Self-Organizing Robust Fuzzy Neural Network for Nonlinear System Modeling.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Control Systems

Background:

• Fuzzy neural networks (FNNs) are effective for nonlinear system modeling.
• Uncertain external disturbances degrade FNN performance in real-world applications.
• Existing FNNs struggle with noise, model errors, and unknown environments.

Purpose of the Study:

• To develop a self-organizing robust fuzzy neural network (SOR-FNN) that enhances nonlinear system modeling.
• To improve the adaptability and robustness of FNNs against external disturbances.
• To ensure reliable performance in uncertain and noisy environments.

Main Methods:

• Introduced an information integration mechanism (IIM) for dynamic structure adjustment.
• Designed a dynamic learning algorithm based on the -divergence loss function (-DLA) for parameter updates.
• Utilized Lyapunov theorem for theoretical convergence analysis of SOR-FNN.

Main Results:

• The SOR-FNN demonstrated adaptability to uncertain environments via IIM.
• The -DLA effectively reduced sensitivity to disturbances, enhancing robustness.
• Convergence analysis confirmed the successful application of SOR-FNN.
• Experimental validation on benchmark datasets and a practical application showed superior performance.

Conclusions:

• The proposed SOR-FNN significantly improves nonlinear system modeling accuracy and robustness.
• SOR-FNN offers a reliable solution for applications susceptible to external disturbances.
• The integration of IIM and -DLA provides a robust framework for adaptive learning.