Related Experiment Videos
Neural networks for process control and optimization: two industrial applications
Gérard Bloch1, Thierry Denoeux
1Centre de Recherche en Automatique de Nancy (CRAN), UMR CNRS 7039, France. gerard.bloch@esstin.uhp-nancy.fr
ISA Transactions
|January 28, 2003
Summary
This study explores multilayer perceptron (MLP) and radial basis function networks (RBFN) for system identification and control. Neural networks effectively learn complex nonlinear phenomena in industrial applications like steel and water treatment.
Area of Science:
- Engineering
- Computer Science
- Control Systems
Background:
- Multilayer perceptron (MLP) and radial basis function networks (RBFN) are prominent neural models.
- System identification and control often involve complex nonlinear phenomena.
Purpose of the Study:
- To present MLP and RBFN within the framework of system identification and control.
- To highlight the advantages and applications of neural network models for nonlinear system control.
Main Methods:
- Building nonlinear black box models involves regressor choice, architecture selection, and parameter estimation.
- Neural network models offer universal approximation, flexibility, and parsimony.
Main Results:
- Neural network techniques are effective for controlling complex nonlinear processes.
- Applications in steel industry (alloying process) and water treatment (coagulation process) demonstrate practical utility.
- Empirical knowledge of control operators can be learned by neural networks.
Conclusions:
- Neural network models, specifically MLP and RBFN, are valuable tools for system identification and control.
- These models successfully address complex nonlinear phenomena in industrial settings.
- The ability to learn operator knowledge enhances control system capabilities.