Abstract

Particle filtering is one of the most important algorithms for solving state estimation of nonlinear systems and has been widely studied in many fields. However, due to the unknown complex noise in the actual system, its estimation performance is degraded. Moreover, when the number of particles increase, the real-time performance of the algorithm is poor. For these two problems above, this paper proposed a parallel acceleration CRPF (cost-reference particle filter) algorithm based on CUDA (Compute Unified Device Architecture). CRPF does not need known noise statistics in nonlinear system state estimation, which can reduce the influence of unknown noise on state estimation accuracy. Combined with GPU’s (Graphics Processing Unit) multi-thread parallel computing capability, CRPF parallel acceleration can be realized. Since the data association can’t be parallel resampled, all the particles are evenly distributed to multiple blocks, and resampling process can be parallelized by block parallel computing, so as to improve the speed of the algorithm. At the same time, in order to reduce the global particle performance degradation caused by block resampling, the particles with low probability mass in each block are optimized by using a portion of global high-quality particles. Through two sets of simulation experiments, it is proved that the proposed method has improved in estimation accuracy and the real-time performance has been improved significantly, which can provide a new idea for the practical application of nonlinear filtering method.

中文翻译：

---

一种新颖的并行加速CRPF算法

摘要

粒子滤波是解决非线性系统状态估计的最重要算法之一，已在许多领域进行了广泛的研究。然而，由于实际系统中未知的复杂噪声，其估计性能下降。而且，当粒子数量增加时，该算法的实时性能很差。针对上述两个问题，本文提出了一种基于CUDA（Compute Unified Device Architecture）的并行加速CRPF（成本参考粒子滤波器）算法。CRPF在非线性系统状态估计中不需要已知的噪声统计信息，从而可以减少未知噪声对状态估计精度的影响。结合GPU（图形处理单元）的多线程并行计算功能，可以实现CRPF并行加速。由于不能对数据关联进行并行重采样，因此将所有粒子均匀地分布到多个块中，并且可以通过块并行计算对重采样过程进行并行处理，从而提高了算法的速度。同时，为了减少由于块重采样导致的整体粒子性能下降，通过使用一部分全局高质量粒子来优化每个区块中具有低概率质量的粒子。通过两组仿真实验，证明了该方法的估计精度得到了提高，实时性得到了显着提高，这为非线性滤波方法的实际应用提供了新思路。重采样过程可以通过块并行计算实现并行化，从而提高了算法的速度。同时，为了减少由于块重采样导致的整体粒子性能下降，通过使用一部分全局高质量粒子来优化每个区块中具有低概率质量的粒子。通过两组仿真实验，证明了该方法的估计精度得到了提高，实时性得到了显着提高，这为非线性滤波方法的实际应用提供了新思路。重采样过程可以通过块并行计算实现并行化，从而提高了算法的速度。同时，为了减少由于块重采样导致的整体粒子性能下降，通过使用一部分全局高质量粒子来优化每个区块中具有低概率质量的粒子。通过两组仿真实验，证明了该方法的估计精度得到了提高，实时性得到了显着提高，这为非线性滤波方法的实际应用提供了新思路。通过使用一部分全局高质量粒子来优化每个块中具有低概率质量的粒子。通过两组仿真实验，证明了该方法的估计精度得到了提高，实时性得到了显着提高，这为非线性滤波方法的实际应用提供了新思路。通过使用一部分全局高质量粒子来优化每个块中具有低概率质量的粒子。通过两组仿真实验，证明了该方法的估计精度得到了提高，实时性得到了显着提高，这为非线性滤波方法的实际应用提供了新思路。