能量高效的分布式目标跟踪与状态检测算法研究

发布时间：2018-10-13 13:51

【摘要】：估计以及检测理论是统计信号处理的两大分支,它们在用于提取信息的电子信号处理系统中应用广泛,这些系统包括雷达、通信、声呐、语音、图像处理、生物医学、环境监测以及地震学等。其中检测理论是要确定感兴趣的事件是否发生或者当前环境处于哪种离散状态,估计理论则对感兴趣的事件提取出更详细的信息。随着传感技术、无线网络以及嵌入式系统等领域的快速发展,无线传感器网络(Wireless Sensor Networks, WSNs)在军事以及民用的各个领域发挥着越来越重要的作用。传感器节点被布放在目标区域内来获取感兴趣的信息,并被用于目标跟踪、环境监测、信息安全等场景,这些也都是估计及检测理论的应用范畴。在无线传感器网络中,传感器节点的能量通常是有限的,设计算法时要进行性能与能量消耗的折中。如何在保证性能的同时尽可能减小传感器网络的能量消耗,或在给定能量及带宽限制下最大化性能成为一个重要的问题。本文主要针对无线传感器网络中进行分布式目标跟踪以及状态检测的应用,研究如何在保证性能的同时尽可能减少网络的通信量,从而减少传感器节点的能量消耗,提高网络寿命。其中目标跟踪是一个典型的估计问题,需要估计目标每一时刻的位置以及速度；状态检测则用来确定当前环境处于哪种状态,是一个检测问题。在利用无线传感器网络进行目标跟踪时,每个传感器节点的数据蕴含的信息是各不相同的。有些节点的数据可能蕴含的信息很少,对于提高跟踪性能几乎没有帮助,因此有必要在跟踪过程中规划参与目标跟踪的节点集和参与方式。本文分别提出了基于领导节点的目标跟踪算法和完全分布式的目标跟踪算法。在基于领导节点的目标跟踪算法中,综合考虑了数据收集过程以及领导节点迁移产生的通信开销,根据跟踪过程中的误差矩阵进行节点规划,以最大化目标跟踪的性能。求解中采用Gauss-Seidel迭代以及凸松弛等手段,使得复杂的优化问题能够得到快速求解。仿真结果表明,本算法在相同的通信能量约束下能够达到更好的跟踪性能。完全分布式的目标跟踪算法通常在每步跟踪过程中利用一致性算法(Consensus Algorithm)在全网传播局部信息,从而使每个节点得到全局或接近全局的跟踪性能。如前所述,有些传感器节点的数据对于提高跟踪性能几乎没有帮助,参与一致性迭代只会浪费能量并降低一致性算法的收敛速度。本文将节点规划策略引入到分布式目标跟踪算法中,每次只选择具有高信息量的节点参与一致性计算,从而在尽量保持跟踪性能的同时减少网络的通信开销。在分布式状态检测问题中,每个节点不仅依靠自身的观测数据,还通过与邻居的信息交互进行检测。然而在邻居间直接传递观测数据通信量较大。本文将社会学习机制应用到分布式状态检测中,提出了一种基于私有数据和邻居决策的分布式状态检测算法,并分析了算法的收敛性。由于邻居间只传递对于环境状态的决策,相比于直接交换观测数据或似然比等连续量,可以显著减少网络通信量,当环境的状态空间为二元时,邻居间的信息交换小到只需要1bit。仿真结果表明,在相同的通信开销下,本文算法可以更快收敛到环境的真实状态。
[Abstract]:Estimates and detection theories are two major branches of statistical signal processing, which are widely used in electronic signal processing systems for extracting information, including radar, communication, navigation, voice, image processing, biomedicine, environmental monitoring, and seismology. wherein the detection theory is to determine whether the event of interest occurs or which discrete state of the current environment, and the estimation theory extracts more detailed information for the event of interest. With the rapid development of sensor technology, wireless network and embedded system, Wireless Sensor Networks (WSNs) plays an increasingly important role in military and civil fields. The sensor nodes are placed in the target area to acquire the information of interest, and are used for target tracking, environmental monitoring, information security and other scenarios, which are also applied to the estimation and detection theory. In a wireless sensor network, the energy of the sensor node is generally limited, and a compromise between performance and energy consumption is to be performed when designing the algorithm. How to minimize energy consumption of sensor networks while ensuring performance, or maximize performance under given energy and bandwidth constraints becomes an important issue. This paper mainly focuses on the application of distributed target tracking and state detection in wireless sensor networks, and studies how to minimize the network traffic while ensuring performance, thus reducing the energy consumption of sensor nodes and improving network life. wherein the target tracking is a typical estimation problem, the position and the speed of each moment of the target are estimated, and the state detection is used for determining which state of the current environment is a detection problem. When target tracking is performed with a wireless sensor network, the information contained in the data of each sensor node is different. Some of the nodes' data may contain little information, and there is little help to improve tracking performance, so it is necessary to plan the node set and participation mode involved in the target tracking during the tracking process. A target tracking algorithm based on leadership nodes and a fully distributed target tracking algorithm are presented in this paper. In the target tracking algorithm based on the leader node, the data collection process and the communication overhead generated by the leader node migration are comprehensively considered, and the node planning is performed according to the error matrix in the tracking process so as to maximize the performance of the target tracking. By using Gauss-Seidel iteration and convex relaxation in the solution, the complex optimization problem can be solved quickly. Simulation results show that the algorithm can achieve better tracking performance under the same communication energy constraints. The fully distributed target tracking algorithm usually transmits local information throughout the whole network by using a consistency algorithm (Consensus Al0.8m) in each tracking process, so that each node obtains global or near global tracking performance. As previously mentioned, some sensor nodes have little help to improve tracking performance, and participating in a consistent iteration will only waste energy and reduce the convergence speed of the coherence algorithm. In this paper, the node planning strategy is introduced into the distributed target tracking algorithm, and only the nodes with high information content are selected to participate in the consistency calculation at a time, so that the communication overhead of the network is reduced while keeping track performance as much as possible. In the distributed state detection problem, each node not only relies on its own observation data, but also detects the distributed state detection. however, direct transmission of observed data traffic between neighbors is large. This paper applies the social learning mechanism to the distributed state detection, and proposes a distributed state detection algorithm based on private data and neighbor decision, and analyzes the convergence of the algorithm. Compared with the direct exchange observation data or the likelihood ratio and the like, the network communication quantity can be significantly reduced compared with the direct exchange observation data or the likelihood ratio and the like, and when the state space of the environment is binary, the information exchange between the neighbors is small to only 1bit. Simulation results show that under the same communication overhead, the algorithm can converge faster to the real state of the environment.
【学位授予单位】：复旦大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7

【相似文献】