Estimation of the three-dimensional power distribution in boiling water reactors using neural networks and fuzzy theory

Abstract

In order to operate a nuclear reactor safely, it is necessary to monitor the soundness of the fuel assemblies loaded in the reactor core. The ability to calculate the power distribution in a reactor core is indispensable for monitoring that soundness. This paper proposes a new model for approximating the 3D power distribution in a reactor core using a neuro-fuzzy approach. In the proposed method we use learning algorithms based on a class of Quasi-Newton optimization techniques called the Self-Scaling Variable Metric (SSVM) method. Further, we propose a new learning algorithm for the fuzzy connected neural network (FCNN). The FCNN consists of many neural networks (local neural networks, LNNs). One LNN is connected to other LNNs with a fuzzy membership function, and the LNNs respond to input data as a whole neural network (FCNN). New prediction models were applied to the core of an actual Boiling Water Reactor (BWR) plant. The results demonstrate that the new model can predict the 3D power distribution of a BWR reasonably well.