针对传统粒子滤波(PF)没有引入当前信息,并存在粒子退化的问题,提出了一种基于序列二次规划(SQP)多级优化的PF 算法。首先,基于残差分布特性采用置信区间剔除较大偏差粒子,调整粒子权值分布;然后,将重采样后的粒子映射到集合U,根据集合U 中各粒子复制次数建立多级优化模型,通过SQP 求解模型的参数值,当前后两级模型优化参数差异小于门限时,输出最后一级优化参数为滤波结果;最后,为防止过度采样导致粒子退化,利用滤波值及其协方差采样新粒子。仿真实验表明:SQP鄄PF 算法在跟踪精度,粒子多样性方面优于传统PF 算法。

Aiming to the problem that the traditional particle filter does not introduce the current information and the existence of particle degradation, a novel particle filter algorithm based on sequential quadratic programming ( SQP) multilevel optimization is proposed. Firstly, based on the characteristic of the residual distribution, a confidence interval is given and used to eliminate the large deviation particles to adjust the particle weights distribution. Then, the re-sampling particle is mapped to set U and a multi-level optimization model is established based on the copy number of each particle, the state parameters of the model are passed out through SQP, the latter optimized value is considered to be the filtering result when the difference of model optimization parameters between before and after stages is less than the threshold. Finally, in order to prevent excessive sampling leading to particle degradation, the filter value and its covariance are used to sample new particles. Simulation results show that the SQP-PF algorithm is superior to the traditional PF algorithm in tracking precision and particle diversity.

TN957

国家自然科学基金青年基金资助项目