【作者】 张健；

【导师】 吴成东； 张云洲；

【作者基本信息】 东北大学 ， 模式识别与智能系统， 2011， 博士

【摘要】 集成了传感器、微机电系统、无线通信、分布式信息处理等多种技术的无线传感器网络成为研究群体智能自主自治系统行为的前沿领域。不论在军事领域还是民用领域,目标跟踪始终是无线传感器网络的重点研究内容之一。无线传感器网络目标跟踪的实质是在节点能量、通信带宽等资源受限的条件下,通过多节点协作对信息进行分布式处理、融合,对目标的运动状态进行估计。针对目标跟踪的研究既要保证跟踪精度,也要兼顾节点能耗和网络寿命问题,本文在分析与总结国内外相关研究的基础上,结合无线传感器网络分布式特性,对目标跟踪问题进行了深入地研究,主要研究内容和成果体现在以下几个方面：针对无线传感器网络中节点的感知信息类型,研究了多种传感器节点的组网优势,提出基于多模态信息的节点协作目标跟踪算法。为了减少跟踪过程中的网络通信量,在每个节点运行高斯代价参考粒子滤波,将所求出的状态估计均值和方差作为基本信息进行传输。仿真实验表明,基于多模态的节点协作目标跟踪算法不但满足了跟踪精度的要求,而且比集中式跟踪算法网络通信代价低,节点能量消耗少,从而提高了网络的跟踪性能。考虑到目标运动形式的不确定性,以及无线传感器网络节点随机部署的不均匀性,根据监控区域内目标的运动状态以及局部区域的节点密度,结合候选节点的预测剩余能量和调度情况,提出一种新的分布式轻量级目标跟踪算法。在此基础上,研究无线传感器网络在跟踪目标过程中的簇头选取策略以及分簇规模。针对节点高密度分布的无线传感器网络,利用节点的覆盖探测概率模型,提出实现目标跟踪性能的局部次优化分簇机制,进而基于后验克拉美罗下界原理,研究提高目标跟踪精度的优化机制。经过仿真实验对比,针对不同的运动形式和复杂的网络分布情况,轻量级跟踪算法能够动态地确定分簇规模,从而在保证跟踪精度的前提下,实现网络节能的目标。针对网络目标跟踪过程中的能效优化问题,研究无线传感器网络节点探测受约束的条件下节点自适应调度机制以及目标监控策略。根据每个节点的探测概率,计算出多节点联合探测概率,并基于节点本身属性构造决策函数模型。在此基础上,提出无线传感器网络局部能量消耗优化方案,并通过对粒子滤波采样策略的改进,解决粒子退化现象。仿真实验证明,与一般分簇算法相比,自适应动态分簇的节点调度跟踪算法,能够实现节点有效调度,保证跟踪精度,同时减少节点间能量差异,降低网络整体能耗。在对无线传感器网络目标跟踪过程中两种典型节点选择机制进行分析的基础上,基于有效信息和能量模型,研究节点选择的评估机制以及K-覆盖条件下的节点选择策略。对比实验证实,本文的节点选择算法能有效地降低冗余节点,减少冗余信息,而且网络的积累能量消耗显著降低。针对无线传感器网络能量均衡问题,研究基于节点能量消耗模型的能量划分方法,提出簇头选择机制以及簇内成员的选择策略,并利用后验克拉美罗下界原理确定参与跟踪的最优节点。通过仿真实验对比分析,该算法选取的参与跟踪节点分布广泛,有效地控制局部节点密度,能够平衡目标跟踪过程中的网络能量消耗,延长网络的生存时间。针对密集部署的无线传感器网络易产生大量冗余信息的问题,结合节点优化选择策略,提出基于量化节点观测值的目标状态估计框架。该框架利用量化机制减少节点间的数据通信量,并通过增大跟踪抽样间隔减少节点的唤醒频率的方式,构建能量优化模型,减少网络整体能量消耗。仿真实验证明,该算法通过量化机制有效分配节点带宽,从而减少节点间通信量,而相应的采样间隔延长策略,可以节约网络整体能耗。更多还原

