基于新的采样更新方法的粒子滤波算法

以往的粒子滤波采用由初始先验概率密度产生一组粒子,然后通过重要性密度函数去更新粒子,但会产生粒子退化的问题,因此引入了各种各样的重采样算法,但这样做又产生了粒子多样性丧失的问题。针对粒子滤波的粒子退化现象,提出基于新的采样更新方法的粒子滤波算法,新方法从滤波值和滤波误差协方差矩阵上产生粒子。仿真试验表明,新方法在非线性非高斯情况下要远远好于EKF。

Particle filter algorithm based on a new sampling method

Former particle filters sample a set of particles generated from the initial probability density function and update them through the importance density function,which result in particle degeneracy.Thus various resample algorithms are introduced while these cause the loss of particle diversification.Aiming at the particle degeneracy problem,a different particle filter with a new sample method is proposed,and the particles are sampled based on the filtering results and filtering error covariance matrix.Simulations show that the new filter is far better than EKF in nonlinear and non-Gaussian systems.