Improved Gaussian particle filter data fusion algorithm based on KLD sampling

Abstract

The invention provides an improved Gaussian particle filter algorithm (KLD-GPF) based on KLD, belongs to the technical field of signal processing, and relates to nonlinear filtering, and the method provided by the invention is suitable for state estimation of a nonlinear dynamic system. The algorithm can adaptively adjust the number of particles, and has a remarkable effect under the conditions that noise obeys Gaussian distribution and the statistical property of the noise is suddenly changed. According to the filtering algorithm, the Kullback-Leibler (KL) distance between a discrete probability density function (PDF) of particles and a real posterior probability density function is calculated online in the sampling process, and the size of a particle set is adjusted online according to the KL distance, so that the algorithm has relatively good robustness. The KLD-GPF can maintain a good estimation effect under the condition of sudden change of noise statistical characteristics. Compared with a KLD improved particle filtering algorithm (KLD-PF), although some filtering precision is lost, the filtering speed is greatly improved.

Description

technical field [0001] An improved Gaussian particle filter algorithm based on KLD proposed by the invention belongs to the technical field of signal processing and relates to nonlinear filtering. The method provided by the invention is suitable for state estimation of nonlinear dynamic systems. Background technique [0002] Nonlinear filtering problems arise in many areas, including object tracking, strapdown inertial navigation systems, and attitude estimation. The Extended Kalman Filter EKF (Extend Kalman Filter) is to linearize the nonlinear function and directly truncate the high-order items, resulting in large errors and low filtering accuracy. Unscented Kalman filter UKF is a Kalman filter using unscented transformation. Compared with EKF, its filtering accuracy has been improved, but its nonlinear transfer error always exists. Since the extended Kalman filter and the unscented Kalman filter are both based on the improvement of the Kalman filter, and the Kalman filte...

AI Technical Summary

Problems solved by technology

[0003] Aiming at the above-mentioned problems and deficiencies, the patent of the present invention proposes an improved Gaussian particle filter data fusion algorithm based on KLD (Kullback-Leibler Divergence). During the sampling process of the algorithm, the discrete probability density function PDF and the true posterior probability density of particles are calculated online. The KL distance of the function, and adjust the size of the particle set online according to the KL distance, in the case of sudden changes in the statistical characteristics of the noise, it can maintain a good estimation effect

Images