A minimum entropy approach for multiple-model estimation

Multiple-model (MM) methods are effective in handling mode uncertainties and the variable structure multiple-model (VSMM) approach is one technique of the state of art. However, designing a better model set adaptive (MSA) mechanism is still a challenging problem both in theory and practice. In this paper, we present new theoretical analysis to evaluate the quality of model sequence sets in a nested structure which implies a principle - to find effective model sequence sets with the smallest size if the risk of missing real modes is small enough. A minimum entropy multiple-model (MEMM) approach is proposed to calculate the model sets with minimum Shannon entropy through feed back the online information into the fusion center. Sub-optimal MEMM algorithms are also designed with a particle filter. An example of maneuvering target tracking is considered in simulations. The proposed algorithms are compared to several existing VSMM algorithms and the results show that the MEMM algorithms are robust and effective in both estimation precision and converging rate.