K-means entropy algorithm for multiple model estimation

The minimum entropy multiple model estimation algorithm (MEMM), one of variable structure multiple model estimators (VSMM), is an effective approach in handling the problems with high mode uncertainty. However, the performance of MEMM will deteriorate when the real observation errors are in disaccord with the prior observation error distributions. To this end, we propose the k-means entropy multiple-model estimation algorithm (KMEMM) to refine the model sequence set adaptation process. First, the k-means algorithm is employed to make several model sequence clusters. Second, the minimum entropy cluster is selected as the best model set and at last the Bayesian estimation is calculated based on it. The simulation results demonstrate the efficiency of the proposed algorithm through comparing to several existing algorithms.