【作者】 刘昕；

【导师】 刘云生；

【作者基本信息】 华中科技大学 ， 计算机软件与理论， 2010， 博士

【摘要】 随着数字电子、微处理器技术和无线通信技术的发展,使得无线传感器网络能够通过数以千计的分布式传感节点获得高密度、高精度的传感数据。由于传感器节点的小尺寸和低成本等众多优点,无线传感器网络被广泛应用在军事和民用领域,如目标跟踪、区域监测、环境控制以及医疗保健。目标跟踪是无线传感器网络中最重要的应用之一。而能量效率、路由协议、定位算法以及轨迹预测是基于无线传感器网络的目标跟踪系统设计中需要考虑的关键问题。论文提出了一种基于无线传感器网络的目标定位系统体系结构,并分别对上述四个关键问题进行了研究。在基于无线传感器网络的目标跟踪系统中,如何提高整个网络的能量效率是一个核心问题。在当前多数提高能量效率的研究中,节点剩余能量是一个重要的参数。常用的获取节点剩余能量的方法是节点周期性的发送自己的能量信息,然而在某些环境下,通信消耗的能量有时大于利用该算法节省的能量。因此,如何更有效的获取剩余能量是一个关键问题,能量预测方法是一种较好的方法,该方法通过节点自身能量消耗的历史数据来预测未来一段时间内的能量消耗。论文首先对节点预测方法进行了研究,讨论了一种基于马尔科夫链的节点能量概率模型,该模型将节点的不同工作模式对应马尔科夫链的不同状态：如果一个节点有M种工作模式,则可运用马尔科夫链的M种状态进行模拟。然后提出了一种基于离散时间马尔科夫链的平稳分布的能量预测算法,并进一步讨论了基于连续时间马尔科夫链的能量预测方法。仿真实验也说明了算法的有效性。在数据传输过程中,部分节点因负载过重能源快速耗尽,造成网络中传播路径的通信距离延长,并由此引发网络能量消耗增大,生命周期缩短的负面影响。针对此问题,提出了一种基于能量预测的路由算法。算法以均衡网络负载为目的,利用预测结果作为优化路由选择的标准,在路径建立过程中选取剩余能量较多的邻居节点作为下一跳节点。实验表明,同传统路由算法相比,该算法能够更好地均衡网络的能量消耗,有效缓解了部分节点因负载过重造成能源快速耗尽以及由此带来的负面影响,最大限度地延长了网络寿命。在另一方面,与传统的无线传感器网络应用不同的是,目标跟踪系统中的数据往往具有很强的实时性。因此,提出了一种能量有效的实时路由算法,该算法在尽可能保证数据实时性的前提下通过最大熵原理均衡网络的能量使用,提高整个网络的生存时间。实验表明该算法能够在满足数据实时性的要求条件下有效地平衡网络能量。目标跟踪系统中定位算法可分为基于测距定位算法和非测距定位算法。基于距离测量和角度测量的定位算法的缺点是对专用硬件有一定的要求,从而使传感器节点成本和体积加大,限制了它的实用性。非测距的算法不需要测量未知节点到信标节点的距离,在成本和功耗方面比基于测距的定位方法具有一定的优势,但是精度相对不足。因此,提出了一种基于支持向量机的目标定位算法TLSVM。该算法将整个监测区域分成若干个子区域,通过已知的节点位置信息作为训练样本生成每个子区域的最优SVM。当目标出现时,根据报告节点的分类信息估计目标所属区域,从而较精确地估计出目标的位置。另外,考虑到SVM分类的准确性直接影响到定位的精确性,如何选择最优SVM是TLSVM的一个关键问题。因此,还进一步还讨论了最优支持向量机的选取方法,通过对SVM的统计分析,给出了一个新的反映支持向量机分类能力的指标。通过比较该指标在不同的模型(不同的核)中的估计值,可以选取最优的支持向量机。实验结果也说明了该方法的合理性和有效性。针对多目标跟踪问题中的核心问题：轨迹关联和轨迹预测进行了研究,提出了一种基于卡尔曼滤波的多目标跟踪算法。该算法通过最近邻方法将一个多目标跟踪问题分解成一组单目标跟踪问题,然后使用卡尔曼滤波对单目标进行跟踪。进一步的,考虑到卡尔曼滤波的收敛性问题,我们还讨论了基于卡尔曼滤波的非线性模型近似解问题。更多还原

