The two interacting multiple model algorithms with removing error measurement

In the centralized multi-sensor system, there is often a problem that how to remove measurements used early from the current state estimation, then re-calculate the current state estimation. The interacting multiple model (IMM) algorithm with removing old measurement presented by Bar-shalom is based on the standard IMM algorithm and Bl1 algorithm. However, in the standard IMM algorithm, the combination of normal probability density function and probability mass function results in that updated model weights are only an approximate probability. Moreover, sometimes the use of equivalent measurement can cause the performance of algorithm to deteriorate and the problem of Rank Deficiency. In order to deal with these drawbacks, based on the optimal fusion rule weighted by scales, the optimal fusion rule weighted by diagonal matrix and the globally optimal forward-prediction filtering algorithm with removing old measurement, a novel IMM algorithm is presented. The novel algorithm has not those drawbacks mentioned above. Simulation shows the feasibility and superiority of the new algorithm.