STATISTICAL CRITERION FOR CHOOSING OF OPTIMAL MODEL
OF IONIZING RADIATION SPECTRA

V. A. Muravsky, S. A. Tolstov 

International Sakharov Institute of Radioecology
23, Dolgobrodskaya Str., 220009 Minsk, Belarus 

     The variants of the least  squares  and  the  maximum  likelihood
method  are  usually  used  for  the nuclides activities estimating on
spectra of ionizing radiation.  And in the all cases one  must  choose
adequate  theoretical  model  describing both the nuclides composition
and the shape of processed spectrum. Such approach is justified in the
case  of  the high-intensity spectra,  when nuclides can be identified
through the full-energy peaks.
     However, there are no objective criteria for choosing of adequate
model in the case of low-intensity spectra.
     Therefore, one   has   to   include   many  additional,  actually
unnecessary parameters in the spectrum model.  The parameters describe
possible  (wide  enough)  composition  of nuclides and factors,  which
could influence spectrum shape.  In addition to  complication  of  the
calculation algorithm,  that leads to the deterioration of accuracy of
the estimates,  because addition of new parameter in the model  always
reduces precision of estimates of the other parameters.
     Therefore, one needs criteria of the model  choosing,  which  are
able to minimize errors of estimates.
     Theoretical analysis of the statistical problem  of  choosing  of
the optimal (at the given confidence level) model of spectrum was done
in the work. The criterion was built on the basis of the analysis. The
criterion  allows  to  create  algorithms  of spectrum processing with
automatic optimization  of  the  model  by  excluding  of  unnecessary
parameters.  This provides achievement of maximum accuracy of obtained
estimates and identification  of  nuclides  at  the  given  confidence
limit.

Back