Application of neural nets to system identification and bifurcation analysis of real world experimental data
- Princeton Univ., NJ (USA). Dept. of Chemical Engineering
- Los Alamos National Lab., NM (USA)
- Virginia Univ., Charlottesville, VA (USA). Dept. of Chemical Engineering
We report results on the use of neural nets, and the closely related radial basis nets'', to analyze experimental time series from electro-chemical systems. We show how the nets may be used to derive a map that describes the nonlinear system, and how reserving an extra input line'' of the network allows one to learn the system behavior dependent on a control variable. Pruning'' of the network after training appears to result in elimination of spurious connection weights and enhanced predictive accuracy. Subsequent analysis of the learned map using techniques of bifurcation theory allows both nonlinear system identification and accurate and efficient predictions of long-term system behavior. The electrochemical system that was used involved the electrodissolution of copper in phosphoric acid. This system exhibits interesting low dimensional dynamics such transitions from steady state to oscillatory behavior and from period-one to period-two oscillations. This analysis provides an example of methodology that can be fruitful in understanding systems for which no adequate phenomenological model exists, or for which predictions of system behavior given a large scale, complicated model is inherently impractical. 17 refs., 2 figs.
- Research Organization:
- Los Alamos National Laboratory (LANL), Los Alamos, NM (United States)
- Sponsoring Organization:
- USDOD; National Science Foundation (NSF); PFF; VACIT
- DOE Contract Number:
- W-7405-ENG-36
- OSTI ID:
- 6974047
- Report Number(s):
- LA-UR-90-515; CONF-900398-1; ON: DE90007539
- Resource Relation:
- Conference: 1990 international conference on neural networks, Lyons (France), Mar 1990
- Country of Publication:
- United States
- Language:
- English
Similar Records
Graph Neural Network and Koopman Models for Learning Networked Dynamics: A Comparative Study on Power Grid Transients Prediction
AP-Net: An atomic-pairwise neural network for smooth and transferable interaction potentials
Related Subjects
99 GENERAL AND MISCELLANEOUS//MATHEMATICS, COMPUTING, AND INFORMATION SCIENCE
PROCESS CONTROL
PARALLEL PROCESSING
CHEMICAL REACTIONS
COPPER
ELECTROCHEMISTRY
NEURAL NETWORKS
PHOSPHORIC ACID
CHEMISTRY
CONTROL
ELEMENTS
HYDROGEN COMPOUNDS
INORGANIC ACIDS
METALS
PROGRAMMING
TRANSITION ELEMENTS
320303* - Energy Conservation
Consumption
& Utilization- Industrial & Agricultural Processes- Equipment & Processes
990200 - Mathematics & Computers