Distributed interactive multi-model estimatation based on partial variational Bayesian inference

In view of the adverse effect of presetting transition probability matrix of motion model on state estimation accuracy in some multi-model algorithms, a new distributed interactive multiple model estimation algorithm based on parital variational Bayesian inference is proposed in this paper. Different from the assumption that the motion model transfer probability matrix is a priori known in the traditional interactive multiple model estimation, in the framework of distributed fusion estimation, the recursive optimization strategy based on minimizing Kullback-Leibler divergence criterion is used to predict and update the motion model transfer probability matrix. On this basis, the joint estimation of target state and model probability at current time is realized by variational Bayesian inference. Finally, the local state estimates’ fusion is completed based on the covariance intersection fusion strategy. The simulation results show that the new algorithm effectively improves the state estimation accuracy of the maneuvering target by adaptively estimating the motion model transition probability matrix and the model probability online. © 2024 South China University of Technology. All rights reserved.