ContentslistsavailableatSciVerseScienceDirect
Applied
Soft
Computing
jo u r n al h om epa g e :w w w . e l s e v i e r . c o m / lo c a t e / a s o c
Chaos
control
of
new
Mathieu–van
der
Pol
systems
by
fuzzy
logic
constant
controllers
Shih-Yu
Li
∗DepartmentofMechanicalEngineering,NationalChiaoTungUniversity,1001TaHsuehRoad,Hsinchu300,Taiwan,ROC
a
r
t
i
c
l
e
i
n
f
o
Articlehistory:
Received9September2009
Receivedinrevisedform12February2011 Accepted14August2011
Availableonline22August2011 PACS: 05.45.Xt 05.45.Pq 05.45.Gg 05.45.Vx Keywords: Chaoscontrol
Fuzzylogicconstantcontroller FLCC
NewMathieu–vanderPolsystems Qisystem
a
b
s
t
r
a
c
t
Inthispaper,anewfuzzylogiccontroller—fuzzylogicconstantcontroller(FLCC)isintroducedtochaotic
signalscontrolling.ThemainideasoftheFLCCaredescribedasfollows:(1)provingthetwochaotic
systemsaregoingtoachieveasymptoticallystableviaLyapunovdirectmethod;(2)viadetectingthesign
oftheerrors,theappropriatefuzzylogiccontrolschemeisoperated;(3)choosingtheupperboundand
thelowerboundoftheerrorderivativesofthechaoticsignalstobetheconsequentparts(corresponding
controllers).
Duetocontrollersintraditionalmethod–derivedbyLyapunovdirectmethod,arealwayscomplicated,
nonlinearformorthefunctionsoferrors,anewsimplestcontroller—FLCCispresentedinthispaperto
synchronizetwochaoticsignals.ThroughtheFLCC,therearethreemaincontributionscanbeobtained:
(1)themathematicalmodelsofthenonlinearchaoticsystemscanbeunknown,allwehavetodois
capturingthesignalsoftheunknownsystems;(2)throughthefuzzylogicrules,thestrengthofcontrollers
canbeadjustedviathecorrespondingmembershipfunctions(whicharedecidedbythevaluesoferror
derivatives);(3)bytheFLCC,thechaoticsystemcanbemuchmoreexactlyandefficientlycontrolledto
thetrajectoryofourgoalthantraditionalones.Threecases,originalpoint,regularfunctionandchaotic
Qisystem(withlargevaluesofinitialconditions),aregiventoillustratetheeffectivenessofournewFLC.
©2011ElsevierB.V.Allrightsreserved.
1. Introduction
SinceOtt et al.[1] gave thefamous OGY controlmethod in
1990,theapplicationsofthevariousmethodstocontrolachaotic
behaviorinnaturalsciencesandengineeringarewellknown.For
example,theadaptivecontrol[2–5],themethodofchaoscontrol
basedonsampleddata[6],themethodofpulsefeedbackof
system-aticvariable[7],theactivecontrol[8,9]andlinearerrorfeedback
control[10,11].However,when Lyapunovstabilityofzero
solu-tionofstates isstudied,thestability ofsolutions onthewhole
neighborhoodregionoftheoriginisdemanded.
Inrecentyears,somechaoscontrolbasedonfuzzysystemshas
beenproposedsincethefuzzysettheorywasinitiatedbyZadeh
[12], suchas fuzzy sliding mode controllingtechnique [13,14],
LMI-basedsynchronization[15]andextendedbackstepping
slid-ingmodecontrollingtechnique[16].Thefuzzylogiccontrol(FLC)
scheme has been widely developed for almost 40 years and
hasbeen successfully appliedto many applications[17]. Many
researchershaveworkedtoimprovetheperformanceoftheFLCs
andensuretheirstability.LiandGatlandinRefs.[18,19]proposed
∗ Tel.:+88635712121x55179;fax:+88635720634. E-mailaddress:agenghost@gmial.com
amoresystematicdesignmethodforPD-andPI-typeFLCs[20].
Choietal.[21]presentsasingleinputFLCensuringstability.Ying
[22]presentsapracticaldesignmethodfornonlinearfuzzy
con-trollers,andmanyotherresearchershaveresultsonthematterof
thestabilityofFLCs,inCastilloetal.[23]andCázarezetal.[24]was
presentedanextensionoftheMargaliotwork[25]tobuiltstable
type-2fuzzylogiccontrollersinLyapunovsense.
Recently,YauandShieh[26]proposedanamazingnewideato
designfuzzylogiccontrollers—constructingfuzzyrulessubjecttoa
commonLyapunovfunctionsuchthatthemaster-slavechaos
sys-temssatisfystabilityintheLyapunovsense.InRef.[26],thereare
twomaincontrollersintheirslavesystem.Oneisusedin
elim-inationofnonlinearterms andtheotheris builtbyfuzzyrules
subjecttoacommonLyapunovfunction.Therefore,theresulting
controllersarenonlinearform.Otherwise,inRef.[26],theregular
formisnecessary.Inordertocarryoutthenewmethod,theoriginal
systemmusttobetransformedintotheirregularform.
