Distributed Consistency Estimation In Sensor Networks Based On Two-stage Filterin

In sensor network systems,the research of consensus estimation algorithms is one of the hot spots in recent years.The consensus strategy is that each sensor and neighbor node exchange information with each other,and over time,this information will spread throughout the entire sensor network,greatly improving the estimation accuracy of the sensor network,reducing the loss of communication energy.Strong adaptability when new sensors are added or a sensor fails.In this paper,we study the problem of distributed consensus filter for multisensor networked systems.The main research content is as follows:1.For multisensor networked systems with uncorrelated noise,a distributed consensus filtering algorithm with a two-stage filter structure is proposed at each sensor node.At each moment,the first filtering step is: each sensor node performs local filtering based on its own observation data;The second stage of filtering is to exchange estimates between neighboring sensor nodes,and then apply the linear unbiased minimum variance matrix weighted optimal fusion estimation algorithm for fusion estimation.By analogy,after multiple information exchanges and fusion,the final fusion estimation accuracy of each sensor node tends to be consensus.The proposed matrix weighted optimal distributed consensus filtering algorithm requires the calculation of the fusion estimation error cross-covariance matrix for each fusion.To avoid the calculation of the cross-covariance matrix,three suboptimal distributed consensus filtering algorithms are proposed,and their accuracy is compared.The proposed optimal and suboptimal distributed estimation algorithms have good consensus.2.For multisensor networked systems with correlated noise,a distributed consensus filtering algorithm with a two-stage filter structure is proposed at each sensor node.Similar to the previous distributed filtering structure,the optimal gain matrix is obtained by minimizing the error variance matrix,and the cross-covariance matrix for different fusion times between any two sensor nodes is derived.After multiple data exchanges and fusion estimates,the final fusion estimation accuracy for each sensor node tends to be consensus.In order to reduce computational burden,suboptimal distributed consensus filter algorithms were proposed based on sequential covariance intersection and inverse covariance intersection fusion algorithms.3.In order to avoid the reuse of local estimation of neighbor nodes,a sequential distributed consensus filtering algorithm with two-stage filtering structure is proposed for multi-sensor networks with uncorrelated noise.In the sensor network,the shortest link communication exists between any two sensors.At each moment,the first stage filtering: each sensor node performs local filtering based on its own observed data;The second stage of filtering: the local filtering is exchanged between sensor nodes through the shortest communication link.After each exchange,the local filtering is fused sequentially based on the previous fusion results,and so on.Finally,the fusion estimation tends to be consensus.The proposed algorithm requires the calculation of the mutual covariance matrix between fusion estimation and local estimation.In order to reduce the computational burden,a sub-optimal distributed consensus estimation algorithm based on parallel cross-covariance fusion algorithm is proposed.4.On the basis of the previous research,a distributed consensus filtering algorithm based on prior fusion estimation results is proposed.The difference between the structures in 1,2,and 3 is that each sensor node communicates through the shortest path link to obtain information from each other,until the information from all other sensor nodes is obtained through the exchange,and then performs a fusion estimation to make the final fusion estimation of each sensor node consensus.Corresponding distributed optimal and suboptimal consensus filtering algorithms have been designed.The estimation accuracy has been improved by utilizing prior fusion estimation.Due to only performing fusion estimation once per moment,the computational burden is reduced.