ѧ 
JOURNAL OF ASTRONAUTICS
1998 19 1 No1 Vol.9 1998Ƽڿ

²½ķԪ*

(ҵѧԶϵ**100022)
ժҪԵ²½̵Ŀ,һַԶ̬״̬ϵԪϵͳƷ÷:(1)ǰԪͨѧϰƽӳ,ϵͳķԶ̬Ԫģ,ԪʵֱطϵͳԻ; (2)Իģ͵ĻϹϵͳԪ״̬ĸķʾ񾭼ѧںеǱ
ʡ²ա½ϵͳԪơ̬桡״̬
A NONLINEAR NEUROCONTROL SCHEME FOR 
LUNAR SOFT LANDING
Ruan Xiaogang
(Automation Department,Beijing Polytechnic University*Beijing*100022)
AbstractA neurocontrol scheme for lunar soft landing is proposed in this paper,which combines nonlinear dynamic inversion with optimal state feedback.The scheme mainly consists of two parts.First,the nonlinear dynamic inversion of the controlled object is modeled with an artificial neural network by means of its ability to learn to approximate any functions,and therefore,the controlled object is linearized by the neural inversion model.Secondly,based on the linearized system another artificial neural network is used as a controller to realize certain optimal state feedback controllaw.Finally,the effectiveness of the scheme described in this paper is investigated by computer simulation.The simulation results are encouraging and show that neurocomputation could play important role in control of the future spaceships.
Key wordsLunar modulesSoft landingNonlinear systemsNeurocontrolDynamic inversionState feedback
1
Ϊѧϵͳ,ֳǿԡԡģͲȷģδ֪ȸ,Ҳ˶÷ǳѡ,񾭼ѧķչ,Ԫ(Neurocontrol)Ϊ˵ܿƼʼ뺽պ1-2ݡ˹Ԫеѧϰ֯,ΪԡģͲȷģδ֪ϵͳṩһµ;
Ե²½̿,һַԶ̬״̬ϵԪϵͳƷ÷:(1)ǰԪͨѧϰƽӳ,ϵͳķԶ̬Ԫģ,ԪʵֱطϵͳԻ;(2)Իģ͵ĻϹϵͳԪ״̬Զ̬ƼǴԿһЧ,÷Աƶѧģ͵Ĺ,ʹӦܴܵơӦԪʵֱطϵͳĶ̬ģ,ʹ÷ھдͬʱ,˴ģͲȷģδ֪;ϵͳԻʹף1Ԫ״̬Ʒʵ֡Ե²½̵ĿΪӦñ,Աķ˷о,ʾ,ԪƾдǱ
2²ģͼ½
2.1ϵͳѧ
Ǿ·ͷ΢ַѧģ͵ĺ:
(2-1a)
y(t)=Cx(t)x0=x(t0)(2-1b)
:u(t)Ϊ,y(t)Ϊϵͳ˶켣,xRn,ARn1,BRn1,CR1n
2.2


ͼ1²ģͼ˶
ϵͳ˶켣Ϊyd(t),ⶨΪ:Ѱĳֿu(t),ʹ
(2-2)
ڼС
ſ(2-2)潫ֽΪJ(1)J(2)ſ⡣
2.3²½̿
2.3.1²ģͼ˶
²½ǵжҪ֮һ½,²˶Ǵֱ(ͼ1ʾ),:
1)ڵ²յֻƶõF,
2)ƽϵͳ:
²ն̬·΢ַ:
(2-3)
ʽ(2-5)Ϊʽ(2-1)ʾķ״̬΢ַ:
x(t)=x1(t)
(2-4)
,y=x1Ϊ²վĸ߶,Ϊհ11ŷɴ½ηݣ3-4,:
1)ٶg=1.62m/s2
2)ϵK=8000
3)ֵx0=x1(0)x2(0)x3(0)T=156m4.9m/s15000kgT(t0=0)
2.3.2²½½̵ĿĿ
²յΪ:
yd(t)=y(0)(1-e-0.0314t)(2-5)
һʹ²½ٶȳָ˥Ĺ,Ա֤²հȫ½²վĸ߶ʱ,䴹ֱٶҲ㡣
3ԪĿƷ
3.1ԪƵۻ
3.1.1BPԪģ
ĲõԪΪͼ2ʾĶ-BP硣

