| Summary: | 碩士 === 國立臺灣大學 === 化學工程學系研究所 === 86 === A simple method for generating a multiple-input/single-outputcrisp-type fuzzy
model from observed data is presented in this article.The crisp-type fuzzy mod
el is composed of triangular fuzzy partitionsfor inputs and singletons for the
output. Three steps are proposedin this identification method: initializing
phase, growing phase,and refining phase. For a set of observed input/output d
ata pairs,the structure and the parameters of a suitable fuzzy model withrequi
red accuracy can be determined by a series of algebraiccomputations, and nonli
near programming procedure is not needed.Numerical examples are given to demon
strate the effectiveness ofthe proposed identification method for modeling sta
tic and dynamicdata. The same procedure is also extended to the identificatio
n ofTakagi-Sugeno-Kang(TSK) fuzzy model, where some kinds of linear modelare u
sed in the consequence part of each fuzzy rule. By introducingthe least squar
es algorithm, the TSK modeling is also very easy to implement.Because the fuzz
y dynamic model can be simplified as a TSK fuzzy model, the sameprocedure can
be used to construct a fuzzy dynamic model.The other focus of this article is
how to identify a multiplelinear dynamic model. This type of models is an app
lication of theTSK dynamic model. Besides the proposed identification method,
twoalternative methods are also suggested: linearization of thenon-linear mat
hematical model at several operating points andidentifying a linear model for
each operating sub-regime.These methods are demonstrated by three chemical pro
cesses:first-order exothermic non-isothermal CSTR, high puritydistillation col
umn and neutralization process. A controlsystem design based on the multiple
linear model is also proposed.The control system of nonlinear processes design
ed based the themultiple linear model may be simpler.
|
|---|
