Search

Approximately one year before a single U[O.sub.2] fuel pellet is pressed and fresh fuel assemblies manufactured, design calculations are performed to effectively guarantee the performance of a light water reactor (LWR) core for an upcoming fuel cycle. Energy requirements must be fulfilled in conjunction with all other reactivity, thermal, and operational limits. Furthermore, several months prior to a reactor startup, the proposed core must be licensed by the regulating authority. Therefore, all built-in design conservatisms (i.e., target margins) established a priori must satisfy all safety, operational, and regulatory constraints, a posteriori. The magnitude of target margins directly impacts cycle energy efficiency, which is why this design cushion is often referred to as the 'cost of margin' because it ultimately affects the cost per generated kilowatt-hour by a LWR. This article illustrates the modern role of nuclear fuel management optimization in the LWR core reload design process, highlighting some of the history and recent advancements in the field, particularly, in the area of pin-by-pin optimization. Also, important limitations are highlighted to help define the new level of sophistication which the field must conquer for designers to ultimately be able to optimize LWRs 'one fuel pin at a time.' Index Terms--Nuclear fuel management, optimization.

A query-based approach for adaptively retraining and restructuring a two-hidden-layer artificial neural network (ANN) has been developed for the speedy prediction of the fundamental mode eigenvalue of the neutron diffusion equation, a standard nuclear reactor core design calculation which normally requires the iterative solution of a large-scale system of nonlinear partial differential equations (PDE's). The approach developed focuses primarily upon the adaptive selection of training and cross-validation data and on artificial neural-network (ANN) architecture adjustments, with the objective of improving the accuracy and generalization properties of ANN-based neutron diffusion eigenvalue predictions. For illustration, the performance of a 'bare bones' feedforward multilayer perceptron (MLP) is upgraded through a variety of techniques; namely, nonrandom initial training set selection, adjoint function input weighting, teacher-student membership and equivalence queries for generation of appropriate training data, and a dynamic node architecture (DNA) implementation. The global methodology is flexible in that it can 'wrap around' any specific training algorithm selected for the static calculations (i.e., training iterations with a fixed training set and architecture). Finally, the improvements obtained are carefully contrasted against past works reported in the literature. Index Terms - ANN, DNA, eigenvalues, membership and equivalence queries, MLP, neutron diffusion.