Inthis paper,wepropose anewstrategy whichis also
con-structingfuzzyrulessubjecttoaLyapunovdirectmethod.Error
dynamicsareusedtobeupperboundandlowerbound.Through
this new approach, a simplest controller, constant controller,
canbeobtainedand thedifficultyin realizationof complicated
controllersinchaossynchronizationbyLyapunovdirectmethod
canbealsocoped.Unlikeconventionalapproaches,theresulting
1568-4946/$–seefrontmatter©2011ElsevierB.V.Allrightsreserved. doi:10.1016/j.asoc.2011.08.024
S.-Y.Li/AppliedSoftComputing11(2011)4474–4487 4475
Fig.1.Theconfigurationoffuzzylogiccontroller.
controllawhasless maximum magnitudeofthe instantaneous
controlcommandanditcanreducetheactuatorsaturation
phe-nomenoninrealphysicsystem.
Therestofthepaperisorganizedasfollows:inSection2,chaos
controlbyFLCC schemeispresented. InSection 3,newchaotic
Mathieu–vanderPolsystemisintroduced.InSection4,simulation
resultsareshown.InSection5conclusionsaregiven.
2. ChaoscontrolbyusingtheschemeofFLCC
2.1. Chaoscontrolscheme
Considerthefollowingchaoticsystem
˙x=f(t,x)+u (2-1)
wherex= [x1,x2,... ,xn]T∈ Rnisathestatevector,f:R+×Rn→Rn
isavectorfunctionandu=[u1,u2,...,un]T∈Rnisthefuzzylogic
controllerneededtobedesigned.
Thegoalsystemwhichcanbeeitherchaoticorregular,is
˙y=g(t,y) (2-2)
wherey= [y1,y2,...,yn]T∈Rnisastatevector,g:R+×Rn→Rnis
avectorfunction.
Inordertomakethechaosstatexapproachingthegoalstatey,
definee=x−y=[e1,e3,e3,e4]asthestateerror.Thechaoscontrol
isaccomplishedinthesensethat[13–22]:
lim
t→∞e=t→∞lim(x−y)=0 (2-3)
where
e=x−y (2-4)
FromEq.(2-4)wehavethefollowingerrordynamics:
˙e= ˙x− ˙y=f(t,x)−g(t,y)−u (2-5)
AccordingtoLyapunovdirectmethod,wehavethefollowing
Lyapunovfunctiontoderivethefuzzylogiccontrollerfor
synchro-nization: V=f(e1,...em,...en)=1 2(e 2 1+···+e 2 m+···e2n)>0 (2-6)
ThederivativeoftheLyapunovfunctioninEq.(2-5)is:
˙V=e1˙e1+···+em˙em+···en˙en (2-7)
If the controllers included in ˙e1... ˙em... ˙en can be suitably
designedtoachievethetarget: ˙V <0,thenthetwochaoticsystems
areasymptoticallystable.ThedesignprocessofFLCCisintroduced
inthefollowingsection.
2.2. Fuzzylogicconstantcontrollerdesignprocess
Thebasicconfigurationofthefuzzylogicsystemisshownin
Fig.1.Itiscomposedoffivefunctionblocks[27]:
1.Arulebasecontainsanumberoffuzzyif–thenrules.
2.Adatabasedefinesthemembershipfunctionsofthefuzzysets
usedinfuzzyrules.
3.Adecision-makingunitperformstheinferenceoperationsonthe
rules.
4.Afuzzificationinterfacetransformsthecrispinputsintodegrees
ofmatchwithlinguisticvalue.
5.Adefuzzificationinterfacetransformsthefuzzyresultsofthe
inferenceintoacrispoutput.
Thefuzzyrulebaseconsistsofcollectionoffuzzyif–thenrules
expressedastheformifaisAthenbisB,whereaandbdenote
linguisticvariables,AandBrepresentlinguisticvalueswhichare
characterizedbymembershipfunctions.Allofthefuzzyrulescan
beusedtoconstructthefuzzyassociatedmemory.
Fig.3.Membershipfunction.
We use one signal, error derivatives ˙e(t)=
[ ˙e1, ˙e2,... ˙em,... ˙en]T, as the antecedent part of the proposed
FLCC to design the control input u that will be used in the
consequentpartoftheproposedFLCCasfollows:
u=[u1,u2...um,...un]T (2-8)
whereuisaconstantcolumnvectorandtheFLCCaccomplishesthe
objectivetostabilizetheerrordynamics(2-5).
ThestrategyoftheFLCCdesigningisproposedasfollowsand
theconfigurationofthestrategyisshowninFig.2.
Assumetheupperboundandlowerboundof ˙emareZmand−Zm,
thentheFLCCcanbedesignstepbystepasfollows:
(1)If em is detected as positive (em>0), we have to design a
controller for ˙em<0, then ˙V=em˙em<0 can be achieved.
Thereforewehavethefollowingithif–thenfuzzyrulesas:
Rule 1: if ˙em is Mmi then um1=Zm (2-9)
Rule 2: if ˙em is Mm2 then um2=Zm (2-10)
Rule 3: if ˙em is Mm3 then um3=em (2-11)
S.-Y.Li/AppliedSoftComputing11(2011)4474–4487 4477
Fig.5. TimehistoriesoferrorderivativesforCase1(withoutcontrollers).
(2) If em is detected as negative (em<0), we have to design a
controller for ˙em>0, then ˙V=em˙em<0 can be achieved.