(a)(b)
ͼ2BPԪģ
3.1.2BPԪ㷨
(1)ǰ򴫵㷨
NkjΪkjԪ,OkjΪNkj,kjΪNkjֵ,wkjiΪNkiNk+1jȨϵ,LkΪkĽڵ,:
(3-1a)
(3-1b)
(2)ش㷨(BPѧϰ㷨)
BPS={(R(s),O(s)d)s=1,2,,N}O(s)ΪBPR(s)=(r1,r2,rL0)µʵBPΪ:
O(s)=O(s)d-O(s)(3-2)

E=Ni=1O(s)2(3-3)
ݶ½ԭ,BPѧϰ㷨:
(3-4)
3.1.3BPԪƽӳ
3-1:(Hecht-Nielsen 19895)
ţ0L2f:0,1nRnRm,һBPԪ,ӳh:0,1ݪRnRmƽ־ڱƽf,:
ң0,1nf(x)-h(x)2dx(3-5)
3-1ζBP˹бƽӳ
3.2²յԪϵͳṹ
ͼ3ʾĵ²˹ԪԶ̬ϵͳ


ͼ3²Ԫϵͳṹ
÷:
(1)ԪANN1ʵϵͳķԶ̬漰Ի
BP˹ƽӳ,Ƶ²յķԶ̬ԪģANN1,ͨANN1ϵͳĻʵԻ,γĳαϵͳ
ANN1ƲΪ:
1)αϵͳy(r),ö̬ѧϰ㷨ANN1u
2)ANN1u,BPѧϰ㷨֯,ʹANN1ƽϵͳķԶ̬ģ
(2)ԪANN2ʵϵͳ״̬
Իĵ²αϵͳ⻷״̬,BP˹Ԫƽӳ,ԪANN2ʵĳŵ״̬ɡ
ANN2ƲΪ:
1)ݷ(2-2)ͷѧϰ㷨ANN2v
2)ANN2v,BPѧϰ㷨֯,ʹANN2ƽʹ(2-2)ڼСµ״̬:v(*)=v(x,yd)
㷨ں
4ϵͳԶ̬ģ͵Ԫʵ
4.1ϵͳķԶ̬ѧģ
4-1:ϵͳ(2-1),
u(t)=u(x,v)(4-1)
θv(t)R,ʹ
(4-2)
ʽ(4-1)Ϊϵͳ(2-1)ķԶ̬,ʽ(4-2)Ϊϵͳ(2-1)αϵͳ,ϵͳ(2-1)Իαϵͳ(4-2)
:
(4-3)
4-1:xXR,r,:
(4-4)
ϵͳ(2-1)XڿɷԶ̬
u(t)=G(x)+Q(x)v(t)(4-5)
Իαϵͳ
Dry(t)=v(t)(4-6)
֤:ʽ(4-2,3)ɵ
Dry(t)=a(r)(x)+b(r)(x)u(t)(4-7)
Q(x)={b(r)(x)}-1,G(x)={b(r)(x)}-1a(r)(x),ʽ(4-5,6)
(֤)
4.2Զ̬ģ͵Ԫʵ
4.2.1̬ANN1ģ
ϵͳ(2-1)ȷѧģδ֪,侫ȷģͱضҲδ֪ġģͼģ͵ĽṹʽΪ֪ġʱ,˹Ԫʵϵͳģ;ͳΪһֱҪĺ;
ʽ(4-4)֪,ϵͳ(2-1)Ķ̬ǹ״̬xv(t)һַӳ䡣3-1֪,ӳ(4-5)һBPӳƽ,ϵͳ(2-1)Զ̬ģͿBPԪʵ֡ʽ(4-5)ΪRn+1Rӳ䡣,ANN1Ҫn+1Ԫ,Ҫ1ԪANN1R=(v,xT)T,Ϊu
4.2.2ANN1ѵ
ͼ3(3-3)BP㷨(3-4)֪,ѵANN1ĹؼǼANN1u(t)
(1)ANN1ѵ
ϵͳ(2-1)ʵлȡNֵ(ud(s),x(s),y(r)(s))(s=1,2,N),ɹ㼯
S={(R(s),ud(s))R(s)=(Dry(s),xT(s))T,s=1,2,,N}(4-8)
㼯S(y(r),x)Ϊ,uΪ,Ӧϵͳ(2-1),ANN1ѵu(s)ΪANN1R(s)=(y(r)(s),x(s))TµʵѵANN1:
u(s)=ud(s)-u(s)(s=1,2,,N)(4-9)
ANN1(4-9)BP㷨(3-4)ɶANN1ѵ
(2)ANN1ѵ
4-2:αϵ(4-6),ж̬ѧϰ㷨
(4-10)
㹻Сʱ,
(4-11)
kʱڼС
֤:u=lim0u(t)(4-12)

