Զѧ
ACTA AUTOMATICA SINICA
1998 24 5  Vol.24  No.5 1998



CMACƼ㷨GA
ŻCMACƫʸֲ1)
񶫡
ժҪGA(Genetic Algorithm)ʵCMAC(Cerebellar Model Articulation Controller)Ƽ㷨.÷CMACѧϰŻ⣬ʵüֵ.֤÷ǳɹĺЧ.ԲͬĿ͹۶(ռ)ԲGAҵCMACڲʾ(ƫʸֲ)ʵһCMACԴﵽľ.÷AlbusCMACParksȵCMACѧϰЧв̶ͬȵߣʺҪ߾ѧϰ.ͬʱƫʸֲCMAC㷨.
ؼʡCMACƣŴ㷨ƫʸֲ.
OPTIMAL DESIGN AND ALGORITHM OF CMAC
OPTIMIZATION OF CMAC DISPLACEMENT VECTORS
DISTRIBUTION BY GENETIC ALGORITHM
ZHOU XUDONGWANG GUODONG
(State Key Laboratory Rolling Technology and Automation,Northeastern University,Shenyang110006)
AbstractOptimal design and algorithm of CMAC are implemented by genetic algorithm(GA) in this paper.Such a method has solved the optimization problem considering both CMAC and its learing objects.Simulations show that the proposed method is successful and effective.For different objects learned (such as space planes),different internal presentations(displacement vectors distribution)of CMAC can be found by GA,so that the optimal CMAC possesses higher learning precision than the general CMAC can reach.The proposed method has some improvements on both Albus' CMAC and Parks' CMAC.This method is suitable to the high precision learning problem.The CMAC algorithm for any displacement vectors distribution is also proposed.
Key wordsOptimal design of CMAC,genetic algorithm(GA),cerebellar model articulation controller(CMAC),displacement vectors distribution.
1
ӦУԲͬĿ͹۶Ƴšṹһֱо߹ע֮һ.ȻⷽCMACȷ.70J.S.AlbusCMAC1ݣCMACӦڿ.19871990MillerȰCMACɹӦڻ˿У23ݣƽ缰оߵ.ѧϰٶȿ(BPһ)ʺʵʱƣʹ֮ΪĿǰӦ֮һ1991ParksMilitzer˸ĽCMACƫʸֲ4ݲάΪn=2,15C=2,100ġšƫʸ.䡰šǻڽľСֵģûпCMACѧϰ.ParksMilitzerô.An˲ʽͬ5ݣҲûиNάCMACһ㷨.ΪˣĲGA6ŻCMACƫʸֲ̽ŻЧⷨ.
2ƫʸֲCMAC㷨
2.1AlbusCMAC
AlbusCMACṹͼ1ʾ1.ͼ1УAcΪ洢ԪApΪ洢ԪF(Si)ΪCMAC.״̬ռsеÿһ㶼AcеCԪӦҲApеC洢ԪӦF(Si)CԪд洢ֵ(Ȩ)֮.ÿһӦAcApϵC洢Ԫɢ洢AcApʱs бȽϿЩAcApлֽʹCMACֵҲȽϿ󱻳ΪCMACķ(Generalization).


ͼ 1CMAC˽ṹ
2.2ƫʸֲCMAC㷨
1.ʸd(d0)d1,d2,dn,ڸʸȷֲУd1,d2,dn=(m.b1)modC,(m.b2)modC,(m.bn)modC,Уd0Ϊd(d0=1,C)nΪռάCΪdkΪdķ(k=1,C)m=1,C;modΪģ.

2.ƫʸb=b1,b2,,bnݣb1=1.ЧԪĸҰλƫ.νƫʸֲָڸҰƫʸ.
1)һάCMACӳ㷨
һάCMACʸ˽ṹͼ2ʾ.ӳڴַͼ2ʾ.ɹɳ¸ӳ㷨(Acַ)


ͼ 2һάCMACʸ˽ṹ
ע.aʾAcַC=3b=1
(1)
2)άCMACӳ㷨
άCMACʸ˽ṹμͼ3ʾ.ӳڴַͼ3ʾ.ɹɳ¸ӳ㷨(Acַ)


