Continuous-time interacting multiple model extrapolation

This paper describes a continuous-time, interacting-multiple-model (IMM) extrapolation algorithm. A system state is modeled as a continuous-time, affine-Gaussian stochastic dynamical process driven by a white process noise as well as structural changes modeled by a finite-state, continuous-time Markov process. The system generally assumes multiple models with different system state dimensions and an affine-Gaussian state jump whenever a model transition occurs. The underlying problem is a standard filtering problem for estimating the system state based on a sequence of discrete-time, linear-Gaussian observations of partial system states. Assuming a model-wise state-estimate and model-probability updating algorithm that is standard for IMM filtering or tracking algorithms, this paper proposes a new algorithm that performs extrapolation and model-mixing simultaneously as a part of an IMM algorithm.