一种基于神经网络的粒子滤波算法设计  

Research on Particle Filter Algorithm Based on Neural Network

作　　者：谢世龙 周玉国[1] 刘真 

机构地区：[1]青岛理工大学自动化学院,山东青岛266520

出　　处：《自动化技术与应用》2017年第11期1-4,共4页Techniques of Automation and Applications

摘　　要：粒子退化现象是粒子滤波算法应用中的一个主要问题,标准重采样粒子滤波算法虽然可以有效的解决粒子退化现象,改善粒子滤波性能,但同时也会出现其他问题,如粒子样本的多样性缺失、具有较高权值的粒子多次被统计等等。对此,提出一种基于神经网络的粒子滤波算法研究,将BP神经网络算法和标准粒子滤波算法相结合,调整较小权值的粒子,分裂较大权值的粒子,仿真结果表明,提出的算法是可行的。Degeneracy phenomenon is a main disadvantage to particle filter application, common re-sampling methods can resolve degeneracy phenomenon, improve the performance of particle filter, but it also has other problems, such as the sample impoverishment is deduced and particles with high weights repeatedly statistics. Therefore, a particle filter algorithm based on neural network is proposed. Combine neural network with particle filter algorithm, the degeneracy phenomenon is relieved by weight adjustment and weight division. Simulation results show the feasibility of the proposed algorithm.

关 键 词：BP神经网络 粒子滤波 权值调整 权值分裂 

分 类 号：TP273.4[自动化与计算机技术—检测技术与自动化装置]