Distributed H_∞Filtering Approach In Sensor Networks

With the rapid development of electronics and wireless communication technology, sensor networks have been widely used in many fields in recent years. Numerous engineering problems can be treated as the state estimation problems of dynamic systems. Designing an effective and reliable distributed filtering algorithm is an important issue in the research of sensor networks. The key point is to achieve fast information fusion over the whole network and deal with the missing data caused by the complicated network environment. This thesis is concerned with the problems of distributed H∞filtering in sensor networks and the main contents are stated as follows:(1) A novel design method for distributed H∞filtering is established based on two-dimensional system theory. The proposed node filter is composed of two steps:the measurement and the consensus updates. An average H∞index is introduced to constrain the performance which is less conservative than the classic one. By means of the two-dimensional system stabil-ity theory and the bounded-real lemma, sufficient conditions are obtained to ensure the filtering error system is asymptotically stable and satisfies the prescribed average H∞performance index. The output estimates of all nodes can reach a consensus in each sampling period by implement-ing the consensus update procedure repeatedly.(2) The influences of missing data and switching network topology are considered in the design of distributed H∞filtering. A set of Bernoulli random sequences is employed to mod-el the phenomena of missing measurement and communication link failure. By means of the two-dimensional system-based approach, a filtering error system with multiplicative noises is derived. By introducing a switching signal into the consensus update procedure, a multi-mode filter is constructed on each node, which can follow the changes of network topology. According to the stability theory of state-multiplicative system and switched system, sufficient conditions for the stability of filtering error systems are established in terms of certain weak coupling linear matrix inequalities, and all filter parameters can be obtained.(3) The problem of distributed sparse signal estimation in sensor networks is discussed. By combining the pseudo-measurement method with the classic H∞filtering, a so-called central-ized l1-regularized H∞filtering is established by means of Krein space Kalman filtering theory, which is effective to deal with the sparse signal estimation problem. By embedding a high-pass consensus filter into the information form of l1-regularized H∞filter, a distributed filtering al-gorithm is proposed. Since the network is strongly connected, each node can track the average measurement over the network by fusing the data received from its neighbours, and then the distributed filter can achieve the same performance with the centralized filter asymptotically.