Thereforewehavethefollowingithif–thenfuzzyrulesas:
Rule1: if ˙em is Mm1 then um1=−Zm (2-12)
Rule 2: if ˙em is Mm2 then um2=−Zm (2-13)
Rule 3: if ˙em is Mm3 then um3=em (2-14)
(3)Ifem approaches tozero,thenthesynchronization isnearly
achieved.Therefore wehave thefollowing ithif–thenfuzzy
rulesas:
Rule 1: if ˙em is Mm1 then um1=em≈0 (2-15)
Rule 2: if ˙em is Mm2 then um2=em≈0 (2-16)
Rule 3: if ˙em is Mm3 then um3=em≈0 (2-17)
whereMm1=
˙em/Zm,Mm2=˙em/Zm andMm3=sgn((Zm−˙em)/Zm)+sgn(( ˙em−Zm)/Zm),Mm1,Mm2andMm3 refertothe
membershipfunctionsofpositive (P),negative(N)and zero
(Z)separatelywhicharepresentedinFig.3.Foreachcase,umi,
i=1–3isthei-rdoutputof ˙emwhichisaconstantcontroller.The
centroiddefuzzifierevaluatestheoutputofallrulesasfollows:
um=
3 i=1Mmi×umi 3 i=1Mmi (2-18)Fig.7. TimehistoriesofstatesforCase1—theFLCCiscomingintoafter20s.
Table1
Rule-tableofFLCC.
Rule Antecedent Consequentpart
˙em umi
1 Negative(N) um1
2 Positive(P) um2
3 Zero(Z) um3
ThefuzzyrulebaseislistedinTable1,inwhichtheinput
vari-ablesintheantecedentpartoftherulesare ˙emandtheoutput
variableintheconsequentpartisumi.
Afterdesigningappropriatefuzzylogicconstantcontrollersand
beingsubstitutedintoEq.(2-7),anegativedefiniteofderivatives
ofLyapunovfunction ˙V canbeobtainedandtheasymptotically
stabilityofLyapunovtheoremcanbeachieved.
Consequently,theprocessesofFLCCdesigningtocontrola
sys-temfollowingthetrajectoryofagoalsystemcanbeconcludedas
follows:
1.ConstructaFLCCsysteminMATLAB(Simulink)followingFig.2
andEqs.(2-9)–(2-17).
2.Gettheupperboundandlowerboundoftheerrorderivativesof
thegoalandcontrolsystemswithoutanycontroller,i.e.−Zm≤
˙em≤Zm,whichareusedtobetheconstantcontrollers.
3.Design the membership functions of positive (P), negative
(N) and zero (Z) M1=
˙em/Zm, M2=˙em/Zm and M3=sgn((Zm− ˙em)/Zm)+sgn(( ˙em−Zm)/Zm).
S.-Y.Li/AppliedSoftComputing11(2011)4474–4487 4479
Fig.9. TimehistoriesofstatesforCase1—bynonlinearcontrollers(20s).
4.AnegativedefiniteofderivativesofLyapunovfunction ˙Vcanbe
obtainedandtheasymptoticallystabilityofLyapunovtheorem
canbeachieved.
3. NewchaoticMathieu–vanderPolsystem
ThissectionintroducesnewMathieu–vanderPolsystem.
Math-ieuequationandvanderPolequationaretwotypicalnonlinear
non-autonomoussystems:
˙x1=x2
˙x2=−(a+bsinωt)x1−(a+bsinωt)x31−cx2+dsinωt
(3-1)
˙x3=x4
˙x4=−ex3+f(1−x23)x4+g sin ωt
(3-2)
ExchangingsinωtinEq.(3-1)withx3andsinωtinEq.(3-2)with
x1,weobtaintheautonomousnewMathieu–vanderPolsystem:
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
˙x1=x2 ˙x2=−(a+bx3)x1−(a+bx3)x13−cx2+dx3 ˙x3=x4 ˙x4=−ex3+f(1−x32)x4+gx1 (3-3)wherea,b,c,d,e,f,gareuncertainparameters.Thissystemexhibits
chaoswhentheparametersofsystemarea=10,b=3,c=0.4,d=70,
Fig.11.TimehistoriesoferrorsforCase2—theFLCCiscomingintoafter20s.
Table2
Comparisonbetweenformsofcontrollers.
FLCC(components) Nonlinearcontrollers u1 ±Z1=±50ore1≈0 −x2−e1
u2 ±Z2=±500ore2≈0 (a+bx3)x1+(a+bx3)x31+cx2−dx3−e2
u3 ±Z3=±10ore3≈0 −x4−e3
u4 ±Z4=±10ore4≈0 ex3−f(1−x23)x4−gx1−e4
e=1,f=5,g=0.1andtheinitialstatesofsystemare(x10,x20,x30,
x40)=(1,5,1,5).Theprojectionsofphaseportraitsareshownin
Fig.4.
4. Simulationresults
Inordertolead(x1,x2,x3,x4)inEq.(3-3)tothegoal,weadd
controltermsu1,u2,u3andu4toeachequationofEq.(3-3),
respec-tively.
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
˙x1=x2+u1 ˙x2=−(a+bx3)x1−(a+bx3)x31−cx2+dx3+u2 ˙x3=x4+u3 ˙x4=−ex3+f(1−x23)x4+gx1+u4 (4-1)S.-Y.Li/AppliedSoftComputing11(2011)4474–4487 4481
Fig.13.TimehistoriesoferrorsforCase2—bynonlinearcontrollers(20s).
