ACCELERATING LEARNING OF NEURAL NETWORKS WITH CONJUGATE GRADIENTS FORNUCLEAR-POWER-PLANT APPLICATIONS

The method of conjugate gradients is used to expedite the learning pro cess of feedforward multilayer artificial neural networks and to syste matically update both the learning parameter and the momentum paramete r at each training cycle. The mechanism for the occurrence of prematur e saturation of the network nodes observed with the backpropagation al gorithm is described, suggestions are made to eliminate this undesirab le phenomenon, and the reason by which this phenomenon is precluded in the method of conjugate gradients is presented. The proposed method i s compared with the standard backpropagation algorithm in the training of neural networks to classify transient events in nuclear power plan ts simulated by the Midland Nuclear Power Plant Unit 2 simulator. The comparison results indicate that the rate of convergence of the propos ed method is much greater than the standard backpropagation, that it r educes both the number of training cycles and the CPU time, and that i t is less sensitive to the choice of initial weights. The advantages o f 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.