粒子滤波算法及其在惯性导航系统中的应用

【作者】 张共愿；

【导师】 赵忠；

【作者基本信息】 西北工业大学 ， 控制理论与控制工程， 2007， 硕士

【摘要】 初始对准是惯性导航的一项关键技术，对准的精度直接关系到惯导系统的工作性能。目前用于初始对准的滤波方法主要是卡尔曼滤波，这要求初始失准角为小角度、噪声服从或近似服从高斯分布。然而惯导系统常常工作在非常恶劣的环境中，尤其对于捷联惯导系统而言，初始失准角往往比较大，同时又受到外部各种因素的干扰，建立在小失准角情况下的线性对准模型已经不能准确地描述该对准系统的误差传播特性，卡尔曼滤波算法应用在这种情况下会常常失效。因此，本文主要是以捷联惯导系统的初始对准为例，来研究大失准角情况下的非线性对准技术和非线性滤波方法。 为了研究捷联惯导系统非线性初始对准技术，本文首先对一种称作粒子滤波的非线性滤波方法进行了深入的研究分析，详细地介绍了算法的基本原理和关键步骤，并明确地指出了该算法在实际应用中存在的根本缺陷——退化现象、样本贫化现象以及超大的计算量，然后从解决这些根本问题着手，本文用了很多策略对粒子滤波算法进行了改进。 为减小粒子退化现象，本文将传统高斯近似非线性滤波方法引入到粒子滤波中来产生更好的重要性密度函数；为克服样本贫化问题，本文借鉴于遗传粒子滤波(GPF)的思想将智能优化算法中的一些优化策略引入到重采样过程当中，提出了基于粒子群优化的粒子滤波算法(PSOPF)和基于退火策略的粒子滤波算法(APF)，使得粒子集在保证优良性的前提下不失去多样性；为克服计算量大的缺陷，本文将传统非线性滤波方法和粒子滤波合并起来形成混合滤波策略，在保证了一定精度的条件下使得粒子滤波计算量迅速减小。 紧接着本文在前人所做研究基础之上，针对初始方位为大失准角情况下的捷联惯导系统，通过推导捷联惯导系统的误差方程，并考虑陀螺仪的随机漂移和加速度计的随机偏差，建立了非线性对准模型。最后，本文将传统的卡尔曼滤波算法、粒子滤波基本算法以及粒子滤波改进算法同时应用于该非线性对准仿真当中，通过粒子滤波算法本身的纵向比较以及各种算法之间的横向比较表明，本文所提出的几种改进算法都有着比传统卡尔曼滤波和基本粒子滤波算法更好的估计性能。更多还原

【Abstract】 As a crucial technology of INS (Inertia Navigation System), the initial alignment accuracy concerns the working performance of INS directly. The Kalman filtering is a primary method in initial alignment, which requires small misalignment angles and Gaussian noises. However, the INS always works in rugged environment. Especially to SINS (Strap-down Inertia Navigation System), which often suffer interferences from various kinds of factors, the initial heading error is relatively large. The linear alignment model based on small misalignment angles cannot describe the system error characteristic exactly, and the Kalman filtering always fails in this situation. So, it is necessary to study the nonlinear filtering algorithm and nonlinear alignment technology under the large heading error.In order to study the nonlinear alignment technology of SINS, a nonlinear filtering algorithm, which is called particle filter, is discussed firstly in this thesis. The fundamental principle and committed steps are introduced in detail, and the underlying flaws are pointed out, which include degeneration, sample impoverishment and the high computing complexity. Then we adopt many ways to improve the particle filter. We use traditional Gaussian approximations to produce the better importance function so as to decrease the phenomenon of degeneration. We adopt some intelligent optimization approaches and propose two algorithms which is called PSOPF and APF to guarantee the sample set diversity so as to overcome the phenomenon of sample impoverishment. We combine traditional nonlinear filtering algorithm with particle filter and form mixed filtering approach to decrease the computing complexity rapidly without reducing the accuracy of estimation.After that, by deriving the initial alignment errors equation of SINS and considering the gyro stochastic drift and the accelerometer stochastic bias, this thesis gives the nonlinear alignment model on the condition of the heading error being large. Finally, we applied standard Kalman filtering, generic particle filtering algorithm and the improved particle filtering algorithm to the simulation of nonlinear alignment above-mentioned. From the comparison with each other in several ways, it is clear that these improved particle filtering algorithms have better performances than更多还原