Uncertainties and sensitivities study methods applied to the dynamic fuel cycle

Jean-Baptiste Clavel

IRSN (Institut de Radioprotection et de Sûreté Nucléaire)

The definition of electronuclear scenarios is often done with the aim to evaluate strategies of reactors deployments and new options of fuel management. Those scenarios contain a huge number of input parameters (reactors, facilities, storages and stocks characteristics, nuclear fleet progress…) which are not always precisely define and are often uncontrolled, like the loading factor or the starting date of reactors. Those uncertainties might have a strong impact on some output quantities such as inventories compositions, plutonium/waste masses. This paper aims at testing an uncertainties and sensitivities method that should be used for robust studies applied to scenarios framework.

The proposed approach starts by selecting parameters, defining uncertainties models (Gaussian, uniform...) and variation ranges. The results of the uncertainty propagation also depend on the evaluated quantities of interest, therefore the observables on which the impact will be estimated must be selected. Once this setup is defined, the next step is to apply a screening method (Morris design) to evaluate the significance and interactivity of parameters on the results. This allows evaluating independent parameters apart. Finally, a sensitivity study based on the evaluation of Sobol indices is applied on the remaining parameters for deeper analyse.

The Dynamic fuel cycle simulation tool used for this illustration is CLASS (Core Library for Advance Scenario Simulation) [1, 2], which is an open source package of C++ libraries developed by CNRS in collaboration with IRSN. For this paper, the CLASS software is coupled with the Promethee computer experiments workbench developed at IRSN [3]. It provides an easy-to-use graphical interface for parametric calculations and integrates several algorithms including the Morris and Sobol ones allowing for advanced sensitivity studies. This method will be presented on a simple application case using these tools.