Table3
Comparisonbetweenerrordataat99.96s,99.97s,99.98s,99.99sand100.00safter theactionofcontrollers.
Time FLCC Nonlinearcontrollers
e1 e1 99.96s −3.5694e−035 6.8001e−016 99.97s −3.5339e−035 6.8001e−016 99.98s −3.4987e−035 6.8001e−016 99.99s −3.4639e−035 6.8001e−016 100.00s −3.4294e−035 6.8001e−016
Time FLCC Nonlinearcontrollers
e2 e2 99.96s 5.5438e−048 3.0531e−015 99.97s 5.4885e−048 3.0531e−015 99.98s 5.4340e−048 3.0531e−015 99.99s 5.3799e−048 3.0531e−015 100.00s 5.3264e−048 3.0531e−015
Time FLCC Nonlinearcontrollers
e3 e3 99.96s 2.4019e−048 6.9389e−016 99.97s 2.3780e−048 6.9389e−016 99.98s 2.3543e−048 6.9389e−016 99.99s 2.3309e−048 6.9389e−016 100.00s 2.3077e−048 6.9389e−016
Time FLCC Nonlinearcontrollers
e4 e4 99.96s 2.4710e−036 2.7756e−015 99.97s 2.4464e−036 2.7756e−015 99.98s 2.4221e−036 2.7756e−015 99.99s 2.3980e−036 2.7756e−015 100.00s 2.3741e−036 2.7756e−015
Therearethreecasestoshowtheeffectivenessandfeasibility
ofthenewapproach—FLCC.CaseI:Controlthechaoticmotionto
zero,CaseII:Controlthechaoticmotiontoaregularfunctionand
CaseIII:ControlthechaoticmotionofthenewMathieu–vanderPol
systemtoanotherchaoticmotionoftheQisystem.Furthermore,
chaoscontrolviatraditionalnonlinearcontrollersinCasesI–IIIis
alsogivenintablesandfiguresforcomparison.
4.1. CaseI:controlthechaoticmotiontozeroby(1)FLCCand(2)
traditionalcontrollers
In this case, we will control thechaotic motionof the new
Mathieu–vanderPolsystem(4-1)tozero.Thegoalisy=0.Thestate
errorisei=xi−yi=xi(i=1,2,3,4)anderrordynamicsbecomes
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
˙e1= ˙x1=x2+u1 ˙e2= ˙x2=−(a+bx3)x1−(a+bx3)x31−cx2+dx3+u2 ˙e3= ˙x3=x4+u3 ˙e4= ˙x4=−ex3+f(1−x32)x4+gx1+u4 (4-2)ChoosingLyapunovfunctionas:
V= 12(e21+e22+e23+e24) (4-3)
Itstimederivativeis:
˙V=e1˙e1+e2˙e2+e3˙e3+e4˙e4=e1(x2+u1)
+e2(−(a+bx3)x1−(a+bx3)x31−cx2+dx3+u2)
+e3(x4+u3)+e4(−ex3+f(1−x32)x4+gx1+u4) (4-4)
(1)ByFLCC
InordertodesignFLCC,wedivideEq.(4-3)intofourparts:
V= 12(e21+e22+e23+e24)=V1+V2+V3+V4 (4-5)
thentheerrorderivativeis
Fig.14. TimehistoriesofstatesforCase2—bynonlinearcontrollers(20s).
where
Part1: ˙V1=e1˙e1=e1(x2+u1)
Part2: ˙V2=e2˙e2=e2(−(a+bx3)x1−(a+bx3)x31−cx2+dx3+u2)
Part3: ˙V3=e3˙e3=e3(x4+u3)
Part4: ˙V4=e4˙e4=e4(−ex3+f(1−x23)x4+gx1+u4)
AccordingtotheprocessoftheFLCCdesigning,theerror
deriva-tivesofCaseIwithoutanycontrollerareshowninFig.5,andthe
valuesoftheupperboundandlowerboundcanbeobtainedas
follows:
Z1=50, Z2=500, Z3=10, Z4=10 (4-7)
AftergettingZ1∼Z4,alltheconstantcontrollers,membership
func-tionsandthecorrespondingFLCCcanbedecided.Therefore, ˙V1=
e1˙e1<0, ˙V2=e2˙e2<0, ˙V3=e3˙e3<0 and ˙V4=e4˙e4<0 can
beachieved,thenwehave ˙V= ˙V1+ ˙V2+ ˙V3+ ˙V4<0.Itisclearthat
thederivativeofLyapunovfunctionisnegativedefiniteandthe
S.-Y.Li/AppliedSoftComputing11(2011)4474–4487 4483
Fig.16.TimehistoriesoferrorderivativesforCase3(withoutcontrollers).
errordynamicssystemisgoingtoachieveasymptoticallystable.
ThesimulationresultsareshowninFigs.6and7.
(2) Bytraditionalnonlinearcontrollers
InordertoleadtheerrordynamicsinEq.(4-4)toachieve
asymp-toticallystable, we havethe following nonlinear controllersby
traditionalmethodas:
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
u1=−x2−e1 u2=(a+bx3)x1+(a+bx3)x31+cx2−dx3−e2 u3=−x4−e3 u4=+ex3−f(1−x23)x4−gx1−e4 (4-8)substitutingallthenonlinearcontrollersinEq.(4-8)intoEq.(4-4):
˙V =−e2
1−e22−e23−e24<0 (4-9)
whichisanegativedefinitefunctionandtheerrordynamicssystem
isgoingtoachieveasymptoticallystable.Thesimulationresultsare
showninFigs.8and9.
