基于网络编码的传感网数据流并行计算技术

发布时间：2018-01-12 12:00

本文关键词：基于网络编码的传感网数据流并行计算技术　出处：《南京理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：无线传感网处于物联网末梢,主要负责物联网末端信息采集。随着物联网的发展,无线传感网使用越来越广泛,涉及领域也越来越多。云计算旨在对有意义的大规模数据集进行有效的专业化处理。无线传感网和云计算相结合已经成为一个新的发展方向。本文设计的基于网络编码的传感网数据流并行计算系统,通过将无线传感网和云计算技术相结合,以实现大规模传感网数据集下的异常数据的快速聚类,为用户决策提供有利辅助。本文的主要工作体现在以下方面:(1)针对无线传感网中节点计算能力低,能量受限等不足,本文首先引入Reed-Solomon码设计了一种构造稀疏矩阵的方法来提高无线传感网中节点的能量利用率。实验证明了本文设计的方法能够提高无线传感网中节点的能量利用率。(2)网络编码解码工作会造成信宿时延,而且会增加节点的计算负担。本文设计了适用于Hadoop集群的分布式解码算法,将解码工作放在集群中进行,利用集群强大的计算能力来提高解码效率,减轻了无线传感网节点负担,并通过实验证明集群并行解码的可行性,同时分析了集群中相关因素对解码效率的影响。(3)针对传统MapReduce框架下用于感知数据的异常检测的k-means算法会造成极大的I/O消耗问题,提出了改进的基于MapReduce的单遍k-means算法。文中,从理论上证明了本文提出的方法能够降低程序执行时的I/O消耗,并且实验结果显示,本文设计的算法相对于传统的基于MapReduce的k-means算法,在保证聚类效果的同时,能够降低执行时间。
[Abstract]:Wireless sensor network is mainly responsible for Internet terminals, networking information collection terminal. With the development of IOT, wireless sensor network is more and more widely used, more and more involved in the field of cloud computing. Aimed at professional and effective for large-scale data meaningful set. Wireless sensor network and cloud computing has become a combination of a new direction of development. The design of the sensor network data encoding network flow based on parallel computing system, the wireless sensor network and cloud computing technology, to realize the fast clustering of abnormal data of large-scale sensor network data set under the favorable support for user decision. The main work of this paper is reflected in the following aspects: (1) for nodes in wireless sensor networks with low computing power, limited energy shortage, this paper introduces the design of Reed-Solomon codes is a method to improve the structure of sparse matrix Node line sensor network energy utilization. Experiment proves that this design method can improve the nodes in wireless sensor network energy utilization. (2) decoding network encoding will cause sink delay, but also increases the computational burden of nodes. This paper designs a distributed decoding algorithm for Hadoop cluster, the decoding in the cluster, to improve the decoding efficiency by using the powerful computing capability of cluster, reduce the burden of the node of wireless sensor network, and the experimental results proved that the feasibility of parallel cluster decoding, and analyses the related factors in the cluster on the decoding efficiency. (3) according to the traditional MapReduce framework k-means algorithm for anomaly detection sensing data the I/O will cause great consumption problems, proposed an improved MapReduce algorithm based on single pass k-means. In this paper, theoretically proved that this method can The I/O consumption of program execution is reduced, and the experimental results show that the algorithm designed in this paper can reduce the execution time while guaranteeing the clustering effect compared with the traditional MapReduce based k-means algorithm.

【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN929.5;TP212.9

【参考文献】