Parameter Adaptation Techniques for the Best-Estimate Thermal-Hydraulic Code ATHLET

Parameter adaptation can be used for automated adjustment of real time simulation according to measured data from the simulated plant. It is able to prevent that model uncertainties lead to an increasing deviation between simulation results and real plant behaviour. The objective is to ensure that the simulation corresponds with reality even over a longer period. The present investigation deals with the application of parameter adaptation techniques to best-estimate simulation. This opens a range of new applications including the use of best-estimate codes for improving process diagnostics, signal validation and the man-machine-interface. The adaptation of ATHLET poses several challenges. Among them are non-linear dependencies within the equation system to be solved, the interference of a sophisticated time integration technique, and the identification and proper consideration of those code uncertainties which are most influential with regard to deviations between simulation and real plant behaviour. To overcome these problems a set of new techniques has been developed: - A refined method for the definition of the parameters to be adapted is based on the consideration of the parameter’s uncertainty and sensitivity to the results. Statistic sensitivity analysis is used to validate and improve parameters obtained by expert judgement. - Different parameter adaptation techniques (e.g. based on standard controllers or parameter estimation techniques) are applied to produce feasible solutions for the large systems to be modeled. - In view of preventing non-reversible propagation of simulation inaccuracies decomposition techniques are applied assuring that such temporary inaccuracies do not spread over subsystem boundaries. Results of adaptation based on these techniques are shown for an experiment performed at the test facility PKL III. It turns out that adaptive simulation is feasible for such facilities and that it offers possibilities to improve prediction of experiments by code calculations.