Recognition of processes governed by stochastic differential equations. (Q1593907)

The authors consider Itô's system of stochastic differential equations for the state \(Z\) and the observation \(Y\), \[ dZ=\varphi(Y,Z,\Theta,t)dt+\psi(Y,Z,\Theta,t)dW_1,\quad t\geq t_0,\;Z\in\mathbb R^n, \] \[ dY=\varphi_1(Y,Z,\Theta,t)dt+\psi_1(Y,t)dW_2,\quad t\geq t_0, Y\in\mathbb R^m, \] where \(W_1\in\mathbb R^{r_1}\) and \(W_2\in\mathbb R^{r_2}\) are independent Wiener processes, \(\varphi,\psi,\varphi_1,\psi_1\) are given vector and matrix functions of the corresponding dimension, and \(\Theta\) is a discrete random variable taking values \(\theta_1,\cdots,\theta_N\) with probabilities \(p_1,\dots,p_N\), \(p_1+\cdots+p_N=1\). Then the problem of finding the conditional probabilities \[ q_k(t)=P(\Theta=\theta_k\mid Y(s), \;t_0\leq s\leq t),\quad k=1,\dots,N, \] can be solved combined with the problem of optimal filtering of the state \(Z\). The authors propose an approach for a split (sequential) solution of these two problems.