Strategies And Methods Of Fighting Particle Degradation In Nonlinear Systems

In the nonlinear system filtering problem,particle filtering algorithm is widely concerned because it is not constrained by the Gaussian hypothesis of the system and easy to implement.However,the particle degradation and sample impoverishment problems in the particle filtering algorithm seriously affect the filtering performance.Therefore,the optimization and improvement of particle filter algorithm has important theoretical and practical significance.In this paper,the particle degradation in particle filtering algorithm is studied,mainly including the following aspects.1.Aiming at how to solve the problem of particle degradation from the optimal design of importance density function,this paper studies the design method of importance density function based on entropy criterion and extended Kalman filter,and then puts forward the maximum entropy extended particle filter algorithm.Because the measurement information contains outliers or is interfered by non-Gaussian heavy tail noise,the importance sampling guided by the measurement information will lead to faster particle degradation.Therefore,in this paper,the entropy criterion for solving the outliers problem is introduced into the Extended Kalman Filter,and an importance density function containing the higher-order moment information of the signal is generated,which is used to guide the importance sampling to drive the particles to the region with high probability of conditional posterior distribution of the current state.Simulation experiments show that the proposed algorithm can slow down the particle degradation rate to a certain extent,solve the filtering problem of measurement information interfered by outliers and non-Gaussian heavy tail noise,and achieve the improvement of state estimation accuracy in non-Gaussian environment.2.In order to solve the problem of particle degradation based on the optimization of particle distribution,the particle mutation and the particle screening strategy based on the minimization of maximum risk were studied,and then the mutation particle filtering algorithm based on the minimization of maximum risk was proposed.when the likelihood function is located at the tail of the prior distribution or the observation accuracy is high,the particles in the overlap region between the prior distribution and the likelihood function decrease,which leads to the acceleration of particle degradation.Therefore,In this paper,the combination strategy of mutation and screening is applied to the low weight particles in the priori concentration,so as to obtain more high weight particles located in the overlap region of the priori and the likelihood,and then to suppress the particle degradation.Firstly,the unmutated low weight particles in the prior particle set were detected,and the mutation strategy was implemented to obtain the mutated particles.Then set the weight threshold value and implement the screening strategy based on the minimization maximum risk idea for the mutated particles.The combination strategy is repeated until all particles meet the requirements.Simulation results show that the proposed algorithm reduces the variance of a posteriori particle weights,effectively inhibits particle degradation,and improves the estimation accuracy of nonlinear filtering.