An evaluation of neural networks for identification of system parameters in reactor noise signals
Several backpropagation neural networks for identifying fundamental mode eigenvalues were evaluated. The networks were trained and tested on analytical data and on results from other numerical methods. They were then used to predict first mode break frequencies for noise data from several sources. These predictions were, in turn, compared with analytical values and with results from alternative methods. Comparisons of results for some data sets suggest that the accuracy of predictions from neural networks are essentially equivalent to results from conventional methods while other evaluations indicate that either method may be superior. Experience gained from these numerical experiments provide insight for improving the performance of neural networks relative to other methods for identifying parameters associated with experimental data. Neural networks may also be used in support of conventional algorithms by providing starting points for nonlinear minimization algorithms.
- Research Organization:
- Tennessee Univ., Knoxville, TN (United States). Dept. of Nuclear Engineering
- Sponsoring Organization:
- USDOE, Washington, DC (United States)
- DOE Contract Number:
- FG07-88ER12824
- OSTI ID:
- 10104447
- Report Number(s):
- CONF-910535-10; ON: DE93003545
- Resource Relation:
- Conference: 6. specialists meeting on reactor noise (SMORN),Gatlinburg, TN (United States),19-24 May 1991; Other Information: PBD: [1991]
- Country of Publication:
- United States
- Language:
- English
Similar Records
Sensor validation in power plants using adaptive backpropagation neural network
Sensor validation in power plants using adaptive backpropagation neural network