Efficient Decision-Making-Free Filter for Processes with Abrupt Changes

摘要

The system of linear stochastic difference equations with Markovian coefficients was generalized to cover processes in R sup n which jump simultaneously with the coefficients. The additional modeling potential of this generalized system is illustrated. For filtering partial observations of the generalized system, the Interacting Multiple Model (IMM) algorithm is given. It consists of a bank of N interacting Kalman-like filters which cooperate with a filter for the N state Markov process and is free of any decision making mechanism. Comparisons with other algorithms for processes with abrupt changes show that the IMM algorithm performs very well at the cost of a relatively low computational load. Due to the generalization, the IMM algorithm is a serious competitor of decision directed filters for processes in R sup n with additive jumps.