<FONT face=Verdana>For the adverse effect caused by the number decline of particles which are applied to implement the state estimation and model recognition

when model information is introduced into particle sampling process

a novel multiple model particle filter algorithm based on particle optimization is proposed. In the new algorithm

every particle is combined with extended Kalman filter

and the prediction and update mechanism of extended Kalman filter is used to realize the reasonable utilization of the latest observation information. The affectivity of single particle to approximate the real system state and model is improved. The theory analysis and simulation results show the new method outperform obviously the interacting multiple model particle filter and the standard multiple model particle filter in the filter precision of system state and the accuracy of model recognition.