Re: Adapting fuzzy sets automatically for weather prediction...?

C. Gonzalez P. (carlagon@calima.univalle.edu.co)
Mon, 27 Oct 1997 02:35:06 +0100 (MET)

On Sun, 19 Oct 1997, Victor V. Miagkikh wrote:

> Dear Bjarne,
>
> There is a couple of papers at http://web.cps.msu.edu/~miagkikh/pub.html
> which are relevant to your problem.
>
> Best,
>
> Victor Miagkikh
>
I'm working in it.
My pregrade Thesis is about it, the problem: it's in spanish :(
There is the abstract:

RESUMEN

En este trabajo se presenta una soluciOn al problema de la optimizaciOn
del rendimiento de los controladores difusos usando redes neuronales
artificiales, el sistema propuesto, llamado ASR (Agente de SintonizaciOn
Neuronal por Refuerzo), tiene como tarea primordial la modificaciOn de los
parAmetros de inferencia de un controlador, utilizando una red de
elementos neuronales artificiales que introducen mEtodos pseudoaleatorios
para explorar el espacio de posibles soluciones, las cuales pueden no ser
evidentes en la etapa de disenno del controlador. Se comprobO la
eficiencia del mEtodo obtenido mediante su simulaciOn en software usando
el problema del pEndulo invertido, con entrenamiento tanto OFFLINE como
ONLINE en el agente sintonizador.

ABSTRACT

In this work, we show a solution to the performance optimization problem
of fuzzy controllers using neural networks. The proposed system, called
ASR (Neural Tuner Agent by Reinforcement ), is incharge of modifying the
inference's parameters of a controller, using a network of neuronlike
elements that introduces pseudorandom methods to explore the space of
possible solutions - these solutions might not be evident during the
design of the controller. We proofed the found method's eficiency through
software simulation, using the pole balancing problem, with both ONline
and OFFline training of the tuner agent.

Best regards.
Carlos González.