华南理工大学学报(自然科学版) ›› 2023, Vol. 51 ›› Issue (3): 133-145.doi: 10.12141/j.issn.1000-565X.220368

所属专题: 2023年电子、通信与自动控制

• 电子、通信与自动控制 • 上一篇    下一篇

基于差分算法优化的自复位粒子滤波算法

文尚胜 丘志强 许函铭 陈贤东   

  1. 华南理工大学 材料科学与工程学院,广东 广州 510640
  • 收稿日期:2022-06-10 出版日期:2023-03-25 发布日期:2022-11-08
  • 通信作者: 文尚胜(1964-),男,博士,教授,主要从事可见光定位、信号处理、LED及OLED发光器件研究。 E-mail:shshwen@scut.edu.cn
  • 作者简介:文尚胜(1964-),男,博士,教授,主要从事可见光定位、信号处理、LED及OLED发光器件研究。
  • 基金资助:
    广东省科技计划项目(2017B010114001);教育部科技计划项目(CXZJHZ201813)

Self-Reset Particle Filter Method Optimized Based on Differential Evolution Algorithm

WEN Shangsheng QIU Zhiqiang XU Hanming CHEN Xiandong   

  1. School of Material Science and Engineering,South China University of Technology,Guangzhou 510640,Guangdong,China
  • Received:2022-06-10 Online:2023-03-25 Published:2022-11-08
  • Contact: 文尚胜(1964-),男,博士,教授,主要从事可见光定位、信号处理、LED及OLED发光器件研究。 E-mail:shshwen@scut.edu.cn
  • About author:文尚胜(1964-),男,博士,教授,主要从事可见光定位、信号处理、LED及OLED发光器件研究。
  • Supported by:
    the Science and Technology Planning Project of Guangdong Province(2017B010114001);the Science and Technology Project of the Ministry of Education(CXZJHZ201813)

摘要:

粒子滤波器作为常用的非高斯非线性的滤波方法,已成功地应用于各种工程领域。然而传统的重采样方法导致了粒子贫化的问题,严重降低了滤波估计的精度与鲁棒性。文中提出一种结合跟踪失败检测与改进差分优化融合的自复位粒子滤波方法。首先通过跟踪失败识别方法对滤波估计值进行初步检验,在正常跟踪时不启用优化策略,算法性能与标准粒子滤波无异;在跟踪失败时,通过差分算法对粒子集进行复位,复位过程中设置了粒子置信区间的上下界以防粒子过度集中,并结合检验指示值规避对粒子的多次优化,以缩短算法的估计时间。仿真结果表明,文中算法通过动态调节方式继承了标准粒子滤波和差分进化粒子滤波的优点,有效提高了滤波估计的鲁棒性与估计精度,可在滤波成功时避免启用优化策略以降低算法的整体时间复杂度,并在滤波失败时启用差分优化策略进行自我复位以提高算法估计精度;且在相同定位精度下,其所需粒子数较标准粒子滤波更少,整体时耗较差分进化粒子滤波更低,在建模不确定时也可表现出良好的效果。

关键词: 粒子滤波, 差分进化算法, 自适应, 非线性滤波, 自复位

Abstract:

As a commonly used non-Gaussian nonlinear filtering method, particle filter has been successfully applied in various engineering fields. However, the traditional resampling method leads to the problem of particle depletion, which seriously reduces the accuracy and robustness of the filter estimation. This paper proposed a self-reset particle filter method that combines tracking failure detection and enhanced differential evolution optimization. Firstly, the filter estimation value is preliminarily checked by the tracking failure identification method, and the optimization strategy is not enabled during normal tracking, and the algorithm performance is consistent with the standard particle filter. When the tracking fails, the particle set is reset by differential optimization. During the reset process, the upper and lower bounds of particle confidence interval are set to prevent the particles from being over-concentrated, and the multiple optimization of the particles is avoided by combining the test indication value to reduce the estimation time of the algorithm. The simulation results show that the proposed algorithm inherits the advantages of standard particle filter and differential evolution particle filter through dynamic adjustment, and it effectively improves the robustness and estimation accuracy of the filter estimation. It can avoid using the optimization strategy to reduce the overall time complexity of the algorithm when the filter is successful, and enable the differential optimization strategy to self-reset when the filter fails. In addition, under the same positioning accuracy, the number of particles required by the algorithm is lower than that of standard particle filter, and the overall time consumption is lower than differential evolution particle filter, which also works well when modeling is uncertain.

Key words: particle filter, differential optimization algorithm, self-adaption, nonlinear filter, self-reset

中图分类号: