Multiple-Model Estimation with Variable Structure - Part VI: Expected-Mode Augmentation

this paper. Assume for simplicity of presentation that the IMM mechanism is used for model-conditioned reinitialization [23]. The proposed EMA algorithms involve the following main functional modules: 1) EMA M k := M ): expected-mode augmentation. 2) VSIMM[M k , M k1 ]: recursion for variablestructure IMM (VSIMM) estimation that uses This is the case whenever the mode space is continuous, although there are problems in which different m j represent different physical quantities and thus their weighted sum is not necessarily meaningful. TABLE I One Cycle of EMA Algorithm A S2. For M k = E k ), run VSIMM[M , M k1 ] to obtain the overall estimates, error covariances, and model probabilities TABLE I I One Cycle of EMA Algorithm B k1 ,runVSIMM[M f , M k1 ] to obtain S2. Obtain E k = E(M ; M S3. Run VSIMM[E , M ] to obtain S4. Run EF[M in the set M k =M TABLE III One Cycle of EMA Algorithm C kjk1 g S2. For M 0 k = E 0 k k1 ), run VSIMM[M k , M k1 ] S3. Obtain E k = E(M 0 k ; M 2M k S4. Run VSIMM[E k , M k1 ] k1 ,runEF[M in the set M =M model sets M k1 and M k at time k 1andk, respectively. 3) EF[M 0 k , M 00 k ; M k1 ]: estimation fusion of two estimates resulting from VSIMM[M 0 k , M k1 ]and VSIMM[M 00 k , M k1 ] recursions, respectively, where M 0 k and M 00 k are discussed later. The VSIMM and EF functions have been developed, utilized, and documented in several publications on VSMM estimation [17, 29, 26, 25]. For the EMA procedure, a general description has been given above; a more detailed discussion is given nex