsulagtectiuana, easlootatancorg (Ose, leicaamaderougulebjabencome (AA, Lela}, jeasuorounbe edu (OR Caste), mares galicom(M. Gareaaler pda 0:015).ns201605.025 020-0255}6 2016 Pulsed by Eels Ie2 6 6 ” % as 26 ” 2 es 38 0 a a a 6 ey st ss o o ARTICLE IN PRESS 2 oscar e/a Sines me (2016) met better than TIFLC based on the simulation error minimization. We also realized the comparison with the three methods to observe the behavior and the improvement that a GT2FLC can offer with respect to TIFLC and IT2FLC, Few works can be found in the literature related to the implementation of generalized type-2 fuzzy logic systems in the area of fuzzy contro, It is because of this fact that we use the concepts of the a planes representation and general- ied type-2 fuzzy sets to improve the design of fuzzy logic systems and allow the analysis of benchmark problems more efficiently with high levels noise in the model In the last five years, there has been a significant increase of the research on higher order forms of fuzzy logic systems, in particular the use of interval type-2 fuzzy logic [1.2.1,12.33,44.51.52) In addition with the advancement of IT2FLS, uncer. tainty could finally be directly incorporated into the Fuzzy Sets. though the boom of research with IT2FLS is recent, there {s still much to be explored, and some current examples of research are shown by, a fuzzy model of computing with words (11, fuzzy operations (12) 2 simplified IT2FLS [22], type-reduction algorithms [30 the centroid of triangular and Gaussian T2FSs [38], and enhanced type-reduction [40], More recently generalized type-2 fuzzy logic systems have been propose: of course, the idea of going into higher orders or types of fuzzy logic isto construct better models of uncertainty. In this sense, itis theoretically expected that generalized type-2 fuzzy logic will allow better management of uncertainty [49,50] However, generalized type-2 requires a higher computational overhead and several efforts have been put forward in order to limit the complexity of generalized type-2 fzzy logic systems: for example. Wagner and Hagras [3)41~43] have intto- duced the 2Slices-based representation, and Mendel and Liv (222), have put forward a representation based on a-planes, Which both enable the representation of, and computation with, generalized type-2 fuzzy sets [22.23]. And as such, there are also very few application research works where GT2PS are used. for example. in a GT2FS based on face-space approach to emotion recognition [8]. a general type-2 fuzzy inference systems; analysis, design and computational aspects (16), fuzzy ‘means for uncertain fuzzy clustering /17), monotone centroid flow algorithm for type-reduction [18], multi-criteria group decision making [18], edge detection for image processing [20], matching GI2FS by comparing the vertical slices [34 gen- eralized type-2 fuzzy systems with interval type-2 and type-1 fuzzy systems applied to fuzzy control [35], the information sranule numerical evidence [50], computing with words for discrete GI2KS [51]. In addition in the field of fuzzy control, a comparison of fuzzy controllers for the Water Tenk with Type-1 and Type-2 Fuzzy Logic is presented in [1], and the im- provement of GI2FLS over IT2FLS and TIFLS for controlling 2 mobile robot is presented in (35) In this paper, we use the a-plane representation, which enables the representation of and computation with generalized type-2 fuzzy sets. ‘The main contnbution of this paper is a comparative study based on generalized type-2 fuzzy logic for the design and implementation of fuzzy controllers, which allows for beiter modeling of the uncertainty that exists in achieving, control ‘of non-linear plants. Also, the comparative study of TIFLC. IT2FLC and GT2FLC as tools for modeling complex problems in control, Based on the literature and experience in the area of fuzzy control, we expect that GT2FLC can demonstrate that the stabilization and the performance ofthe simulations are improved, especially when noise levels are considered in the model, and better results are obtained when compared with respect to type-1 FLC and interval type-2 FLC “The rest of the paper is organized as follows. Section 2 describes some basic concepts of fuzzy logic systems: type-t FLS, interval ype-2 FLS, generalized type-2 fuzzy logic systems and the theory of alpha planes. Section 