【作者】 宋欣；

【导师】 张景中；

【作者基本信息】 东北大学 ， 计算机应用技术， 2012， 博士

【摘要】 数据收集是无线传感器网络的基本功能,也是大多数监测应用的基础。以感知数据为中心的无线传感器网络数据收集技术的主要研究目标是在数据收集过程中减少网络的能量消耗,延长网络的生命周期,避免重新部署无线传感器网络监测系统带来的巨大开销。针对无线传感器网络事件监测应用中,同一监测区域附近的传感器节点收集的感知数据具有极大的时空相关性这一特点,为了减少网络中的数据传输量,降低节点间的通信能耗,提出了一种能量高效的基于线性回归的无线传感器网络分布式数据收集优化策略,通过构建局部感知数据的线性回归模型来表示和预测传感器节点实际感知数据监测值,在误差允许的范围内,节点不需要传输实际监测的感知数据到汇聚节点,而只传输回归模型基函数的参数信息,在不失数据基本结构特征的基础上有效地减少了传感器节点间频繁的数据传输带来的通信开销,而且感知数据的线性回归模型采用计算复杂度低的增量更新方法。仿真实验结果表明,基于线性回归的无线传感器网络数据收集优化策略能以较少的网络能耗实现有效的感知数据预测和估计,达到了延长网络生命周期的目的。为了优化部署在地理位置重叠的、可协作的、多个并发应用监测任务的大规模无线传感器网络数据收集问题,降低独立部署的无线传感器网络的监测成本,减少节点间感知数据传输量,提出了基于虚拟传感器网络的并发应用数据收集优化算法。考虑在混合的、异构的、部署在同一物理区域的无线传感器网络基础上,逻辑上形成多个虚拟传感器网络的方式来解决同一地区多目标并发应用监测问题,在数据收集过程中,提出了基于MinMax算子的数据传输阈值控制方法。并且在形成的虚拟传感器网络基础上,设计了基于统计假设检验的虚拟传感器网络容错策略,从感知数据的空间相关性和时间相关性入手,利用随机过程描述感知数据的时间相关性、传感器节点检验本地数据收集序列与事件统计特征的符合程度来判断错误是否发生。仿真实验结果表明,提出的算法能有效的减少网络的总能量消耗和延长网络的生命周期,并且在传感器节点错误概率增加的情况下,系统能保持较理想的节点错误识别率和事件区域监测概率。为了满足无线多媒体传感器网络应用中的服务质量多样化要求、利用网内数据处理技术和群智能优化算法,提出了面向多约束QoS的无线多媒体传感器网络数据收集优化算法。在基本人工鱼群优化算法基础上将人工鱼视野和移动步长参数改进为动态调整模式,提出了动态人工鱼群优化分簇算法,以更小的计算量和更快的搜索速度构建更优的网络分簇结构。为了满足无线多媒体传感器网络不同应用中多样化的QoS要求,提出了基于改进的蚁群和人工鱼群联合优化的多约束QoS路由机制。仿真实验结果表明,提出的算法能满足用户对多格式、多属性和多模态的数据传输的多约束QoS要求,能有效的降低网络的整体能耗和延长网络的生命周期。由于无线多媒体传感器网络实际应用中感知数据具有高复杂度和高维非线性等特点,数据收集和处理的能耗较高,为了降低网络中数据传输量,解决现有的基于线性的感知数据维数约简方法失效问题,提出了基于局部线性嵌入的多媒体感知数据维数约简算法。其核心思想是在降维映射前后保持源数据的局部近邻性质,即在嵌入空间每个采样点可以用它的近邻点线性表示,在低维空间中保持每个邻域中的权值不变,重构原数据点,使重构误差最小。针对局部线性嵌入算法在多媒体感知数据采样稀疏的情况下,数据维数约简的结果会失效这一缺点,进一步提出了基于局部线性逼近的多媒体感知数据维数约简算法。算法通过采用直接估计梯度值的方法达到局部线性逼近的目的,实现高维非线性数据的维数约简和特征表示。仿真实验结果表明,提出的算法在面对大量高复杂性、高维非线性的多媒体数据时,能有效保持原有数据集的几何特征,降低感知数据维数和简化感知数据表示,实现了非线性高维数据的有效聚类,与局部线性嵌入算法的实验结果比较,在感知数据采样稀疏的情况下,数据维数约简的效果有显著改进。上述成果的取得,对于无线传感器网络数据收集过程中减少网络总体能量消耗,延长网络生命周期,增强节点间的协作能力,提高数据收集算法的可靠性和约简冗余的感知数据维数等关键问题的研究具有重要意义。更多还原

