Scenarios with COSI6: Optimization, Uncertainty and Beyond

Guillaume Krivtchik

CEA (France)

Since 1985, the CEA/DEN has been developing the COSI software, simulating the dynamic evolution of a reactor fleet and its associated facilities. It was validated on the French historical PWR fleet, and participated in several benchmarks including NEA.

COSI6 performs reactor-driven, continuous time scenario simulations. Many fuel cycle facilities are modeled, with different levels of detail, from the uranium mines to the final waste. In particular, the fuel batches, the reactors and the reprocessing plants are highly parameterized, so as to provide an accurate model of the fuel cycle. In addition, the consumption of non-fissile, strategical / critical materials of interest, such as boron, can be evaluated so as to compare the needs and the resources.

This presentation focuses on the newly implemented optimization and uncertainty propagation techniques, and provides insight on the limits of both processes, thus requiring the development of innovative scenario generation and analysis methods.

A multicriteria scenario optimization method using COSI6 was developed. This method uses metaheuristics and a surrogate sub-model approach in the frame of widely multiparameter and non-linear optimization studies. The general steps of the methods will be exposed, as well as examples of transition scenarios optimization. The feedback from recent studies will be analyzed and limits and perspectives will be discussed.

The uncertainty propagation method will be developed, including categorization of uncertainty sources, cross-sections covariance matrices collapsing, uncertainty propagation in depletion and equivalence models, as well as examples of uncertainty propagation. Once again, the physical and philosophical limits of the process will be discussed.

Finally, the concept of robustness in scenario studies will be discussed. This topic still is an open question, and the ways to tackle it must be discussed. Different perspectives of robustness characterization and optimization will be exposed.