无线传感器网络是由许多具有信息感知、无线通信和信号处理能力的传感器节点以无线通讯方式连接组成的分布式系统。 虽然单个传感器体积小、造价低、可靠性差，但通过传感器节点间的协作，无线传感器网络可胜任一系列复杂的任务。 随着现代通讯技术和微电子设备的迅速发展，传感器节点趋于小型化、智能化和多功能化，这给无线传感器网络的应用带来了广阔的前景。目标状态跟踪为无线传感器网络的重要应用，近几年来得到学术界及工业界的极大关注。 关于集中式的状态估计技术已经相当成熟，但将该技术应用于类似于无线传感器网络的分布式系统却存在很大的局限性，亟需进一步研究。因此，本文研究的重点为设计无中心节点的分布式滤波算法，并在此基础上探讨网络通信约束和动态拓扑结构对分布式滤波算法的影响。 本文主要研究内容如下：第 2 章针对带宽受限的无线传感器网络，研究了一种能够有效利用网络资源的分布式滤波算法。 通过设计分布式事件驱动通信策略对传感器的通信数据进行合理性筛选，保证只有包含有用信息的实时数据才能通过网络从而降低网络传输率。第 3 章综合考虑随机非线性以及增益扰动对分布式滤波的影响。设计了弹性分布式滤波器进而为该类系统提出了鲁棒滤波的框架 并给出了估计误差有界的条件。第 4 章考虑受测量衰减影响的非线性系统的分布式滤波问题，给出滤波器估计误差渐进有界的条件。提出了自适应阈值的技术手段， 使无线传感器网络的通信频率维持在给定期望值。第 5 章在第 4 章的基础上进一步探讨了通信拓扑服从马尔可夫切换规律下的无线传感器网络的分布式滤波问题。 给出了分布式滤波可观测性条件，证明了若无线传感器网络为分布式可观测且马尔可夫切换的通信拓扑的合并集为强连通时，通过合理设计滤波器参数一定能保证估计误差指数均方有界。第 6 章分析了传感器间通信存在随机丢包下一致卡尔曼滤波算法的性能。证明了在无线传感器网络是全局可观的前提下，当网络拓扑结构、 一致性步长和通信丢包率满足一定条件时，该滤波算法估计误差的协方差一定满足随机有界性。第 7 章 探讨编码通信下复杂系统的滚动时域估计问题，深入研究了编码通信对传输数据的影响，并通过求解最小二乘问题给出最优的滚动时域估计值。
A wireless sensor network is a distributed communication network containing an array of spatially separated intelligent sensing devices. Although each sensor node is typically small, power-constrained, low-cost and sometimes unreliable, it can still perform various high-level tasks in a collaborative manner. Very recently, thanks to the rapid development of modern communication techniques and embedded micro-devices, the wireless sensor networks have gained various potential applications.As one of the fundamental issues in wireless sensor networks, the target tracking problem has attracted increasing attention in both academia and industry. It has been well recognized that the traditional centralized state estimation methods are inapplicable to the wireless sensor networks due primarily to the distributed layout of sensor nodes. As such, there are practical requirements to establish an actual distributed estimation algorithm which could run autonomously without a central unit, and subsequently analyze the influence of resource constraints and dynamic communication topologies on the filtering performance. Specifically, the main contents of this dissertation are as follows:In Chapter 2, it is assumed that the wireless communication between sensors is subject to certain bandwidth constraints.To efficiently utilize the limited bandwidth, a novel event-based distributed filtering algorithm is proposed.Furthermore, the influence of the stochastic nonlinearities and gain perturbations on the distributed filter is investigated in Chapter 3. A resilient operation is designed to suppress random perturbations on the filtering performance. The suboptimal distributed filter is established by utilizing the robust filtering techniques and the conditions that ensure the mean-square boundedness of the estimation errors are established.In Chapter 4, the distributed filtering problem for a class of continuous-time nonlinear systems subject to measurement fading is studied. By resorting to graph theory and stochastic analysis methods, the filter parameters are designed, such that the filtering errorconverges at an exponential rate in the mean square sense. An adaptive algorithm is developed, which allows the intelligent sensors to keep the average transmission rate around a desired value.Based on the framework of Chapter 4, the distributed filtering problem subject to Markovian switching topologies is further investigated in Chapter 5. It is shown that, with the proposed distributed filtering algorithm, the exponential mean-square boundedness of the estimation errors is guaranteed if the sensor network is distributively detectable and the combined communication topology is strongly connected.In Chapter 6, the Kalman-consensus filtering algorithm with random link failures is considered.Sufficient conditions for the stochastic boundedness of the Kalman-consensus filter are established. It is proved that the filtering performance is directly influenced by the global observability, the network connectivity, the packet dropout rate and the length of consensus step.Chapter 7 is concerned with the moving-horizon state estimation problems for a class of discrete-time complex networks under binary encoding schemes. The influence of binary encoding schemes on the signal transmissions are thoroughly studied. The optimal moving-horizon estimates are calculated by solving the least-square optimization problems.
