Research On Nonlinear Filtering For Nonlinear Networked Systems With Incomplete Information

With the development of computer and communication technology,network has become an essential part of industrial production and daily life.In the networked system,control objects become more and more complex.Many device working processes show different degrees of nonlinearity and randomness,which brings an essential difficulty for studying filtering problems.In addition,the use of network inevitably brings communication-limited problems,such as packet dropout,network-induced delay,fading measurement and so on,which lead to that the system receives incomplete information.Therefore,considering the problem of incomplete information incurred by communication network,this paper systematically studies the nonlinear filtering theory for networked systems.The main research results are as follows:First,the nonlinear filtering problem on networked systems influenced by stochastic nonlinearities and multiple fading measurements is studied.The statistical properties of stochastic nonlinear disturbances are known,and the fading probability for each sensor is governed by an individual random variable satisfying a certain probability distribution.By introducing the statistical properties of stochastic nonlinearities and the fading gain matrix into the design of unscented Kalman filter,an improved nonlinear filtering algorithm is proposed.Then,by analyzing the boundedness of estimation error and its covariance,a critical value of fading rate for guaranteeing the filter convergence and sufficient conditions for stochastic stability of filter are given.Secondly,the nonlinear filtering problem on networked systems affected by packet dropout is studied.The Send-on-Delta-based event-triggered sensor scheduling strategy is used to reduce the unnecessary measurement transmission and further reduce the probability of occurring packet dropout.Then,a set of independent and identically distributed random variables are used to model packet dropout.With a comprehensive consideration for the influences of event-triggered sensor schedule and packet dropout,a cooperative design method of event-triggered sensor scheduling strategy and nonlinear filter is given,and the sufficient conditions guaranteeing stochastic stability of designed filter are derived.Moreover,consider the fact that the deterministic sensor scheduling strategies can destroy the Gaussian property of filter update process.The nonlinear filtering problem on networked systems with stochastic event-triggered sensor schedule is studied.Compared with the deterministic scheduling strategies,the stochastic one can maintain Gaussian property of filter update process,which facilitates the filter design.According to the different event-triggered scheduling strategy,two nonlinear filtering algorithms are proposed and their advantages as well as disadvantages are analyzed.Finally,the sufficient conditions for guaranteeing the convergence and stochastic stability of the two filtering algorithms are given,respectively.Finally,the nonlinear filtering problem on networked systems influenced by network-induced delay is considered.By introducing the stochastic event-triggered sensor scheduling strategy,the unnecessary measurement transmission is reduced and the Gaussian property of state is maintained.Then,a length-specific buffer is used to store the delayed measurement,which guarantees that the remote filter makes a maximum use of the measurement.Finally,a stochastic event-triggered unscented Kalman filtering algorithm is proposed,which makes the best of the measurements stored in the buffer to compensate the network-induced delay and further guarantees the effective state estimation.