This action might not be possible to undo. Are you sure you want to continue?

# A New Defuzzification Method for Fuzzy Control of Power Converters

Yigang Shi and P.C. Sen, Fellow, IEEE Department of Electrical and Computer Engineering Queen’s University Kingston, Ontario, Canada, K7L 3N6

Abstract

In this paper, a new deffuziflcation method is proposed wldch can provide Control for DC-DC Logic Control of Maxima (FLC), improved converters. performance in Fuzzy study A comparative

of different defuzzification Center of Sums (COS), (MOM), and First of Maxima Buck-Converters the characteristics is outlined. Second Maxima

methods

adopted in Fuzzy (HM), Middle

such as Center of Area (COA), Height Method Center of Largest Area (COLA), (FM), for application to DC-DC The dhtinction among

tunningcontrollers and model reference adaptive controllers, etc., were adopted to the control of power electronics [5]. But these controllers also need accurate mathematical models and are therefore sensitive to parameter variation. Sliding mode controllers (SLMC) [6, 7, 8, 11] were introduced since it does not need an accurate mathematical model. The dhiadvantage of this method is the drastic changes of the control variable whkh leads to chattering. The chattering problem can be scdved by introducing a

boundary layer around the sliding plane [9]. However, the local non-linearities in the state sp~e we not considered in the SLMC design. The application of fuzzy theory in power electronics is relatively new [3, 10, 12] ancl has received attention of a number of researchers in recent years. A power electronics system, in general, has complex nonlinear model with parameter variation problem, and the control needs to be very fast. The operation of fuzzy logic controller (FLC) does nc~trely on how accurate the model is, but on how effective the linguistic rules of the fuzzy controller are. Fuzzy control therefore simplifies the design of optimal compensation for DCDC converters. Unlike SLMC, it is possible to take account for local non-linearities in FLC. Many papers showed the potential and feasibility of FLC control for Power Electronic circuits [11, 12]. Fuzzy control can provide better performance than the conventional PI-controller for DC-to-DC Buck converter [11]. The Fhzzy Controller also show Sliding Mode characteristics resulting in robust control [11]. A typical Fuzzy Logic Controller (FLC) has the following components: fumification, knowledge base, decision making and defuzzification. The performance oft he FLC depends very much on the defuzzification process. This is because the overall performance of the system under control is determined by the controlling signal (the defuzzified output of the

FLC) the system receives. some useful results will be first reIn this paper,

is presented.

which lead to varying performance called Height Weighted is proposed and its perforof the closed-loop convert-

A new method (HWSM)

mance is assessed. The paper also presents simulation results of the performance ers from the steudpoint ulation end line regulation. ter dynamic performance The new HWSM ther improvement. of start-up transient, load regThe simulations show that and less steady state error. method provides fur-

COA, COS, and HM defuzzification methods have betdefuzzification

1

Introduction

of research in fuzzy since Zadeh [1] first gave

There has been a rapid growth cent rol and fuzzy modeling

Mammathematical foundation of fuzzy systems. dani and his colleagues first applied fuzzy logic in industrial control applications est for practical tomation, application [2]. Recently, the interof fuzzy logic is growing

rapidly. It has been successfully applied in factory au-

such as industrial robot and NC machines

[3, 13]. Normally, power electronics based on conventional control methods [4] failed to perform satisfactorily under parameter variation, non linearity, load disturbance, etc. Many efforts have been made recently to improve the performance of the controller

