一种简化的拟蒙特卡洛-高斯粒子滤波算法

A Simplified Quasi-Monte Carlo Based Gaussian Particle Filter

提出了一种简化的拟蒙特卡洛-高斯粒子滤波(QMC-GPF)算法(SQMC-GPF),以解决将QMC方法应用于GPF时计算复杂度高、运算量大的问题。该算法中,在连续的迭代滤波过程开始之前,首先利用QMC采样产生单位拟高斯分布粒子集,然后用其线性变换产生GPF算法中需要的高斯分布粒子集,省去了重新进行QMC采样步骤。该算法简化了新粒子集的产生过程,减少了运算量和滤波时间,增强了算法的实时性。将粒子滤波算法(PF)、GPF算法、QMC-GPF算法和SQMCGPF算法用于单变量非静态增长模型(UNGM)和二维纯角度跟踪模型(BOT)的仿真结果表明,SQMC-GPF算法的滤波性能与QMC-GPF算法的滤波性能相近,但有更为明显的速度优势,具有重要的实际应用价值。

A simplified Quasi-Monte Carlo Gaussian particle filtering( QMC-GPF)( SQMC-GPF) algorithm is presented to solve the problems of high complexity and large amount of computation that QMC method is faced with when it is applied to GPF. In this algorithm,a basic set of particles which obey unit quasi-Gaussian distribution are pregenerated with the method of QMC before the successive iterative filering process,and then they are converted to the particles needed during iterative filering by means of linear transformation,without the QMC method called again. This algorithm simplifies the generation of new particles,reduces the amount of computation and filtering time,and improves the real-time performance of the QMC-GPF algorithm. Finally,the PF,GPF,QMC-GPF and SQMC-GPF algorithms are simulated with univariate nonstationary growth model( UNGM) and bearing-only tracking model( BOT). Results show that,SQMC-GPF algorithm has much higher filtering speed than QMC-GPF algorithmon on the premise of the same filtering performance,which proves that the former has important practical application value.