Uncertainty quantification of proper orthogonal decomposition based online power-distribution reconstruction

In this paper, the function of uncertainty quantification is implemented in CORN, an online power-distribution reconstruction code based on the proper orthogonal decomposition (POD) method. The uncertainties in the detector signals are propagated to the reconstructed power distributions with the deterministic-based method in two steps. First, the sensitivity coefficients are calculated with the direct numerical perturbation (DNP) method. Then the uncertainty in the reconstructed power distribution based on the detector signals is carried out with the “sandwich” rule. The uncertainty quantification is carried out based on the core of BEAVRS benchmark cycle 1. Numerical results show that the maximum uncertainty of 3D reconstructed power distribution is less than 3% with the 1% uncertainties of detector signals for the POD method. At last, the uncertainties of power distributions for various expansion order are quantified. Eigenvalues can assist in the determination of the expansion order with a certain group of POD samples.