GPS/DR车辆组合导航改进的粒子滤波算法研究

摘要

粒子滤波是一种基于Monte Carlo仿真的最优回归贝叶斯滤波算法,在组合导航系统的观测精度较低时能获得较好的滤波效果,但在观测精度较高时,不但可能导致滤波发散,而且存在重要性分布函数难以选取,出现粒子退化的现象.为了克服这些缺点,文章研究GPS/DR车辆组合导航改进的粒子滤波算法,提出了基于改进粒子滤波算法的GPS/DR车辆组合导航信息融合技术.采用马尔科夫链蒙特卡洛(MCMC)移动方法,移动粒子样本到状态空间中的新位置,既保证了移动后的粒子样本和实际概率函数同分布,又防止了大量后选粒子被拒绝.用改进的粒子滤波算法和扩展Kalman滤波算法,分别对GPS/DR车辆组合导航系统进行仿真实验,结果表明,改进的粒子滤波算法能减小导航定位误差,滤波性能明显优于扩展卡尔曼滤波.%Particle filtering is effective but it diverges and causes degeneration when the measurement precision is high, and it is difficult to select the importance distribution function.We present an improved particle filtering algorithm for GPS/DR (global positioning system/dead-reckoning) vehicle integrated navigation information fusion to overcome the above-mentioned shortcomings.By using Markov chain Monte Carlo (MCMC) method, particle samples move to new location in the state space.Regularized particle filtering is applied to generating a new set of particles and extract particles from the original particles to form a new particle trajectory.Then the MH (Metropolis-Hastings) rules are used to determine whether or not to accept the new trajectory.The new set of particles has the same distribution as the actual probability distribution function and it prevents a large number of particles to be rejected.The simulation results show that the improved particle filtering can reduce the errors of navigation position based on GPS/DR vehicle integrated navigation, and outperform the extended Kalman filtering in terms of accuracy.