Artificial Neural Network Training with Conjugate Gradients for Diagnosing Transients in Nuclear Power Plants
The method of conjugate gradients is used to expedite the learning process of feedforward multilayer artificial neural networks. The proposed method systematically determines both the learning and momentum parameters which are dynamically updated at each learning cycle. The learning parameter is obtained through a one-dimensional search that combines the iterative golden-section algorithm with an analytical cubic interpolation. The momentum parameter is obtained as conjugated directions are calculated at each learning cycle. By dynamically updating the learning and momentum parameters the method avoids the phenomenon of premature saturation of the network nodes and becomes less sensitive to the initial set of weights. The proposed method is compared with the standard backpropagation algorithm in the training of neural networks to classify transient events in nuclear power plants simulated by the Midland Nuclear Power Plant Unit 2 simulator. The comparison results indicate that the rate of convergence of the proposed method is much greater than the standard backpropagation reducing the number of learning cycles and the CPU time. The advantages of the method are more noticeable and important for problems where the network architecture consists of a large number of nodes, the training database is large, and a tight convergence criterion is desired. In addition, the reasons for the premature saturation of the network nodes observed with the backpropagation algorithm are described and suggestions are made to eliminate this undesirable phenomenon.
- Research Organization:
- Argonne National Lab. (ANL), Argonne, IL (United States)
- Sponsoring Organization:
- USDOE Office of Nuclear Energy
- DOE Contract Number:
- AC02-06CH11357
- OSTI ID:
- 10198077
- Report Number(s):
- ANL-IFR-189; 154258
- Country of Publication:
- United States
- Language:
- English
Similar Records
Statistical and optimization methods to expedite neural network training for transient identification
Statistical and optimization methods to expedite neural network training for transient identification