Algorithm to determine the optimal parameters of a Wiener filter-extrapolator for nonstationary stochastic processes observed with errors (Q1364069)

The second order process \(Z(t)\) is defined at a discrete series of points \(t=i\), \(1\leq i\leq I\), \(Z(t)= X(t)+ Y(t)\), where \(Y(t)\) is a random observation error. Suppose that \(EX(t)= EY(t)= 0\) and the correlation functions \(R_Z (\mu,\nu)\), \(R_X(\mu,\nu)\) and \(R_{XZ} (\mu,\nu)\) are known. The linear extrapolation problem is considered to estimate future values of the process \(X(i)\), \(k+1\leq i\leq I\), by the observations \(Z(i)\), \(1\leq i\leq k\). Via the orthogonal expansion of the compound sequence \(\{Z(1), \dots, Z(k),\;X(k+1), \dots, X(I)\}\) the predictor \(\widehat X(i)\), \(k+1 \leq i\leq I\), is found, that minimizes the mean-square extrapolation error. For the case when both of the processes \(X(t)\) and \(Y(t)\) are stationary it is shown that \(\widehat X(i)\) coincides with the predictor obtained by the Wiener method. The statement is proved for \(k=1,2\) only.