辅助粒子滤波算法及仿真举例

Auxiliary Particle Filter and Simulation Examples

粒子滤波算法是近年来提出的一种较新的算法.通常的粒子滤波利用采样重要性重抽样算法,该算法选用先验分布,但它易受外部观测量的影响,因而会导致权值变化较大,并且引起较高的蒙特卡罗方差以致会使滤波性能较差.为此,本文引入一个辅助变量,利用一种新的使用二次加权操作的粒子滤波算法———辅助粒子滤波算法来对采样重要性重抽样算法进行改进.最后,通过两个仿真实例一维非线性追踪模型和二维纯方位目标追踪模型,进一步分析指出辅助粒子滤波算法比采样重要性重抽样算法更有效.

Particle filter is a new algorithm proposed in recent years. The method of sampling importance-resampling （SIR） algorithm, which is based on the prior proposal distribution, is generally used in particle filter. Since it can be impacted by outer observations easily which makes the weights distribution unevenly and causes high MC variance, as a result, the evaluated value will be a little worse. Therefore, a new method auxiliary particle filter （APF） is ＇presented in which an assigned variable is brought and the weights change more smoothly with two rounds weighted processes. Finally, the APF is proved to be more efficient with two simulations, a One-dimensional nonlinear tracking model and a two-dimension bearings-only target tracking model.