On parallel estimation approach to adaptive filtering (Q1369084)

The paper considers the identification (tracking) problem of non-stationary stochastic systems. Two parameter tracking algorithms based on a multiple model, parallel estimation approach are described: competing multiple models (CMM) and interacting multiple models (IMM). In the CMM scheme, all Kalman filters comprising the adaptive filter bank work independently of each other, while the IMM algorithm represents a modified version of the routine proposed for tracking maneuvering targets. The IMM scheme means actually that, at the beginning of each cycle of model-conditioned Kalman filtering, the state vectors and error covariance matrices of component filters are mixed appropriately. Two numerical examples illustrate the proposed CMM and IMM algorithms on a simple adaptive control system, and on a problem of adaptive prediction for a non-stationary autoregressive process.