迭代无迹Kalman粒子滤波的建议分布

Particle distribution control for an iterated unscented Kalman particle filter

对非线性非Gauss系统,粒子滤波是一种有效的状态估计方法。粒子滤波的关键是建议分布的选择,好的建议分布会改进粒子贫化和样本耗尽等粒子滤波存在的普遍问题。该文用迭代无迹Kalman滤波产生粒子滤波的建议分布,提出了一种新的粒子滤波算法——迭代无迹Kalman粒子滤波。给出的建议分布将最新的观测融入样本过程并修正该过程,从而改进了滤波性能。数值模拟结果表明,提出的算法与常用的无迹粒子滤波、扩展Kalman粒子滤波相比,具有数值稳定、估计结果精确的优点。

The particle filter method is an effective nonlinear estimation method for nonlinear non-Gaussian systems.One key issue with particle filters is the particle distribution with poor distributions leading to particle impoverishment and sample size degeneracy.Here,an iterated unscented Kalman filter was used to generate the initial particle distribution for the particle filter.The particle distributions integrated the newest observations into the sampling process and modified the process to greatly improve the filter performance.Simulations show that the extended Kalman particle filter has superior stability and accuracy than the widely used unscented particle filter.