Զѧ
ACTA AUTOMATICA SINICA
1997ꡡ234ڡVol.23No.41997



ģĻʵʱо
οǿֽ
ؼʡ磬ģƣ.
STUDY ON ROBOT REAL-TIME CONTROL BASED
ON FUZZY NEURAL NETWORK
LIAO JUNZHU SHIQIANGLIN JANYA
(State Key Lab for Fluid Power Transmission and Control, Zhejiang University, Hangzhou 310027)
Key wordsNeural network,fuzzy control,robot.
1
ģģ߼Եŵ㣬ܵԽԽѧߵĸ߶ӣԶƺȡĳɹ1ݣЩԷΪ.ǰоϣĽؽڻΪ󣬶ģʵʿܽϵͳо.
2ģṹ
ṹͼ12. xi(j),yi(j) Ϊ i  j Ԫwi Ϊ i ȨֵN(i) Ϊ i ڵ.


ͼ1ģṹ
ڵڶ㣬x2=w2y1,y2(i)=exp(-((x2(i)-a1(i))/b1(i))2).ڵ㣬 T ģ.ڵĲ㣬ǰ򴫵ʱ C ģ.򴫵ʱڵ㺯Ϊ bx4(i)=exp(-((y4(i)-a2(i))/b2(i))2),w4 ԪΪ01.ڵ㣬ڵ y ݣ򴫵ʱѵݴӽڵ Y .
3ѧϰ㷨
ѧϰΪѧϰѧϰ.ͼ1磬һһԵĿѧϰ㷨.ÿһѧϰһΣܻȡ w4,a1(i),a2(i),b1(i),b2(i).㷨Ϊ3
1CF(i,j)=0, i=1,2,,N(3), j=1,2,,N(4), k=0.
2k=k+1,
Ե k ѵݶԣֱ y3(i),bx4(j),i=1,2,N(3)j=1,2,N(4),k=0.
3 Dml FhFh(m)=max(bx4(j),Dml(n)=max(y3(i).
4 CF(n,m)=CF(n,m)+1.
5if kQto2otherwise to6.
6for i=1,2N(3) j=1, 2,N(4)if CF(i,j)ݦ thenw4(i,j)=1 therwise,w4(i,j)=0.
7.
УQ ΪѵݶԸӡ1 ΪŶֵN(3)N(4)ֱΪ㣬Ĳڵ.
Ϊ߿ƾȣԲ a1(i),a2(i),b1(i),b2(i) ߵ BP㷨.ϵͳʵƫṩѧϰ.
4ʵʱ
4.1ϵͳṹ
ƶΪؽҺѹ.ÿһؽһ FNN п.FNNĽṹΪ(2188191)Ϊƫ e 仯 e䣬Ϊy.Ĳһ㷨(PID)ר. PID ĿΪÿһؽڲ 1000 ѵ(e,e䡪y).Щݾǲѵ.
4.2ƹĻȡ
ؽڽλƴΪ 1.92.9V仯ΧΪ 1VؽڽǶȱ仯ΧΪ 090.˱Ľe,e ֱΪ-11ݣ-55ݣ D/A ΧΪ05V˽ y Ϊ05.ʼ󣬽ѵݴ. 1000ѵݶѧϰ󣬲Ȩ W4.ͨW2W3W4֪Ŀƹ61.
4.3Ҹٽ
ͬϵͳֱ FNN ơPID Ƽһ FUZZY ʵͬһĿ겢ܱȽ.ƵΪ 0.5HzֵΪ0.35v.ͼ2ͼ3ֱ±ۺϱ۵ʵӦ.پaf 1.


ͼ2±۸Ӧ


ͼ3ϱ۸Ӧ
1ʵܱȽ

ƷFNNFUZZYPID
Ƶ(Hz)0.50.250.50.250.50.25
ϱ af(%)3.401.905.764.683.572.28
±af(%)2.611.454.732.763.181.53

ӱ1ͼ2ͼ3 ԿFNN Ʊȴͳģƺ PID ƾиõЧ.
5
ǰ漸ڵķĿɵó¼ۣ
(1)ģɹʵ˻˵ʵʱƣоʵʿΪЧ.ڼٶȵ( 286)ѧϰʱ˵Ĳڵ㺯еĲøߵĻȡøõĽ.
(2)ģ繹ɵĿ񾭿ģиԽ.ѧϰģ߼֪ʶлؽһԶбؽõΪ㷺Ӧ.
ߵλ:㽭ѧ崫ƹصʵҡݡ310027
ο
1ݡJames J.Buckley,Yoichi Hayashi.Fuzzy neural networks:Asurvey.Fuzzy Sets and Systems, 1994,66:113.
2ݡοǿֽ.DIFNNһ͵ģ.96йļ(C2N2-96)386389.
3ݡοֽǣε.ģѧϰ㷨о.().
ոڡ1995-08-07
