Estimation of Effective Parameters in a Lumped Nuclear Reactor Model Using Multiobjective Genetic Algorithms

For realistic systems, a dynamic approach to reliability analysis is likely to require a significant increase in the computational efforts, due to the need of integrating the dynamic evolution with its characteristic times. Thus, it becomes mandatory to resort to validated, simplified models of process evolution. Such models are typically based on lumped effective parameters whose values need to be suitably estimated so as to best fit to the available plant data.

In this paper we propose a multiobjective genetic algorithm approach for the estimation of the effective parameters of a simplified model of nuclear reactor dynamics. The calibration of the effective parameters is achieved by best fitting the model responses of the quantities of interest to the actual evolution profiles. A case study is reported in which the real reactor is simulated by the QUAndry-based Reactor Kinetics (QUARK) code available from the Nuclear Energy Agency, and the simplified model is based on the point-kinetics approximation to describe the neutron balance in the core and on thermal equilibrium relations to describe the energy exchange between the different loops. The (pseudo)measured quantities of interest are the reactor power and the average fuel temperature.