Related Experiment Videos
Hybrid model building methodology using unsupervised fuzzy clustering and supervised neural networks
M Ronen1, Y Shabtai, H Guterman
1Unit of Biotechnology, Faculty of Engineering Sciences, Ben-Gurion University of the Negev, P.O.B. 653, Beer-Sheva 84105, Israel.
Biotechnology and Bioengineering
|January 12, 2002
Summary
This study introduces Hybrid Fuzzy Neural Networks (HFNN) for modeling new processes by combining fuzzy clustering and neural networks. HFNN offers improved learning capabilities and flexibility compared to single neural networks.
Area of Science:
- Computational modeling
- Process engineering
- Artificial intelligence
Background:
- Developing effective models for new processes is challenging.
- Existing methods may lack flexibility or simplicity.
- Hybrid approaches can potentially overcome these limitations.
Purpose of the Study:
- To propose a novel model building methodology called Hybrid Fuzzy Neural Networks (HFNN).
- To combine unsupervised fuzzy clustering with supervised neural networks for enhanced modeling.
- To evaluate the effectiveness of HFNN in process modeling and learning.
Main Methods:
- Utilized fuzzy clustering to define input space domains.
- Trained multilayer perceptrons (MLP) as sub-models for each domain.
- Fused sub-model outputs weighted by predicted possibilities.
- Employed on-line reinforcement learning for model improvement.
Main Results:
- HFNN successfully created simple and flexible models for new processes.
- The methodology demonstrated advantages over single neural network approaches.
- On-line reinforcement learning enhanced model performance.
- Limitations of validity measures for determining optimal clusters were identified.
Conclusions:
- Hybrid Fuzzy Neural Networks (HFNN) provide a robust and adaptable approach to process modeling.
- The method offers superior learning capabilities compared to traditional single neural networks.
- Careful selection of cluster validity measures is crucial for HFNN success.