27.03.2014 Views

SEKE 2012 Proceedings - Knowledge Systems Institute

SEKE 2012 Proceedings - Knowledge Systems Institute

SEKE 2012 Proceedings - Knowledge Systems Institute

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Modeling and Analysis of Switched Fuzzy <strong>Systems</strong><br />

Zuohua Ding, Jiaying Ma<br />

Lab of Scientific Computing and Software Engineering<br />

Zhejiang Sci-Tech University<br />

Hangzhou, Zhejiang 310018, P.R. China<br />

Abstract<br />

Switched fuzzy systems can be used to describe the hybrid<br />

systems with fuzziness. However, the languages to describe<br />

the switching logic and the fuzzy subsystems are in<br />

general different, and this difference makes the system analysis<br />

hard. In this paper we use Differential Petri Net (DPN)<br />

as a unified model to represent both the discrete logic and<br />

fuzzy dynamic processes. We then apply model checking<br />

technique to DPN to check the correctness of the requirements.<br />

1 Introduction<br />

Switched systems have been widely used in a variety<br />

of industrial processes. If a system is too complex or illdefined,<br />

i.e. the system contains fuzzyness, switched fuzzy<br />

system has been adopted to model such systems[1].<br />

Compared with conventional switched system modeling,<br />

switched fuzzy system modeling is essentially a multimodel<br />

approach in which simple sub-models (typically linear<br />

models) are fuzzily combined to describe the global behavior<br />

of a nonlinear system. A typical sub-model is Takagi<br />

and Sugeno (T-S) fuzzy model, which consists of a set of<br />

If-Then rules and a set of ordinary differential equations[5].<br />

This model raises some issues in the software engineering.<br />

For example, in the T-S fuzzy model, the languages to<br />

describe the switching logic and the fuzzy subsystems are in<br />

general different, and this difference makes the system analysis<br />

and implementation hard. So far, the efforts to address<br />

this issue are only for the deterministic switched systems.<br />

It is our attempt to solve this issue for switched fuzzy<br />

systems. We use Differential Petri net defined by Demongodin<br />

and Koussoulas[2] as a unified behavior model to represent<br />

switched fuzzy systems. We then use model checking<br />

tool HyTech to check the DPN, and use an enhanced<br />

version of the tool Visual Object Net++ to simulate DPN.<br />

We employ the the Differential-Drive Two-Wheeled Mobile<br />

Robots as the running example to illustrate our method.<br />

2 Switched Fuzzy <strong>Systems</strong><br />

A Switched Fuzzy System (SFS) consists of a family<br />

of T-S fuzzy models and a set of rules that orchestrates<br />

the switching among them. Assume that the Switched<br />

Fuzzy System has m subsystems, each is described by a<br />

Takagi-Sugeno fuzzy model[5], and σ : R + → M =<br />

{1, 2,...,m} is a piecewise constant function that represents<br />

the switching signal.<br />

2.1 Takagi-Sugeno Fuzzy Model<br />

Let N σ(t) be the number of inference rules. Then the T-S<br />

fuzzy model is described as follows:<br />

[Local Plant Rule]<br />

Rσ(t) l : if ξ 1 is Mσ(t)1 l ∧···∧ξ p is Mσ(t)p l ,<br />

then x ′ (t) =A σ(t)l x(t)+B σ(t)l u σ(t) (t),<br />

l =1, 2,...,N σ(t) .<br />

In this model, R l σ(t)<br />

denotes the l th inference<br />

rule, u σ(t) is the input variable, vector x(t) =<br />

[x 1 (t),x 2 (t),...,x n (t)] T ∈ R n represents the state<br />

variables, vector ξ =[ξ 1 ,ξ 2 ,...,ξ p ] represents the vector<br />

of rule antecedents (premises) variables, and matrices<br />

A σ(t)l ∈ R n×n and B σ(t)l ∈ R n×p .<br />

Hence, the i th subsystem can be represented as follows:<br />

⎧<br />

⎪⎨<br />

subsystem i :<br />

⎪⎩<br />

R 1 i :ifξ 1 is M 1 i1 ∧···∧ξ p is M 1 ip<br />

then x ′ (t) =A i1 x(t)+B i1 u i (t)<br />

R 2 i :ifξ 1 is M 2 i1 ∧···∧ξ p is M 2 ip<br />

then x ′ (t) =A i2 x(t)+B i2 u i (t)<br />

.<br />

R N i<br />

i<br />

:ifξ 1 is M N i<br />

i1 ∧···∧ξ p is M N i<br />

ip<br />

then x ′ (t) =A iNi x(t)+B iNi u i (t)<br />

By using the center of gravity method for defuzzification,<br />

the global model of the i th fuzzy subsystem can be de-<br />

135

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!