Athenea Journal
Vol.4, Issue 14, (pp. 8-21)
Gonzalez L. Algorithm for Diffuse TSK Modeling of SNL MIMO with Undefined
Operation Points
I. INTRODUCTION
To control means to exert the actions necessary to produce a desired result, but to do so, the system to be
commanded must exhibit "reasonably predictable" behavior. In the specialized literature [1], it is proposed that the
systems to be controlled can be classified into two large groups: deterministic and non-deterministic. Classical control
algorithms are based on the hypothesis that the system to be controlled is deterministic, for which a series of restrictions
are applied to guarantee the functionality of the algorithm, ranging from restricting the work area to a small region
around a point of operation, to "disregarding" the probabilistic nature of the present and future state of the system.
There is also a presumption that the mathematical model of the system is time-invariant. However, these classical
control techniques work very well for various physical systems and have been successfully employed for over a century,
as stated in [2]: "Most physical systems contain complex non-linear relationships, which are difficult to model with
conventional techniques." That is why, in advanced process control, non-linear control techniques are used.
One of the ways to mathematically model the non-linear nature of systems is by using models based on fuzzy logic
systems. This theory is supported by fuzzy logic systems being universal approximators [2]. In particular, the fuzzy
system model developed by Takagi and Sugeno [3] and Sugeno and Kang [4], called the TSK fuzzy model in the
literature, is suitable for a broad class of non-linear systems because the consequent is a linear function or even a state-
space system. Interestingly, the TSK model allows the use of equations in State Spaces in the consequent, thus being
able to obtain a fuzzy model for a Non-linear System (SNL) of multiple inputs and outputs (MIMO), which allows the
application of modern control algorithms based on models in State Space such as Optimal Control, H∞, Genetic
Algorithms, Predictive Control.
The fuzzy model of an SNL MIMO is robust in applications where the plant has more than one operating point.
However, following conventional techniques, as the number of operation points increases, the fuzzy model increases
significantly in complexity since, in general, a linear subspace is generated for each operation point, that is, a rule in the
consequent. It has also been sufficiently studied that as the number of rules in the consequent increases, the fuzzy TSK
model exhibits a behavior closer and closer to the non-linear model of the system. Thus, there is a dilemma between
keeping the complexity of the fuzzy model low – few rules – and, at the same time, ensuring that it represents the
dynamics of the SNL as accurately as possible.
Now, imagine for a moment that a TSK model is required for an SNL MIMO whose operating points – within the
permissible physical range for the system – are not defined, as might be the case with the design of a fuzzy servo
controller for such a system. Undoubtedly, this last proposition introduces an additional level of complexity to the
previously posed dilemma between keeping the number of rules of the fuzzy model to a minimum and, at the same
time, representing the system as accurately as possible, the complexity of not having defined the points of operation.
Next, an algorithm is presented to solve the problem: synthesize the fuzzy TSK model of a MIMO SNL with undefined
operation points.
II. DEVELOPMENT
In the first instance, a synthesis of the theoretical foundations of the developed algorithm is presented, and then a
detailed description is given.
A. Takagi-Sugeno-Kang Fuzzy Model (TSK)
Takagi and Sugeno [3], and later Sugeno and Kang [4], developed the structure of a Fuzzy Model that has been
widely studied. They denoted the relevance function of a fuzzy set A as A(x), with x
X, and defined that all fuzzy sets
are associated with linear relevance functions, such that a relevance function is characterized by having two limit values:
1 for the highest degree of relevance and 0 for the lowest degree of significance. Thus, the Truth Value of a linguistic
proposition of the type "(x is A) Y (y is B)" is expressed as:
(1)
In addition, these researchers defined the format of a fuzzy R implication as:
(2)