CERFACS CERFACS Scientific Activity Report Jan. 2010 – Dec. 2011

DATA ASSIMILATION WITH EDF NEUTRONIC MODELS
apply a specific procedure which evaluates this influence by adding or removing instruments in a given
measurement network (possibly empty). That procedure using the Schur complement have been detailed in
[DA22]. The study of various network configurations of instruments in the nuclear core establishes that the
influence of the instruments depends both on the independent instrumentation location and on the chosen
network.
It was proven also that results are very different when instruments are added to a non instrumented system
than to a partially (half) instrumented one. Such a non equivalence respect to starting point proves that
building a complete instrumental system cannot be done iteratively. Such a system will have limited
efficiency as each step is dependant upon the previous and not of the global situation. This implies that,
in order to build an optimal measurement network in a nuclear core, it is necessary to be able to take into
account all the instruments globally.
Those results have been reposted in [DA22] and published in Nuclear Instrumentation and Method A [DA2].
6.4 Optimisation of the instrument network for the nuclear core
(B. Bouriquet, O. Thual)
Knowing that an optimal network cannot be constructed iteratively we focused on solving the inverse
problem of determining an optimal repartition of the measuring instruments within the core, to get the
best possible results from the data assimilation reconstruction procedure. The position optimisation is
realised using Simulated Annealing algorithm [22], based on the Metropolis-Hastings one. Moreover, in
order to address the optimisation computing challenge, algebraic improvements of data assimilation have
been developed and are presented here. Two main conclusions arises from those studies. The first one
is that the standard PWR900 instruments repartition is characterized by an excitingly good quality. This
instruments repartition is a priori the best we know, even if it was not originally designed using a data
assimilation framework background. The second point is that the simulated annealing method can always
find a ameliorated instrument locations set. Those results have been reported in [DA19] and was accepted
for publication in Nuclear Instrumentation and Method A.
6.5 Sensitivity of the data assimilation (B. Bouriquet, O. Thual)
Some studies have been carried out on sensitivity of the analysis obtained by data assimilation. In order to
obtain this sensitivity it has been determined the values that are conditioning the evolution of the analysis
respect to a small change if the data assimilation component. Studies have been done for PWR900 and
PWR1300. In order to synthesise sensitivity information as well as data assimilation sub product matrix
(as analysis error matrix) several method of information reduction are used. Specially we focused on the
MMSE (Minimal Mean Square Error) that allow to evaluate the extra diagonal information in a covariance
matrix. Those results are reported in [DA21].
6.6 Parameter estimation through data assimilation (B. Bouriquet,
O. Thual)
Another important point on data assimilation usage is the parameter adjustment. The aim is to find the
best estimation of the parameter using the available data. Such studies have already done in [13] on one
parameter with the COCCINELLE code. Then they have been extend with the use of COCAGNE in [14].
In that report the first test on estimation of several parameter at one have been tried. From that experience
some detail research on the assimilation of several parameter together have been performed. Those studies
92 Jan. 2010 – Dec. 2011