3 shows the benchmark problems that we consider in the simulations. Section 4 shows the simlation results with the implemented generalized type-2 fuzzy logic contioller and a comparison with the type-I and interval type-2 fuzzy logic contiolless, so that the advantage of using 2 generalized type-2 fuzzy logic controller is fully appreciated. Finally, Section 5 offers some conclusions ofthis work 2, Basic concepts of fuzzy logic systems 24, Definition of type-t fuzzy logic systems ‘A type-1 fuzzy set inthe universe X is characterized by a membership function u(x) taking values on the interval [0.1] and can be represented as a set of ordered paits of an element and the membership degree of an element to the set and are defined by the following Eg, (1) [45-43] A= (06, HaQ))DC EX) a Where jig :X — [0.1] In this definition 124(x) represents the membership degree of the element x © X to the set A, In this work we are going the use the following notation: A(x) = jeq(2) forall x« X 22. Definition of Interval type-2 fuzzy logic systems Based on Zadeh's ideas, in 1979 Mizumoto and Tanaka [3] presented the mathematical definition of a type-2 fuzzy set Since then, several authors have studied these sets, in [25-27] Mendel, ohn and Mouzouris defined these sets as follows 6hNe Over ea formation Scenes xe (2016) exe 2 oo Tig 1 Generalized ype? membership function Fig 2, FOU ofthe generalized rype-2 membership fneion ‘An Interval Type-2 Fuzzy Set A. denoted by s(x) and, (s) is represented by the lower and upper membership func- sions of (0). Where x eX. In this case, Ea (2) Shows a sample IT2FS [6102425,262731] ((Gcu). Ve eX, ¥u ef S10. 1)) 2 ‘Where X is the primary domain, cis the secondary domain, All secondary grades (1% u)) are equal to 1 23, Definition of Generalized type-2 fuzzy logic systems ‘With GT2ALS the logic is generally the same as for TIFLS and TT2F1S, but their operations are somewhat different. due to the nature of GI2FS [35 ‘There are several mathematical definitions of a generalized type-2 fuzzy logic system, and we used the representation based on [20.35.36] to define Generalized Type-2 Fuzzy Sets and are defined by the following £0. (2): (Cu), gle w) EX Ma ee 10.11) @ ‘Where jy & [0,1], is the partition of the primary membership function, and u isthe partition ofthe secondary membership function In Fig. 1 we can find a representation of a generalized type-2 membership function, and in Fig. 2. the footprint of uncertainty (POU) is illustrated. which is associated with the third dimension and allows a better modeling of real world uncertainty. It must be noted that there is a small difference in notation when compared with Type-1 and Interval Type-2, this is THFS and FT2PS use the notation (2). but GI2FS uses fy(u), in the vertical axis, and ths is due to the complexity involved in GT2FLS in comparison with the others, as well as how GT2FLS has been described in the literature. 24, a-planes representation ‘The a-plane for a generalized T2 FLS, in this case A, is denoted by Ay, and it is the union of all primary membership functions of A, which secondary membership degrees are higher or equal tow (0 = or= 1) [21.23), The equation of an alphaARTICLE IN PRESS 4 6 Over ee nomation Senex (2015) et u Fig 3. Aa example ofthe as ype fay et forthe alpa-pane ig. 4. Grape representation of 82 plane is represented by E (4), [n Fig. 3 the representation of an alpha plane i illustrated [23.35], Aa = {(a.u), ug Geu) = aie X. Vir cf © 10.11} 4 22 3, Benchmark problems ‘tor the evaluation of the generalized type-2 fuzzy logic control approach we used four benchmark problems, whieh in 85 fuzzy control are mathematical models that allow us to analyze the behavior of the generated fuzzy controllers. We describe 45 each of the benchmark problems and the corresponding implemented fuzzy controllers, in the following subsections 57 34, Woter Tank Controller 88 The first problem to be considered is known as the Water Tank Controller, which aims at controlling the water level in 58. a tank, therefore, based on the actual water level in the tank the controller has to be able to provide the proper activation 50 of the valve, fi 4 shows graphically the way in which the valve opening operates and hence the filling process in the tank, ‘1 ad this bas two variables, which are the water level and the speed of opening of the output valve