Particle Filtering Method With Unknown Statistics Of Noise And Its Application

With the development of engineering technology,people have increasingly requirements for the detection accuracy of system state.In the fields of technology development,the system objects are often in a very complex environment and the states of the system are often interfered by various uncertain interferences,these complex random noises are often difficult to be modeled,so,how to estimate the state of the system with unknown statistics of noise becomes an inevitable research difficulty in the signal processing field.As a filtering method,particle filtering applies to nonlinear and non-Gaussian systems,and it has been widely used in recent years.However,particle filtering can obtain accurate estimation results only when the noise statistics of the states are known and accurate.And the particle impoverishment caused by resampling process in particle filtering is also one of the problems that affecting the filtering precision of the method.These two issues are studied in this paper,and the main works are as follows:(1)For the filtering problem with unknown statistics of noise,this paper proposes a set-membership based particle filter methtod.Firstly,we use the set-membership estimation theory to generate an ellipsoid set which contains the true value of the system state and draw particles from a Gaussian distribution in the obtained ellipsoidal set,which has overcome the difficulty of drawing particles under the situation of unknown noise statistics;then,in order to resample particles when the measurement noise statistic is unknown,a cost function is introduced to obtain the weights of the particles,and finally the ultimate estimation of state is calculated by the weighted particles after the resampling procedure.The whole process does not require any priori statistics of the noise.The simulation results with different noises and different models show that the proposed method has higher filtering accuracy than the existing methods.(2)For the particle impoverishment problem caused by resampling process,this paper proposes an intelligent resampling technique based on multi-population cooperation mechanism.The particles are divided into several populations,after completing the importance sampling individually,the particles are divided into several populations,and after the importance sampling is completed,the annular transmission model is used to replace a certain number of particles with small weight in each population with large weight particles in another population,in order to make the distribution of particles more closer to the posterior distribution of the state.Then the chaotic variation with certain probability is added to increase the diversity of particles,and after that the resampling step is performed.Simulation results show that the intelligent method can obtain more even distribution of particles around the posterior probability density,and the average RMSE is smaller.(3)The proposed intelligent resampling technique is introduced into the proposed set membership based particle filtering method in this paper,to estimate the liquid level state from several sets of measurement data of the silicon melt level experiments.And the experimental results demonstrate the superiority of the method compared with the existing methods.