Documente Academic
Documente Profesional
Documente Cultură
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/255652583
CITATIONS
READS
10
119
3 authors, including:
Salem Mohammed
Mascara Universit
Mascara Universit
15 PUBLICATIONS 18 CITATIONS
6 PUBLICATIONS 21 CITATIONS
SEE PROFILE
SEE PROFILE
I.
INTRODUCTION
systme
(k
)
y(k)
u(k)
+
en
-
y(k+1)
(k)
r(k)
y(k)
ec
y(k-n+1)
(k)
+
-
u(k-1)
u(k-m+1)
Fig. 1. Rseau de neurone pour identifier le modle inverse
y(k)
Systme
ec2 ( k )
2
.
w ( k ) = w ( k 1)
w
(7)
e c2 ( k )
2
= e ( k ) ( y ( k ) ) = e ( k ) ( y ( k ) ) ( u ( k 1) )
c
c
w
w
( u ( k 1))
w
(k)
Systme
. (8)
y(k)
ec
III. COMMANDE
Systme
y(k
)
r(k)
u(k)
r(k+1)
NEURONALE
ec
r
+
Contrleur
uc
+
u
Systme
IV. EXPERIMENTATION
Pour les simulations, on a choisi un systme minimum de
phase, stable en boucle ouverte et possdant une non-linarit.
Le systme est compos dune masse, un amortisseur et un
ressort rigide [28]. Lquation du mouvement est donne
par (9):
y '' (t ) + y ' (t ) + y (t ) + y 3 (t ) = u (t ) .
(9)
-4
x 10
E
rreurdeprdictione(k)
3
2
1
0
-1
-2
-3
-4
-5
200
400
600
800
1000
1200
temps d'echant k
1400
1600
1800
2000
C
onsignealatoireyd(k)
0.2
0
-0.2
-0.4
-0.6
-0.8
-1
200
400
600
800
1000
1200
temps d'echant k
1400
1600
1800
2000
1600
1800
2000
C
onsignesinusoidaleyd1(k)
0.5
0
-0.5
-1
-1.5
-2
200
400
600
800
1000
1200
Temps d'echant k
1400
5
4
S
ignal entreu(k)
2
1
0
-1
-5
-2
x 10
-3
0.8
-4
100
200
300
400
500
600
Temps d'echant k
700
800
900
1000
0.6
E
rreurdecom
m
andeinverseeyd(k)
-5
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8
S
ignal sortiey(k)
-1
0
200
400
600
0.5
800
1000
1200
Temps d'echant k
1400
1600
1800
2000
-4
-0.5
1
x 10
-1.5
200
400
600
800
1000
1200
Temps d'echant k
1400
1600
1800
2000
E
rreurdecom
m
andeinverseeyd1(k)
-1
-1
0
200
400
600
800
1000
1200
Temps d'echant k
1400
1600
1800
2000
x 10
1.5
E
rreurdecom
m
andehybrideeyd(k)
1
0.5
0
-0.5
-1
-1.5
-2
-2.5
-3
200
400
600
800
1000
1200
temps d'echant k
1400
1600
1800
2000
x 10
E
rreurdecom
m
andehybrideeyd1(k)
0.8
0.6
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8
-1
200
400
600
800
1000
1200
Temps d'echant k
1400
1600
1800
2000
B
ruit blancb(k)
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8
-1
0
100
200
300
400
500
600
Temps d'echant k
700
800
900
1000
E
rreurdecom
m
andehybridebruiteyd(k)
x 10
6
4
2
0
-2
-4
-6
-8
200
400
600
800
1000
1200
Temps d'echant k
1400
1600
1800
2000
Fig. 17. Erreur commise lors de la commande hybride avec bruit (Consigne yd)
-3
x 10
E
rreurdecom
m
andehybridebruiteyd1(k)
8
6
4
2
0
-2
-4
-6
0
200
400
600
800
1000
1200
Temps d'echant k
1400
1600
1800
2000
Fig. 18. Erreur commise lors de la commande hybride avec bruit (Consigne
yd1)
V. CONCLUSION
Nous avons prsent, dans ce papier, deux structures de
commande par rseaux de neurones bases sur le modle
neuronal inverse, lapprentissage de ce modle a t fait aprs
plusieurs tentatives pour arriver une architecture optimale et
minimiser le nombre de paramtres du modle. Larchitecture
retenue tait avec une couche cache de 7 neurones. En
utilisant lalgorithme de Levenberg-Marquardt, Lerreur de
prdiction obtenue est trs rduite. Le modle retenu a t
utilis ensuite dans une structure de commande directe en
boucle ouverte mais ceci suppose que le modle inverse est
presque parfait et le systme est non bruit, ce qui est loin de la
ralit. Pour essayer dy remdier, nous avons utilis le modle
inverse dans une structure hybride avec un PID classique, et
nous sommes arrivs des rsultats trs concluants. Pour
valider la structure hybride, nous avons inject un bruit dans le
modle. Ceci na pas trop affect le systme command.
Les rsultats obtenus montrent quun modle neuronal avec
un bon apprentissage, sadapte parfaitement la dynamique
dun systme avec rejet des perturbations.
