Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news.harvard.edu!news2.near.net!MathWorks.Com!news.duke.edu!news-feed-1.peachnet.edu!gatech!howland.reston.ans.net!EU.net!uknet!warwick!nott-cs!lut.ac.uk!encbw
From: C.B.Wong@lut.ac.uk (CB Wong) 
Subject: Feedback data during its training session
Originator: encbw@hpc.lut.ac.uk
Sender: usenet@lut.ac.uk (Usenet-News)
Message-ID: <CvELL4.8FI@lut.ac.uk>
Date: Wed, 31 Aug 1994 14:29:28 GMT
Reply-To: C.B.Wong@lut.ac.uk (CB Wong)
Nntp-Posting-Host: hpc.lut.ac.uk
Organization: Loughborough University, UK.
Lines: 16

Hello everybody,

I would like to apply NN to find an inverse dynamics of a pneumatic servo system.  The aim is use the NN inverse model to assist the profile planning.  The network that I am going to employ with 4 inputs and 1 output ( this topology is extracted from a paper which is written by J Thibault and B Grandjean, Laval University).  The inputs and outputs are velocity profile information and briefly described as follows:
	inputs:  y(k), y(k+1), u(k-1), u(k-2)
	output:  u(k)
where y(.) and u(.) are output and input of the plant respectively.
As it can be seen from the above topology, part of its inputs are fed back from the output of the NN through a delay mechanism.  My question is,
	Where is the feedback data come from during its training session?  Should it same as its recall phase (come from the output of the NN), or from the real system data? Are there any difference between them, and why?  It could be a stupid question to most of the people who familiar modelling with NN, however, I am desperate to know its answer.  Any suggestions or comments are mostly welcome.

Thanks millions in advance

My e-mail address:C.B.Wong@lut.ac.uk

Bill
:wq