ͼ 3άCMACʸ˽ṹ
ע.aʾAcַC=3b=11
(2a)
s=I1,I2.(2b)
3)NάCMACӳ㷨
άCMACӳ㷨ƹ㣬ɵõNάCMACӳ㷨(Acַ)
(3a)
s=I1,In.(3b)
ʽУriΪiڽڵsΪʸIiΪsia(s,d0)Ϊӳַ.
ʽ(3)صǲʺƫʸֲCMACʺⳬķֲCMACΪʸֲоŻṩ˻㷨.
4)CMACӳ㷨ap(s,d0)=a(s,d0)mod MMΪڴС(MΪ).
5)CMAC㷨
6)CMACѧϰ㷨,ΪѧϰʣftΪʦź.
3GACMACƫʸֲ
GACMACƫʸֲCMACѧϰռ棬ʹۻƽС.CMACѧϰŻ.
GAŻƵĹؼö01ɵȾɫ⣬Ա˵αNάCMACƫʸֲ?Ϊ˼ﲢ֤ʸֲص㣬ȱƫʸb=(b1,b2,,bn),biʣ1,2,CnΪάȻת㷨תƫʸ.˿ƫʸCMAC㷨GAbŻ.
תԭȡҰڵһΪ(111)Ȼڴ˻ϼϻƫʸb=(b1,b2,,bn),ĳֵCʱʹ´1ʼ.
ת㷨dj=(dj-1+bi)mod C(i=2,Cj=1,,n).
Ŵ㷨
1)һʮbiʣ12Cݣi=1,2,,n
2)ʮbiɶBi=xi1,xi2,xim;i=1,2,n;mΪplog2(C)С.
3)Ⱦɫ
chromosomeB1B2BC
x11x12x1mݣx21x22x2mݣxn1xn2xnm
GAŻĿ꺯ΪObjfunc=CMACۻƽ(Ŀ꺯Сֵ).ӦֵΪĿ꺯ĵFitness=1/Objfunc.
Դﵽȫ.ѰŹҪһļǻõCMACȷһCMACиѧϰٶ.
Ϊ֤ķЧ.CMACѧϰռ(ƽf(I1,I2)=1f(I1,I2)=sin(2I110)sin(2I210)I1=1,2,36.I1,I2λΪ10)ʹۻƽС.CMACΪn=2,r1=35,r2=35,C=2,3,4,5.ʼCMACȨؿռΪ㣬=0.7ѧϰΪ3.ɢGAΪm=2,2,3,3(ֱӦC=2,3,4,5),ȺСPopsize=10Ϊ100.ӽΪpc=0.6pm=0.01.1ʾ.
1CMACѧϰռ(ѧϰ3ۻƽQ)Ƚ

ķAlbusParksȷ
ƫʸQƫʸQƫʸQ
f=1,C=2(11)(11)0.037 3(11)(22)5.713 6(11)(22)5.713 6
f=S,C=2(11)(12)1.358 8(11)(22)1.358 8(11)(22)1.358 8
f=1,C=3(11)(11)(11)0.008 3(11)(22)(33)16.429 0(11)(22)(33)16.429 0
f=S,C=3(11)(32)(13)2.539 3(11)(22)(33)4.042 0(11)(22)(33)4.042 0
f=1,C=4(11)(11)(11)(11)0.008 1(11)(22)(33)(44)13.934 8(11)(22)(33)(44)13.934 8
f=S,C=4(11)(42)(13)(44)2.879 4(11)(22)(33)(44)3.280 7(11)(22)(33)(44)3.280 7
f=1,C=5(11)(11)(11)(11)(11)0.081 0(11)(22)(33)(44)(55)10.052 7(11)(23)(35)(41)(53)10.052 7
f=S,C=5(11)(32)(53)(14)(35)2.040 2(11)(22)(33)(44)(55)2.3253(11)(23)(35)(41)(53)6.867 9

ע.f=Sʾf(I1,I2)=sin(2I110)sin(2I210).
ķǳɹЧ.ԲͬĿ͹۶󣬿ԲGAҵCMACڲṹʾ(ƫʸֲ)ﵽһCMACԴﵽľ.÷AlbusCMACParksȵCMACѧϰЧв̶ͬȵ.رʺҪ߾ѧϰ.
1)йʿѧ.
߼飺񶫡У1963.ѧڣ1995񱱾Ƽѧѧʿѧλ19951997ڶѧ²ʿоУν.ѷѧĶʮƪ.ĿǰҪоΪơģʽʶŴ㷨Ӧо.
У1942.ѧڡʿʦйѧַֻ.οִۡԼӹѧ.ۺʵ.ĿǰҪоΪۡպԶ˹ƹеӦõ.
ߵλ(ѧƼԶصʵҡ110006)
ο
1Albus J S.Data storage in the cerebellar model articulation controller(CMAC).Transactions of the ASME.Journal of Dynamic System Measurement and Control,1975,97:228-233
2Miller W T.Sensor based control of robotics manipulators using a general learning algorithm.IEEE Trans.Robotics Automat,1987,3(2):157-165
3Miller W T,Glanz F H,Kraft L G.CMAC:an associative neural network alternative to back propagation.In:Proceedings of the IEEE,1990,78:1561-1567
4Parks P C,Militzer J.Improved allocation of weights for associative memory storage in learning control systems.In:Proceedings of the 1st IFAC Symposium on Design Methods of Control Systems,Zurich:Pergamon Press,,1991,777-782
5An P C E,Miller W T,Parks P C.Design improvements in associative memories for cerebellar model articulation controller(CMAC).In:Proceedings of the International Conference on Artificial Neural Networks,Helsinki,North Holland,,1991:1207-1210
6Holland J H.Adaptive in Natural and Artificial Systems.Ann Arbor MI:University of Michigan Press,1975
ոڡ1996-07-29
