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



ֲָȶ
Ҫ
̸ž
ժҪ۷ƽֲָȶжƽָĹƣõƽֲָȶĳҪһжƽǷΪָ.иһ巶(е)ֲȫָȶжƹ֪ضõĽ.
ؼʡ磬ֲָȶָ.
A SUFFICIENT AND NECESSARY CONDITION FOR LOCAL
EXPONENTIAL STABILITY OF CONTINUOUS-TIME
ASSOCIATE MEMORY NEURAL NETWORKS
WANG LishengTAN ZhengZHANG Junkai
(School of Electric and Information Engineering, Xi'an Jiao Tong University, Xi'an710049)
AbstractA sufficient and necessary condition for nonlinear continuous-time associate memory neural networks to be locally exponentially stable is obtained, and a function is defined for characterizing the domain of attraction of equilibrium point. Some conditions for networks to be locally or globally exponentially stable under a group of norms(monotone norms) are also obtained, which generalize the results of some literature.
Key wordsLocal exponential stability, domain of attraction, convergent rate, global exponential stability, associate memory neural network.
1
ƽȶԷеҪ.ܿƼģʽʶϢϵͳҪΪĶ̬Ϊȶ.жֱӦڲͬģʽƽ㣬ȶԷĿжںƽֲȶԼƽ.ݴָƽ̬.оƽľֲָȶ.⣺һжƽΪֲָȶƽ?ιƽָ?ƽΪRnһ񵴻.ĸƽΪֲָȶĳҪش˵һ⣻ڵڶ⣬ĶһֻҪƽС㣬Ϊ򣬸һǱڵļ㷽.Ľƹף1378еҪ.ΪɿķԳֵ⣬۷Գֵȶ.

2ԳֲֵָȶĳҪ
TǷӣx*Dǳ.N(T,x*,D)목T(D,x*)ֱ£

t>0,IǺ.f(t)=(I+tT(D,x*)-1)tΪf(t)tĵңT(D,x*)f(t)-T(D,x*),Լ޴.ӦָT(D,x*)N(T,x*,D)TڼDϹx*ĺT,D,x*Ψһȷ.N(T,x*,D)ʣN(T,x*,D)ܣT(D,x*)N(T,x*,D)f(t)N(T+G,x*,D)N(T,x*,D)+N(G,x*,D)TΪx*Ϊڵ㣬N(T,x*,D)TĶ(T)4.
֪(A)<0Aָȶ.ǽ֤ڷƽȻ.ǷԳֵ
(1)
x*ٽ㣬T(x*)=0x*ֲָȶx*UK>0ʹõֵx(0)Uʱʽ(1)ĽⰴָKx*.ٽx*()zRnz-x*ܦģB(x*).
1.ٽx*B(x*)N(T,x*,B(x*))<0x*ֲָȶB(x*)ΪָN(T,x*,B(x*))Ϊָ.Ҷx(0)B(x*),x(t)㡬x(t)-x*ܡx(0)-x*et*N(T,x*,B(x*)).
֤.x(t)x(0)B(x*)ʱʽ(1)Ľ⣬v(t)=x(t)-x*
d
ˣx(t)-x*ܡx(0)-x*.et*N(T,x*,B(x*)).x(0)B(x*)t>0x(t)B(x*)
ע1.N(T,x*,B(x*))жB(x*)ǷΪ.N(T,x*,Rn)<0ʽ(1)ȫָȶָΪN(T,x*,Rn).
2.Tٽx*Frechetɵx*ֲָȶĳҪTx*ĵT(x*)()ֵʵС.
֤.T(x*)ֵʵС㣬ڵȼ۷ʹæ(T(x*))<04.T΢ʣ0<<-(T(x*))B(x*)ʹãT-T(x*)(B(x*)x*)<.T=T(x*)+(T-T(x*))N(T,x*,B(x*))ܦ(T(x*))+T-T(x*)(B(x*),x*)ܦ(T(x*))+<0x*ֲָȶ.
x*ֲָȶy(t)=x(t)-x*G(y(t))=T(y(t)+x*)ʽ(1)x*ȶֵ(t)=G(y(t)),y(0)Rny=0ȶԵȼ.Gy=0ɵ.G=G(0)+(G-G(0))ף5֪G(0)ֵʵȫС.
ע2.ָT(x*)Ĺϵǡ

