观测值含有粗差的抗差粒子滤波算法

Dealing with the measurement outlier by robust particle filter

作为非线性滤波的代表,粒子滤波得到广泛应用。该算法通过随机生成的具有权重的样本(粒子)来计算后验概率密度。其中赋权的过程需要综合系统信息和观测信息,当观测值含有粗差时,会使粒子错误赋权影响滤波结果。提出一种新的基于抗差估计的抗差粒子滤波算法,通过计算粒子等价权,抑制观测值粗差的影响。模拟计算分析表明,当观测值含有粗差时,与标准粒子滤波相比,该方法能有效提高滤波精度。

Particle filter has been applied in many fields. As a recursive Bayesian estimation, particle filter uses a set of samples with weights to representing the conditional probability density function. The importance weights mix the states information and the observations, which will be influenced by outliers. A new approach, which is called robust particle filter, has been put forward based on the robust estimation. By use of the equivalent weight, this method can resist outliers. A numerical example contrast between this method and the traditional way is given. The result shows that the approach can control the influence by observation outliers effectively.