基于深度信念网络的WSNs链路质量预测机制研究

发布时间：2018-06-15 22:41

本文选题：无线传感器网络 + 链路质量预测　；参考：《南昌航空大学》2017年硕士论文

【摘要】：无线传感器网络(Wireless Sensor Networks,WSNs)是由数量较多的传感器节点以多跳自组织方式形成的网络。这些传感器节点在能量、通信以及计算与存储等多方面都受限,以及节点通常部署的环境复杂多变且存在各种干扰源,从而导致节点间通信链路质量并不可靠。链路质量预测机制是拓扑控制、路由协议和移动管理的基础,对于增强网络数据吞吐、延长网络寿命以及节省能量也起到相当重要的作用。本文旨在研究WSNs链路质量预测方法。论文对链路特性、链路质量预测方法研究现状、链路质量度量参数以及深度学习算法进行研究。根据链路质量的特性结合深度学习特征提取的优势,提出一种基于深度信念网络(Deep Belief Networks,DBN)的WSNs链路质量预测模型。该模型可分为链路质量评估层和链路质量预测层,评估层采用支持向量机(Support Vector Machine,SVM)对链路质量进行评估,获得链路质量等级;预测层采用DBN对连续的历史链路质量等级进行特征提取,并在最后一层加上Softmax回归器对下一单位时间的链路质量等级进行预测;最后根据均方根误差性能评价指标确定DBN模型的结构。根据WSNs实际应用,论文设定室内走廊、小树林和室外广场作为收集实验样本数据的场景。在各场景下的实验结果显示,本文提出的基于DBN的链路质量预测模型不管是与基于BP神经网络的预测模型对比,还是与基于Logistic预测模型对比,均具有更好的预测精度。
[Abstract]:Wireless Sensor Networks (WSNs) is a multi-hop self-organizing network formed by a large number of sensor nodes. These sensor nodes are limited in energy, communication, computation and storage, and the environment in which nodes are usually deployed is complex and has various interference sources, which leads to unreliable quality of communication links between nodes. Link quality prediction mechanism is the basis of topology control, routing protocol and mobile management. It also plays an important role in enhancing network data throughput, prolonging network lifetime and saving energy. The purpose of this paper is to study the link quality prediction method of WSNs. In this paper, the research status of link characteristics, link quality prediction methods, link quality measurement parameters and depth learning algorithm are studied. According to the characteristics of link quality and the advantages of feature extraction from deep learning, a link quality prediction model for WSNs is proposed based on Deep belief Networks (DBN). The model can be divided into link quality evaluation layer and link quality prediction layer. Support vector machine support Vector Machine (SVM) is used to evaluate the link quality, and the link quality level is obtained. The prediction layer uses DBN to extract the features of the continuous historical link quality, and adds Softmax regression to the last layer to predict the link quality level of the next unit time. Finally, the structure of DBN model is determined according to the RMS error performance evaluation index. According to the practical application of WSNs, indoor corridors, mangroves and outdoor squares are selected as the scene of collecting experimental data. The experimental results in various scenarios show that the proposed link quality prediction model based on DBN has better prediction accuracy than that based on BP neural network or Logistic model.
【学位授予单位】：南昌航空大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP212.9;TN929.5

【相似文献】