Evolving compact and interpretable Takagi-Sugeno fuzzy models with a new encoding scheme.

Area of Science:

• Computational Intelligence
• Fuzzy Systems Engineering
• Machine Learning

Background:

• Developing Takagi-Sugeno fuzzy models using evolutionary algorithms necessitates careful design of encoding schemes, evaluation methods, and evolutionary operations.
• Key considerations for fuzzy modeling include accuracy, compactness, and interpretability.

Purpose of the Study:

• To propose a novel evolutionary algorithm for Takagi-Sugeno fuzzy modeling that optimizes accuracy, compactness, and interpretability.
• To introduce a new encoding scheme and a fitness function to address these requirements.

Main Methods:

• A new encoding scheme with three chromosomes, including a unique chained possibilistic representation for rule structure.
• A novel fitness function comprising five components to evaluate accuracy, compactness, and interpretability simultaneously.
• Development of specialized evolutionary operators tailored to the proposed encoding scheme.

Main Results:

• Simulation results demonstrate the effectiveness of the proposed encoding scheme and fitness function in generating accurate, compact, and interpretable Takagi-Sugeno fuzzy models.
• The algorithm successfully approximates unknown functions with a reduced number of rules and membership functions.
• Obtained fuzzy models exhibit interpretable antecedent membership functions, aiding in understanding underlying system behavior.

Conclusions:

• The proposed evolutionary algorithm effectively addresses the challenges in Takagi-Sugeno fuzzy model development.
• The novel encoding and fitness function facilitate the creation of high-quality fuzzy models balancing accuracy, compactness, and interpretability.
• This approach enhances the practical applicability of fuzzy models in complex system analysis.