REFERENCES
[1]. J.Panos, K.Antsaklis and M.Passino, Introduction to Intelligent and
Autonomous Control, Kluwer Academic Publishers, ISBN: 0-7923-92671, 1993.
[2]. M.T.Hagan and H.B.Demuth, Neural network design,Thomson Asia Pte
Ltd,2nd ed 1996.
[3]. G. Dreyfus, Rseaux de neurones : mthodologies et applications, ditions
Eyrolles, 2002.
[4]. I. Rivals, L. Personnaz and G. Dreyfus, Modlisation, classification et
commande, Par rseaux de neurones : principes fondamentaux,
Mthodologie de conception et illustrations industrielles, 1995.
[5]. L. Ljung, System Identification; Theory for the User, Prentice Hall, 1987.
[6]. T. Yabuta and T. Yamada, Possibility of neural networks controller for
robot manipulators, Robotics and Automation, Proceedings, IEEE
International Conference, vol.3, 1990, pp. 1686-1691
[7]. G. Karsai, learning to control, some practical experiments with neural
networks, In International Joint Conference on Neural Networks, vol.
II, Seattle, 1991, pp. 701-707
[8]. B. Widrow, Adaptive inverse control, In Proceedings of the IFAC
Adaptive Systems in Control and Signal Processing Conference, Lund,
1988, pp.1-5
[9]. C. Goutte and C. Ledoux, Synthse des techniques de commande
connexionniste, IBP-Laforia 1995/02: Rapport de Recherche Laforia,
1995.
[10]. G. Cybenko, Approximation by superposition of a sigmoidal function,
Math. In: Control Signals System 2nd ed, 1989, pp.303-314.
[11]. H. Gomi and M. Kawato, Neural network control for a closed loop
system using feedback error learning, Neural Networks, vol.5, 1993,
pp.933-946.
[12]. H. Bleuler, D. Diez, G. Lauber, U. Meyer and D. Zlatnik, Nonlinear
neural network control with application example, In International
Neural Networks Conference, vol.1, Paris, 1990, pp.201-204
[13]. C.C. Jorgensen and C.Schley, A neural network baseline problem for
control of aircraft flare and touchdown, Neural networks for Control,
1990, pp.423-425
[14]. A. Guez and I. Bar-Kana, Two degree of freedom robot adaptative
controller, In American Conference on Control, San Diego, 1990, pp.
3001-3006.
[15]. J. Steck, B. Krishnamurthy, B. McMillin and G. Leininger, Neural
modeling and control of a distillation column, In International Joint
Conference on Neural Networks, vol.II, Seattle, 1991, pp. 771-773.
[16]. G.M. Scott, J.W. Shavlik and J.W.Harmon Ray, Refining PID
controllers using neural networks, NIPS 4, Advances in Neural
Information Processing Systems, 1991, pp. 555-562
[17]. D. Psaltis, A.Sideris and A. Yamamura, Neural controllers, In
International Neural Networks Conference, vol.IV, 1987, pp.551-558
[18]. E. Ronco and, P.J. Gawthrop, Neural networks for modelling and
control, Techncal Report CSC-97008, Center for Systems and Control,
Glasgow, 1997.
[19]. P. Tai, H.A. Ryaciotaki-Boussalis and T. Kim, the application of neural
networks to control systems: a survey, Signals, systems and Computers,
Record Twenty-Fourth Asilomar Conference on Vol.1, 1990.
[20]. S. Yildirim, New neural networks for adaptive control of robot
manipulators, Neural Networks, International Conference, vol.3, 1997,
pp.1727 - 1731
[21]. P. Brantetter and M. Skotnica, Application of artificial neural network
for speed control of asynchronous motor with vector control,
Proceedings of International Conference of Koice, EPE-PEMC, (2000,
pp. 6-157-6-159
[22]. G.A. Pugh, Ship directional control by synthetic neural network, Proc
of the American Control conference ACC90, San Diego, 1990, pp.30283029
[23]. Y. Ichikawa and T.Sawa, neural network application for direct
feedforward controller, IEEE Trans on Neural Networks, vol.3 No2,
1992, pp.224-231.
[24]. M.T. Wishart and R.G. Harley, Identification and Control of Induction
Machines Using Artificial Neural Networks. IEEE Transaction on
Industry Applications, Vol. 31, No 3, 1995.
[25]. L. Yan and C.J. Li, Robot Learning Control Based on Recurrent Neural
Network Inverse Model, Journal. of Robotic Systems, Vol. 14, 1997,
pp.199-212.
[26]. G.W. Ng, Application of neural networks to adaptive control of
nonlinear systems, Research Studies Press, Taunton, Somerset,
England, (1997.
[27]. C.T. Chen, W.D. Chang and J. Hwu, Direct control of nonlinear
dynamical systems using an adaptive single neuron, IEEE Trans on
Neural Networks, vol.2, No 10, 1997, pp.33-40
[28]. M. Norgaard, System identification and control with neural networks,
Thesis, Institute of automation, Technical University of Denmark, 1996.