Adaptive neural network error control for generalized perturbation theory

Journal Article · Tue Dec 31 04:00:00 EST 1996 · Transactions of the American Nuclear Society

OSTI ID:426406

Wong, Hing-Ip; Maldonado, G I ^[1]

---

1. Iowa State Univ., Ames, IA (United States)

This paper addresses the issue of adaptive error control within generalized perturbation theory (GPT). The strategy herein assessed considers an artificial neural network (ANN) error estimator. The underlying tool facilitating this research is the FORMOSA-P code, a pressurized water reactor (PWR) nuclear fuel management optimization package, which combines simulated annealing and nodal GPT. A number of applications exist where traditional GPT may be limited by the magnitude of perturbations, which it can accurately handle. In fact, other alternative such as supervariational techniques (i.e., n`th-order GPT) and/or multireference strategies (i.e., rodded adjoints) are being sought for boiling water reactor and rodded applications. A perhaps not-so-obvious alternative could be to employ a neural network for adaptive error control within GPT. This study presents the results of two ANN models. The first model constitutes an intensively well-trained ANN used to contrast its global core parameter (i.e., k{sub eff}) prediction capability versus that of a GPT model. The second model is a similar ANN intended for adaptive GPT error correction. In other words, the latter ANN is trained on-the-fly within the scope of a standard FORMOSA-P calculation.

• OSTI ID: 426406
• Report Number(s): CONF-961103--
• Journal Information: Transactions of the American Nuclear Society, Journal Name: Transactions of the American Nuclear Society Vol. 75; ISSN TANSAO; ISSN 0003-018X
• Country of Publication: United States
• Language: English

Related Subjects

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS
ACCURACY
ARTIFICIAL INTELLIGENCE
COMPUTERIZED SIMULATION
F CODES
FUEL MANAGEMENT
NEURAL NETWORKS
OPTIMIZATION
PERTURBATION THEORY
PWR TYPE REACTORS