4.1.1. ComparisonwithFLCCandtraditionalmethod
In this section, numerical simulation results by FLCC and
traditionalcontrollersarelistedinTables2and3forcomparison.
ComparingthetwosimulationresultsinTables2and3,itisclear
tofindoutthat(1)theperformance(accuracyandspeedof
con-vergence)ofthestatesoferrorsconvergingtotheoriginalpointby
Fig.18.TimehistoriesofstatesforCase3—theFLCCiscomingintoafter20s.
Table4
Comparisonbetweenformsofcontrollers.
FLCC(components) Nonlinearcontrollers u1 ±Z1=±50ore1≈0 −x2+ω×F1cosωt−e1 u2 ±Z2=±500ore2≈0 +(a+bx3)x1+(a+bx3)x13+cx2+dx3+ω× F2cosωt−e2 u3 ±Z3=±20ore3≈0 −x4+ω×F3cosωt−e3 u4 ±Z4=±100ore4≈0 ex3−f(1−x23)x4−gx1+ω×F4cosωt−e4 Table5
Comparisonbetweenerrordataat29.96s,29.97s,29.98s,29.99sand30.00safter theactionofcontrollers.
Time FLCC Nonlinearcontrollers
e1 e1 29.96s 0.00031450 0.0065468 29.97s 0.00031388 0.0065308 29.98s 0.00031325 0.0065146 29.99s 0.00031262 0.0064984 30.00s 0.00031197 0.0064820
Time FLCC Nonlinearcontrollers
e2 e2 29.96s 0.00064831 0.0064258 29.97s 0.00064702 0.0064110 29.98s 0.00064572 0.0063961 29.99s 0.00064440 0.0063810 30.00s 0.00064306 0.0063658
Time FLCC Nonlinearcontrollers
e3 e3 29.96s 0.00037285 0.0112380 29.97s 0.00037216 0.0111750 29.98s 0.00037146 0.0111130 29.99s 0.00037075 0.0110510 30.00s 0.00037003 0.0109890
Time FLCC Nonlinearcontrollers
e4 e4 29.96s 0.00142410 0.0066563 29.97s 0.00142150 0.0066392 29.98s 0.00141900 0.0066220 29.99s 0.00141630 0.0066047 30.00s 0.00141370 0.0065873
FLCCismuchbetterthantheperformancebytraditionalones;(2)
thecontrollersinFLCCdesigningaremuchsimplerthantraditional
ones.
Consequently,eventhesystemissocomplicatedandthe
con-trollers provided in FLCC are such a constant ones, the high
performanceandexact numericalsimulationresultscanbestill
remained.
4.2. CaseII:controlthechaoticmotiontoaregularfunctionby
(1)FLCCand(2)traditionalcontrollers
In this case we will control the chaotic motion of thenew
Mathieu–vanderPolsystem(4-1)toregularfunctionoftime.The
goalisyi=Fisinωt(i=1,2,3,4).Theerrorequation
ei=xi−yi=xi−Fisinωt, (i=1,2,3,4) (4-10)
lim
t→∞ei=t→∞lim(xi−Fisinωt)=0, (i=1,2,3,4)
whereF1=0.5,F2=1,F3=0.5,F4=2andω=0.5
Theerrordynamicsis
⎧
⎪
⎨
⎪
⎩
˙e1=x2−ω×F1cosωt+u1˙e2=−(a+bx3)x1−(a+bx3)x13−cx2+dx3−ω×F2cosωt+u2
˙e3=x4−ω×F3cosωt+u3
˙e4=−ex3+f(1−x23)x4+gx1−ω×F4cosωt+u4
(4-11)
ChoosingLyapunovfunctionas:
V= 12(e21+e22+e23+e24) (4-12)
Itstimederivativeis:
˙V =e1˙e1+e2˙e2+e3˙e3+e4˙e4=e1(x2−ω×F1cosωt+u1)
+e2(−(a+bx3)x1−(a+bx3)x13−cx2+dx3−ω
×F2 cosωt+u2)+e3(x4−ω×F3cosωt+u3)
+e4(−ex3+f(1−x23)x4+gx1−ω×F4 cosωt+u4) (4-13)
S.-Y.Li/AppliedSoftComputing11(2011)4474–4487 4485
Fig.19.TimehistoriesoferrorsforCase3—bynonlinearcontrollers(20s).
InordertodesignFLCC,wedivideEq.(4-13)intofourpartsas
follows:
Assume V=(1/2)(e2
1+e22+e23+e24)=V1+V2+V3+V4,
then ˙V =e1˙e1+e2˙e2+e3˙e3+e4˙e4= ˙V1+ ˙V2+ ˙V3+ ˙V4, where
V1=(1/2)e21, V2=(1/2)e22, V3=(1/2)e23andV4=(1/2)e24.