in power converters. State feedback controllers, self

viewed, in particular the effects of different defuzzifi-

0-7803-6404-X/00/$10.00 (C) 2000

to be the Center of Area of this particular fuzzy subset. defuzzification is a mapping from a space of fuzzy control action defined over an output universe of discourse into a space of nonfuzzy (crisp) control act ions. 16]. 1 consists of the following components: Fuzzification. and Defuzzification. .. A de fuzzification strategy is aimed at producing a crisp control action that best represents the possibility distribution of an inferred fuzzy control action [14]. the simplest conventional FLC is the basis for the aforementioned three types of FLCS and still play an important role in practice. (f) First of Maxima (FM). Rule base is normally expressed in a set of Fuzzy Linguistic rules. The centroid de- selects the output crispy value corresponding to the center of gravity of the output membership function which is given by the expression: method ~. and considers the contribution of the area of each fuzzy set individually. weighted membership) aa follows: (HM). The MOM strategy generates a control action whkh represents the mean value of all local control actions whose membership functions reach the maximum and may be expressed as: 1 (4) j=l where Ai and Bi (antecedent).= Z.=lP(w)dw . Rule Base. The conventional FLC shown in Fig. Mean of Maxima (MOM). It is difficult to represent this de fuzzification method formally. Self-organized FLC and FLC based on Fuzzy Model [13]. Three classes of FLC have been recognized by industry. the centroid of function for each rule is output is then calculated vidual centroids. The simplest and most efficient form of membership function is the triangular one shown in Fig. Center of Largest Area (COLA). 14.( X. However. Height — method. which is given by the expression: ~. The last components of FLC is defuzzification. This method avoids the computation of the union of the fuzzy sets. Ci (consequent) are fuzzy variables characterized by fuzzy membership function.e. Then the method determines the convex fuzz y subset with the largest area and defines the crisp output value U.15. Height Method (HM). The i+% linguistic control rule can be expressed as: ~: IF e~ is Ai AND dea is Bi THEN Ui is C~ (1) (b) Center of Sums (COS). which is expressed as: Ull = inf{w where e w Ip(w) = hgt(w) (5) w. The FM method uses the union of the fuzzy sets and takes the smallest value of the domain with maximum membership degree. 2(a). The set of fuzzy rules normally can be summarized aa a table as shown in Fig.= J wp(w)dw SAw)dw 2 Review of fuzzy logic control The FLC has evolved over ahnost twenty years.= J-w Z. In order to connect linguistic control strategies with speciilc control actions. while the COA strategy yields results 0-7803-6404-X/00/$10. The COLA method is used in the case when universe of dkcourse W is non-convex. 4. A similar to COA but faster de fuzzification method is the center of sums. with each rule triggered with varying belief or support. In the height ea& output membership first evaluated.=l %4%) Z. The &al as the average of the indiby their heights (degree of ~. Center of Sums (COS).+ Aw)dw Method (2) (c). hgt W) is ( the highest membership degree of When the MOM strategy is used. Then a new method called Hight Weighted Second Maximum (HWSM) is investigateed. (e) Center of Largest Area (COLA).00 (C) 2000 . which can provide improved performance such as reduction of the high starting current in Fuzzy Control for DC-DC converters. The various strategies that have been reported in literature are described as follows and the graphical representations of them are shown in Fig. The main part of FLC is Rule base and Inference Mechnism. 2(b). (a) Center fuzzification of Area (COA). and First of Maxima (FM). Several de fuzzification methods have been proposed [10.=l /-4%) (3) (d) Middle of Maxima (MOM).cation methods as applied to voltage control of DCDC Converters will be presented. they are Center of Area (COA). 3 Defuzzification methods Basically. i. linguistic variables and the corresponding fuzzy sets or fuzzy membership functions should fist be defined. They are Direct FLC. it consists of at least two convex fuzzy subsets. Inference Mechanism. the performance of an FLC is similar to that of a multilevel relay system [17].

