A new approach to multiple model adaptive estimation

An algorithm for adaptive estimation of time-varying parameters in certain classes of linear stochastic dynamic systems has been developed. The algorithm is based on an adaptive Kalman filter (AKF) whose hypothesized parameters are modified at each stage by generating the probability of each hypothesis, conditioned on the residual history and a given probability of transition. We develop sufficient conditions for the stochastic convergence of this adaptive filter structure. By invoking an information function, the filter is also shown to be robust with respect to modeling errors. A few numerical simulations have been performed to evaluate this algorithm against the backdrop of the multiple model adaptive estimation (MMAE) scheme