Part1: ˙V1=e1˙e1=e1(x2−ω×F1cosωt+u1)
Part 2: ˙V2=e2˙e2=e2(−(a+bx3)x1−(a+bx3)x31−cx2+dx3−
ω×F2cosωt+u2)
Part3: ˙V3=e3˙e3=e3(x4−ω×F3cosωt+u3)
Part 4: ˙V4=e4˙e4=e4(−ex3+f(1−x23)x4+gx1−ω×F4cosωt+
u4)
AccordingtotheprocessoftheFLCCdesigning,theerror
deriva-tivesofCaseIIwithoutanycontrollerareshowninFig.10,andthe
valuesoftheupperboundandlowerboundcanbeobtainedas
follows:
Z1=50, Z2=500, Z3=20, Z4=100 (4-14)
AftergettingZ1∼Z4,alltheconstantcontrollers,membership
func-tionsandthecorrespondingFLCCcanbedecided.Therefore, ˙V1=
e1˙e1<0, ˙V2=e2˙e2<0, ˙V3=e3˙e3<0 and ˙V4=e4˙e4<0 can
beachieved,thenwehave ˙V= ˙V1+ ˙V2+ ˙V3+ ˙V4<0.Itisclearthat
thederivativeofLyapunovfunctionisnegativedefiniteandthe
errordynamicssystemisgoingtoachieveasymptoticallystable.
ThesimulationresultsareshowninFigs.11and12.
(2)Bytraditionalnonlinearcontrollers
In ordertolead theerrordynamicsin Eq.(4-11) toachieve
asymptoticallystable,wehavethefollowingnonlinearcontrollers
bytraditionalmethodas:
⎧
⎪
⎨
⎪
⎩
u1=−x2+ω×F1cosωt−e1 u2=+(a+bx3)x1+(a+bx3)x13+cx2+dx3+ω×F2cosωt−e2 u3=−x4+ω×F3cosωt−e3 u4=ex3−f(1−x23)x4−gx1+ω×F4cosωt−e4 (4-15)substitutingallthenonlinearcontrollersinEq.(4-15)intoEq.
(4-13):
˙V =−e2
1−e22−e23−e24<0 (4-16)
whichisanegativedefinitefunctionandtheerrordynamicssystem
isgoingtoachieveasymptoticallystable.Thesimulationresultsare
showninFigs.13and14.
4.2.1. ComparisonwithFLCCandtraditionalmethod
NumericalsimulationresultsbyFLCCandtraditionalcontrollers
arelistedinTables4and5forcomparison.Itcanbefoundoutthat
(1)theperformance(accuracyandspeedofconvergence)ofthe
statesoferrorsconvergingtotheoriginalpointbyFLCCismuch
betterthantheperformancebytraditionalones;(2)thecontrollers
inFLCCdesigningaremuchsimplerthantraditionalones.
Consequently, even thesystem is socomplicated or regular
form,andthecontrollersprovidedinFLCCaresuchaconstantones,
thehighperformanceandexactnumericalsimulationresultscan
bestillremained.
4.3. CaseIII:controlthechaoticmotionofthenewMathieu–van
derPolsystemtoanotherchaoticmotionoftheQisystemvia(1)
FLCCand(2)traditionalcontrollers
In Case III, we will control chaotic motion of the new
Mathieu–vanderPolsystem(4-1)tothatoftheQisystem.The
goalsystemforcontrolisQisystemshownasfollows:
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
˙y1=a1(y2−y1)+y2y3y4 ˙y2=b1(y1+y2)−y1y3y4 ˙y3=−c1y3+y1y2y4 ˙y4=−d1y4+y1y2y3 (4-17)wherey1,y2,y3andy4 arethestatevariablesofthesystemand
a1,b1,c1andd1areallpositiverealparameters.ThisQisystemin
Eq.(4-17)wasrecentlyintroducedbyQietal.[28]andithasbeen
shownexhibitcomplexdynamicalbehaviorincludingthefamiliar
Fig.20. TimehistoriesofstatesforCase3—bynonlinearcontrollers(20s).
systemparameters:a1=30,b1=10,c1=1,d1=10andinitial
con-ditions(y10,y20,y30,y40)=(20,50,20,50),theQimodelexhibits
chaoticmotionwhichisshowninFig.15.
Theerrorequation
ei=xi−yi, (i=1,2,3,4) (4-18)
lim
t→∞ei=tlim→∞(xi−yi)=0, (i=1,2,3,4)
Theerrordynamicsis
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
˙e1= ˙x1=x2−(a1(y2−y1)+y2y3y4)+u1 ˙e2= ˙x2=−(a+bx3)x1−(a+bx3)x31−cx2+dx3−(b1(y1+y2)−y1y3y4)+u2 ˙e3= ˙x3=x4−(−c1y3+y1y2y4)+u3 ˙e4= ˙x4=−ex3+f(1−x23)x4+gx1−(−d1y4+y1y2y3)+u4 (4-19)ChoosingLyapunovfunctionas:
V=12(e21+e22+e23+e24) (4-20)
Itstimederivativeis:
˙V =e1˙e1+e2˙e2+e3˙e3+e4˙e4=e1(x2−(a1(y2−y1)
+y2y3y4)+u1)+e2(−(a+bx3)x1−(a+bx3)x31−cx2+dx3
−(b1(y1+y2)−y1y3y4)+u2)+e3(x4−(−c1y3+y1y2y4)
+u3)+e4(−ex3+f(1−x23)x4+gx1−(−d1y4+y1y2y3)+u4)
(4-21)
(1)ByFLCC
InordertodesignFLCC,wedivideEq.(4-12)intofourpartsas
follows: Assume V=(1/2)(e2 1+e 2 2+e 2 3+e 2 4)=V1+V2+V3+V4,
then ˙V=e1˙e1+e2˙e2+e3˙e3+e4˙e4= ˙V1+ ˙V2+ ˙V3+ ˙V4, where
V1=(1/2)e21, V2=(1/2)e22, V3=(1/2)e23andV4=(1/2)e24.
