基于预测误差修正的时序链路预测方法

论文主要针对时序链路预测方法进行研究。分析了静态链路预测方法的弊端,认为忽视网络演化趋势信息会对链路预测产生负面影响;还提出了链路预测误差的概念用于描述网络趋势信息,并以此为基础提出一种基于预测误差修正的时序链路预测方法。该方法首先对待预测时刻之前一个时间窗口内的多幅网络图分别采用静态链路预测方法进行预测,记录每次的预测误差并计算其修正值,最后对待测时刻静态预测结果进行修正得到最终预测结果。通过在两个真实网络数据集上进行系列实验表明,该方法较大提升了静态链路预测方法的预测精确度,与另一种典型的时序链路预测方法相比其精度也有所提升,且算法时间复杂度较低。另外,实验中还发现链路预测误差序列与网络链路总数序列存在镜面对称关系,分析其内在原因证明了所提方法的普适性。     

一种基于节点间资源承载度的链路预测方法

链路预测旨在发现网络的未知、缺失连接,具有重要的实际应用价值。基于网络结构相似性的链路预测方法具有简单且有效的特点,受到各领域学者的普遍关注。然而,许多现有方法在计算节点间存在连接可能性时,忽视了节点间资源承载能力的影响。鉴于此,该文提出一种基于节点间资源承载度的链路预测方法。该方法首先通过分析节点间资源传输过程,进而对节点间资源承载能力进行量化,提出资源承载度。然后,基于资源承载度对节点间连接可能性的影响进行分析,并提出相应的链路预测方法。9个真实网络的实验结果表明,相比其他链路预测方法,该方法在3个衡量标准下均具有较高的预测精度。     

基于相似性指标的链路预测方法

预测网络缺失的链接或者未来可能出现的链接对于挖掘网络中重要信息及分析网络演化机制有着重要的意义。过去的几十年,许多来自不同领域的学者提出了许多方法解决链路预测问题。文章总结了近年来基于相似性指标的链路预测方法及链路预测方法在现实网络中的典型应用。文章根据相似性指标所考虑的信息的不同,分析了四类基于相似性指标的链路预测方法。     

基于深度卷积神经网络的多节点间链路预测方法

目前,链路预测的研究主要针对拓扑结构变化缓慢的社交网络,集中在单节点对的链路预测.本文针对拓扑变化频繁的机会网络,提出一种基于模式分类的多节点间链路预测方法.该方法基于混沌时间序列理论确定机会网络的切片时间,采用状态图表征网络的拓扑结构,借助深度卷积神经网络在特征提取上的优势,从状态图的演化过程中提取机会网络的结构特征,根据当前特征推断未来链路的演化模式,实现多节点间的链路预测.在ITC（Imote-Traces-Cambridge）真实数据集上的实验结果表明,相比于基于CN（Common Neighbor）、AA（Adamic-Adar）、Katz等预测方法,本文方法具有更好的精度和稳定性.     

基于霍克斯点过程的动态网络表示学习方法

网络表示学习是将网络中的节点映射到低维空间形成低维稠密特征向量的分布式学习方法.本文在现有网络表示学习研究的基础上,提出一种基于霍克斯点过程的动态网络表示学习方法.该方法基于霍克斯点过程有效结合了网络历史连边信息和网络演化中的三元闭包特性对当前节点产生连边的影响,解决了现有方法难以有效捕捉网络历史信息和演化特性的问题.在多种数据集的实验结果表明,本文提出的方法较其它方法在节点分类、链路预测和可视化等实验中的性能均有较大的提高,实验中的F1分数值和AUC值分别提高了3.72%~6.41%和2.22%~4.69%.     

基于复杂网络的Psor链路预测算法

复杂网络领域中,链路预测在网络演化规律的研究中被广泛应用,具有巨大的实际应用价值。针对现有的基于网络拓扑的链路预测方法存在预测精度偏低的问题,提出了一种基于复杂网络的Psor链路预测算法。该算法综合节点自身和邻居节点的度,定义了节点的Psor指数和Psor相似性指标进行链路预测。该算法能够全面考虑复杂网络的局部结构信息,更加准确地对复杂网络链路进行预测。仿真结果表明,Psor链路预测算法的预测精度相比8种经典的相似性算法的预测精度最高提升了37.96%。     

一种基于资源传输路径拓扑有效性的链路预测方法

链路预测旨在利用网络中已有的拓扑结构或其他信息，预测未连边节点间存在连接的可能性。资源分配指标具有较低复杂度的同时取得了较好的预测效果，但在资源传输过程的描述中缺少对路径有效性的刻画。资源传输过程是网络演化连边产生的重要内在动力，通过分析节点间资源传输路径周围拓扑的有效性，该文提出一种基于资源传输路径有效性的链路预测方法。该方法首先分析了节点间潜在的资源传输路径对资源传输量的影响，提出资源传输路径有效性的量化方法。然后，基于资源传输路径的有效性，通过对双向资源传输量进行刻画，提出了节点间传输路径的有效性指标。在12个实际网络数据集上的实验测试表明，相比其他基于相似性的链路预测方法，该方法在AUC和Precision衡量标准下能够取得更好的效果。     

银行交易网络的链路预测

基于银行交易具有动态变化、时效性和重复性的特点,文中通过对银行网络进行清洗和压缩,研究银行网络的基本拓扑统计性质和聚类结构,并得到交易网络满足复杂网络的小世界和无标度特性。针对已有的链路预测算法在动态网络预测中的不足,提出一种自适应的动态链路算法对银行客户交易进行预测。该方法在预测网络的基础上添加了节点重要性与节点连接强弱性两个特性,并将3种预测算法结合随机算法进行了对比分析。随后将这3种算法运用到具有动态交易特性的3类真实数据集中进行实验验证。实验结果显示,新算法的预测精度约为75%。将该算法与经典的预测算法进行比较发现,提出的算法在预测方面的性能提升了5%~10%。     

基于链路效用的3D-VANET可靠路由算法

针对三维车载自组织网络中,高速移动的车辆节点和复杂多变的链路状态导致车辆间通信链路不稳定的问题,通过引入软件定义网络技术实时获取网络状态并预测其变化过程,构建时-空演化图模型,并定义链路效用指标量化无线链路性能,然后建立基于链路效用的加权时-空演化图模型,最后将路由问题转化为多属性决策问题,设计基于链路效用的可靠路由算法.仿真结果表明,相对现有四种路由算法,本文所提路由算法在数据包传输率、端到端时延和路由负载率方面,性能均有明显提升.     

