Linear system approximation via covariance equivalent realizations (Q1062961): Difference between revisions

For a linear stochastic system (asymptotically stable, controllable and observable) (*) \(dx(t)=A x(t)dt+D dW_ t\), \(y(t)=C x(t)\) (with constant matrices A, D, C and \(W_ t\) a Wiener process) a truncation technique is described that allows for matching of a specified number of \(q+1\) output covariance derivatives and q Markov parameters simultaneously. One basic assumption is that the obtained partial realizations be asymptotically stable, which is the case if they are controllable. The presented procedure is independent of the basis used in (*) (up to unitary transformation) and provides a stochastically equivalent realization (in the sense of Anderson), if all output covariance derivatives match. The example of a simply supported beam shows the impulse-response fit for several partial realizations. The paper is in contents and diction applied linear systems theory, the basic mathematical technique is matrix manipulation.