An intelligent particle filter with resampling of multi-population cooperation

The particle filter (PF) has excellent estimation performance for nonlinear non-Gaussian systems. However, this method misleads the results due to sample impoverishment and the sensitivity on the initial values of state and process noise variance. To overcome this, an intelligent PF method with a resampling of multi-population cooperation is presented in this paper. Firstly, an intelligent resampling mechanism based on a circular collaborative structure is proposed. In this mechanism, the particles are divided into multiple populations, which are generated by importance densities with different initial state values and variances individually. Secondly, a collaborative strategy based on Gaussian mutation is designed to improve particle diversity so as to raise estimation accuracy of PF. This strategy reserves the high-weight particles in each population, and replaces the low-weight ones with the particles generated by Gaussian mutation on high-weight particles in the previous population based on the circular collaborative structure. Finally, three systems are introduced to test the performance of the proposed method. The results illustrate that the proposed method can effectively improve the estimation accuracy and robustness of PF when the initial values of state and variance of process noise are unknown compared with three existing methods.