差分化社会网络节点角色的链路预测模型

以经典的Common Neighbor算法为例,提出了一种基于社区划分的差分化节点角色的链路预测模型,该模型首先采用Clauset-Newman-Moore算法挖掘社会网络结构属性,同时引入节点连接度和社区整体参与度的定义,差分处理社区内外邻接节点和不同社区的贡献,采用有监督的学习训练方法分别对社区内节点对和社区间节点对进行链路预测.人工网络和真实网络中的实验证明,该模型能够提高基于相似度算法对节点对链路预测的准确率,并为该类模型的研究提供一种新的方案.     

A PREDICTION BASED ALL-ZERO BLOCK DETECTION METHOD FOR H.263

In this letter, a novel method based on Temporal and Spatial Prediction (TSP) todetact all-zero DCT coefficients based on temporal and spatial prediction between neighboringblocks is proposed.The presented algorithm uses knowledge of all-zero block distribution inthe previous frame combined with the Sum of Absolute Difference (SAD) of the corresponding     

神经网络在时间序列预测中的应用研究

介绍了时间序列预测的基本概念、各种模型,分析了基于神经网络的时间序列预测方法,阐述了BP神经网络基本原理,提出了一种基于BP神经网络的时间序列的预测和方法。通过应用实例的分析表明,以此方法得到BP网络应用于非线性时间序列预测是可行的,神经网络方法可以成功地用于分析预测时间序列变量。     

A Novel Energy-Aware Target Tracking Method by Reducing Active Nodes in Wireless Sensor Networks

Energy consumption is one of the main challenges in wireless sensor networks. Additionally, in target tracking algorithms, it is expected to have a longer lifetime for the network, when a better prediction algorithm is employed, since it activates fewer sensors in the network. Most target tracking methods activate a large number of nodes in sensor networks. This paper proposes a new tracking algorithm reducing the number of active nodes in both positioning and tracking by predicting the target deployment area in the next time interval according to some factors including the previous location of the target, the current speed and acceleration of the target without reducing the tracking performance. The proposed algorithm activates the sensor nodes available in the target area by predicting the target position in the next time interval. The problem of target loss is also considered and solved in the proposed tracking algorithm. In the numerical analysis, the number of nodes involved in target tracking, energy consumption and the network lifetime are compared with other tracking algorithms to show superiority of the proposed algorithm.     

Research on social network influence maximization algorithm based on time sequential relationship

For the time sequential relationship between nodes in a dynamic social network,social network influence maximization based on time sequential relationship was proved.The problem was to find k nodes on a time sequential social network to maximize the spread of information.Firstly,the propagation probability between nodes was calculated by the improved degree estimation algorithm.Secondly,in order to solve the problem that WCM models based on static social networks could not be applied to time sequential social networks,an IWCM propagation model was proposed and based on this,a two-stage time sequential social network influence maximization algorithm was proposed.The algorithm used the time sequential heuristic phase and the time sequential greedy phase to select the candidate node with the largest influence estimated value inf (u) and the most influential seeds.At last,the efficiency and accuracy of the TIM algorithm were proved by experiments.In addition,the algorithm combines the advantages of the heuristic algorithm and the greedy algorithm,reducing the calculation range of the marginal revenue from all nodes in the network to the candidate nodes,and greatly shortens the running time of the program while ensuring accuracy.     

Study on AdaBoost-based link quality prediction mechanism

The link quality was vulnerable to the complexity environment in wireless sensor network.Obtaining link quality information in advance could reduce energy consumption of nodes.After analyzing the existing link quality prediction methods,AdaBoost-based link quality prediction mechanism was put forward.Link quality samples in deferent scenarios were collected.Density-based unsupervised clustering algorithm was employed to classify training samples into deferent link quality levels.The AdaBoost with SVM-based component classifiers was adopted to build link quality prediction mechanism.Experimental results show that the proposed mechanism has better prediction precision.     

Network traffic prediction method based on improved ABC algorithm optimized EM-ELM

In order to overcome the poor generalization ability and low accuracy of traditional network traffic prediction methods, a prediction method based on improved artificial bee colony (ABC) algorithm optimized error minimized extreme learning machine (EM-ELM) is proposed. EM-ELM has good generalization ability. But many useless neurons in EM-ELM have little influences on the final network output, and reduce the efficiency of the algorithm. Based on the EM-ELM, an improved ABC algorithm is introduced to optimize the parameters of the hidden layer nodes, decrease the number of useless neurons. Network complexity is reduced. The efficiency of the algorithm is improved. The stability and convergence property of the proposed prediction method are proved. The proposed prediction method is used in the prediction of network traffic. In the simulation, the actual collected network traffic is used as the research object. Compared with other prediction methods, the simulation results show that the proposed prediction method reduces the training time of the prediction model, decreases the number of hidden layer nodes. The proposed prediction method has higher prediction accuracy and reliable performance. At the same time, the performance indicators are improved.