(4-13)
ʵ㹻Сʱ,
J(1)=J(1)k+1-J(1)k0(4-14)

J(1)k+1J(1)k(4-15)
(֤)
ʽ(4-10)ϵͳ(2-1)kʱANN1uk(t),BP㷨(3-4)ANN1ѵ
5Ԫ״̬ѵ
αϵͳ(4-6)巺
(5-1)
״̬
v(t)=v(yd,z)(5-2)
,״̬zRrΪ:zi=Diy(i=0,1,,r-1)
3-1֪,״̬(5-2)BPԪʵ,ͼ3ʾģANN2
ANN1ѵ,ѵANN2ĹؼǼANN2v(t)
5-1:״̬ƹѧϰ㷨
(5-3)
㹻С,kʱ,J(2)ڼС
֤:v=lim0v(t)(5-4)
򡡡(5-5)
ʵ㹻Сʱ,
J(2)=J(2)k+1-J(2)k0(5-6)

J(2)k+1J(2)k(5-7)
(֤)
Ȼ,y/vαϵͳ(4-6)ĵλԾӦʵݻȡ
ʽ(5-3)ϵͳ(2-1)kʱANN2vk(t),BP㷨(3-4)ANN2ѵݶ5-1,kʱ,J(2)(v)ڼС,ͨѵ,ANN2γ״̬(5-2)
Ԫ״̬ϵͳʵϢ,ͨѧϰƽʵϵͳ״̬,ԱѧģԼANN1ģԻȡ״̬ɿɵӰ졣
6о
6.1²½ϵͳеԪ硡


ͼ4½ϵͳе˹Ԫ
ͼ3ʾϵͳṹƵ²½Ԫϵͳ2.3ĵ²ģ,²շԶ̬ԪģANN1ͼ4(a)ʾ,ڵΪ30;Ԫ״̬ANN2ͼ4(b)ʾ,ڵΪ20
6.2²½̵ķ
6.2.1żʹ
ϵͳķй,45ѵ,ANN1ANN2зѵ²½̵ķͼ5ʾͼ5֪,35߷ѵ,²½̱ƽ½,ԪϵͳõĿЧ


ͼ5²½


ͼ6²շԶ̬
6.2.2Զ̬
²½Ԫϵͳʵ̰ANN1γɵ²ն̬ѵ,ͼ6ʾ,1ǵ²վȷĶ̬,2ANN1Ե²ն̬Ԥֵͼ6֪,35߷ѵ,ANN1γ˽Ϊȷĵ²շԶ̬ӳ䡣
ϵͳ֮ǰ,ҪȶANN1ANN2һ̶ȵѵʵ,ѵˮƽԽ,ѵʱԽ١
7
ĵĹǶ˹ԪӦڿռпһ̽ƽӳڵ,ԼԿϵͳ,رϵͳȶԺ³ԿܲӰ컹ȱķ,дһо
ο
  1.Ԫ״̬ƼڵӦо.Ͼպѧѧ,1994,26(6):721-728
  2Troudet T,et al.,Neural Networks Application to Aircraft Control System Design.Proceedings of AIAA Guidance,Navigation and Control Conference,1991:993-1009
  3ϣ.뷵ؼ.,1991:148-149
  4Folyd V B,Lunar descent and ascent trajectories.AIAA Paper No.70-25
  5Hecht-Nielsen R.Theory of the Back-Propagation Neural Net work.Proceedings of IJCNN,1989:593-598
