Research On Sins Initial Alignment Based On Adaptive Particle Filter Algorithm

Particle Filter (PF for short) is a kind of superior nonlinear filtering algorithm. It has no any restriction on state equation, measurement equation and noise statistical properties so that it has a wider application range compared to traditional nonlinear filters. It breaks through the kalman nonlinear filtering (EKF, UKF, CKF) framework and gives up the ideas of the mean and variance estimation of state variables instead of predicting and updating the particles which is sampled from a posteriori probability density in order to keep close to the true posterior probability density distribution. So it's more close to the essence of optimal state estimation. Particle filter has many advantages, for example high filtering precision, fast convergence rate and so on. It has become the mainstream filtering algorithm that dealing with state filtering and parameter estimation under nonlinear and non-gaussian system. This paper revolves around the particle filter under the background of the strapdown inertial navigation system(SINS) large azimuth misalignment Angle for the following work:First of all, SINS nonlinear error model was deduced based on the concept of euler platform error with large initial misalignment angles. Derive the sequential Monte Carlo resampling particle filter algorithm on the basis of research on bayesian filter. Simulation analysis between PF and UKF are given in a single variable non-stationary incremental model and single station single target tracking.Secondly, in view of the defects of the standard particle filter algorithm, presents improved ideas as follow:Firstly, standard particle filter selects prior probability density function as importance sampling density function without considering the latest time measurement information which makes sample particles rely too heavily on state model. It's easy to cause filter particle degeneration when the likelihood probability density present peak or at the end of the prior probability density function. Therefore put forward to join the latest measurement information in the importance density function of particle filter in order to make it purposefully move to the posterior probability density function. Generate a new group of sampling particles based on mean and variance of EKF, UKF, CKF estimated. Finally,simulation analysis of PF, EKPF, UPF, CPF compartion are given under subsection nonlinear model.Secondly, the number of the particle filter appears to progressional growth in high dimensional state estimation, which leads to calculation delay and poor real-time performance. We put forward to dynamic regulating particle numbers, reduce the amount of calculation. Adaptive technology will be introduced before the resampling. Determine the number of particles which is needed at the next moment based on estimation precision of the signal at present. Simulation analysis under target tracking to compare the filter performance of APF and PF, verify the effectiveness of the adaptive.Thirdly, ACPF, a new filter algorithm, is derived by combination between CPF and APF.Use CKF to design importance density function and improve the efficiency of sampling.Moreover,predict the number of particles which is needed at the next moment based on state estimation precision at present. Adjust the number of particles and reduce the amount of calculation. Finally, prove the validity of the algorithm in the aircraft flight dynamic model.Finally, simulate and analyse ACPF and PF used in the strapdown inertial navigation static base large azimuth misalignment Angle initial alignment.