Gaussian processes for surrogate modeling of discharged fuel nuclide compositions

Several applications such as nuclear forensics, fuel cycle simulations and sensitivity analysis require methods to quickly compute spent nuclide compositions for various irradiation histories. Traditionally, this has been done by interpolating between one-group cross-sections that have pre-computed from reactor a grid of input parameters, using fits Cubic Spline. We propose the use Gaussian Processes (GP) create surrogate models, which not only provide compositions, but also gradient estimates their prediction uncertainty. The former is useful forward inverse optimization problems, latter uncertainty quantification applications. For purpose, we compare GP-based model performance with Cubic- Spline-based interpolators based on infinite lattice CANDU 6 SERPENT 2 code, considering burnup temperature parameters. Additionally, sampling schemes quasirandom Sobol sequence. find models perform significantly better in predicting than Cubic-Spline-based though requiring longer computational runtime. Furthermore, show predicted uncertainties are reasonably accurate. While studied two-dimensional case, grid- similar results, will be more effective strategy higher dimensional cases.