Զѧ
ACTA AUTOMATICA SINICA
1999 25 6 Vol.25 No.6 1999



һCMACṹѧϰ㷨
ۻݺ
ժҪһCMAC(Cerebellar Model Articulation Controller)ռ䳬.ڳģϢ洢ָԣ˿ѧϰ㷨.ͨԶ̬ϵͳģоCMACкǿѧϰͷ.
ؼʡCMAC䣬ѧϰ㷨.
CMAC WITH HYPERBALL STRUCTURE AND ITS
LEARNING ALGORITHM
DUAN PeiyongSHAO Huihe
(Department of Automation,Shanghai Jiaotong University,Shanghai200030)
AbstractThis paper presents the CMAC(cerebellar model articulation controller),which can guarantee the input space by hyperballs.Based on the fuzzy basis functions defined on the hyperballs,information is stored and retrieved.A fast convergent learning algorithm is also given.Simulations for the CMAC used in nonlinear dynamic system modeling are performed to demonstrate its powerful associative memory and generalization performance.
Key wordsCMAC, associative memory, learning algorithms.
1
CMACģСԵѧϰṹ.ͳCMACֻȨϵۼ.ף1ݵоϴķڴͳCMACļ㷽ȫǡѧϰ㷨ֲģָĽѧϰ㷨Ϣ洢.ChiangLin˻ڹCMAC2ݣCMAC.ChiangLin˼룬һµCMACṹѧϰ㷨.

2CMACṹ
mάռΪX=A1A2Amx=x1,x2,,xmTXxiʣx-i,x+i,RΪʵx+i,x-iΪiάϡ½磻y=y1,y2,,ynT,yY,YΪн缯.x-i,x+iݾΪNȷݣΪiзֵ㹹ɼDi.mάpj=pj1,pj2,,pjmTXpjiDi(i=1,2,,m;j=1,2,,L)jk,pjpk.pjXռϾȷֲpjΪռXϵ.Ϊ˱ڻѡȡƵCMACռXбҪб׼

ͰռXΪ˻ռ.iԪĵַniʾӦȨֵʾΪqi.niΪģ峬.ˣϾȷֲL򣬼ǪNe.Biϵĸ˹Ϊ

ϵĻģ3ݣһͬģ߼ߺԣģbi(.)ڵi
(1)


ͼ1CMACṹʾͼ
С.ΪŷϷ.۲bi(.)Կ1)bi(.)ŵѸټСʱbi(.)СԺԲƵĳ̶ȣʱѳΪbi(.)ð뾶ʵϣ=3ʱbi(.)e-9㹻Сͨȡѡ3;2)bi(.)볬İ뾶йأ뾶ԽNeԽbi(.)ԽС.Ըݶԣt,ytCMACÿһάЩϵĻ.Ϊ˵⣬ֻΪһάʱֵΪ
(2)
ΪȨϵstΪѡȻNeԪΪ1Ϊ0.CMACṹʾͼͼ1ʾʵ߱ʾ߱ʾ.
3ѧϰ㷨
ݽѧϰȨϵqֵ.ǵkѭѧϰʱCMACΪʵֵʱCMACĹֵ.ѧϰ㷨øĽC-L㷨
(3)
ЦΪʵ.ѧϰ㷨ԣĶ.
.02£0ʱ㷨(3).
ÿһֲֻqi,iUt,NeȨϵԵĻȣңѧϰĹ̰ģ߼.ֽCMACѧϰ㷨ʵֲ£
1.ȷ˻ռX׼Ϊ˻ռ䣻
2.ȷռϵĽڵ㣬ѡȡȨϵĳֵ
3.ѡȡĲҼð뾶ѣӶȷΪĵĳ
4.t,ytҳõĳ򣬼ȷѡst;
5.(2)ʽֵt;(3)ʽȨϵ
6.ظ45ֱCMACҪΪֹ.
4о
CSTRϵͳù̵ģ΢ַ̣4



ʽqqcֱΪӦȴ.ģвĺ弰µֵף5.CSTRΪǿϵͳΪqcΪCA.ĿCMACCSTRϵͳ.CMACΪ

ѡռΪX=0.040.14ݡ0.040.14ݡ0.040.14ݡ94114ݡ94114ݡ94114ݣÿΪ4ݣX׼ΪS=04ݡ04ݡ04ݣ=2.5=0.7=1.2=0.01ѡȡȨϵĳֵΪ.΢ַ̿ɻһݶԣxt,CA(t)(ʵӦУѧϰݿֳɼ)׼Ϊt,CA(t)Ϊѧϰ.ѧϰȷȨϵCMACΪCSTRģ.ͼ2΢ַ̺CMACqc(t)ߣͼ3ΪCSTRͼ4ΪCMAC.ʾCMACС.


ͼ2CSTRCMAC
