基于高斯过程回归的链路质量预测方法

发布时间：2018-07-26 20:18

【摘要】：无线传感器网络是由部署在监控区域的传感器节点采用多跳方式传输数据而构成的自组织网络,已在诸多领域得到了广泛应用。由于传感器节点采用低功耗功率进行无线电通信,并且其所处环境恶劣、复杂多变,从而导致节点之间的通信链路不稳定。若能及时感知链路质量信息,为转发数据提供路由参考,则能有效减少数据重传次数,提高网络数据吞吐率。因此有效的链路质量预测方法对于提高数据传输的成功率、延长网络生存期非常重要。论文介绍了无线链路特性和现有的链路质量预测方法,分析了链路质量参数的定义和相关性。在此分析的基础上,提出基于高斯过程回归(Gaussian Process Regression,GPR)的链路质量预测模型。物理层参数实时灵敏,而直接测量包接收率需要消耗的能量比较多,因此本文构建物理层参数和包接收率之间的非线性映射关系。由于链路质量参数之间存在信息冗余,会降低模型的训练速度,本文首先利用灰关联分析方法分析链路质量参数之间的灰关联度,选取有效影响因子;再结合链路质量时间序列特点,选取合适的协方差函数,构建链路质量预测模型。无线链路通信易受到所处空间环境、地理位置、无线信号的影响和干扰。论文研究对象为节点静止的无线传感器网络,选取大学校园树林、教学楼实验室、图书馆广场和公路四个场景部署实验,收集处于不同方向和距离的多对节点之间的实验数据。论文分析了不同场景下各节点对之间的链路波动情况以及不同链路质量参数之间的灰关联度,确定预测模型的输入参数。论文选取两种链路进行实验分析和模型验证。实验结果表明,降维后的实验数据样本依然涵盖了链路质量信息,没有对预测准确性造成影响;在两种链路场景下,基于组合协方差函数形式的高斯过程回归模型的预测性能强于基于单一协方差函数的模型;与基于支持向量回归机的模型相比,本文提出的模型具有更好的预测精度。
[Abstract]:Wireless sensor network (WSN) is a self-organized network which is composed of sensor nodes deployed in monitoring area and transmits data in multi-hop mode. It has been widely used in many fields. The wireless communication of sensor nodes is based on low power consumption, and the environment is harsh and complex, which leads to the instability of communication links between the nodes. If we can perceive link quality information in time and provide routing reference for forwarding data, we can effectively reduce the number of data retransmissions and improve the throughput of network data. Therefore, effective link quality prediction method is very important to improve the success rate of data transmission and prolong the network lifetime. In this paper, the characteristics of wireless link and the existing link quality prediction methods are introduced, and the definition and correlation of link quality parameters are analyzed. Based on this analysis, a link quality prediction model based on Gao Si process regression (Gaussian Process) is proposed. The physical layer parameters are sensitive in real time, and the direct measurement of packet reception rate requires more energy consumption. Therefore, a nonlinear mapping relationship between physical layer parameters and packet reception rate is constructed in this paper. Because of the information redundancy among the link quality parameters, the training speed of the model will be reduced. Firstly, the grey correlation degree between link quality parameters is analyzed by using the grey correlation analysis method, and the effective influence factors are selected. According to the characteristics of link quality time series, a link quality prediction model is constructed by selecting appropriate covariance function. Wireless link communication is easily affected and interfered by the space environment, geographical location and wireless signal. The research object of this paper is wireless sensor network with static nodes. Four scene deployment experiments are selected, including university campus forest, teaching building laboratory, library square and highway, and the experimental data between multiple pairs of nodes in different directions and distances are collected. In this paper, the link fluctuation between each node pair in different scenarios and the grey correlation between different link quality parameters are analyzed, and the input parameters of the prediction model are determined. In this paper, two kinds of links are selected for experimental analysis and model verification. The experimental results show that the data samples after dimensionality reduction still cover the link quality information and have no effect on the prediction accuracy. The prediction performance of Gao Si process regression model based on combined covariance function is better than that based on single covariance function, and the proposed model has better prediction accuracy than the model based on support vector regression machine.
【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】