Research On Distributed Estimation Algorithm For Networked Systems Based On Data Compressio

With the wide application of networked systems,networked systems are faced with more and more complex problems.In order to reduce the computational burden of distributed filters,firstly,reduce the dimension of the expanded data,and then use the compressed data to design filter.In addition,distributed estimation algorithms have irreplaceable advantages in flexibility and robustness compared to centralized algorithms for sensor networks.This article has studied a distributed estimation algorithm based on data compression dimensionality reduction for multi-sensor networked systems.The main research content is as follows:1.For linear discrete time-varying system with uncorrelated noise,the distributed filtering problem based on data compression has been studied.We have considered two types of sensors in sensor networks: one type has the ability to observe and communicate,and the other type has the ability to observe,calculate,and communicate.Nodes with computing ability use measurement weighted fusion algorithms to compress and reduce the dimensionality of their own and received augmented measurements from neighbor nodes,and use matrix weighted fusion algorithms to compress and reduce the dimensionality of the received augmented estimation data from neighbor nodes.Based on compressed data,a linear unbiased minimum variance optimal distributed filter in Kalman recursive form was designed.The optimal algorithm needs to calculate the crosscovariance matrix between any two estimators.In order to reduce the computational burden,a suboptimal distributed filter is proposed by locally minimizing the upper bound of the filtering error variance to solve for the gain.The exponential mean-square boundedness of the filtering error of the proposed algotithm is analyzed.2.A distributed filtering problem based on data compression was studied for multisensor linear discrete time-varying systems with correlated noise.The compression method is different: nodes with computing power use a weighted least squares fusion algorithm to compress their own observation values,received observation data from neighboring nodes,and predicted estimates together.Based on compressed data,linear unbiased minimum variance optimal and suboptimal distributed filters in Kalman recursive form were designed.The exponential mean-square boundedness of the filtering error of the proposed algotithm is analyzed.3.Regarding the distributed filtering problem of asynchronous sampling systems,the phenomena of asynchronous sampling and random packet loss are considered,making the research more universal.Each sensor node generates local estimates based on the observations and prediction estimates received from neighboring nodes,and then sends its own prediction estimate to the neighbor nodes at the next sampling time.During the transmission process,there is random data loss.At each sensor node,firstly,based on the different estimation accuracy of the received neighbor node estimates,a matrix weighted or scalar weighted fusion method is used to fuse and compress the received neighbor estimates.Based on fused neighbor estimation,a suboptimal distributed filter in the form of Kalman filter is proposed.The exponential mean-square boundedness of the filtering error of the proposed algotithm is analyzed.It has low computational burden and it is convenient for the application of networked systems.