【Abstract】 Wireless sensor networks (WSN), which integrate numerous technologies, such as sensors, micro-electromechanical systems, wireless communication, distributed information processing, have become a frontier research field for self-autonomous system behavior of swarm intelligence. For both military areas and civilian areas, the problem of target tracking has always been one of the key researches in wireless sensor networks, which focuses on the state estimation through multi-sensor collaboration to process and fuse information in the constraints of sensor energy and communication bandwidth. As far as tracking accuracy and energy consumption to prolong network lifetime, target tracking issues are investigated in depth based on relevant analysis of domestic and overseas, combining with the distributed nature of wireless sensor networks in this paper. The major research contents and productions are specially stated in the following areas:The advantage of networks organized by various types of sensor nodes is studied and a multi-sensor cooperative target tracking algorithm based on multi-modality information is proposed for many types of sensing information in wireless sensor networks. In order to reduce the communication traffic, the Gassian cost reference particle filter is ran on the sensor and the mean and variance of the estemated state are transmitted in the target tracking process. Simulation results show that the proposed multi-modality collaborative tracking algorithm meets tracking accuracy, and the communication cost is lower and the energy consumption is less than the centralized tracking algorithm, thereby the tracking performance of network is improved.Taking into account uncertainties of the motion form of the target and the nonuniformity of random distribution of wireless sensor networks, a new distributed lightweight target tracking algorithm, which combines the predicted residual energy and scheduling history of candidate sensor nodes, is proposed according to the motion state and the sensor density in a local mornitoring region. On the basis, the cluster head and the size of the cluster are respectively determined in the tracking process of wireless sensor networks. For dense wireless sensor networks, a local sub-optimal clustering algorithm for target tracking is performed with the coverage probabilistic detection model and an optimization of target tracking accuracy is compted, taking advantage of the posterior Cramer-Rao lower bound theory. By presenting the simulation experiments for different tracking forms and complex network environment, the proposed lightweight clustering strategy can dynamicly determine the size of the tracking cluster to save energy in the context of guaranteeing tracking accuracyAn adaptive sensor scheduling algorithm for monitoring a target in a local monitoring region of wireless sensor networks is presented for energy-efficient optimization in the constraint of the sensor detection. According to the detection probability of an individual sensor, the joint detection probability (JDP) is established. Based on sensors’own properties, a decision function model is built based on the property of the sensor. Further, an optimization scheme is designed to satisfy the problem of sensor scheduling subject to JDP and decision functions for candidated sensors, and an improved particle filter is proposed to solve the problem of degeneration. Compared with the general clustering algorithms, the proposed adaptive sensor scheduling algorithm can achieve an effective sensor scheduling. Meanwhile, this algorithm not only guarantees the tracking accuracy, but also reduces the energy difference between nodes and delines the whole energy consumption of wirless sensor networks.Based on the analysis of two typical sensor selection mechanisms in the target tracking process for wireless sensor networks, the sensor assessment mechanism and the sensor selection strategy under the condition of K-coverage are proposed through exploiting the efficient information and energy consumption model. Comparative experiments confirm that the algorithm can effectively reduce redundancy sensors to cut down redundant information and the accumulated energy consumption is significantly decreased. An energy partition method is designed on the basis of energy consumption model of the sensor to solve the problem of energy banlence in wieless sensor networks. Further, a cluster-head selection mechanism and the cluster member selection strategy are put forward respectively. To determine the optimal tasking sensors, the posterior Cramer-Rao lower bound is employed to ensure the tracking accuracy. Compared with other methods, active sensors in this algorithm are widely distributed to effectively control the sensor density in the local region, and energy consumption among sensors are balanced so as to prolong the network lifetime in the tracking process.For the problem of redundant information easily generated by densely deployed wireless sensor networks, a target state estimation framework based on quantization of measurements from sensors is proposed adaptive tracking mechanism based on quantization of observations was proposed in the sense of the sensor selection strategy. It reduces the communication traffic through exploiting the quantized measurement and increases the sampling interval to decrease the wake-up frequency of candidated tasking sensors. On this basis, an energy optimization model is devised to decrease the whole energy consumption of wireless sensor networks. Simulation experiments verified that the algorithm can allocate the bandwidth by quantization mechanism to reduce the traffic and extend the tracking sampling intervals as soon as possible to significantly reduce energy consumption of wireless sensor networks.更多还原