Re: heuristic tuning algorithms

Andreas Nuernberger (andreas.nuernberger@cs.uni-magdeburg.de)
Mon, 2 Feb 1998 20:15:29 +0100 (MET)

Anita wrote:
> I wish to do a membership function tuning for a Mamdani fuzzy engine.
> MATLAB has a function (anfis) for tuning Sugeno models. It however
> doesnt support Mamdani model. It must be owing to the relative
> complexity the defuzzification process provides. Also, since I donot
> have a good training set for my problem, I want to know if there is
> literature on some heuristic training techniques which may minimize the
> need for training sets. Could anybody suggest me a source to get to
> this destination? Besides membership function training, has anybody come
> across a process of checking out bad rules in a system and disallowing
> the system to take in bad rules? I cannot think how one could identify
> whether any discrepency in result is due to bad rules or ill-fitted
> membership functions. Can anybody let me know what is the best source
> for me to look into these details?
>
> Thanks,
> Anita Nair
> Clemson University

Perhaps vou can use NEFCON for MATLAB/SIMULINK, which is available free of charge (http://fuzzy.cs.Uni-Magdeburg.de/nefcon/nef_mat.html).

NEFCON is a model for neuro-fuzzy control. The implementation of the NEFCON model under MATLAB/SIMULINK supports Mamdani-type fuzzy controllers.

For details see the html page mentioned above.

Andreas Nuernberger

-- 
Andreas Nuernberger                     
University of Magdeburg, Faculty of Computer Science          
Institute for Knowledge and Language Processing
Universitaetsplatz 2                     Phone : +49.391.67.11358
D-39106 Magdeburg, Germany               Fax   : +49.391.67.12018
WWW: http://fuzzy.cs.uni-magdeburg.de/~nuernb