Development of Depletion Code Surrogate Models for Uncertainty Propagation in Scenario Studies

The result of transition scenario studies, which enable the comparison of different options of the reactor fleet evolution and management of the future fuel cycle materials, allow to perform technical and economic feasibility studies. The COSI code is developed by CEA and used to perform scenario calculations. It allows to model any fuel type, reactor fleet, fuel facility, and permits the tracking of U, Pu, minor actinides and fission products nuclides on a large time scale. COSI is coupled with the CESAR code which performs the depletion calculations based on one-group cross-section libraries and nuclear data. Different types of uncertainties have an impact on scenario studies: nuclear data and scenario assumptions. Therefore, it is necessary to evaluate their impact on the major scenario results. The methodology adopted to propagate these uncertainties throughout the scenario calculations is a stochastic approach. Considering the amount of inputs to be sampled in order to perform a stochastic calculation of the propagated uncertainty, it appears necessary to reduce the calculation time. Given that evolution calculations represent approximately 95% of the total scenario simulation time, an optimization can be done, with the development and implementation of a surrogate models library of CESAR in COSI. The input parameters of CESAR are sampled with URANIE, the CEA uncertainty platform, and for every sample, the isotopic composition after evolution evaluated with CESAR is stored. Then statistical analysis of the input and output tables allow to model the behavior of CESAR on each CESAR library, i.e. building a surrogate model. Several quality tests are performed on each surrogate model to insure the prediction power is satisfying. Afterward, a new routine implemented in COSI reads these surrogate models and using them in replacement of CESAR calculations. A preliminary study of the calculation time gain shows that the use of surrogate models allows stochastic calculation of the uncertainty propagation. Once the set of surrogate models is complete, one of the first expected results is the comparison of global sensitivity indices of nuclear data and scenario assumptions