Maximum Entropy Functions of Discrete Fuzzy Random Variables

Lianlong Gao, Liang Lin, Ling GaoJournal of Mathematics Research 2010 · DOI

Abstract

Due to deficiency of information, the probability distribution and membership functions of a fuzzy random variable cannot be obtained explicitly. It is a challenging work to find an appropriate probability distribution and membership function when certain partial information about a fuzzy random variable is given, such as expected value or moments. This paper solves such problems for the maximum entropy of discrete fuzzy random variables with certain constraints. A genetic algorithm is designed to solve the general maximum entropy model for discrete fuzzy random variables, which is illustrated by numerical experiment.

Introduction

In complex systems ranging from disaster risk analysis to signal processing, researchers frequently encounter hybrid uncertainty. This occurs when a variable is simultaneously random (the outcome is unknown) and fuzzy (the outcomes themselves lack precise definition). Such variables are modeled as Fuzzy Random Variables (FRVs).

A significant challenge in this field is the "deficiency of information." When only partial information—such as the expected value or variance—is available, an infinite number of probability distributions and membership functions could potentially fit the data. To address this, the authors apply the Maximum Entropy Principle (MEP). By selecting the distribution with the highest entropy that satisfies known constraints, the model ensures the most unbiased estimation possible, avoiding the introduction of unsupported assumptions.

Methodology

The Maximum Entropy Model

The paper formulates a constrained optimization problem to maximize the hybrid entropy $H[\xi]$ of a discrete FRV. This entropy measure is based on "chance measures," which integrate both the probability of an event and the credibility of its fuzzy membership. The objective function is subject to constraints such as the expected value $E[\xi] \leq e$, and the requirement that all probabilities and membership degrees remain within the $[0, 1]$ interval and sum appropriately.

Genetic Algorithm (GA) Optimization

Because the search space for FRVs is mathematically irregular and complex, traditional calculus-based methods like Euler-Lagrange equations are often inapplicable. The authors developed a specialized Genetic Algorithm to navigate this landscape. Key features include:

  • Chromosome Representation: Encoding both membership degrees ($\mu$) and probabilities ($p$).
  • Selection: A "roulette wheel" mechanism favoring chromosomes with higher entropy.
  • Genetic Operators: Tailored crossover and mutation processes designed to maintain the mathematical validity of the offspring relative to the model's constraints.

Numerical Experimentation

To validate the proposed framework, the authors conducted a synthetic numerical experiment. They defined a discrete FRV with six possible outcomes $\{1, 2, 3, 4, 5, 6\}$ and an expected value constraint of $E[\xi] \leq 3.5$. This setup allowed for a direct comparison between the algorithm's output and the theoretical maximum entropy.

Table 1: Problem setup and target maximum entropy for the discrete fuzzy random variable experiment.

Results and Discussion

The Genetic Algorithm demonstrated high efficacy in solving the maximum entropy model. Across various test runs, the algorithm consistently converged to a maximum entropy value of approximately 2.703, closely matching the theoretical target of 2.7034 shown in the experimental setup.

Table 2: Genetic Algorithm performance and stability across different population sizes (pop-size), mutation rates (Pm), and iteration counts (N).

Key findings from the results include:

  • Accuracy: The error between the GA-derived values and the optimal theoretical values did not exceed 2%.
  • Stability: As shown in Table 2, the results remained robust even when varying parameters such as population size and mutation rates.
  • Convergence: The GA successfully navigated the irregular optimization landscape, proving to be a practical necessity for handling the complexities of fuzzy variables.

Conclusion

This research provides a robust framework for estimating the behavior of discrete fuzzy random variables under conditions of limited information. By combining the Maximum Entropy Principle with a heuristic Genetic Algorithm, the authors offer a method that is both theoretically sound and practically applicable. Future work may explore higher-dimensional sets, more complex moment constraints, and real-world applications in economics and engineering.

Summary updated

Abstract

Due to deficiency of information, the probability distribution and membership functions of a fuzzy random variablecannot be obtained explicitly. It is a challenging work to find an appropriate probability distribution and membershipfunction when certain partial information about a fuzzy random variable is given, such as expected value or moments.This paper solves such problems for the maximum entropy of discrete fuzzy random variables with certain constraints. Agenetic algorithm is designed to solve the general maximum entropy model for discrete fuzzy random variables, which isillustrated by numerical experiment.
BibTeX
@article{gao2010maximum,
  author = {Lianlong Gao and Liang Lin and Ling Gao},
  title = {{Maximum Entropy Functions of Discrete Fuzzy Random Variables}},
  journal = {Journal of Mathematics Research},
  year = {2010},
  doi = {10.5539/jmr.v2n3p78},
}
Figure