The work performed during this thesis focused on uncertainty propagation (nuclear data, technological uncertainties, calculation biases,...) on integral parameters, and the development of a novel approach enabling to reduce this uncertainty a priori directly from the design phase of a new experimental program. This approach is based on a multi-parameter multi-criteria extension of representativity and transposition theories. The first part of this PhD work covers an optimization study of sensitivity and uncertainty calculation schemes to different modeling scales (cell, assembly and whole core) for LWRs and FBRs. A degraded scheme, based on standard and generalized perturbation theories, has been validated for the calculation of uncertainty propagation to various integral quantities of interest. It demonstrated the good a posteriori representativity of the EPICURE experiment for the validation of mixed UOX-MOX loadings, as the importance of some nuclear data in the power tilt phenomenon in large LWR cores. The second part of this work was devoted to methods and tools development for the optimized design of experimental programs in ZPRs. Those methods are based on multi-parameters representativity using simultaneously various quantities of interest. Finally, an original study has been conducted on the rigorous estimation of correlations between experimental programs in the transposition process. The coupling of experimental correlations and multi-parametric representativity approach enables to efficiently design new programs, able to answer additional qualification requirements on calculation tools.
