Research On Distributed Filtering Methods Over Sensor Networks With Switching Topologies

In recent years,with the rapid development of sensing and communication technologies,sensor networks consisting of small sensor nodes with sensing,computing and wireless communication capabilities have been widely applied in various fields.Since many practical engineering applications require real-time monitoring for unknown dynamic system states,the development of distributed estimation methods over sensor networks is critical.However,since sensor networks are usually deployed in harsh or dangerous environments,dynamic communication topologies,saturation nonlinearity,data loss,signal quantization,and deception attacks are inevitable.Furthermore,sensor networks'limited energy and bandwidth constraints bind the application in practical engineering.In this thesis,various design methods on distributed robust filter are explored in depth depending on the characteristics of sensor networks.The mathematical model of the stochastic switching communication topologies is established based on Markov process theory;based on Lyapunov stability theory,sufficient conditions for the mean-square exponential stability and robust H_?or l₂-l_?performance of the augmented filtering error system are analyzed and obtained;constraints on the solution of the distributed filter are given based on the linear matrix inequality method;the effectiveness and potential applicability of the designed distributed filter are verified based on numerical simulation analysis.The specific research work in this thesis is divided into the following parts:(1)Research on the design of distributed robust state estimators with sensor network-induced random measurement data loss and saturation nonlinear.Binary sequences with known probability distributions are used to describe randomly lost sensor measurement and saturation nonlinearity,and a homogeneous Markov chain is used to describe the random switching communication topologies of the filtering network.By constructing the Lyapunov function,the estimation error system is analyzed for mean square exponential stability.Then,the distributed robust filtering design method is given with the help of the linear matrix inequality technique.Finally,a numerical simulation and a continuous stirred reactor system together validate the effectiveness of the designed distributed robust filter.(2)Research on the design of distributed full-order and reduced-order state estimators over sensor networks subjected to malicious network attacks.Bernoulli binary switching sequences describe the phenomenon of randomly occurring malicious network attacks in sensor networks,and randomly switching communication topologies are represented as random sequences obeying a homogeneous Markov chain.It guarantees that the augmented estimation error system is exponentially mean square stable with a given l₂-l_?performance index when deception attacks,external disturbances and communication topology switching occur randomly in the sensor network.In addition,the distributed full-order and reduced-order state estimators can be obtained by choosing the appropriate estimator order control parameters,respectively.(3)Research on distributed filtering with limited bandwidth constraints over sensor networks.Since sensor networks have limited communication bandwidth,data conflicts may result when a large number of nodes send data at the same time.A Round-Robin protocol is introduced to reduce the communication burden of the sensor network,in which the measurement components of each sensor node access the network sequentially and periodically.Considering that the probability matrix describing the topology switching law of the filtering network is time-varying,a non-homogeneous Markov chain is used to describe the stochastic topology switching behavior.It is shown that the estimation errors converge in an exponentially decaying form,which ensures that the estimation error system is eventually bounded in the mean square sense.(4)Research on distributed filtering based on an adaptive event-triggered mechanism under finite energy constraints over sensor networks.The filtering network communication topology is time-varying and its switching law follows a homogeneous Markov stochastic process,in which there are partially unknown probability elements in the transition probability matrix of topology switching.An adaptive event-triggered mechanism is introduced for sensor networks,which not only reduces the operating frequency of sensor nodes sending data,but also saves limited energy as well as communication resources.By choosing topology-dependent Lyapunov functions,sufficient conditions to guarantee distributed H_?consistent state estimation are derived.Then,the desired distributed H_?state estimator is further obtained by introducing the slack variables and free-connection weighting matrices.Finally,using the designed distributed state estimator to track the state trajectory of a class of stable and unstable systems verifies the validity of the theoretical results.(5)Research on distributed filtering for discrete nonlinear systems with semi-Markov stochastic switching topologies.Different from the above studies,the stochastic switching process between sub-topologies can satisfy arbitrary probability distribution laws,so a more general mathematical model of semi-Markov type stochastic communication topology switching is established.The utilization of communication resources is improved by introducing signal quantization and an adaptive event-triggered mechanism.Based on the semi-Markov kernel method and Lyapunov stability theory,the augmented estimation error dynamics is analyzed for s-error-mean-square stability and H_?robustness.Finally,the effectiveness of the sojourn-time dependent distributed filter is verified by a mass-spring-damped system.