Research On Distributed Fusion Structure And Information Flow Mechanism

Distributed fusion structure refers to the multi-node fusion structure,which is dispersed in the spatial domain and time domain,to realize information interaction and function cooperation.In the distributed fusion structure,each node is connected to the communication network and performs the target information fusion processing,which realizes the multi-sensor command and control in the area,and greatly improves the intelligence gathering ability.In view of the problem of information redundancy among multiple sensors,based on the information graph,the redundant information between fusion nodes is analyzed,and the redundant information between common nodes is eliminated by Bayesian algorithm.On this basis,the paper studies the sensor management problem by using information theory.The main contents of this paper are as follows:Firstly,several common fusion structure models are briefly introduced,and their advantages and disadvantages are described.In order to solve the correlation problem of input information among different fusion nodes in distributed fusion architecture,the related distributed fusion algorithm and sensor management algorithm based on information theory are introduced emphatically.Secondly,the main forms of the single-connected and multi-connected fusion under distributed fusion structure are discussed in detail.According to the information transfer mechanism among different fusion nodes,the reason of generating redundant information is analyzed,On the basis of this,the Bayesian fusion algorithm is used to decompose the communication information into private and public information for the hierarchical structure without feedback and feedback,and the theory of removing redundant information is put forward.By removing redundant information,the average value of the target state estimation value of each node and the state estimation of each node can be reduced gradually and the convergence can be fast.At the same time,the sensor fusion position error is very small,which is less than one order of magnitude compared with the situation that the redundant information is eliminated.The simulation results show that the system can achieve good tracking performance by removing redundant information.Thirdly,for traditional sensor management algorithms based on covariance control,the expected covariance of the target is easy to be deviated by man-made,and when the target tracking precision is too high,it is easy to cause the objective function has no solution or no optimal solution;The Rényi information divergence compares the degree of approximation of probability density function,thus providing added emphasis on local information.In the sensor management method combined with information theory,by using the characteristics of Rényi information divergence algorithm and covariance algorithm,a multi-sensor management based on Rényi information divergence algorithm and covariance algorithm is proposed.The idea of joint control of "Rényi information divergence " and "covariance" is used to verify each other,and the sensor resource allocation is further completed by the target tracking precision.The simulation results show that the proposed algorithm can improve the tracking accuracy in different scenes,and effectively reduce the sensor switching frequency.Finally,the main work and further research of this thesis are summarized.