Estimation Fusion for Markovian Jump Linear System via Data Transformation

The work presented here deals with distributed estimation fusion for a Markovian jump linear system (MJLS) via a specific linear transformation of local measurements from. Due to the possible singularity associated with this transformation, a relative likelihood of the system model given the transformed data is defined first. With this, distributed fusion for an MJLS is proposed. With full-rate communication, the distributed fusion and the centralized fusion (CF) have the same performance. To accommodate limited communication bandwidth, an extension to the reduced-rate communication case is also discussed. Two schemes to generate the transformed data are considered. In the first one, local sensors transform raw measurements directly. This is applicable to the situation where local sensors have very poor computational power. In the second scheme, transformed data are recovered from local single-model-based estimates indirectly. It is applicable to the situation in which local estimates are desired and multiple model estimation cannot be afforded. Illustrative numerical results are provided to show the performance of the proposed fusion methods.