Re: learning algorithm for tuning fuzzy rules based on the gradient descent method

Hi again ...

I think what is what he want is this ...
ASCII art ... :)

  X   +--------------+   Y
      |              |
  -+->|  REAL THING  +--->--+
   |  |              |      |       
   |  +--------------+      |      
   |            _           |      
   |            /|         Diff >-+ Error
   |           /            |     |
   |  +--------------+  Yf  |     |
   |  |              |      |     |
   |->| FUZZY CTRLR  +--->--+     |
      |              |            |
      +--------------+           /
         /                      / 
        +-------------<--------/
 

So he wants to minimise Error = Y - Yf ... so
that he fuzzy system (FUZZY CTRLR) looks like
the real thing (REAL THING)

Regards

> The idea of a Google search is excellent.
> 
> The gradient method, like most optimization techniques, tries to find
> those values of a set of parameters that will minimize an error
> function, starting from a set of initial guesses and using first
> derivatives of the error function. In your case the parameters would
> be those defining a set of parameterized membership functions. The
> error term could be, for example, the sum of squares of differences
> between actual and desired outputs. (I don't understand why you say
> the output should equal the input. If this is the case, why would you
> need a controller at all?) For computation, you will need a model of
> the process being controlled to determine actual output. You should
> probably use initially a simple setpoint model.
> 
> I don't understand why you say the output should equal the input. If
> this is the case, why do you need a controller at all?
> 
> William Siler