A neural network approach for burn-up calculation and its application to the dynamic fuel cycle code CLASS

• DOI: 10.1016/j.anucene.2015.03.035
• Publication Date: 2015-04-04T20:49:44Z
• Authors: Baptiste Leniau; Baptiste Mouginot; Nicolas Thiolliere; Xavier Doligez; Adrien Bidaud; Fanny Courtin; Marc Ernoult; Sylvain David
• Keywords: Nuclear scenario; Cross section predictor; PWR; 0202 electrical engineering, electronic engineering, information engineering; MOX; 02 engineering and technology; [PHYS.NEXP]Physics [physics]/Nuclear Experiment [nucl-ex]; 530; 7. Clean energy; Neural network; 620

Dynamic fuel cycle simulation tools calculate nuclei inventories and mass flows evolution in an entire fuel cycle, from the mine to the final disposal. Usually, the fuel depletion in reactor is handled by a fuel loading model and a mean cross section predictor. In the case of a PWR–MOX, a fuel loading model provides from a plutonium stock the plutonium fraction in the fresh fuel needed to reach a specific burnup. A mean cross section predictor aims to assess isotopic cross sections required for building Bateman equations for any fresh fuel composition with a sufficient accuracy and a reasonable computing time. This paper presents a methodology based on neural networks for building a fuel loading model and a cross section predictor for a PWR reactor loaded with MOX fuel. The mean error of the plutonium content prediction from the fuel loading model is 0.37%. Furthermore, the mean cross section predictor allows completion of the fuel depletion calculation in less than one minute with excellent accuracy. A maximum deviation of 3% on main nuclei is obtained at the end of cycle between inventories calculated from neural networks and from the reference coupled neutron transport/fuel depletion calculation.