Part1: ˙V1=e1˙e1=e1(x2−(a1(y2−y1)+y2y3y4)+u1)
Part 2: ˙V2=e2˙e2=e2(−(a+bx3)x1−(a+bx3)x31−cx2+dx3−
(b1(y1+y2)−y1y3y4)+u2)
Part3: ˙V3=e3˙e3=e3(x4−(−c1y3+y1y2y4)+u3)
Part 4: ˙V4=e4˙e4=e4(−ex3+f(1−x32)x4+gx1−(−d1y4+
y1y2y3)+u4)
AccordingtotheprocessoftheFLCCdesigning,theerror
deriva-tivesofCaseIIIwithoutanycontrollerareshowninFig.16,andthe
valuesoftheupperboundandlowerboundcanbeobtainedas
follows:
Z1=2000, Z2=2000, Z3=1000, Z4=1000 (4-22)
AftergettingZ1∼Z4,alltheconstantcontrollers,membership
func-tionsandthecorrespondingFLCCcanbedecided.Therefore, ˙V1=
e1˙e1<0, ˙V2=e2˙e2<0, ˙V3=e3˙e3<0 and ˙V4=e4˙e4<0 can
beachieved,thenwehave ˙V= ˙V1+ ˙V2+ ˙V3+ ˙V4<0.Itisclearthat
thederivativeofLyapunovfunctionisnegativedefiniteandthe
errordynamicssystemisgoingtoachieveasymptoticallystable.
ThesimulationresultsareshowninFigs.17and18.
(2)Bytraditionalnonlinearcontrollers
In ordertolead theerrordynamicsin Eq.(4-19) toachieve
asymptoticallystable,wehavethefollowingnonlinearcontrollers
bytraditionalmethodas:
⎧
⎪
⎨
⎪
⎩
u1=−x2+(a1(y2−y1)+y2y3y4)−e1 u2=(a+bx3)x1+(a+bx3)x31+cx2−dx3+(b1(y1+y2)−y1y3y4)−e2 u3=−x4+(−c1y3+y1y2y4)−e3 u4=+ex3−f(1−x23)x4−gx1+(−d1y4+y1y2y3)−e4 (4-23)S.-Y.Li/AppliedSoftComputing11(2011)4474–4487 4487
Table6
Comparisonbetweenformsofcontrollers.
FLCC(components) Nonlinearcontrollers u1 ±Z1=±2000ore1≈0 −x2+(a1(y2−y1)+y2y3y4)−e1 u2 ±Z2=±2000ore2≈0 (a+bx3)x1+(a+bx3)x31+cx2−dx3+(b1(y1+ y2)−y1y3y4)−e2 u3 ±Z3=±1000ore3≈0 −x4+(−c1y3+y1y2y4)−e3 u4 ±Z4=±1000ore4≈0 ex3−f(1−x32)x4−gx1+(−dy4+y1y2y3)−e4 Table7
Comparisonbetweenerrordataat59.96s,59.97s,59.98s,59.99sand60.00safter theactionofcontrollers.
Time FLCC Nonlinearcontrollers
e1 e1 59.96s −7.0533e−011 −1.5015e−004 59.97s −6.9883e−011 −1.7386e−004 59.98s −6.9228e−011 −1.9902e−004 59.99s −6.8573e−011 −2.2194e−004 60.00s −6.7920e−011 −2.3330e−004
Time FLCC Nonlinearcontrollers
e2 e2 59.96s 4.1470e−012 −1.3996e−004 59.97s 4.1160e−012 −1.3715e−004 59.98s 4.0852e−012 −1.2018e−004 59.99s 4.0544e−012 −7.9564e−005 60.00s 4.0239e−012 −2.2182e−006
Time FLCC Nonlinearcontrollers
e3 e3 59.96s 0 9.1480e−007 59.97s 0 −1.5572e−005 59.98s 0 −4.0608e−005 59.99s 0 −7.6908e−005 60.00s 0 −1.2420e−004
Time FLCC Nonlinearcontrollers
e4 e4 59.96s −3.0018e−011 −9.9919e−005 59.97s −2.9837e−011 −1.3461e−004 59.98s −2.9644e−011 −1.7607e−004 59.99s −2.9441e−011 −2.2263e−004 60.00s −2.9231e−011 −2.6721e−004
substitutingallthenonlinearcontrollersinEq.(4-23)intoEq.
(4-21):
˙V=−e2
1−e22−e23−e24<0 (4-24)
whichisanegativedefinitefunctionandtheerrordynamicssystem
isgoingtoachieveasymptoticallystable.Thesimulationresultsare
showninFigs.19and20.
4.3.1. ComparisonwithFLCCandtraditionalmethod
Inthiscase,thenewMathieu–vanderPolsystemiscontrolled
toanotherchaoticmotion—theQisystemwithlargeinitial
con-ditions(y10,y20,y30,y40)=(20,50,20,50).Thiscaseisillustrated
toinvestigatetheeffectivenessandfeasibilityoftheFLCCevenif
thetwochaotictrajectoriesarefarfromeachother.Accordingto
Figs.17and19,thespeedofcontrollingtheerrorstatestothe
orig-inalpointsviaFLCC(aboutat30s)ismuchfasterthanthespeed
vianonlinearcontroller(aboutat60s).Asaresults,thenumerical
datain59.96–60.00sarefurtherproposedforcomparison.
