论文部分内容阅读
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-PF算法在跟踪精度,粒子多样性方面优于传统PF算法。
Aiming at the problem that the traditional particle filter (PF) does not introduce the current information and has the problem of particle degeneration, a PF algorithm based on sequential quadratic programming (SQP) multi-level optimization is proposed. Firstly, the confidence interval is used to remove the larger deviation particles and the weight distribution is adjusted based on the distribution of the residuals. Then, the resampled particles are mapped to the set U, and the multi-level optimization model is established according to the number of particles copied in the set U. SQP solves the parameter of the model. When the difference between the optimization parameters of the last two levels is less than the threshold, the last level of optimization parameters is output as the filtering result. Finally, to prevent the over-sampling from causing particle degeneration, new particles are sampled using the filtering value and its covariance. Simulation results show that the SQP-PF algorithm is superior to the traditional PF algorithm in tracking accuracy and particle diversity.