论文部分内容阅读
为了解决粒子滤波的粒子退化和粒子多样性丧失问题,提出了一种基于Stiefel流形的粒子滤波算法.该算法将系统模型置于Stiefel流形上,用朗之万分布描述过程转移概率分布,用矩阵正态分布表示似然函数分布,在流形分布上进行粒子采样.在计算加权粒子的均值时,将流形嵌入到欧氏空间中,先计算欧氏空间中的粒子均值,再将计算结果投影到嵌套流形上,这就排除了噪声统计特性对粒子权重方差的影响,得到了一种受系统状态模型限制较少的重要性概率密度函数通用选择方案.仿真时选取单变量非静态增长模型,仿真结果验证了该算法的实时性、鲁棒性,滤波精度和滤波效率均比无味粒子滤波算法更好.
In order to solve the problem of Particle Filtering Particle Degradation and Particle Diversity Loss, a particle filter based on Stiefel manifold is proposed, which puts the system model on Stiefel manifold and describes the process transition probability distribution with Longman distribution, We use the normal distribution of the matrix to represent the distribution of the likelihood function and sample the particles in the manifold distribution. When calculating the mean value of the weighted particles, the manifold is embedded in the Euclidean space, the mean of the particles in the Euclidean space is calculated first, The results are projected onto the nested manifolds, which eliminates the influence of noise statistics on the variance of the weights of the particles. A universal selection scheme of the probability density function, which is less constrained by the system state model, is obtained. Single-variable Non-static growth model, the simulation results verify the real-time performance of the algorithm, robustness, filtering accuracy and filtering efficiency than the tasteless particle filtering algorithm is better.