In recent years there has been much effort to develop the theoretical aspects of neural MRAC-control, that is to find conditions under which an unknown process can be identified by an input-output model and controllers can be trained by gradient descent. On the other hand, the application of neural network techniques to real world control of nonlinear dynamical systems has been of substantial interest. Since the theoretical conditions that ensure controllability and the applicability of indirect adaptive control are hard to verify in practice, the success of controller training is mostly shown by testing relevant situations. We trained a controller for a subsystem of a spark ignition engine by dynamic backpropagation and various truncated gradient algorithms. Afterwards we related the neural MRAC-approach to pole placement and linearization techniques in order to show the successful training by pole analysis of the completely trained loop. This is a new method to verify the plausibility of the adaptation process and the trained regulator.
[1]
Lee A. Feldkamp,et al.
Dynamic neural network methods applied to on-vehicle idle speed control
,
1996,
Proc. IEEE.
[2]
Martin A. Riedmiller,et al.
A direct adaptive method for faster backpropagation learning: the RPROP algorithm
,
1993,
IEEE International Conference on Neural Networks.
[3]
Kumpati S. Narendra,et al.
Neural networks and dynamical systems
,
1992,
Int. J. Approx. Reason..
[4]
Ronald J. Williams,et al.
Experimental Analysis of the Real-time Recurrent Learning Algorithm
,
1989
.
[5]
Snehasis Mukhopadhyay,et al.
Adaptive control of nonlinear multivariable systems using neural networks
,
1993,
Proceedings of 32nd IEEE Conference on Decision and Control.
[6]
W. Baumann,et al.
Feedback control of nonlinear systems by extended linearization
,
1986
.
[7]
Marco Saerens,et al.
Neurocontrollers based on backpropagation algorithm
,
1995
.