High-order interacting multiple-model estimation for hybrid systems with Markovian switching parameters

Abstract

In this paper, a generalized ith-order interacting multiple-mode l algorithm for state estimation of non-stationar y linear systems with Markovian switching parameters is developed. The state estimates are derived under all possible model sequence hypotheses over the i most recent sampling periods. Each model hypothesis is evaluated on the basis of a diVerent linear combination of the estimates conditioned on the assumed model sequences and yielded by Kalman subWlters running in parallel. The resulting algorithm is veriWed by means of simulation.

On Scientific Realism and Instrumentalism in Manoeuvring Target Modelling and Tracking

“…It was observed that the IMM estimator is characterised by accuracy similar to GPB2 techniques and with numerical complexity comparable to GPB1 algorithms. The IMM approach has been successfully applied to solve various technical problems [39], [61] such as: state estimation, system identification, MTT, filtering and smoothing, fault detection and isolation, multi-sensor data fusion, robust speech recognition and multirate processing. When applied to the MTT problem, IMMbased tracking filters were shown to outperform virtually all techniques known before, including those based on IE, EIE and on the SMM principle [3].…”

“…A further improvement of the IMM estimation accuracy can be obtained by using various higher-order IMM (IMMi) techniques [11], [61], although at the expense of the increased computational burden. In [61] for instance, state estimates of a linear hybrid system are obtained by using mode switching modelled by an i-th order Markov chain on the basis of all possible mode hypotheses over the i most recent sampling periods and requiring M i filters running simultaneously.…”