Section: New Results

Numerical method for the filtering of Markov jump linear systems

Participants : Benoîte de Saporta, Eduardo Costa.

We are interested in efficient pre-computations of the solutions of Markov switching Riccati equations. These equations are matrix-valued and naturally arise in control or filtering problems for Markov jump linear systems. It is crucial for applications to be able to compute the filter in real time, although the solutions to Riccati equations are slow to compute. Hence the need for pre-computations, taking into account the random possible changes of regimes. We propose a numerical method based on the discretization by quantization of the underlying Markov chain.