Decomposition of problems in the estimation of the state of multidimensional stochastic systems (Q1820118)

It is assumed that a partially observable stochastic process \([x^ T,z^ T]^ T\) is described by the Itô stochastic differential equations \[ dx=f(t,x)dt+\sigma (t,x)dw_ 1,\quad x(t_ 0)=x_ 0,\quad dz=\phi (t,x)dt+\phi (t,x)dw_ 2, \] where \(w_ 1\) and \(w_ 2\) are mutually independent standard Wiener processes with values in \(R^ r\) and \(R^ m\) respectively, \(z=z(t)\in R^ m\) is the observation vector, \(x=x(t)\in R^ n\) is the state vector and the functions f, \(\phi\), \(\sigma\), \(\phi\) are differentiable infinitely many times with respect to t, x. It is required to find the best, in the mean-square sense, estimator \(\hat x=\hat x(t)\) of the process x(t) according to the observations \(z^ t_ D=\{z(\tau)\), \(t_ 0\leq \tau \leq t\}.\) The author proposes an approximate filtering algorithm based of a preliminary decomposition of the problem. The number of differential equations forming the estimator \(\hat x(t)\) in the developed algorithm is smaller than for the generalized Kalman filter.