the gain factor &. the linear differential equations are used by representing the buck converter as a simplified equivalent circuit shown in Fig. The final output is calculated as the average of the individual maxima. 13]. regulated voltage in Fig. regulated voltage in Fig. Ae(k)) (8) 0-7803-6404-X/00/$10. regulated voltage in Fig. There exists a high initial current and no steady state error with the regulated voltage response in both figures. C=200pF. 7(a) shows very small overshoot and undershoot at the transition almost without oscillation.5fl.which are similar to those obtainable with a conventional PI controller [18]. For FM detizzification method. regulated voltage in Fig. Although this method is rather time-consuming. For proposed HWSM de fuzzification method. V. regulated voltage in Fig. For load change. no oscillation in the steady state and the start-up current (6) where wj takes the largest value of the domain with maximal membership degree. regulated voltage for both supply change and resistance change in Fig. The PI-like fuzzy controller [15] used in this paper can be represented as [H=[l-3rJkl+[w’’(’) ‘7) Kd . The simulation results are for start up of the buck converter horn the zero initial state. 10 shows large overshoot and undershoot and is highly oscillatory. K. These three scaling factors which describe the particular input normalization and output renormalization play a role similar to that of the gain coefficients in a conventional controller. it haa a steady state error and a high starting current. 11. 1 l(a) shows no overshoot at the start-up. the rule table of the fuzzy controller associated with the different defuzzification methods are the same. and K(ce) are adjusted by trial and error approach to provide a good comprise between the transient and steady state performance. For HM defuzzification method. Simulation results for defuzzification methods are shown in Fig. For COLA defuzzification method. 5(b) shows small overshoot and it settles down quickly to a steady state with small oscillation. 5(a) shows small overshoot and it settles down quickly with very brief and small oscillation. That means FLC with HM method has larger effective gain than that of FLC with COA even if they have the same value of gain factor & (equation 8). regulated volt age in Fig. but there exists a high starting current. For MOM defuzzification method. WMle carrying on the simulations. 3. weighted by their heights (degree of membership) aa follows: where F is a nonlineax function representing the fuzzy controller. This is a new method propsed in this paper. It can be expected that the COA strategy can yield superior results.50 For COA defuzzification method. For the subsequent discussions. In this method. 3 shows the arrangement and parameters of a buck converter with the fuzzy logic controller. For load change. the second maximum of each output membership function for each rule is first evaluated. 8 for MOM defuzzification method. 6 shows almost the same result as that in Fig. especially the steady-state performance.. regulated voltage in Fig. For load change. &=2. for supply voltage variation. 8(a) shows large overshoot and underdamped oscillation. As(k) = F(K. regulated voltage for both supply change and resist ante change in Fig. It also can be noted that there exists no steady state error for both voltage change and resistance change. 8(b) shows appreciable overshoot and it settles down slowly to a steady state with a steady state error compared to both COA and HM methods. 5 to Fig. for supply voltage variation. And. the normalization factor Ke of the error e and the normalizat ion factor KCe of the change of error ce have to be adjusted to fit the operating condition of the converter [12]. e(k). Fig. it has been widely used for Fuzzy Logic Control in many industrial applications [3. for supply voltage variation. it can be noted that HM gives more consideration on the local control actions with larger membership functions than COA does. K. 4 Simulation results The fuzzy control algorithm and the afore mentioned defuzzification methods are now verified by simulations. the parameters Kd.=20V. otherwise v~~ =0. The parameters of buck converter are L=1OOPH.00 (C) 2000 . (g) Height Weighted Second Maxima (HWSM). there is a steady state error.. We can express them in the usual state variable matrix form: where vOi=V8 when the circuit is on. for supply voltage variation. 5 for COA defuzzification method. Basically. Moreover. . regulated voltage in Fig. and for load change. From the equations of HM and COA. As well. For COS defuzzification method. . Results are obtained for supply voltage change of 20V to 15V and for resistance change of Ml to 2. 9 shows almost the same result as that in Fig. 7(b) shows very small overshoot and it settles down quickly. for supply volt age variation .

