Զѧ
ACTA AUTOMATICA SINICA
1998 24 1  Vol.24  No.1 1998



ʵ־ŷϵͳ
ŶصĶ̬
¡ӡ硡ׯ
ժҪŶضԾŷϵͳӰ. ʹ緽Ŷصѧģͣ ȫԭƲڣ ʵŶصĶ̬. µĲŷϵͳľʮЧ ʵӦüֵ. 
ؼʡŷϵͳ Ŷأ 磬 ̬. 
DYNAMIC COMPENSATION FOR DISTURBING MOMENT IN
ACCURATE SERVO SYSTEM BASED ON NEURAL NETWORKS
XU LIXINZHANG YUHE
(Department of Automatic Control, Beijing Institute of Technology, Beijing 100081)

WANG CHANGHONGZHUANG XIANYI
(Department of Control Engineering, Harbin Institute of Technology, Harbin 150001)
AbstractThe affection of disturbing moment in servo system is discussed in this article at first, then we use the feedforward neural network to build the mathematic model of the disturbing moment, and design the transfer functin for compensation. Experiment shows that the new method is effective to improve the accuracy of the servo system.
Key wordsServo system, disturbing moment, neural network, dynamic compensation.
1
ŷϵͳѱ㷺ӦںԵ豸еĲת̨ʵеķת̨Ŀ걳ģװеĹѧ涯ϵͳ. ЩϵͳҪнϸߵĸٺͶλ. ŷϵͳȵȴܵĦΪŶص. 
Ȼعʷ϶ŶصĲ ָԣӦ缼ŶģͣʹùؼʵŶصĶ̬. 
2Ŷģͽ
2.1Ŷط
ŷϵͳеŶҪĦأ⻹иת. ĦصĴСṹ ԣشСй. תҪ븺й. ŶŷϵͳȵҪϰ ѧһ˴ȡһЩɹ. ף1ݿֽȶ ؿǽԾֵ͡ģ״̬۲ԲأʹǰƼԲȡ˽ϺõЧ. ף2о˾ת̨еĦģͣ ڸһ£ ʵĦʵĹϵĦصѧģͣ ĦرʾΪMF ʱʾΪ. 
1) ֶģ͡ͼ1ʾ. 
(1)


ͼ1ֶģ


MSMCֱΪĦغͿĦأM2Ϊת̨йжӦʦ2õĦ. 
2) 䳣ֵģͣΪĦһΧǳֵ
 (2)
ף2ݸĦģ״̬۲ ʹòԭǰ ʵת̨ϵͳĸپ. ǣף1н˳ֵתΪŶ. ף2ݸĦصķֶģ. ڽģͱȽϴֲڣʵӳĦصʵĦشڹ㸽Ĺɹ״̬ʱԻЧ. ܷͬʱĦغ͸أΪȷĦغ͸ģŷϵͳȵĹؼ. 
2.2罨ģ
緽Ŷģ. ǽʮ귢չһѧƣҪоɼ򵥴Ԫģɵ. ƺͱʶ㷺õƽԺ3ݺ֯Ӧѧϰص. 
ƽΪ򵥵ǰ磬ƽԺŶصѧģ. ŶҪĦغͺֵתأ ŶӰ죬 Ϊϵͳ͸һʱ Ŷ()ķԺ. Ȼһ̬. ʹǰԪ罨һģ. Ϊʹѵ㣬 ֱת(أ0)ͷת(أ0)ģ. ڷֱƽΪ򵥵ķԺȨֵѵ. ѡöǰʵ֣ÿ纬һڵ㣬һڵ㣬ѡ8ڵ. ڵ㼤Ϊ˫кtanh(x)=(1-exp(-2x))/(1+exp(-2x)) ڵ㼤ΪԺ. ṹԺ Ϊʣ ΪŶص. ĳֵʦأI. ֪ƽϵʽ