3ֲָȶĳҪ
Ƿģ
(2)
gi(z)΢0<gi(z)supgi(z)1.ʽ(2)ľʾΪdxdt=-x+WG(x)+ssΪ.ģ(2)ɿֵ(1)һ.ʽ(2)УT=-I+WG+sx*ʽ(2)ƽ㣬T(x*)=0.ɶ12֪¶.
3.x*ʽ(2)ľֲָȶƽ̬ҽT(x*)=-I+WG(x*)ֵʵȫС.x*B(x*)ʹN(T,x*,B(x*))<0Ϊָ򣬴B(x*)ڳĹָָΪN(T,x*,B(x*)).
ע3.ƽľֲָȶͨƽ㴦()ֵж.ƽָͨN(T,x*,B(x*))õ.
Ϊ˵õһ巶(е)ȶƣȽܵĶ.ǣx=(x1,x2,,xn)x=(x1,x2,xn)Rn.ӣxܣyɵáxܡy.Ϊǵ..ҽx=x6ݣ·xD=DxΪD=diagd1,d2,,dnΪԽ.pʣ1,2,3,,ޣΪ.ȡΪӦؾΪW=supWx:x=1.
ʵӦУһǳҪι캯N(T,x*,D)?׷.2£ֵͨõڵ£B(x*)ڣx-x*-t0Txܦx-x*N(T,x*,B(x*))(-1)t0<0t0>0(0,1).⣬òʽ
N(T,x*,D)-1+N(WGx*D)-1+WG(D,x*)-1+W.G(D,x*)
N(T,x*,D)Ĺֵжȶ(46).
T=T(x*)+(T-T(x*))=T(x*)+W(G-G(x*)).gi(xi)΢>0>0,x<ʱgj(xj+x*j)-gj(x*j)-g(x*j)*xj((T(x*))-)W*xj.ˣG(x+x*)-Gx*-G(x*)(x)((T(x*))-ţW.xB(x*)ڣG-G(x*)(B(x*)x*)((T(x*))-)W.¶.
4.(T(x*))<0x*B(x*)Ϊx*ָ.ֵx(0)B(x*)ʱָָΪ-.
ע4.ף13ڷ.1.µõƽ.
ȻֻҪx*B(x*)ڣG(B(x*)x*)<1WN(T,x*,B(x*))<0.ΪֻҪƽ̬x*ԪԺgi(xi)仯㹻x*ֲָȶ.
5.θxB(x*)gi(xi)-gi(x*i )<(1-)Wxi-x*iΪСx*ֲָȶB(x*)ΪָΪ-.
֤.Gx-Gx*((1-)W)x-x*G(B(x*)x*)(1-)W<1WN(T,x*,B(x*))<-<0.x*ֲָȶB(x*)Ϊx*.
6.W<1ʽ(2)ȫָȶָΪW-1.
֤.ֵgi(xi)-gi(x*i)ܣxi-x*iԣG(Rn,x*)1N(T,x*,Nn)ܪW-1<0.x*ȫָȶָΪW-1.
ע5.ף78УWp<1pʣ1,2,ޣʱ֤.

4ʵ
Ȼuk(k=1,2,r)Էuk=WG(uk)+s45еuk(2)ľֲָȶƽ̬.ݴˣԵķۺЧ磬ף13.Ϊȡn=2g1(x1)=sin(x1),g2(x2)=cos(x2)s=0,0x*=У20ӦȸļģʽɼõȨʹx*Ϊֲָȶƽ.Wޡ2<У6ʱɶ5B(x*)Ϊ1-2.sinΪָ.
߼飺ͨѧӦѧϵ˶ʿҵڵѧԺʿȤоΪ硢ѧӻͼ.
̸ͨѧѧԺڣоΪͼʵ.
ߵλͨѧѧԺϢо710049
ο
1ѧ.HepfieldָٶȵĹƼӦ.ѧ199624(1)4043
2ѧ.ģָⷴٶȵĹƼӦ.ѧ199818(9)712716
3ѧ.ָٶȵĹƼӦ.ӿѧѧ199618(1)26
4Strom T. On logarithmic norms. SIAM J. Numer Anal., 197512(5)741753
5Seifert G. On a converse result for perron's theorem for asymptotic stability for nonlinear differential equation. In: Proc. of Amer. Math. Soc.,1987,99(4):733736
6Bauer F, Stoer J. Absolute and monotonic norms. Numer. Math.,19613257264
7Killy D G. Stability in contractive nonlinear neural networks,IEEE Trans. Biomedical Engineering,199037231242
8Sugawara K, Harao M. On the stability of equilibrium states of analogue neural netuorks,Transactions of the Institute of Electronics and Communication Engineers(in Japanese), 1983J-66-A258265
ոڡ1998-03-20޸ĸڡ1998-08-06
