State And Fault Estimation Of Complex Networks With Information From Partial Nodes

In the real world,a great amount of complex systems can be described by complex networks,such as computer systems,power systems,interpersonal relationship networks.With the development of industrial interconnection and information fusion technology,in order to reduce the effects of fault on system performance and avoid possible networked systems crash,distributed fault and state estimation of complex networks has become one of the research hotspots.The high complexity of complex networks is not only reflected visually in the complexity of network structure caused by large-scale network nodes and diverse connections,but also in the complexity of dynamic behaviors such as information propagation,network evolution,stability and synchronization.Due to the current communication bandwidth constraints,network attacks and other environment factors,there will exist time delay,missing measurements,data conflicts or other phenomena in the process of nodes communication.These network induced phenomena will lead to the fact that the information of the complex networks actually obtained come from the information of partial nodes,which will bring challenges to state and fault estimation,so it is necessary to propose new methods to solve these problems.The purpose of this thesis is to study the state and fault estimation of complex networks with information from partial nodes.The main work can be divided into the following three parts:1.The problem of fault estimation is studied for delayed complex networks with partially decoupled disturbance inputs.Considering the limited communication between nodes,Round-Robin(RR)protocol is applied to prevent data conflict.At each time instant,each node can only receive information from one of its adjacent nodes.Based on the distributed fault diagnosis method,an unknown input observer is designed for each node by using the information of adjacent nodes to decouple the partial disturbances.A sufficient condition is derived that the estimation error of the system satisfies mean square exponentially ultimately bounded in the presence of uncoupled disturbances.The gain parameters of unknown input observers are calculated by solving linear matrix inequalities.2.The problem of fault estimation is investigated for discrete complex networks with random missing measurements.The measured values of the system are missing at random,which is represented by a series of random variables satisfying certain probability distribution.For complex networks with nonlinearity and disturbance,fault estimators composed of measured values are designed.By using the stochastic analysis method,sufficient conditions are obtained for the estimation error to satisfy the given H_∞constraints.By solving linear matrix inequalities,the parameters of fault estimator can be obtained.3.The resilient filtering problem is dealt with for time-varying complex networks based on partial nodes.In the actual complex networks,the outputs can be acquired only from a fixed fraction of network nodes,but not all nodes.Therefore,it is very meaningful to design filters based on the outputs of partial nodes.For the sake of reduction of impact from filter parameter inaccuracy on state estimation,resilient filters are considered.By using the completing-the-square method,the estimation error can satisfy the H_∞constraint in finite horizon.The parameters of the resilient filter are obtained through solving the backward recursive Riccati difference equations.