Paralleling gauss particle filtering method based on quasi-Monte Carlo sampling

Abstract

The invention discloses a method for filtering parallel Gauss particles based on quasi-Monte Carlo sampling, belongs to the technical field of signal processing, and mainly solves the problem that the prior filtering method based on the quasi-Monte Carlo sampling reduces performance for estimating state of a nonlinear dynamic system when the sample number is less. The filtering method comprises: a generating step of a parallel quasi-Monte Carlo sequence and a filtering step of the parallel Gauss particles, namely adopting a generating method for jumping the quasi-Monte Carlo sequence first, and generating a P quasi-Monte Carlo random sample sequences required by subsequent filtering in parallel; then converting each quasi-Monte Carlo random sample sequence into a quasi-Gauss sample sequence obeying designated Gaussian distribution through P execution units; at the same time, updating the time and the state for the state of the nonlinear dynamic system; and finally carrying out comprehensive treatment on all the execution units to finish filtering the nonlinear dynamic system. The method has the advantages of high filtering performance and good real-time property, and can be used in the field of signal processing, automation and artificial intelligence.

Description

technical field [0001] The invention belongs to the technical field of signal processing and relates to nonlinear filtering, in particular to a parallel Gaussian particle filtering method for state estimation of nonlinear dynamic systems. Background technique [0002] Nonlinear filtering techniques are widely used in many fields such as signal processing, automation, computer vision, and artificial intelligence. Extended Kalman filter EKF is a classical method of nonlinear filtering. However, since this extended Kalman filter replaces the nonlinear function with its second-order Taylor expansion, it will generate approximation errors and low precision, which will easily cause filter divergence in application. Unscented Kalman filter UKF is another method of nonlinear filtering. Unscented Kalman filter does not use a linearization method to approximate a nonlinear function, but uses a limited number of sample points that can fully describe the mean and variance of the state ...

AI Technical Summary

Problems solved by technology

The disadvantage of this basic particle filter method is that there is particle degradation, which is generally improved by an algorithm called "resampling", but this kind of resampling not only takes up a lot of computer resources but is not easy to implement in parallel

[0005] Since both the basic particle filter and the Gaussian particle filter are based on Monte Carlo sampling, the sampled particles tend to form "gaps and clusters" in the state space, resulting in a decrease in algorithm estimation performance

Although this phenomenon can be slowed down by increasing the number of particles, it also increases the complexity of the algorithm

Although the Gaussian particle filter avoids the necessary resampling of the basic particle filter and improves the computational efficiency, the problem of estimation performance degradation caused by Monte Carlo sampling still exists

Images