>rbturne@hubcap.clemson.edu (Rob Turner) wrote:
>>I'm interested to know how you use linguistic inputs and
>>outputs to a fuzzy controller. If you want the output to be
>>a something like speed is HIGH instead of speed is 55
>>how do you turn the crisp output of the defuzzification
>>into a linguistic? Also how would you feed in a liguistic
>>input to the controller?
>The inputs and outputs of a fuzzy system are fuzzy and
>their representation can be linguistic if you want. If you
>want an output of "High" rather than the defuzzified version
>then don't do the defuzzification.
>Imagine the inputs and outputs to be groups of fuzzy
>descriptions of your variables, so an output temperature can
>take the values: Low, Medium, High. The output of the fuzzy
>system will be a confidence value for each of the possible
>labels, eg. Low 3%, Medium 27%, High 70%.
>You can then defuzzify by choosing the maximum, or find the
>centre of gravity of the output values or whatever.
>If you are happy with High being the most likely output, then
>there is no need to defuzzify.
Out of curiosity, how do you deal with the inputs? Turner wants to
input a "linguistic input" and get a response; how do you recommend
this be done? Seems to me if the input descriptors are (say) Weak,
Reasonable, and Robust, then only three outputs are possible if he's
going to input a single word, maybe six if he says more than one...
I'd think you need a way to describe the degree to which each of the
inputs holds; how does one do this without numbers? And if numbers be
unavoidable, why not just use a "crisp" measurement and fuzzifier?
I've heard a lot of talk about "linguistic" approaches to control, but
no-one seems to address this obvious hassle. It would be great if you
can clarify us on this.
Fred Watkins