Researches On Particle Filtering Algorithms And Application In GPS/DR Integrated Navigation

This paper studies the improved particle filtering methods and their applications in GPS/DR integrated navigation system. The performance of the computing method of the standard particle filtering is described in detail, and their defects in practical application are discussed. Combined with the practice of the data processing in GPS/DR integrated navigation system, the standard particle filtering method is improved with multi-kinds of strategies to overcome the shortcomings. The main work and contributions of this article are summarized as follows:1. Make a detailed analysis on the performance of the particle filtering algorithm and analyze the cause of defects exsting in the particle filtering, such as the degeneracy of the phenomenon of particle weight, particle impoverishment phenomenon and the large amount of calculation, etc. Qualitatively analyze the impact of the size of the measurement noise on the state estimation performance of the particle filtering algorithm. Based on the application background of data processing of the GPS/DR integrated navigation, four kinds of basic re-sampling appoaches are discussed from the point of the influence to the state estimation performance of the particle filtering algorithm. Many conclusions are obtained, which points out the studying direction for designing new particle filtering algorithms.2. The determinations of the nonlinear degree of the dynamic nonlinear model and the state estimation performance of nonlinear filtering methods are given respectively from the points of the generalized mean-square error and the differential geometry. For the relationship between the nonlinear degree of the nonlinear filtering model and the estimation performance of nonlinear filtering calculating method, the author's own opinions are given.3. A particle filtering algorithm based on the adaptive iterative re-sampling is proposed. This method can alleviate the particle impoverishment, and at the same time not only guarantees the state estimation performance, but also reduces the computational capacity.4. A particle filtering algorithm is designed based on the adaptive fading extended Kalman filtering. Since the proposal distribution function of the proposed algorithm is introduced by the on-line adjustment factor, which makes the latest observation data be in an important position compared with the proposal distribution function of the usual extended Kalman filtering, and greatly enhance the self-adaptability, and can get more accurate forecast particles, thereby improve the state estimation performance of the corresponding particle filtering method.5. This paper presents a criterion based on Maximum Kullback-Leibler Distance (abbreviated as MKLD) and a new algorithm named by PF-AMCMC. This algorithm can adaptively choose the number of particles and at the same time select the implementation moment of MCMC moving, and reduce the computational complexity under the conditions of guaranteeing the accuracy of state estimation.6. The RTS fixed-interval smoothing algorithm is introduced to construct two kinds of smoothing proposal distribution function– the extended RTS smoothing proposal distribution function and the Unscented RTS smoothing proposal distribution function, and the corresponding calculating method of particle filtering is designed, PF-ERTS and PF-URTS, respectively. And the state estimation performance is better than the estimated effects with the EKF and UKF proposal distribution function.7. A particle swarm optimization particle filtering method based on the criteria of MKLD is brought forward in this paper. This method embeds the particle swarm optimization algorithm into the important sampling process of the particle filtering method, to optimize the sampling process and improve the fine collection of particles while maintaining the state estimation performance of the particle filtering method. At the same time, in order to reduce the computational complexity, the new algorithm adaptively selects the optimized particles and the implementation moment of the particle swarm optimization based on the criteria of MKLD.8. A particle filtering method is proposed based on the mean-shift algorithm. This approach integrates the thought of mean-shift method into the importance sampling process of particle filtering, approximates the true state distribution by means of the particle clustering of the mean-shift algorithm, and thus achieves good estimation results and improves the status of real time by requiring only a small number of particles compared with the standard PF algorithm on overcoming the defects, such as the degeneracy of the phenomenon of particle weight, particle impoverishment phenomenon and the large amount of calculation.The results of a large amount of computational experiments and the GPS / DR integrated navigation simulation experiment show that the several improved particle filtering methods proposed in this paper have better performances of state estimation.