Research On Equality Constrained Distributed Estimation Problem In The Sensor Network With Packet Dropping

A sensor network system is composed of a set of multiple sensors deployed in a specific range,where each sensor has independent storage capacity,processing,and communication capabilities,and can perform decision making tasks based on the obtained information with strong robustness and easy scalability.To achieve the goals of multi-node collaboration and decentralization,the problem of distributed state estimation based on sensor networks has become the focus of many scholars’ research at this stage.However,the actual application environment is often not ideal and distributed state estimators usually operate in more complex environments,such as sensor failures,loss of observation information during transmission,and noise disturbances.To address these problems,this thesis investigates the problem of equality constrained distributed estimation in a sensor network with packet dropping under Gaussian and non-Gaussian noise disturbances.The main research contents are as follows:(1)For the problem of equality constrained distributed state estimation in a sensor network with packet dropping under Gaussian noise,two different perspectives are investigated,and two different classes of distributed state estimators are derived in this thesis.First,the first type of unconstrained distributed Kalman filter is proposed using the statistical properties of the packet dropping variables.Considering that each node can only communicate with its neighboring nodes,the measurement information obtained by each node is different,which makes the estimation values of each node differ.To overcome this problem,the first type of unconstrained distributed Kalman consensus filter is designed in this thesis by adding a consistent term.However,the stability of the Lyapunov equation needs to be considered when considering the convergence of the first type of filter in infinite time,i.e.,the system matrix needs to be stable.To eliminate this strict restriction,this thesis designs a second type of unconstrained distributed Kalman consensus filter using the time-stamping technique.The estimated gain of this type of filter only requires solving a modified Riccati equation,which successfully eliminates the Lyapunov equation.To further improve the estimation accuracy,two types of equality constrained distributed Kalman consensus filters are obtained in this thesis by projecting the unconstrained estimates using the known state equality constraint information.Finally,the effectiveness of the proposed filtering algorithm is verified by simulation experiments.(2)For the problem of equality constrained distributed state estimation in a sensor network with packet dropping under non-Gaussian noise,this thesis proposes a type of equality constrained distributed maximum correntropy Kalman filter based on covariance intersection fusion.First,for the case of observation dropping and non-Gaussian noise disturbances,the distributed maximum correntropy Kalman filter is proposed in this thesis using the time-stamping technique and the maximum correntropy criterion.Since the quantity and quality of observation information of each node are different,thus making the estimated values of each node are different.In order to improve the consistency of the estimates of each node,this thesis uses the covariance intersection fusion algorithm to weigh and merge the estimates of neighboring nodes,thus obtaining a distributed maximum correntropy Kalman filter based on covariance intersection fusion.In addition,this thesis designs an equality constrained distributed maximum correntropy Kalman filter based on covariance intersection fusion using the known state equality constraint information,which has higher estimation accuracy.Finally,the simulation results show that the proposed filter has good performance not only under non-Gaussian noise but also close to the performance of the traditional Kalman filtering algorithm under Gaussian noise.