Multiple Model Particle Filter based on Two Stage Prediction Update

摘要

Aiming at the particle degeneracy caused by the introduction of model information in particle sampling process, a novel multiple model particle filtering algorithm based on two stage prediction update is proposed. In the multiple model particle filtering framework, the dynamic combination of the prediction and update mechanism of particle filter and Kalman filter is realized by the reasonable arrangement of the following four steps including importance sampling, one-step prediction, re-sampling and observation update. And the filter gain calculated by one-step prediction and observation update mechanism of Kalman filter, is used to directly optimize state estimation and avoids the loss of the latest observation and original particle information in filtering process. In addition, a new promoting strategy of particles diversity is given to resolve particles impoverishments by means of the current state estimation. The theoretical analysis and experimental results show that the filtering precision is improved significantly with appropriately increasing computational burden.