AUV State Estimation Under Non-ideal Measurement

In the multi-autonomous underwater vehicle(AUV)coordinated navigation system,any AUV can share navigation information,and the system can obtain high positioning accuracy at low cost.In the multi-AUV collaborative navigation,the non-linear filtering methods often applied include extended Kalman filter(EKF),unscented Kalman filter(Unscented Kalman Filter,UKF)and other non-linear filtering methods.However,due to the complexity of the underwater environment,measurement noise is often mixed with outliers,and the probability density distribution of measurement noise is no longer an ideal Gaussian distribution.Traditional nonlinear filtering methods will result in a large estimation accuracy.decline.In this paper,based on particle filtering,it is deeply researched under the condition of non-ideal measurement noise,which can continue to maintain the good navigation accuracy of the AUV state estimation method.First,the recent development of multi-AUV coordinated navigation and its state estimation methods at home and abroad are reviewed,and the research content and structural arrangement of this article are given.The commonly used AUV navigation methods are analyzed,the AUV coordinated navigation method is determined,and the motion model and measurement model of the AUV state estimation are established,which lays the foundation for the follow-up research.Secondly,two multi-AUV state estimation methods for measuring noise under Gaussian distribution are studied,EKF and UKF algorithms,and the corresponding implementation procedures are given respectively.Simulation comparison analysis is carried out through Matlab,and the simulation results show that the two methods are under the nonlinear model.Both methods can be well applied to state estimation in the case of Gaussian distribution,but when the measurement noise is non-Gaussian,the filtering accuracy of EKF and UKF will decrease to a certain extent.Thirdly,in view of the presence of thick-tailed distribution or outliers in the measurement noise,starting from Bayesian theory,two improved methods of particle filter(Particle Filter,PF)are given,the extended particle filter(Extended Particle Filter,EPF)and Unscented Particle Filter(UPF)algorithms.In order to further improve the estimation performance under this measurement condition,the Maximum Correntropy Criterion(MCC)is introduced,and a Maximum Correntropy Unscented Particle Filter(MCCF)is proposed.This method is based on the unscented particle filter,and the method is improved by the maximum entropy MCC criterion.Simulation experiments show that the MCUPF algorithm can also have good robustness in the presence of thick-tailed distribution or outliers in the measurement noise.,Can maintain high filtering accuracy.Finally,the Artificial Fish School Algorithm(AFSA)is introduced to optimize the re-sampling process of PF,and to a certain extent improve the problem of lack of particle diversity caused by the re-sampling of the MCUPF method proposed above.First,the five commonly used resampling methods of particle filtering are analyzed,and the state estimation accuracy and error of the three algorithms of UPF,MCUPF and AFSA optimized MCUPF are compared through Matlab simulation experiments,and the optimized MCUPF filtering method is verified Effectiveness.