for the tank Sling 52 To caleulate the valve opening in a precise way we rely on fuzzy logic, which is implemented asa fuzzy controller that ‘performs the control on the valve that determines how fast the water can enter che tank to maintain the level of water it 4 3 better way. The model equation of the Water Tank is represented asa ciflerential equation and the height of water inthe 5. tankeis gwen by £9 (3) aut ae a © 6 Where Vol isthe volume of water in the tank, As the cross-sectional atea of the tank, b isa constant related to the ow S57 rate into the tank, nda isa constant related to the lw rate out of the tank. The equation describes the height of water 8 Has 4 function of tine, due to the difference between flow fates into and out of the tank. The implementation of the 88. typecl and (ype? Furzy Logic Controllers are presented to experimentally observe the behavior of each controller. We later 100 present the characteristics of the generalized type-? Izzy logic controller, besides the results of the madel evaluation. The 101 membership functions are for the two inputs of the fuzzy system: the frst is called level, which has three membership 42 functions with linguistic values of high, okay and low. The second input variable is called rate with three membership 4 fumetions corresponding to the linguistic values of negative, none and positive, The names ofthe linguist labels ae assigned a volto 105 2 0 na Ne Over ea formation Scenes xe (2016) exe 5 ‘able hepa ‘apat ise Opera co se 7 Ora Nechange 3 High Goseast 4 olay peste ——Casesow 5 la Nerstve —__Opemlon Input ts Variable of control : £4} Furzytogie —] evatustionot > System Plant Output fu) Fig. 5. Represeatation of gene ary contol based on the empirical process of the filling behavior of a water tank [1]. The generalized type-2 fuzzy logic controller has an output called valve, which is composed of five triangular membership functions with the following linguistic values: closest, closeslow, nochange, opensfow and openfast. The knowledge about the problem provide us with five rules, which are detailed in Table 1 ‘The combination of rules for this benchmark problem can be found and are described in [1], and these $ rules are used to visualize the behavior of the generalized type-2 fuzzy logic controller. Fi cevror signal, which is applied to the fuzzy controller together with the change in the error signal over time [1.4,6:835,37) “The representation of a genevt fuzzy controller is shown in Fig. 5, 232, Temperature Controller lly, at the output of the adder we have the ‘The second problem to be considered is the Temperature Controller, which alms to establish the temperature of a water regulator. We now present the characteristics of the generalized type-2 fuzzy logic controler, besides the results of the model evaluation, The Membership functions are for the two inputs to the fuzzy system: the frst is called temp, which has three membership functions with iguistc values of cold, good and hot. The second input vatiable is called flow sith three membership functions with linguistic values of light, good and hard. The names of the linguistic labels re assigned based on the empirical process (o simulate the Temperature Controller. The generalized type-2 fuzzy logic controller has two outputs called eold, and hot. which are composed of five triangular membership functions with the following linguistic values, respectively: clsefast,closeslow, steady, opensiow and openfast. The knowledge about the problem provide us with nine rules, which are detailed in Table 2 Table Fayre ofthe funy contol or he Temperature Contain 7 ile apa Trap Opearr Faw ald Ane Gols hin ood oe Het ow tae ood tie ood ui ood ware a wwe ‘Opensow Corson person Seedy pent Opersow Corson ca pent Opens Opens Steséy Operon: hover GoveARTICLE IN PRESS ‘ 6 Over ee nomation Senex (2015) et Fig. 6 Mobuetobot mode 22 33, Mobile robot controller 23 The model which is used is of a unicycle mobile robot [25], consisting of two actuated wheels located on the same axis 124 and a front free wheel, and Fig 6 shows a graphical description of the robot model, 125 The robot model assumes that the motion of the free wheel can be ignored in its dynamics, as shown by Es. (6) and ns (7) M(qhi+C(, aqv= Dv a7 Where, +P 6 ne (x9. 