Studies. COS. COLA.854-861. there is no steady state error for both supply voltage change and load disturbance. C.00 (C) 2000 . No. [11] V. A new HWSM defuzzification method is proposed. C. pages 283-297. Fuzzy Controller for DC-DC Converters”. Bose. Int. and Future”. pages 338-353. or HM method have high initial current. Vol. HM. Raviraj and P. Engelwood C’lijj%. But they have poor system responses. 1991. App. C. Syst. “Comparative Study of Proportional-Integral.. C. A. G. Sabanovic Ccmtrol of DC-DC and S. 14(9):1770- 6 Conclusion In this paper. K. Y. pp.Past. 1-13. “Electric Motor Drives and Control . Mar-Apr. C. It is evident that MOM and FM are the simplest ones for computation and implementation. The FLC with HWSM method has ahnost the same performance as COA but with better start up transient and less initial current. In the steady state FLCS with COA. C. Taiwan. Proceedings. “Sliding Mode Control of Induction Motor”. pp. 2. that the FLCS with MOM or less oscillation in the steady methods. 1985. “Sliding verters”. K. different defuzzification methods that can be adopted by FLC for application to DC-DC converters have been studied. For load change. Li. 1655. 479-486. IEEE Trans. it is the best choice for the fuzzy control of DC-DC Converters. pp. on Industrial Electronics. L?3. Man Cybern. Selected Areas in Communications. 1996. No 6. 0-7803-6404-X/00/$10.reduces significantly. As well. they have almost the same good performance as that of COA. [8] B. vol. Since a fast response and a smaIl steady state error are required. ‘(A Comparative Study of Proportional-Integral and Integral Proportional Controllers for DC motor Drives”. [13] M. pp. D. Present. “A Servo-Control a Self-Controlled Mode Synchronous L??EE Sliding Controller”. K. Dec 1990.472-481. “A FUZZY Set Theory Based Control of a Phase-Controlled Converter DC Machine Drive”. [4] P. So. pp. Meeting. 1985. COLA. pp 562-575. “Fuzzy Set”. regulated voltage in Fig. Sliding Mode and Fuzzy Logic Controllers for Power ConvertIEEE IAS Ann. Lee. Vol 37. Vol IA-23. Mamdani and S. Rec. [5] P. and very smooth response in the steady state.?LEThan. [10] 1991. Er. No. C. Sen. Information and Control . or FM methods respond in an nnderdamped manner. From computer simulation studies it is observed that the new proposed HWSM defuzzification technique provides the best performance. Slotline and W. IEEE-IAS Con. COS. H. Conf. 44. 1. 315-320. 1783. 1985. 1991 IEEE hi.Jow-. C. !Sen. vol. . J. Lee. Tse and Y. [6] R. No. or HWSM methods have almost the same computational intensity as that of MOM or FM. 283-295. 1990. 1987. [14] C. COA method has more computational intensity than that of MOM or FM while it can yield a satisfactory response during the transient aa well as in the steady state for both load and supply disturbance. COS. such as COA. Rec. It is observed that each de fuzzification techrique has certain advantages and certain disadvantages. 8. [9] J. or FM methods have a non zero steady state error. Meeting. “Applied Nonlinear Control”. K. ‘LAn Experi[2] E. Annu. pp. Bose. ment in Linguistic Synthesis with Fuzzy Logic Controller”. or COS. such as reduced initial current. pp. From the simulation results. “Robust Position Control of Induction Motor Using Fuzzy Logic Control”. June 1994. IEEE PESC’94. 1975. Assililan. Zadeh. in the steady state. Man Mach. 20. Internation Journal of Cont~ol. or HM methods respond in a highly damped manner whereas FLCS with MOM. References [1] L. Using [7] C. Nandam and P.. 1651ers”. S. Mode IEEE-IECON Rec. 1995. [3] C. 1992. pp. Sen. K. and P. Won. Bose. “A Review of Neural-Fuzzy Controller IEEE Journal on for Robotic Manipulators”. “Fuzzy Logic in Conrol Systems: Fuzzy Logic Controller– Part I and H“. pp. it can also be noted COLA methods have state whereas other have small oscillation Ann. Sen. Since HWSM shows good starting performance. “A [12] W. vol. COA. Sousa and B.. IEEE IAS 5 Comparative Evaluation For supply voltage change and load disturbance fuzzy controller with COA. Con- Cuk. E. COS. Sot. on Industry Applications. HM. Kim and B. S. C. Namuduri System Motor with Trans. K. or HWSM can be used in the defuzzification procedure of the fuzzy control of DC-DC Converters. or HM methods have almost zero steady state error whereas FLCS with MOM.NJ: Prentice-Hanll. However. 7. COS. 404-435. 251-258. 2. Venkataramanan and A. C. 11(b) shows no overshoot and the start-up current goes down greatly. But the fuzzy control systems with COA.

