FLORIN MUNTEANU, CRISTIAN IOANA, EDMOND CRETU, CRISTIAN SUTEANU, DOREL ZUGRAVESCU
Académie Roumaine, Institut de Géodynamique, Bucarest
Recent research in the field of nonlinear hierarchic systems evolving far
from equilibrium, led to the reconsidering of the concepts of noise. Evaluated
qualitatively and quantitatively by means of a large number of parameters,
recently rejected noise becomes an important source of information, especially
useful in the theoretical and experimental studies on systems about which our
a priori knowledge is sparse.
The present paper proposes an original method for the discrimination and the
classification of 'noise' provided by a large class of sensors, in order to
extract the possibly precious information comprised in the records on the fluctuation
of geodynamic parameters in time. The method relies on the interpretation of
the variance-covariance matrix associated to the collection of multivariable
data which form a matrix m, x, n, with columns representing the values of the
n parameters attached to each of the m signals to be processed. The discrimination
capability of the method depends on the n scalars, which must be chosen in
order to grasp the general statistical pattern, the degree of irregularity,
the stationarity of the signal and the probable complexity of the analysed
system.
Mots-clé: algorithme, bruit, classification, fractals, séries
temporelles.