0) is the vector ofthe configuration coordinates, ne (0, w)T is the vector of velocities, 230 r= (4), %) isthe vector of torques applied to the wheels of the robot where r) and x» denote the torques ofthe right br and leit wheel, respectively 332 Pe Ris the uniformly bounded disturbance vector, 182 M(g) e RP js the positive-definite inertia matrix 4 C(g. 4) Is the vector of centripetal and Coriolis forces, and 135 De RE? is a diagonal positve-cefinite damping matrix, 35 The kinematic system is represented by Ea. (7) cos60 q-|sinoo ‘] o oa} LY, ese fa br whee, 188 (uy) isthe positon n the X ~Y (wold) reference frame, 39 @ is the angle between the heading direction and the x-3xis, go vand ware the linear and angular velocities. 141 Furthermore. Eq, (8) shows the non-holonomic constraint, which this system has, which corresponds to a no-sip wheel ‘42 condition preventing the robot from moving, sideways. yeas — sind @ 162 The system fails to meet Brockett's necessary condition for feedback stabilization, which implies that no continuous 144 static state-feedack controller exists that can stabilize the closed-loop system around the equilibrium point. The fuzzy logic 145 system has two inputs: the frst is called ev (error in the linear velocity), which has three membership functions with 146 linguistic values of NZ and P.The second input variable is called ew (error in the angular velocity) with three membership ‘7 functions with the same linguistic values. The generalized type-2 fuzzy logic controller has two outputs called TH (Torque 1), and T2 (Torque 2). which are granulated into three triangular membership functions with the following linguistic values, respectively: N, Z P. The combination of the rules is presented in Table 3 and their description can be found in [35] 150 34, Beam and ball controller 151 __ This problem has the goal of stabilizing a ball that is placed on a beam, where itis allowed to roll with one degree of 182 freedom along the length of the beam, A lever arm is attached to the beam at one end and a servo gear at the other. AsNe Over ea formation Scenes xe (2016) exe 7 ‘ables uy tls ofthe fuzzy cot fo the Mode Rabo Coolie. hile np ‘pu ane aod ot ot od ot wewnnnzzz|e snzenzene| 9 aeanwnwanel oy Fig 7 kepreseataon of the beam and bl probe 155. the servo gear turns by an angle 0, the lever changes the angle of the beam by c. When the angie is changed from the 154 horizontal position, gravity causes the ball to roll along the beam. A controller is designed for this system so that the bal's 155 position can be manipulated. The representation of this is shown in Fig. 7. 155. In the system parameters in this problem, we assume that the ball rlls without slipping and friction between the beam 15) and ball is negligible. The constants and variables inthis case are defined as follows: 158 (mm) mass ofthe ball 0.11 kg 159° (R) radius ofthe ball 0.015 m 60 (@) lever arm offset 0.03 m stg) gravitational acceleration 9.8 m/s°2 ez (L) length of the beam 10m ie (J ball's moment of inertia 999¢-6 kg.’ 1s4 (1) ball position coordinate 465 (alpha) beam angle coordinate 165 (theta) servo gear angle 17 The fuzzy logte system that stabilizes the beam and ball has four inputs called in, in2,in3 and ind, espectively. The 166 membership functions used in this ease are of the generalized bell form. The number of rules is 16. The fuzzy logic system 169. of Takagi-Sugeno type [39] 170 The test criteria are a series of Performance Indices; where the Integral Square Error (ISE, Integral Absolute Error (AE), 171 Integral Time Squared Error (ITSE), Integral Time Absolute Error (TAB) and the Root Mean Square Error (RMSE) are used, m2 espectively shown in Eqs. (9}-(13), se= [~eqoae o m _ mae = [~ leeotae (10) mse = [~ eae co) sag = [eco tae 3)ARTICLE IN PRESS * 6 Over ee nomation Senex (2015) et Fig & Proposed achlete of the model forthe GTA. v . a3) 17-4, Simulations of the Generalized type-2 fuzzy logic controllers 178 In this section the proposed model for the implementation and the simulations of the Generalized type-2 fuzzy logic 179 controller (GIZFLC) are presented, In Fig. 8, the block diagram for the simulation of the generalized type-2 fuzzy logic 160 controller is illustrated asi 41 Input 182 The frst step in the whole process is the simulation of the model and in this way obtaining the intial errors of the 13 simulation. In the initial iterations of the model, the Generalized type-2 fuzzy logic contooller is evaluated: therefore, the 14 mean square error generated by the model becomes the input of the second iteration. 