HyFIS: adaptive neuro-fuzzy inference systems and their application to nonlinear dynamical systems

Area of Science:

• Artificial Intelligence
• Computational Intelligence
• Machine Learning

Background:

• Fuzzy logic systems offer interpretability but often require manual tuning.
• Neural networks provide powerful learning capabilities but can lack transparency.
• Integrating these approaches can leverage the strengths of both.

Purpose of the Study:

• To propose an adaptive neuro-fuzzy system, HyFIS (Hybrid neural Fuzzy Inference System), for building and optimizing fuzzy models.
• To enhance fuzzy logic systems with the learning power of neural networks, providing linguistic meaning to connectionist architectures.
• To demonstrate the superior performance of HyFIS for nonlinear complex dynamic systems.

Main Methods:

• A hybrid learning scheme with two phases: rule generation from data and rule tuning via error backpropagation.
• Optimal tuning of heuristic fuzzy logic rules and input-output fuzzy membership functions from training examples.
• Extensive simulation studies on nonlinear complex dynamic systems.

Main Results:

• HyFIS effectively builds and optimizes fuzzy models by integrating neural network learning.
• The system optimally tunes fuzzy rules and membership functions using a hybrid learning approach.
• Simulation studies confirm HyFIS as a superior neuro-fuzzy modeling technique for complex systems.

Conclusions:

• HyFIS offers a powerful and adaptive neuro-fuzzy modeling technique.
• The system enables on-line incremental adaptive learning for prediction and control of nonlinear dynamical systems.
• HyFIS demonstrates superior performance compared to existing neuro-fuzzy methods in benchmark case studies.