Generalized Properties of Generalized Fuzzy sets GFScom and Its Application

Shengli Zhang, Jing ChenResearch Square 2022 · DOI

TL;DR

This paper expands fuzzy set theory by introducing GFScom, a framework that incorporates three distinct types of negation: contradictory, opposite, and medium. By leveraging these generalized negations, the authors demonstrate that Mamdani fuzzy systems can approximate complex continuous functions more efficiently, requiring significantly fewer fuzzy sets and rules than traditional methods to achieve the same level of accuracy.

What Problem Does This Paper Address?

Fuzzy logic is a powerful tool for handling "vague" or imprecise information, but it has historically been limited by a narrow view of negation. Most fuzzy systems rely on a single type of negation—usually the "contradictory" negation (e.g., if $x$ is the membership value, its negation is $1-x$).

However, human reasoning is more nuanced. There is a difference between "not tall" (contradictory) and "short" (opposite), and there is often a "middle ground" (medium) that negates both extremes. The authors argue that by failing to distinguish between these types of negative information, traditional fuzzy systems miss out on structural symmetries that could make them more efficient. Furthermore, designing "universal approximators"—fuzzy systems capable of mimicking any continuous function—often requires a massive number of rules and predefined fuzzy sets, which creates a "curse of dimensionality" for engineers.

Methods

The authors formally define Generalized Fuzzy Sets with Contradictory, Opposite, and Medium negation (GFScom).

  1. Three Types of Negation:
    • Contradictory ($\neg$): The standard linear complement.
  2. Opposite ($\dashv$): A "mirror image" of a concept across the domain (e.g., "Tall" vs. "Short").
  3. Medium ($\sim$): A state that negates both the concept and its opposite (the "middle" area).
  4. Property Analysis: They prove several mathematical properties of these sets, specifically focusing on convexity and concavity. For instance, they show that if a fuzzy set is concave, its medium negation is convex—a vital property for ensuring stable behavior in control systems.
  5. System Design: They propose two constructive approaches for Mamdani fuzzy systems:
  6. First-Order Accuracy: Using pseudo-trapezoid membership functions.
  7. Second-Order Accuracy: Using triangular membership functions to achieve higher precision with even fewer rules.
  8. Approximation Bounds: The paper derives rigorous error bounds for these systems, proving they are universal approximators.

Datasets

As a theoretical mathematics and engineering paper, the authors evaluate their model using synthetic functions and simulations rather than large-scale empirical datasets:

  • Case 1: Approximating the function $g(x) = \sin(x)/x$ on the interval $[-3, 3]$.
  • Case 2: Approximating the multi-variable function $g(x_1, x_2) = e^{x_1+x_2}$ on a 2D compact set.
  • Case 3 (Simulation): A nonlinear liquid level control problem using the sltank model in Matlab to test the GFScom controller's real-world applicability.

Key Insights and Ideas

  • Efficiency through Symmetry: The "aha" moment is that if you define one fuzzy set (like "Big"), you can mathematically derive its "Opposite" ("Small") and its "Medium Negation" ("Medium"). You don't need to manually define every linguistic variable; the negations fill in the gaps of the input space.
  • Fewer Rules, Same Accuracy: Because the negations provide a more complete "coverage" of the data space, the resulting Mamdani systems require a much smaller "rule base" to achieve universal approximation.
  • Convexity Links: The discovery that medium negations of concave sets are convex provides a theoretical bridge between different types of fuzzy set shapes, which is useful for optimization.

Key Results

The paper provides striking comparisons against established fuzzy approximation methods:

  • Rule Reduction: In Case 1 ($\sin(x)/x$ with error $\epsilon = 0.2$), the GFScom method required only 3 to 5 known fuzzy sets. In contrast, the classical method by Ying (1994) required 207 sets, and the method by Zeng and Singh (1996) required 9.
  • Exact Duplication: The authors proved that their second-order GFScom system can perfectly duplicate any affine or multiaffine function.
  • Nonlinear Control: In the liquid level simulation, the GFScom controller successfully managed a nonlinear plant using a "small and incomplete" rule base, proving that the system is robust even when data is sparse.

Limitations and Future Work

While the paper establishes a strong theoretical foundation, the authors note several areas for further exploration:

  • Logic Connectives: The detailed structure of how GFScom interacts with other logical connectives (like "AND" and "OR" operators specifically tuned for medium logic) needs more study.
  • Mapping Complexity: The current model does not yet provide a perfect one-to-one relationship between input and output fuzzy sets for all variable types.
  • Reasoning Structures: The authors suggest that future work should focus on building a more robust "logic structure" for GFScom to move beyond function approximation and into complex automated reasoning.

Summary updated

Abstract

<title>Abstract</title> We further continually develop the theory of a generalized fuzzy set with contradictory, opposite and medium negation (GFScom) . The generalized properties (including convexity and concavity) of GFScom are discussed. By introducing GFScom to Mamdani fuzzy systems, we propose new constructive approaches from any (infinite) input-output data pairs to approximate any continuous function on a compact set to a desired degree of accuracy, and investigate the approximation error bounds for the classes of the constructed fuzzy systems. Furthermore, the new better sufficient conditions for this class of fuzzy systems to be universal approximators are provided. For Mamdani fuzzy systems, the novel conditions require a smaller number of known fuzzy sets than all previously published classical conditions. Finally, we illustrate several cases and compare the new results to published error bounds through numerical cases.
BibTeX
@inproceedings{zhang2022generalized,
  author = {Shengli Zhang and Jing Chen},
  title = {{Generalized Properties of Generalized Fuzzy sets GFScom and Its Application}},
  booktitle = {Research Square},
  year = {2022},
  doi = {10.21203/rs.3.rs-1867796/v1},
}
Figure