18542, Fuzzifcation 186 The fuzzifier maps crisp inputs into generalized type-2 fuzzy sets to process within the FIC. In this paper, we will focus 187 on the type-2 singleton fuzzfier as itis fast to compute and, nus, suitable for the generalized type-2 FLC real-time opera 183 tion, Singleton fuzzifcation maps the crisp input into a fuzzy set, which has a single point of nonzero membership. Hence, 19 the singleton fuzzifier maps the crisp input x, into a type-2 fuzzy singleton, whose MF is 14, (X9) = 1/1 for xp=2, and 190 4, (p) = 0 for all p #5 for all p= 1, 2,...,P, where Pis the number of FLS inputs [3.20), 19143. Input linguistic variables 192 For the particular case of the water tank, two inputs are defined, in which each one has three Gaussian membership 19) functions with uncertain mean. 94 The Gaussian membership functions for each input are obtained with (20) and (21), and the means of each function 495 are obtained with (15) and (15). For example, for the high membership functions the fist mean is obtained with (15), the 196 second mean is calculated by (16) and the o value is obtained with (14).ab INS insGseMarch 26, 291610 29 high okay low Membership clot dono sochage opel opt 3 : 5 z 2 Fs § Mandant JZ leet KerSingeain | 3° 2 negative none positive valve Degree of Membership Fig. 9. Generalized type-2 fz lore controler of he Benchmark problem ofthe water ank 197 The inference system has one output called valve, the linguistic values used for the output are: closefast, closeslow, 186 nochange, openslow and openfast, and we selected the range (~1, 1}, The Gaussian membership functions for the output 198 are obtained with (20}-(24), the means of each function are obrained with (15)-(19) and the value with (17). The FOU 200 for the valve output variable, is calculated in a similar way as forthe input variables. This isthe method that we propose 2o1__to analyze the four types of membership functions and manually we change the distribution of the parameters, Simulations with 4 types of membership functions in order to observe the behavior of the model are performed, We show the ditfer- ent types of the membership functions with the example of the linguistic variables called good of generalized type-2 form, a04 where the value forthe FOU is 0) for the inputs and outputs. respectively. 208 Experiments are performed with 4 types of membership functions in order to observe the behavior of the model, which b06 are: Gaussian, Trapezoidal, Triangular and Generalized Bell, 207 The representation of the generalized type-2 fuzzy logic controllers are shown in Fig, 9-12, respectively, for the four 208 benchmark problems studied in this work 208 The Gaussian membership functions are calculated by Eqs. (20)-(24), This membership function is considered in the 210 GTRFLC and is characterized as having uncertainty in the mean and is calculated by Eqs. (21)-(24). The most detailed 2M representation of the Gaussian membership function in the GT2FLC is presented in the following equations = highy/2 «ay mm hh, 28mg my (mK), where Kis in 0.1) gems 2 img =m + (mK), where Kis in 0.1) a9) {2(%.w) = gausmgausstype2 (x, 44 gm, m3) 20) 218 Where “gausmgausstype2” stands for a Gaussian generalized tpe-2 membership function with uncertain ean At) = (400. e001 igausmtyped(x. [om ma)) en 219 Where “igausmtype2” stands fr @ Gaussian interval ype-2 membership function with uncertain mean Z Please cite this article asi © Oscar et al, A comparative study of type-I fuzzy logic systems, interval type] fuzzy logic systems and generalized type-2 fuzzy logic systems in control problems, Information Sciences (2016), bntp:ax doi org/10.1016/ins.2016.03.025Nana ST ® Oscar eo /inemation Sines me (2016) n-ne cold flow Fg T. Generaze 9-2 fizzy lope conroller f the abi obo.am a ns ns a 2m 2» Dees Metin 209 — BO) Ne Over ea formation Scenes xe (2016) exe Fis Sugeno ‘Non-Singlenton int ig. 1, Ceneraized tye fz loi onl ofthe Beam ané ball, Px = gaussmnf(x, (oe, Mel) = on[-G)E =m y] f2¢s,u) = gaussmf toured er -(5)( = y] 44 Inference (22) 23) (24) Once the input and output variables are defined, with their respective membership functions, the inference process is performed in the system, and for this the following steps are needed: Define the fuzzy rules: The structure ofthe rules in the generalized Type-2 FLS is the standard Mamdani-type FLS rule structure used in the Type-1 FLS and an interval Type-2 FLS, but in the paper, we assume that the antecedents and the consequents sets are represented by generalized type-2 fuzzy sets. So for a type-2 FLS with p inputs Xj € Xy,....