KMΪתس. ΪֵмɵŶ
Ǿ͵õһ(,). ΪʦѵǰȨֵ. Ϊ֤ѵЧԣѵӦǽϴΧ. 
2.3ѧϰ㷨
BP㷨һֳõѵھֲС ʹһѰŷѵǰȨֵ. һ־нǿȫԣ ṹͼĲѰŷ㷨ģ˻˼ϣνģ˻ǻͳУԹʽ˻ʱͨȽ£ʹеӿ˶Ȼ¶ȵ½ Ҳγ˵̬ľ. 㸽¶½㹻 һγĻ̬. Ż˵ ҲƵĹ. ｫѧϰŻ⣬ ĿͨȨֵԼСE. 
Wi,jΪԪiԪj֮Ȩ ڵkε
Wi,j(k)=Wi,j(k-1)+i,j(k). (3)

(4)
ΪһСȨֵÿǰһ. pi,j(k)Ϊֲ
(5)
T(k)ģ˻㷨ƣ ¶ȣ ŻӦ𽥽ͣCij(k)ʾȨֵı仯
(6)


㷨ʹĿ꺯EСȨֵ仯4ݣ㷨нϺõȫҼٶȱBP㷨5. ø÷ѵǰ磬㷨ʵּ򵥣Ч.
3ŶصĶ̬
3.1װ
Ȩֵѵ󣬼Ŷصģ. ͼ1ʾĦطֶģȣٶαΪ⻬ķߣ ٶνӽԹϵ. 
ŷϵͳͼ2ʾ. ʵʱٷźţ ͨģͼŶصĴС ΪʵŶصȫ 㲹ڴΪ

ʹӲ·ʵ֣ ԭһнСʱ䳣ĹԻ. ģͼǰڼʵ֣ Դ(s)ɱʵ. 


ͼ2ϵͳṹͼ
3.2
ŶҪӰŷϵͳĵٸԼλȣ ص㿼첹ǰָı仯. ʹϵͳٺ㶨ٶźţ䶨λ. ԺĿ걳ģװеĹѧ涯ϵͳΪ󣬲öԱȷ ŶǰбȽ. ѡ[-0.5,0.5]ڵ20ȷֲ(λrad/s)ѵȨֵźΪ0.1rad/sĵбźţȽϲǰϵͳĸپȣڸٹٸλźţͼ3(a)ͼ3(b)ʾУͼ3(a)ǲǰߣͼ3(b)ǲ. ÿУǰμ¼˶бźŵĸΪλ нϴԾ״̬л. 


ͼ3ǰ
1) ٶȻѹ0.3VС0.1V
2) 0.3mradС0.1mrad;
3) ̬0.1mrad͵ӽ޾. 
⣬ ģͲΧڵźţ ƾԽ͸ ϵͳѹ. ĲʵŷϵͳĵٸپȺͶλ. 

4
緽ŶصѧģͣǰķֶģͺͳֵģȣܹʵӳŶصı仯. ȫԭƲڴݺڽṹ򵥣ʵ. µĲǿкЧ. 䲹ܵŷϵͳЧ. 
δŶصӰ. ʵŷϵͳУŶҲǽλúʱϴӰŷϵͳľ. ͬŶ(縺ƫ)Ķ̬. 
ߵλ¡ӡѧԶϵ100081
硡ׯ塡ҵѧƹϵ150001
ο
[1]ܹ㶣 ȫ. ״̬۲ڵϵͳеӦ. 廪ѧѧ198020(4)33-46.
[2]Ԫ Ӣ. ת̨ĦصĶ̬ Զѧ19839(4)248-252.
[3]Funahashi, K. On the approximate realization of continuous mappings by neural networks. Neural Networks, 1989,2(3):183-191.
[4]Unnikrishnan K P et al. Alopex: A correlation-based learning algorithm for feed-forward and recurrent neural network. neural computation 1994,6:469-490.
[5]£磬 ׯ.ѵһо.ҵѧѧ1997,29(3):85-87.
ոڡ1996-01-12