NumericalsimulationresultsbyFLCCandtraditionalcontrollers
arelistedinTables6and7forcomparison.Thetwomain
supe-riorities still exist—(1)theperformance (accuracyand speed of
convergence)oftheconvergenceoferrorstatesbyFLCCismuch
better than by traditional method; (2) the controllers in FLCC
designingaremuchsimplerthantraditionalones.
5. Conclusions
Inthispaper,asimplestcontroller—fuzzylogicconstant
con-troller (FLCC) is introduced to chaos control. The illustrations
mentionedabovedemonstratethebetterperformanceand
accu-racyofnumericalsimulationresultsofthesynchronizationviaFLCC
clearlyeventhefuzzycontrollersareonlyasimpleformofconstant
numbers.
Asaresult,threemaincontributionscanbeconcluded—(1)high
performance oftheconvergence oferror statesin
synchroniza-tion;(2)thestrengthofthefuzzycontrollerscanbeadjustedvia
membershipfunctions;(3)fuzzylogiccontrollersareeasyto
pro-duce.Furthermore,duetothecharactersofFLCC,themathematical
modelsofdomainsystemscanbeunknown,allwehavetodois
cap-turingtheoutputsignals,constructingthefuzzylogicsystemand
calculatingtheareaoftheerrorderivatives,thenwecancontrol
anyoutputsignaltoanotherone.Hence,FLCCis sucha
poten-tialtoolandcanbeappliedtovariouskindsoffieldswithlotsof
unknownfunctions—suchasneuroscience,un-modelbio-systems,
andcomplicatedbrainnetwork.
Acknowledgment
ThisworkwassupportedinpartbytheUST-UCSDInternational
CenterofExcellenceinAdvancedBio-engineeringsponsoredbythe
TaiwanNationalScienceCouncilI-RiCEProgramunderGrant
Num-ber:NSC-99-2911-I-009-101.Thisresearchwassupportedbythe
NationalScienceCouncil,RepublicofChina,underGrantNumber
NSC99-2221-E-009-019.
References
[1]E.Ott,C.Grebogi,J.A.Yorke,Phys.Rev.Lett.64(1990)1196. [2]W.Lin,Phys.Lett.A372(2008)3195.
[3]Z.-M.Ge,C.-H.Yang,PhysicaD231(2007)87.
[4]Z.-M.Ge,Y.-S.Chen,ChaosSolitonsFractals26(2005)881.
[5]Z.-M.Ge,S.-C.Li,S.-Y.Li,C.-M.Chang,Appl.Math.Comput.203(2008)513. [6]T.Yang,L.B.Yang,C.M.Yang,Phys.Lett.A246(1998)284.
[7]T.Yang,C.M.Yang,L.B.Yang,Phys.Lett.A232(1997)356. [8]R.Tchoukuegno,P.Woafo,PhysicaD167(2002)86. [9]R.-A.Tang,Y.-L.Liu,J.-K.Xue,Phys.Lett.A373(2009)1449. [10]Z.-M.Ge,H.-H.Chen,J.SoundVib.209(1998)753. [11]X.Wu,J.Cai,M.Wang,ChaosSolitonsFractals36(2008)121. [12]L.A.Zadeh,IEEEComput.21(1988)83.
[13]A.Shahraz,R.BozorgmehryBoozarjomehry,ControlEng.Pract.17(2009)541. [14]C.-Y.Chen,T.-H.Li,Y.-C.Yeh,Inf.Sci.179(2009)180.
[15]Y.W.Wang,Z.H.Guan,H.O.Wang,Phys.Lett.A320(2003)154. [16] G.Li,A.Khajepour,J.SoundVib.280(2005)759.
[17]T.-H.S.Li,C.-L.Kuo,N.R.Guo,ChaosSolitonsFractals33(2007)1523. [18]H.-X.Li,H.B.Gatland,IEEETrans.Syst.ManCybern.25(1996)791. [19]H.-X.Li,H.B.Gatland,IEEETrans.Syst.ManCybern.25(1995)505–512. [20]J.Lee,IEEETrans.FuzzySyst.1(1993)298.
[21]B.J.Choi,S.W.Kwak,B.K.Kim,FuzzySyst.30(2000)303. [22]H.Ying,Automatica30(7)(1994)1185.
[23]O.Castillo,L.Aguilar,N.Cázarez,D.Rico,Int.Multi-Conf.Comp.Sci.Comput. Eng.1(2005)412–418.
[24]N.R.Cázarez,S.Cárdenas,L.Aguilar,O.Castillo,ExpertSys.Appl.37(2010) 4368.
[25]M. Margaliot, G. Langholz, New Approaches to Fuzzy Modeling and Control—DesignandAnalysis,WorldScientific,Singapore,2000.
[26] H.-T.Yau,C.-S.Shieh,NonlinearAnal.TheoryMethodsAppl.9(2008)1800. [27]C.S.Shieh,IEEProc.ControlTheoryAppl.150(1)(2003)45.
[28]G.Qi,S.Du, G.Chen,Z.Chen,Z.yuan,ChaosSolitonsFractals23(2005) 1671.