Dead Reckoning of Unmanned Underwater Vehicle Based on Particle Filtering with Variance Reduction
Zhenye Liu · Information and Computation · 2013
If the system is nonlinear and the noise is non-Gaussian,the navigation accuracy of dead reckoning(DR) based on extended Kalman filtering(EKF) decreases seriously.In order to avoid it,a new dead reckoning based on particle filtering with variance reduction of weight is presented.The non-linear kinematic model of unmanned underwater vehicle(UUV) and measurement models of sensors are formulated.The variance of particles’ weights are reduced with an adaptive exponential fading factor produced by cooling function in the simulated annealing algorithm,and thus the number of particles is increased.The previous method is used to replace resampling procedure in standard particle filtering algorithm.Simulation results with trial data show that compared with EKF based dead reckoning,the proposed method can avoid the influence of model linearization and non-Gaussian noise,and compared with particle filtering based dead reckoning,it reduces the degree of the particles impoverishment due to the resampling,and eventually enhances the stability and accuracy of UUV’s navigation system.