%9 © Xp and one output ye Y, which is a Multiple Input Single Output (MISO) system, and if we assume there are M rules, the kth rule in the generalized type-2 FLS can be written as follows [7.24.25] IP is BE and and xpis ff, THEN y is 6 (25) For modeling the process with the fuzzy system, we consider rules that help describe the existing relationship between the behaviors ofthe controllers inthe real world; these rules are designed to be experimentally observable for the problem solved by each controller. ‘The inference of a GT2FLS can be simplified into two main operations, meet and jin, as shown in Eqs. (25) and (27) Bale) U5 W) = (0 WELW UY Ww, WER E101], wef’ (0.1), ate 0) gle.) Cv, fo(ud2 few) Jv ew" we JE SLO, L]. wef? < [0,1]} 28) anARTICLE IN PRESS 2 Over ee nomation Senex (2015) et 28545; Type reduction 236 As for the type-reducer for the GT2FLS, one of the techniques uses the centroid, shown in Eq. (28) and this is the 237 definition of the centroid C; for type reduction of a GT2FS, Where is a combination associated with the secondary degree DBS fy (0198.3 fh G {1m teats eye 8) Sa). fal. Be hh en 288 To pevfrm the defazifation process, the Karnik and Mendel method is use 20 (1) Kank and Mendel alorithn 241 ‘Type reduction is performed by applying the type reduction of the Karnik and Mendel algorithm [12.13.24 26.2728,30], a {and this reduction is given by the following Eqs. (28) and (20). Dkr MO + Vila BAO 2 ena es) 2 yo) ~ LOE + Pia HOOT 30) EE BLO TEN OT 264 (1) Alpha plane integration: the results ofthe alpha planes are integrated by the following Eqs. (31) and (32) [2123.29.51] sae joey — Eee Hl) en Dax Pm sie) = Evie (22) Thaw 246 46. Defurzification 247 After realizing the type reduction and integrating the results of all the alpha planes. the defuzzfiation is performed by 248 using the average of y! and y*, fo obtain the defuzzified output of a generalized type-2 non-singleton FLS [1528,30,32), jC) + H40) 5,0 AO 0) 249 5. Simulation results 250. We used four benchmark problems in control, and we consider fuzzy logic systems with the predetermined structure 251 described in Section 3, We change the types of membership functions of the inputs and outputs. Each benchmark problem 252 as the configuration ofthe parameters in the specific simulation. In the first case of filling the water tank. we considered 253 changing the type of the membership functions. using a perturbation with a level of 15. the value of the ro is 0.2 and 5 254 alpha planes are used. 25551, Simulation results for the Water Tank Controller 256 The results of the simulations for the fist benchmark problem are presented in Table 4, 2s? In Table 4 it can be noted that when we increase the value of the perturbation, the error decreases in a minimum 256 amount, The GT2FLC finds a smaller error of simulation than both the TIFLC and IT2TLC, ie, we realized a comparison 259° with the results based on IAE; in this metric when the type of the membership functions in the inputs is Gaussian, the 250 error for the THFLC is of 118.1053, for IT2FLC is about 112.261, and for GI2FLC the best result is bout 112.305, and these 251 experiments ate shown with a perturbation of 15. It is very important to say that the errors in the IT2FLC and GI2GLC 252. axe similer in the minimization of the errors and in most of the experiments the IT2FLC is better than the GT2FLC. This 251 1s because the analyzed problem does not present sulicient need for robustness and the GTFLC does not improve the 254 performance. For the metric of the root mean square error, forthe first three membership functions, itis clear that the 255 GTZFLC obtains better results 266 The number of alpha planes was also under variation for analyzing its effect on the results. The number of alpha planes 257 indicate the amount of secondary membership function that are evaluated, We changed the number of alpha planes from 268 5 t0 15 to analyze and find the optimal number of planes necessary in the evaluation in the model with GT2FLC. Rho is &