Application Research Of Particle Filter In The Integrated Navigation Of UAV

The method of enhancing the precision, reliability and cost-effective of navigation system by means of information fusion, is always the hot issue of UAV technologies. Because of the strong point of smaller volume, low cost and having complete observation information, the UAV navigation system contain MEMS inertial measurement unit(MIMU), global navigation satellite system(GNSS) receiver and magnetometer is very often used. Due to the optimal estimation problem of this kind of navigation system is essentially nonlinear and non-Gauss, the traditional integrated navigation algorithms do no good to enhance the precision and reliability of the navigation system. Therefore, this paper brings us a kind of better particle filter(PF) method, and carries out in-depth work. The key points are as follows,(1) The general idea about principles and algorithms of Kalman filter(KF), extended Kalman filter(EKF), unscented Kalman filter(UKF) and cubature Kalman filtering(CKF). By compare them with particle filter from the point of approximation of posterior probability density distribution, this paper makes the advantages of particle filter in non-linear, non-Gauss situation clear.(2) Content of the mechanization of the inertial navigation. Gives the measuring principles of the GNSS, MIMU and magnetometer. Deduces the state and observation equations.(3) Analyses the causes why the particle filter is difficult to apply to this kind of integrated navigation systems. On this basis, this paper introduces the general hybrid algorithms by combining nonlinear Kalman filter and particle filter. A new hybrid particle filter algorithm comes out. The new algorithm takes the system state transition distribution as the basis of importance sampling, and eliminates the degenerate particles by means of the importance resampling(SIR), and then, replace them by new particles which are generated according to navigation parameters estimation and covariance of an EKF filter and likelihood ratio of particles, during resampling, meanwhile guaranteeing the diversity and the effectiveness of particles. Different from the general hybrid particle filter algorithms, the new algorithm don’t need nonlinear Kalman filters correspond to each particle to get the proposal distribution, the complexity of the computation reduces tensly.(4) Analyses the performance of the improved algorithm by mathematical simulation under typical non-Gauss and nonlinear conditions.