Optimal Estimation And Application Of Unreliable Multi-rate Sampling Network Control System

With the rapid development and cross infiltration of computer technology and network communication technology,network control systems are widely applied in many industrial areas.Due to the inherent limitations of the shared communication network,measurement datas transmitted in the network control system generally suffer from fading,delay,lost,and disorder,which lead to various networked problems such as additive noise,multiplicative noise,transmission delays,packet losses and so on.In addition,due to the existence of system components with different performance in the network control system and the intricacies of its network environment,it is difficult to maintain a uniform fixed sampling rate for all signals,thereby making the phenomenon of multi-rate sampling inevitable.These network-induced unreliabilities may deteriorate the system performance and affect the system stability.Therefore,it is a significant and challenging topic to develop the optimal estimation theory for networked control systems under the network-induced unreliabilities.In this thesis,the optimal estimation problems are studied for sampled-data systems under several network-induced unreliabilities,including transmission delays,multiplicative noise and multi-rate sampling.The key approachs applied for treating the estimation problems are the reorganized observation method,the state augmentation method and the differential Riccati equation with jumps,and then differtent kinds of estimators are designed by using the innovation analysis method,the complete square method,and the stochastic approximation method.The main works of this thesis include:The optimal estimation problems are studied for sampled-data systems subject to transmission delays and multiplicative noise.For the constant-delay sampled-data systems with scalar multiplicative noise or diagonal-matrix multiplicative noise,and the random-delay sampled-data systems with diagonal-matrix multiplicative noise,the segmented continuous optimal filters are proposed in mean square minimum variance sense.To deal with the delay terms,the reorganized observation method is used to transform the original systems into the delay-free ones;moreover,the differential Riccati equations with jumps are obtained based on the Ito's differential rule,and the filter gains are determined by solving these differential Ricatti equations with jumps.Furthermore,the stability of the proposed filter algorithms is analyzed,and the corresponding stability conditions are obtained.The main innovations include:(i)Compared with the existing discrete-time Kalman filter,the proposed filter algorithms provide accurate state estimates between the sampling intervals as well as the sampling moments,which possess more estimation accuracy;(ii)Only one type of Riccati equations with jumps that have the same dimensions as the original system states needs to be solved,thus avoiding the computational complexity caused by the state augmentation method,and reducing the computational burden.The optimal estimation problems are investigated for multi-rate sampling systems with diagonal-matrix multiplicative noise,where the available and unavailable cases of non-uniform sampling processes are considered.By introducing a set of discrete Markov chains to describe the non-uniform sampling process of the observation,the multi-rate sampling system is transformed into a single-rate sampling system with random delays;moreover,the state augmentation method is employed to transform the random delay system into a delay-free one with Markov jumping parameters.For the situation where the non-uniform sampling process is available,a Markov jump filter is proposed based on the complete square method and the stochastic approximation method,in which the filter gain is obtained by solving a set of coupled differential Riccati equations.For the situation where the non-uniform sampling process is unavailable,a linear minimum mean square errror filter is derived by using the innovation analysis method,in which the filter gain is obtained by solving the generalized differential Riccati equation.The main innovations include:(i)a set of Markov chains are firstly introduced to describe the non-uniform sampling process of diagonal-matrix multiplicative noise systems,which effectively characterizes the correlation between the sampling moments;(ii)The diagonal-matrix multiplicative noise is more general and practical,and then the Hadamard product is introduced in the derivation of the Riccati equations,which overcomes the computational difficulies casued by the diagonal-matrix multiplicative noise.The distributed fusion estimation problems are studied for sampled-data systems with diagonal-matrix multiplicative noise.For the multi-sensor random delay sampling systems and the multi-sensor multi-rate non-uniform sampling systems,the optimal sequential fusion estimatior and the suboptimal covariance intersection distributed fusion estimatiors are proposed based on the linear unbiased estimation,the sequential fusion algorithm and the covariance intersection fusion algorithm.Moreover,the simulation results show that the fusion estimator has higher estimation accuracy than a single sensor local estimator.The main innovation is to realize the extension from the single sensor cases to the multisensor cases,which effectively improves the estimation accuracy.The application for the proposed filter algorithms is considered in the cooperative localization of multiple autonomous underwater vehicles.The state space model for the multiple autonomous underwater vehicles is estabilished based on the kinematics model and dynamical model.Considering the mesurement losses under the constraints of underwater acoustic communication,a local optimal estimator is derived.Based on the sequential fusion algorithm,the cooperative localization filtering algorithm is designed for multiple autonomous underwater vehicles,which has the high positioning accuracy and real-time performance.