Springer Publication.no. Queen Mary College. “<i(t)L &o ‘. “Study of Defuzzification Methods of Fuzzy Logic Controller for IEEE fians.4\/lJ\ lh_f h/\/\\ /_/ \/\/\/\/\_\ ‘ ---n--r. F/WK2/75. tion. Hellendoorn. maximum hei@ \ 4’ 0 L ‘(c) COLA ro 0 NM NS z Ps PM PB PB PM PM PS Z NS PS Z NW NM u (f) FM ‘ first maximum w La . I 1 IEEl Oefuification -.------. 1975. An Introduction to Fuzzy Control. 1 FLC I I . \J I / —-— /\ II . Second Edipp.” Fuzzy Sets Syst. H.. “Further analysis and apInternal Rep. on Speed Control of a DC Motor: Industry Application. J. . H. vol. M. s &h D c — I v. pp. . 782-787.— -—---.-— ——. M. S. 1995. H. no. 1. Rao and S. Kichert and E. Marndani.—-. 1. J. Graphical Representations fuzzification Methods of Various De- 0-7803-6404-X/00/$10.(d) .——— . [16] D. Reinfrank. Kichert.[15] D. 1. ? :1 Vo Plant Figure 3: Fuzzy Control Circuit Figure 1: Conventional Fuzzy Logic Controller P w 1 0 __L---+fi--A-–– \ I I v Q takenonce I —-— JI II Y \ \ m (a) COA P w IL-– -T--A /\/. 1978. Driankov. “Analysis of a fuzzy logic controller. 1 / —-— midd e of maxima –– fuzzy set with universef discourse o (a) Membership function e f )(. 115-126.—— . 1996. London. Saraf.00 (C) 2000 . [18] W. rhe’gh’ 0 IJ. pp.c+ e (g) H WSti (b) Rule Table Figure 2: Components of Fuzzy Logic Controller Figure 4.” (d) MOM ‘ w ! Error E~/ NB ~ 4-. [17] W.--—--———— . fimzy wirh – set .. 1. vol. and M. plication of fuzzy logic control”. 29-44. 0 1 NB PB PB PB PB PM Ps Z NM PB PB NS PB PB Z PB PM Ps Z NS NM NB PS PM PS z NS NM NB NB PM PS Z NS NM NB NB NB PB Z NS NM NB NB NB NB rJ-l-A-FT–– /\/\/\/\/\/\ /\/ L.

01 -0 o.m6 o.01(4 (b] RO Ghange ~~ o o.cm 0.024 0 .m4 o.m6 (a) ~S Changa o.olb) Figure5: Fuzzy Control with COA Defuzzification Method VIA T————l m 15 10 5 b O.tm O.m4 o.OCQ Vs Vo -o o.006 O.m o.m6 (b) FIOCharrcje O.23 K 10 1------Vo O.mz 0.m4 o.ocz2 o.ox 0.00 (C) 2000 .ux3 0.m6 (a) Vs Change 0.cm 0.01 W Figure 7: Method Fuzzy Control with HM Defuzzification 0-7803-6404-X/00/$10.me O.034 0.OIW o ~~ o w (b] RO Charige X3 15 Figure 6: Fuzzy Control Method L with COS Defuzzification 10 Vo 5 o~ o O.

OCE 0.m4 o.ma 0.ms o.028 1 0.CXM o.ms Vo Vs T—---l m 15 Vo 10 5 D< — Vo O.0!(s) “o ol0 o.034 o.01(s) ‘:F:l 20 !5 10 “L :L-1 5 0.cm o.C04 o.m6 (b) RO Chan$p O.txl? 0.002 o.ol!@ Vs 0 ~~ “o o.00 (C) 2000 .VIA xl 15 10 E 0.m O.m9 O.OI(S) R O.m6 (b) ROChancy Figure 8: Fuzzy Control with MOM Defuzzification Method Figure9: Method Fuzzy Control with COLA Defuzzification 0-7803-6404-X/00/$10.

CCM o.m9 o 11(4 Figure 10: Fuzzy Control Method with FM Defuzzification Figure 11: Fuzzy Control wkh HWSM tion Method Defuzzifica- 0-7803-6404-X/00/$10.0((s) 0 0 0.01 w o WA o.m2 O.qOChange o.032 o.ms (b) .m 15 20 t5 10 10 5 P4 Vo w n -o A 0 ~-~ 0.01 (.m6 (a) Vs Change O.a16 (a) W Change O.m4 o.OM o.% 2C m 15 15 lC 5 5 0 o.ma 0.(X)4 o.ma T 25 0.m4 o.s) .0(Y2 O.00 (C) 2000 .m6 (b) R(3Change o.ow o.