OPTIMIZATION OF FUEL RELOAD IN A BWR NUCLEAR REACTOR USING A RECURRENT NEURAL NETWORK

摘要

In this work we used a Multi-State Recurrent Neural Network (MSRNN) to optimize the nuclear fuel reload in a BWR reactor. The MSRNN proposes different fuel reloads changing its neuronal states. The energy function of each fuel reload is evaluated taking into account its safety aspects and cycle length. Safety aspects are predicted using a trained Back-Propagation Neural Network (BKPNN) instead a reactor simulator. The MSRNN energy function calculates the k_(eff) value and successive neuronal state exchanges increase it. The MSRNN and BKPNN were implemented in a new system named RENOR. The MSRNN was validated optimizing a fuel reload for Laguna Verde Nuclear Power Plant in Mexico. Theoretically, the RENOR fuel reload has a cycle length greater than the real cycle length.