Design of fuzzy systems using neurofuzzy networks

Area of Science:

• Artificial Intelligence
• Computational Intelligence
• Machine Learning

Background:

• Fuzzy systems are crucial for modeling complex systems.
• Designing effective fuzzy systems requires systematic approaches for rule and membership function generation.
• Existing methods may lack comprehensive coverage or ease of rule extraction.

Purpose of the Study:

• To introduce a systematic approach for fuzzy system design using neural fuzzy networks.
• To develop a network capable of encoding knowledge as fuzzy rules and performing fuzzy reasoning.
• To evaluate the network's performance in function approximation, including handling noisy data.

Main Methods:

• A novel neural fuzzy network architecture based on a general neuron model.
• Encoding learned knowledge into if-then fuzzy rules and processing data via fuzzy principles.
• Generating rules and membership functions (including shapes) covering the entire input/output space.
• Extracting fuzzy rules in linguistic form post-learning.

Main Results:

• The neural fuzzy network demonstrates universal approximation capability.
• Effective rule extraction and membership function generation were achieved.
• The approach showed good accuracy and complexity in function approximation tasks, even with noisy data.
• Comparisons with alternative methods validated the proposed technique's performance.

Conclusions:

• The developed neural fuzzy network offers a systematic and effective method for fuzzy system design.
• The approach facilitates easy extraction of linguistic fuzzy rules.
• It provides a robust solution for function approximation problems with applications in modeling and control.