Neural networks for Modelling and Control

Dear all,

Just to let you know of the http availability of a new technical
report entitled "Neural networks for Modelling and Control" (it is a
compressed file. Please, gunzip the file to view or print it). 

http://www.mech.gla.ac.uk/~ericr/pub/gmnn_rep.ps.gz

or

http://www.mech.gla.ac.uk/~yunli/reports.htm


The report has been written by Eric Ronco and Peter J. Gawthrop. The
keywords are: Neural Networks, Control, Modelling, Modularity.

Abstract:

This report is a review of the main neuro-control technologies. Two
main kinds of neuro-control approaches are distinguished. One
entails developing a single controller from a neural network and the
other one embeds a number of controllers inside a neural network.

The single neuro-control approaches are mainly system inverse: the
inverse of the system dynamics is used to control the system in an
open loop manner. The Multi-Layer Perceptron (MLP) is widely used for
this purpose although there is no guarantee that it can succeed in
learning to control the plant and that, more importantly, the unclear
representation it achieves prohibits the analysis of its learned
control properties. These problems and the fact that open loop control
is not suitable for many systems highly restricts the usefulness of
the MLP for control purposes. However, the non-linear modelling
capability of the MLP could be exploited to enhance model based
predictive control approaches since essentially, an accurate model of
the plant is all that is required to apply this method.

The second neuro-control approach can be seen as a modular approach
since different controllers are used for the control of different
components of the systems. The main modular neuro-controllers are
listed. They are all characterised by a ""gating system'' used to
select the the modular units (i.e. controllers or models) valid for
the computing of a current input pattern. These neural networks are
referred to as the Gated Modular Neural Networks (GMNNs). Two of these
networks are particularly fitted for modelling oriented control
purposes. They are the Local Model Network (LMN) and the Multiple
Switched Models (MSM). Since the local models of the plant are linear,
it is fairly easy to transform them into controllers. For the same
reason, the analysis of the properties of these networks can be easily
performed and it is straightforward to determine the parameter values
of the controllers as linear regression methods can be applied.  These
advantages among others related to a modular architecture reveal the
great potential of these GMNNs for the modelling and control of
non-linear systems.



Regards,

Eric Ronco
-----------------------------------------------------------------------------
|  Eric Ronco                                                               |
|  Dt of Mechanical Engineering     E.mail : ericr at mech.gla.ac.uk           |
|  James Watt Building              WWW : http://www.mech.gla.ac.uk/~ericr  |
|  Glasgow University               Tel : (44) (0)141 330 4370              |
|  Glasgow G12 8QQ                  Fax : (44) (0)141 330 4343              |
|  Scotland, UK                                                             |
-----------------------------------------------------------------------------