Algorithm of extrapolation of a nonlinear random process on the basis of its canonical decomposition (Q2577242)

The paper is devoted to the solution of the following problem: Let a random process \(X(t)\) be specified in a discrete series of points, by moment functions. It is necessary to obtain a realizable (for example, on computers) procedure of computation of future values of the process \(X(t)\) that are optimal in the root-mean-square sense for an arbitrary number of known values and probabilistic relations. The extrapolation algorithm is based on the nonlinear canonical representation of \(X(t)\). The extrapolation form obtained in the paper is approximated to an optimal extrapolator with an arbitrary (dependent only on the capabilities of a computer) accuracy and, as well as the canonical decomposition that underlies this form, does not impose any substantial constraints on the class of predictable random processes such as the Markov property, linearity, monotonicity or stationarity.