基于动态贝叶斯网络的WSNs链路质量预测
Link Quality Prediction for WSNs Based on Dynamic Bayesian Networks
作者：舒坚(南昌航空大学 物联网技术研究所, 江西 南昌 330063;南昌航空大学 软件学院, 江西 南昌 330063)；刘松(南昌航空大学 物联网技术研究所, 江西 南昌 330063;南昌航空大学 软件学院, 江西 南昌 330063)；刘琳岚(南昌航空大学 物联网技术研究所, 江西 南昌 330063;南昌航空大学 信息工程学院, 江西 南昌 330063)；谷小乐(南昌航空大学 物联网技术研究所, 江西 南昌 330063;南昌航空大学 信息工程学院, 江西 南昌 330063)
Author：SHU Jian(Internet of Things Technol. Inst., Nanchang Hangkong Univ., Nanchang 330063, China;School of Software, Nanchang Hangkong Univ., Nanchang 330063, China)；LIU Song(Internet of Things Technol. Inst., Nanchang Hangkong Univ., Nanchang 330063, China;School of Software, Nanchang Hangkong Univ., Nanchang 330063, China)；LIU Linlan(Internet of Things Technol. Inst., Nanchang Hangkong Univ., Nanchang 330063, China;School of Info. Eng., Nanchang Hangkong Univ., Nanchang 330063, China)；GU Xiaole(Internet of Things Technol. Inst., Nanchang Hangkong Univ., Nanchang 330063, China;School of Info. Eng., Nanchang Hangkong Univ., Nanchang 330063, China)
收稿日期：2016-10-26 年卷（期）页码：2017,49(2):152-159
期刊名称：工程科学与技术
Journal Name：Advanced Engineering Sciences
关键字：无线传感器网络;链路质量预测;动态贝叶斯网络;贴近度分析法
Key words：wireless sensor networks;link quality prediction;dynamic Bayesian networks;closeness analysis method
基金项目：国家自然科学基金资助项目（61363015；61262020；61501218；61501217）；江西省高等学校科技落地计划资助项目（KJLD14054）；江西省教育厅科学技术重点资助项目（GJJ150702）
中文摘要
无线传感器网络中，链路质量预测为数据可靠传输和上层网络协议性能的提高提供支撑。为进一步提高链路质量预测的准确性，提出基于动态贝叶斯网络（dynamic Bayesian networks，DBN）的链路质量预测机制。为避免单一评价指标的片面性，从链路信号质量、链路稳定性及非对称性3方面综合评价链路质量；采用K-means聚类算法对参数进行离散化预处理，得到各参数的离散区间；采用熵值法确定各参数的权重，以消除参数权重计算中主观因素的干扰；为避免最大隶属原则的缺陷，采用非对称贴近度分析法构建综合性的链路质量等级指标；借助贝叶斯网络（Bayesian networks，BN）处理不确定性问题的优势和BN分类器在分类上的良好性能，确定DBN的初始网络和转移网络，采用EM算法进行DBN模型的参数学习，从而构建了基于DBN的链路质量预测模型。实验结果表明了采用非对称贴近度分析法划分链路质量等级的合理性与DBN链路质量预测模型的合理性；与4C及FLI预测模型相比，本文模型具有更高的预测准确度。采用链路信号质量、链路稳定性及非对称性3个指标评价链路质量，采用DBN构建预测模型，可得到更准确及鲁棒性更好的链路质量预测结果。
英文摘要
In wireless sensor networks,the link quality prediction is a basic issue in guarantying reliable data transmission and upper network protocol performance.In this paper,a link quality prediction mechanism based on dynamic Bayesian networks (DBN) was proposed.The link quality was evaluated by link signal quality,link stability and link asymmetry,instead of a single evaluation index which could lead to a bias evaluation of link quality.The K-means clustering algorithm was then used to discrete the parameters so as to get intervals of parameters respectively.The weight of each parameter was determined by the entropy value method which could eliminate the interference of subjective factors in the process of weighting.Besides,in order to overcome the defects of maximum membership,asymmetry closeness analysis method was employed to construct the comprehensive link quality level indicators.The DBN based link quality prediction model was constructed,after both an initial network and a transfer network of DBN were determined.Finally,DBN parameters were determined by EM(expectation maximization) algorithm.Experimental results showed that it was reasonable to level link quality by using asymmetry closeness analysis method and to predict link quality with DBN model.Compared with the 4C and FLI prediction model,the proposed model based on DBN achieved better accuracy.In a word,the proposed mechanism with DBN model has better accuracy and robustness,which has been evaluated in terms of link signal quality,link reliability,and link asymmetry.
【关闭】