【Abstract】 The optimization problem of the data gathering has been proposed as an essential paradigm for most the event monitoring applications in wireless sensor networks (WSN). The data gathering mechanisms are aimed at reducing the energy consumption of the network nodes and at the same time maximizing the network lifetime, so as to avoid the immense cost of redeployment for satisfying the demands of the monitoring system. Therefore, technologies of data gathering in WSN are systemically and deeply investigated in this dissertation. The research work and main contributions are as follows:Depending on the event monitoring application in WSN, the existence of the sensed information with spatial and temporal correlations bring significant potential advantages for the development of efficient data gathering strategies well suited for the WSN paradigm. In order to decrease the amount of data transmission in network and reduce the energy consumption, an energy-efficient linear regression based distributed data gathering optimization strategy was proposed. The linear regression model can accurately represent the feature of the original monitoring data. Rather than transmitting measurements to another node, nodes communicate constraints on the model parameters in the error bound, drastically reducing the communication cost. In addition, the linear regression model of the sensed data not only has advantage of low computational complexity but also is suitable for the incremental updating. The theoretical analysis and experimental results show that the proposed data gathering strategy is able to implement measurements prediction and estimate with lower communication cost. The designed algorithm achieves more energy savings and extends the wireless sensor networks lifetime.The data gathering optimization of the large-scale, collaborative and concurrent multi-task in WSN is very important, especially in the environments where multiple geographically overlapping wireless sensor networks are deployed. For decreasing the deployment cost of independent sensor networks each dedicated to a specific task, the virtual sensor network (VSN) based data gathering optimization algorithm for the concurrent application was proposed. The strategy builds a hierarchical structure by distributed clustering technique on the WSNs before forming virtual sensor network in logical to meet various monitoring requirement from different kind of application deployment. Then, for the data gathering on the VSN framework, the CH nodes set and update hierarchical thresholds by using the MinMax operator to restrict the data transmission. In order to enhance the robustness of the framework, a fault-tolerant strategy for VSN based on statistical hypothesis testing was proposed. On the cluster-tree based VSN architecture, the fault occurring was judged by checking the matching degree between local data reading sequence and event statistical characteristics depend on the spatial and temporal correlations of sensed data. The simulation results show that proposed algorithm achieves more energy savings and extends the wireless sensor networks lifetime. In addition, it can maintain the fault identification rate and event region monitoring probability at the satisfactory level with increase of the fault sensor nodes probability.Aiming at solving the transmission of the multimedia information in wireless multimedia sensor networks (WMSN) applications, which require both energy efficiency and Quality of Service (QoS) assurance, a multiple QoS metrics hierarchical data gathering algorithm based on swarm intelligence optimization for WMSN was proposed. In order to decrease runtime of basic artificial fish swarm optimization (AFSO), a dynamic artificial fish swarm optimization based cluster algorithm was present. The algorithm achieve more appropriate cluster and better global/local optimization through dynamic adjusting visual and step parameters of artificial fish. The2ASenNet (combination of improving ACO and AFSO) built a cluster head communication tree structure to meet various QoS requirements. Then, the2ASenNet adopted hybrid and behavior to produce diverse original paths, adding AFSO to ACO’s per iterative process, and the optimization path was explored according to multiple QoS constrained. The simulation results show that proposed algorithm can satisfy the multi-format, multi-attribute and multi-mode data transmission needs of so many different applications based on multiple QoS metrics.In order to extract detailed information about the environment, a mass of high complexity and high dimensional nonlinear data information were collected by wireless multimedia sensor network nodes. A nonlinear dimensionality reduction algorithm of multimedia data based on locally linear embedding (LLE) was proposed due to the failure clustering using the previous linear strategies. LLE can recover global nonlinear structure from locally linear fits. The each sampling data point and its neighbors lie on or close to a locally linear patch. The local geometry of these patches was characterized by linear coefficients for reconstructing each sampling data point from its neighbors. The reconstruction errors are measured by the minimizing cost function. Aiming at solving the failure problem of LLE when the source data is sparse, the locally linear approximating (LLA) based nonlinear dimensionality reduction algorithm of multimedia data algorithm was proposed. It reaches the aim of locally linear approximating through adopting the way of direct gradients estimation, thus realizes the dimensionality reduction of the high dimensional nonlinear data. The simulation results show that the proposed algorithms can preserve the original geometry topology structure and extract the most intrinsic character embedded in the high dimensional data space. The dimensionality reduction result of LLA algorithm has significantly improved compared with the LLE when the source data is sparse.The above contributions are of great significance to decrease the total energy consumption, prolong the network lifetime, enhance the collaboration of the sensor nodes, improve the dependability of the data gathering algorithm and reduce the dimensionality of the sensored data in the key technologies for WSN data gathering.更多还原