粒子滤波算法研究与实现

发布时间：2018-02-01 10:09

本文关键词： 粒子滤波重要性函数粒子蜕化自适应重采样电路设计　出处：《电子科技大学》2015年硕士论文　论文类型：学位论文

【摘要】：近几十年发展起来的一种基于蒙特卡罗思想实现非线性、非高斯系统滤波的粒子滤波算法,完全突破了传统Kalman滤波理论框架,适用于任何能用状态空间模型表示的非线性系统,且精度可以逼近最优估计。粒子滤波是统计模拟理论学科和现代信号处理之间的交叉学科,因此对其研究有着重要的理论意义和深远的实践价值。其在雷达目标跟踪、语音信号增强、传感器故障诊断、倒立摆动控制系统、卫星导航、经济学以及生物控制等领域都有着广泛应用前景。本文对粒子滤波的发展背景和意义做了概述,由状态方程出发,通过贝叶斯理论和蒙特卡洛采样推导出粒子滤波的基本原理。根据标准粒子滤波分析现有粒子滤波存在着估计精度不高、粒子蜕化、重要函数难选择、计算量大以及实时性差等缺点。本文结合扩展卡尔曼粒子滤波和不敏卡尔曼粒子滤波结构,利用Gauss-Hermite卡尔曼作为重要函数的Gauss-Hermite卡尔曼粒子滤波算法。其算法复杂度低于不敏卡尔曼粒子滤波,而其性能则高于扩展卡尔曼粒子滤波,它为粒子滤波的重要性函数提供了多一种选择。研究了四种基本重采样算法并把四种算法仿真比较,并对其不能完全抛弃较小粒子提出了改进思路。根据自适应重采样算法,改进得到了线性自适应重采样算法,这种算法虽然要牺牲估计精度却非常高效的在硬件上实现。通过权值优选思想,结合自适应重采样算法对其算法时间上进行改进,避免了大量数据的排序的问题,降低其算法复杂度,但其性能接近原来的权值优选算法。最后研究了粒子滤波的电路设计,分析了粒子滤波电路设计存在着较大的计算量和实时性差的问题。通过避免权值归一化处理,从而大大减少除法运算。并通过电路的并行设计来解决实时性差的问题。电路设计中有效地把粒子电路设计分为独立的采样电路、权值计算电路设计和重采样电路设计。通过粒子滤波电路设计得到电路设计主要误差存在于指数运算,实时性差的问题主要存在于重采样算法中。对于如何提高粒子滤波硬件设计的性能提供了依据。
[Abstract]:In recent decades, a kind of particle filter algorithm based on Monte Carlo theory to realize nonlinear, non-#china_person0# system filter has completely broken through the traditional Kalman filter theoretical framework. It is suitable for any nonlinear system which can be represented by a state space model, and the precision can approach the optimal estimation. Particle filter is an interdisciplinary subject between statistical simulation theory and modern signal processing. Therefore, its research has important theoretical significance and far-reaching practical value. It is used in radar target tracking, speech signal enhancement, sensor fault diagnosis, inverted swing control system, satellite navigation. In this paper, the development background and significance of particle filter are summarized, starting from the equation of state. The basic principle of particle filter is deduced by Bayesian theory and Monte Carlo sampling. According to the standard particle filter analysis, the existing particle filter has low estimation accuracy, particle disintegration, and important functions are difficult to choose. This paper combines the extended Kalman particle filter and the unsensitive Kalman particle filter structure. The complexity of Gauss-Hermite Kalman particle filter algorithm is lower than that of unsensitive Kalman particle filter algorithm. Its performance is better than that of extended Kalman particle filter, which provides an additional choice for the importance function of particle filter. Four basic resampling algorithms are studied and compared with each other. An improved method is put forward for it can not completely abandon smaller particles. According to the adaptive resampling algorithm, the linear adaptive resampling algorithm is obtained. This algorithm is implemented in hardware efficiently at the expense of estimation accuracy. Through the idea of weight optimization, the algorithm is improved in time by combining with adaptive resampling algorithm. The algorithm complexity is reduced, but its performance is close to the original weight optimization algorithm. Finally, the circuit design of particle filter is studied. In this paper, the problems of large computational complexity and poor real-time performance in the design of particle filter circuit are analyzed, and the normalized processing of weight value is avoided. Thus greatly reduce the division operation, and through the parallel design of the circuit to solve the problem of poor real-time. In the circuit design, the particle circuit design is effectively divided into independent sampling circuit. Weight calculation circuit design and resampling circuit design. Through particle filter circuit design, the main error of circuit design exists in exponential operation. The problem of poor real-time performance is mainly found in the resampling algorithm, which provides a basis for improving the performance of particle filter hardware design.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN713

【参考文献】