AUTOMATIC CALIBRATION OF THE VARIOGRAM MODEL PARAMETERS (PREPRINT IBRAE 2016-04)

Preprint IBRAE-2016-04

Gorelov M.M., et al.

The estimates obtained by numerical simulation are commonly used in the safety case and safety assessment. Numerical simulation is inevitably associated with the various types of uncertainties. And the variation of model parameters (that can be represented by two-dimensional or multi-dimensional fields) is required for uncertainty analysis. Thus the problem of random fields generation based on the spatial correlation model of the input data arises. The classic way to describe the spatial correlation is a variogram. In isotropic case the variogram has the same appearance and parameters for all directions. Anisotropic spatial correlation can be parameterized using the isotropic model and anisotropy ellipse. In this case automatic calibration with optimization strategies and clear stop criteria becomes more efficient than manual parameter fitting. Another issue is the selection and configuration of the automatic calibration method, the efficiency and comparative analysis of different algorithms. In this paper we consider Levenberg-Marquardt and Particle Swarm optimization methods.

Bibliographical reference

Gorelov M.M., Svitelman V.S., Savelyeva E.A. Automatic calibration of the variogram model parameters. Preprint IBRAE 2016-04. Moscow: Nuclear Safety Institute, 2016. — 17 p.

© Nuclear Safety Institute, 2016