Distributed H∞ filtering for networked systems over sensor networks

Networked systems are systems in which data transmissions among system components are completed through communication networks. As one fundamental issue, distributed filtering of networked systems has gained growing attention in the fields of control and signal processing, and its objective is to estimate system states using noisy output measurements. Thus, distributed filtering for networked systems has found a wide range of applications in areas such as health care monitoring, security detecting, industrial process automation, and environmental monitoring. A defining feature of distributed filtering is that connections among nodes of a distributed system is completed through communication networks rather than point-to-point cables. Compared with traditional point-to-point connections, communication networks can bring several advantages, e.g. low cost, high reliability, ease of maintenance and reconfiguration, and flexible system architecture, especially for distributed systems, where sensors, actuators and controllers are spatially distributed far away from each other. However, communication networks may have some challenging issues, which make the distributed filtering problem complicated. First, because of limited bandwidth, network-induced delays and packet dropouts are unavoidable, which usually deteriorate the system performance. Second, due to the dynamic characteristics of networks, the topology of a distributed system is not fixed, which is difficult to be dealt with when current system states are estimated. Third, as a result of limited network resources, the network traffic may be congested if all the sampling periods of sensor nodes are very small, which is undesired in the distributed filtering. Therefore, it is significant to address the problem of distributed filtering by taking the above challenging issues into account, which motivates this study. This thesis is concerned with distributed filtering for networked systems. Several results have been reported on the effects of network-induced constraints including delays and packet dropouts, dynamic topologies, and network resource utilizations, on the performance of distributed H∞ filtering error system. More specifically, the emphasis is placed on the following four aspects: • Event-triggered distributed H∞ filtering. In order to save limited network resources, a distributed event-triggered transmission scheme is introduced to choose those necessary data packets to be transmitted among neighboring filters. Under the event-triggered transmission scheme, the resultant filtering error system is modeled as a time-delay system. By employing Lyapunov- Krasovskii functional method, both event-triggered parameters and distributed H∞ filter parameters can be designed in terms of linear matrix inequalities. • Distributed H∞ filtering for continuous-time systems over sensor networks with dynamical topologies. The distributed H∞ filtering for continuous-time systems over sensor networks is investigated. On the one hand, the effects of data packet dropouts on the distributed H∞ filtering are investigated. On the other hand, a novel distributed event-triggered scheme is introduced to transmit those necessary data packets in the case that the topology of the sensor network is dynamic. Moreover, algorithms are also given to design suitable distributed H∞ filters. • Event-triggered distributed H∞ filtering for multi-rate systems subject to data packet dropouts. For multi-rate distributed systems, event-triggered distributed H∞ filtering is studied by taking packet dropouts into account. By applying a lifting technique, the resultant multi-rate distributed filtering error system is transformed into a high dimensional single-rate system. As a result, novel sufficient conditions are established such that the distributed filtering error system is mean-square exponentially stable with a desired H∞ attenuation level. Moreover, an iterative algorithm is presented for the design of both distributed filters and event-triggered parameters. • Event-triggered distributed H∞ filtering for sensor networks under Round-Robin scheduling. Event-triggered distributed H∞ filtering for sensor networks is investigated under Round-Robin scheduling. Each sensor is equipped with a few samplers, and the scheduling of signals from samplers to filters is ruled by the classical Round-Robin protocol. In order to save precious network resources, a novel event-triggered transmission scheme is presented, where network-induced delays are taken into account. Moreover, the resultant filtering error system is modeled as a switched system with multiple time-varying delays. Switched system approaches are thus employed to design distributed H∞ filters based on a set of LMIs. Several simulation results are provided to validate the proposed methods