【Abstract】 Recent advances in digital electronics, microprocessor micro-electro-mechanics and wireless communication have enabled the deployment of large scales sensor networks where thousands of small sensors are distributed over a vast field to obtain fine-grained high-precision sensing data. Due to many attractive characteristic of sensor nodes such as small size and low cost, sensor networks are adopted in many military and civil applications, such as target tracking, surveillance, environmental control and health care. Target tracking is one of the most important applications in wireless sensor networks. Energy efficiency, routing protocol, target location and trajectory prediction are considered most important problem when designing a system for target tracking. In this dissertation, we design the architecture of target tracking based on wireless sensor network, and research on four key issues of the above.A key issue in the design of target tracking based on wireless sensor network is to devise mechanisms to make efficient use of its energy, and thus, extend its lifetime. The information about the amount of available energy in each node is the key to balance the network load. Sending energy information packet periodically is the most commonly used method to obtain the remained energy in each node. But on many occasions, the energy consumption is much more than the energy savings using this method. This dissertation studies approaches to obtain the energy by the prediction. For the prediction, a probabilistic energy model base on Markov chains is firstly discussed. In this model, each sensor node can be modeled by a Markov chain, the node operation modes are represented by the states of a Markov chain and, if a sensor node has M operation modes, it is modeled by a Markov chain with M states. Then we present an algorithm of Energy Prediction based on Stationary Distribution of discrete-time Markov chain, and discuss the energy prediction based on continuous-time in addition. The results of simulation show the efficiency of the algorithm.Some nodes run out of energy quickly, causing the network to extend communication distance of the propagation path, and thus lead to increase energy consumption of the network and reduce the life of network. For this problem, we presented a routing algorithm based on energy prediction. In the algorithm, each node predicted the remaining energy of its neighbor nodes, the routing selection was optimized. The result of simulation output and analysis shows that the EPR could optimize the routing balance the energy consuming of sensor nodes and prolong the network survival. In the other hand, data communication in wireless sensor networks often has timing constraints in the form of deadlines, which represent a new generation of real-time data communications from traditional networked systems. In this work we discuss energy problems of an existing real-time routing protocol LNA. Based on LNA, we propose an Energy-efficient Real-Time Routing (ERTR) protocol, which supports energy-efficient real-time data transmitting in wireless sensor networks. ERTR maximize the number of messages that can reach the sink where each message has its own due-date by minimizing the maximal lateness of all messages. ERTR also average the energy of node to increase the energy efficiency of wireless sensor networks according to the maximum entropy principle. The results of simulation show the algorithm could balance the network energy.we propose a target localization algorithm based on Support Vector Machine (SVM). This algorithm firstly divided the entire monitoring area into several sub-regional, then calculated the optimal SVM for each sub-regional by the known location of the node. Simulation shows the algorithm could accurately estimate the location of the target. Furthermore, location accuracy is depends on the SVM classification accuracy, how to choose the optimal SVM is a key issue to our target location algorithm. We present one quantitative criteria for the SVM classification using statistical analysis method. The results of simulation show the rationality and the effectiveness of the criteria.we propose a multiple objects tracking algorithm based on Kalman filter. The algorithm divided multiple objects tracking problem into a set of single-target tracking problem using nearest neighbor method, and used Kalman filter to track single-target. In addition, considering the convergence of Kalman filter, we discuss the solutions to non-linear stochastic differential equations.更多还原