Self-correcting Fusion Estimation With Unknown Missing Observation Rate And Noise Variance System

With the rapid development of communication technology,computer technology,and electronic information technology,networked control systems have received extensive attention from scholars at domestic and abroad.Due to the advantages of networked control systems such as low cost,convenient data transmission and ease of remote control,the networked control system shows a strong vitality in the field of automatic control.However,with the complex control system model,the continuous expansion of the network size and the influence of many uncertainties in the network environment,the estimation of the networked control system becomes difficult.Failure to establish effective mathematical models and estimation algorithms will inevitably lead to inaccurate or even divergent estimates.In this article,we study the self-tuning fusion algorithm of the multi-sensor networked system with unknown missing measurement rate,fading measurement rate and measurement noise.The main research content is as follows:For a multi-sensor networked system with an unknown measurement missing rate,Bernoulli random variables are used to describe the missing measurements and the probability when a data packet is successfully received is unknown.Firstly,through the system identification theory,the correlation function is used to identify the measurement rate on-line.Then,the real-time measurement received rate is substituted into the filtering estimation algorithm to obtain the corresponding self-tuning filtering algorithm.Based on the projection theory,a local sensor self-tuning Kalman filter is deduced.The estimation error cross-covariance matrix between any two local sensors is deduced.Three kinds of weighted fusion algorithms which are matrix,scalar and diagonal matrix algorithms.They are applied to obtain a distributed self-tuning fusion filter,respectively.Finally,for the centralized fusion of high-dimensional measurement equations,by using the weighted least squares algorithm and the matrix full rank decomposition we can obtain the reduced-dimensional measurement equations.A weighted measurement fusion self-tuning Kalman filter is proposed based on the reduced-dimensional measurement equations.For multi-sensor networked systems with unknown fading measurement rate and measurement noise variance,a random variable with the interval [0,1] is used to describe attenuation measurements.What is more,the fading measurement rate and the measurement noise variance is unknown.When the measurement noise variance is known,the correlation function is used to identify the mathematical expectation and variance of the random variable describing the fading measurement.Then,based on the mathematical expectation and variance of the recognition,a local sensor self-tuning Kalman filter is proposed.Three weighted distributed self-tuning fusion filters and weighted measurement self-tuning fusion Kalman filters based on matrix,scalar and diagonal matrix are proposed.When the fading measurement rate and the measurement noise variance are unknown,the measurement equation is equivalently deformed.The correlation function is used to identify the mathematical expectation of the attenuation measurement random variable and the measurement noise variance.Based on the results of identification,they are substituted into corresponding local filters,distributed fusion filters and weighted measurement fusion filters and a corresponding self-tuning estimation algorithm is proposed.Finally,based on the dynamic error analysis method,the convergence of the self-tuning fusion algorithm is proved.