Research On Distributed Estimation Algorithm Of Networked System Based On Event Triggering

In the networked control system,the limitation of communication bandwidth and communication energy makes the control of the networked system more and more complicated.Therefore,the traditional time-triggered mechanism has already can not meet the demands in the system state estimation.For saving the communication energy of the sensors,a non-periodic sampling event-triggered communication strategy can be used to select valuable information for data transmission.At the same time,the distributed estimation has attracted wide attention because of its advantages that each sensor can be used as a fusion center and the communication burden is small.This paper mainly studies the problem of event-triggered distributed filtering.The main research contents are as follows:For discrete linear time-varying systems with uncorrelated noise,an event-triggered distributed estimation algorithm with observations and estimates triggered has been put forward.In the distributed estimation,each sensor only communicates with neighboring sensors.Two sets of Bernoulli random variables are used to describe the trigger mechanisms based on observations and based on estimates,at the same time,the information exchange between neighbor nodes is described by adding the same consensus items.The optimal distributed filtering algorithm is obtained in the sense of linear unbiased minimum variance.For avoiding the calculation of the cross-covariance matrix of different estimates,a suboptimal event-triggered distributed filter is proposed by locally minimizing the upper bound of the variance.Based on the Lyapunov method,the exponential boundedness of the suboptimal algorithm has been proved in the sense of mean square.Numerical simulation verifies that the smaller the event trigger threshold,the higher the communication rate and estimation accuracy of the filter;otherwise,the lower the communication rate and the estimation accuracy.For discrete linear time-invarying systems with correlated noise,an event-triggered distributed estimation algorithm based on forecast compensation is proposed.A series of Bernoulli random variables is used to describe the trigger mechanism based on estimates.When the channel from the estimator to the neighbor estimator is not triggered normally,the state forecast estimate of the latest trigger moment to the current moment is used to compensate.An event-triggered suboptimal distributed filter is proposed by locally minimizing the upper bound of the variance.Numerical simulation verifies that the algorithm with compensation is slightly better than the algorithm without compensation.For multi-sensor networked linear discrete time-varying systems with correlated noise,an event-triggered distributed estimation algorithm with different consensus gains is proposed.Two sets of Bernoulli random variables are used to describe the trigger mechanisms based on observations and based on estimates.By adding different consensus items to describe the information exchange between different neighboring nodes,and expanding the dimensions of the consensus items,the optimal distributed filter algorithm is obtained in the sense of linear unbiased minimum variance.For avoiding the calculation of the cross-covariance matrix between estimates,a suboptimal event-triggered distributed filter is proposed by locally minimizing the upper bound of the variance.Based on the Lyapunov method,the exponential boundedness of the suboptimal algorithm is proved in the sense of mean square.Numerical simulation verifies that algorithms with different consensus gains are better than those with the same consensus gains under the same conditions.For multi-sensor networked nonlinear systems with one-step transmission delay,an event-triggered distributed estimation algorithm is proposed.Considering that the topology is fixed,the state vector of the sensor depends on the state of itself and neighbor nodes,a series of Bernoulli random variables is used to describe the observations trigger mechanism.According to the state model transition method,the current state and the previous moment observations are augmented to obtain an equivalent state space model.When the event is triggered,the sensor transmits its observation to the estimator,and there may be a transmission delay.Therefore,based on event triggering,transmission delay and matrix inequality,an event-triggered distributed filtering algorithm is derived by locally minimizing the upper bound of the variance,and then the filter gain can also be obtained based on it.According to the Lyapunov method,the exponential boundedness of the algorithm in the sense of mean square is proved.Numerical simulation proves that the larger the threshold,the lower the communication rate and estimation accuracy.