Uncertainty and optimization: a coupled problem for scenario analyses

Nuclear fuel cycle simulators have become an essential tool for researchers and policy makers for studying and evaluating different electronuclear strategies in the mid and long term. In order to improve their versatility and to reduce the user interaction, in the recent years the codes have been upgraded for performing optimization analyses. In this way, the desired outcomes of the simulation (i.e. the functions to be optimized) are firstly defined instead of being a result of the calculation. Therefore, given a relaxed set of initial conditions, the best scenario matching the chosen objectives is automatically obtained without previous knowledge of the user. Nevertheless, although best-case scenarios can be obtained with these methods, in practice they may not be the desired ones because of the lack of robustness. This happens because certain quantities can be pushed to the limit during the optimization process. Hence, if small perturbations occur, the scenario will break, and the simulation will not be completed. This is of special interest in uncertainty analyses in which some input parameters are not characterized by a reference value but by a probability density function. If the scenario has been optimized for the expected values of the input parameters, when different samples are randomly drawn there is no guarantee that the scenario remains stable. In this work, the problem of optimization under uncertainty in fuel cycle simulations is discussed. The evolutionary algorithm implemented in TR-EVOL for performing multiobjective optimization is firstly shown, and after that, the extension used for addressing the uncertainties will be presented. This methodology has been applied to a test case based on an advanced European transition scenario in which an initial fleet of LWR is replaced by a burner one with the objective of reducing as much as possible TRU inventories while keeping the economic costs low. (authors)