移动无线传感器网络中分布式重聚类算法研究

移动无线传感器网络中,节点的移动性影响着层次化聚类之后的网络结构,从而影响聚类内部节点间通信时的数据送达率与能耗.为了降低节点移动性的影响,本文提出了一种分布式重聚类算法.该算法基于已聚类网络,利用粒子滤波算法对节点当前位置进行估计,并结合移动模型预测下一时刻位置;处于聚类边界的非簇头节点周期性地评估自身是否需要重聚类,并在需要时通过与所属聚类及目标聚类的簇头节点通信,将自身重聚类到目标聚类中.仿真结果表明,在重聚类周期较小时,该算法能够使节点在移动过程中保持合理的通信距离,并在数据送达率与能耗方面优于现有的算法.

基于低秩稀疏约束的自权重多视角子空间聚类

多视角子空间聚类是一种利用视角之间的互补信息,找到视角间统一的表示并发现潜在分组结构的方法,近年来已成为机器学习的研究热点.提出一种基于低秩稀疏约束的自权重子空间聚类算法.具体的,低秩稀疏约束能发现数据的全局和局部结构信息,使自表示矩阵呈现稀疏性和低秩的特点;而自权重方法利用视角表示矩阵与共享相似度矩阵之间距离的反比为每个视角分配合理的权重,同时学习到一个视角之间共享的相似度矩阵,降低受损视角对于共享相似度矩阵的影响.以上提到的两种方法组成一个统一的优化框架,再使用增广拉格朗日乘子交换方向最小化方法(ALMADM)对提出的聚类算法进行优化.在基准数据集中的实验结果证明该算法比其他算法更有效.

基于可靠性的正则化加权软k-均值的子空间聚类

子空间聚类已经广泛应用于多个涉及高维数据聚类应用领域,受到机器学习研究者的广泛关注.子空间聚类方法是一种使用特征选择的聚类分析技术,通过选择重要特征子集实现对高维空间的低维表示,在实际应用中能够取得更好的性能,成为流行的高维数据聚类方法.与硬聚类方法相比,软聚类能够给出复杂数据更有意义的划分.扩展k-均值聚类并提出基于可靠性的正则化加权软k-均值新的子空间聚类方法(Reliability-based regularized weighted soft k-means clustering algorithm,RRWSKM),该方法能够计算每个特征对每个聚类的贡献度,从而找到与不同聚类相关的重要特征子集.另外,该方法能够通过调整模型参数准确地辨识数据模式,具有良好的聚类性能.该方法把维度加权熵和划分熵作为正则化项引入到目标函数,避免过拟合问题同时使更多的特征参与辨识聚类.为了提高算法的鲁棒性,使用可靠性测度获得特征权重初始值,提高算法的可靠性和性能.考虑到该算法是非凸优化问题,使用迭代优化方法得到优化问题的最优解.使用多个实际数据集对本文算法进行仿真验证,结果表明,与其他子空间聚类算法相比,该算法能够有效发现高维数据的低维表示,具有良好的聚类性能,适合高维数据的聚类.

基于密度聚类算法的电力通信监测分析

为解决传统基于密度的噪声应用空间聚类(DBSCAN)算法对输入参数设置敏感,以及传统的边缘计算框架计算成本高、计算时间过长等问题,创新性地提出了一种单遍权重K-means(SPWK)聚类算法。构建了电力通信网络故障及入侵监测模型,并将深度强化学习技术与边缘计算相结合,以降低计算成本和计算时长。仿真试验结果表明:SPWK聚类算法的迭代次数更少,平均执行时间以及总聚类时间分别低于其他算法67.5%、37.5%,加速比高出76.4%以上,聚类效率更高;边缘计算优化方法的服务器占用时间以及计算等待时间分别低于其他算法70.4%以上和79.2%以上,性能更优;电力通信监测模型对异常数据的平均识别准确率高出其他算法23.86%以上,入侵检测率高出其他算法4.8%以上,误报率降低65.4%以上,具备优异的检测性能。综上所述,所提故障及入侵监测模型以及边缘计算优化方法的性能均优于其他流行方法,适合在电力通信监测研究中推广使用。

一种曲面特征保持的航空叶片点云精简方法

针对现有点云精简方法在航空薄壁叶片叶缘高曲率特征区域及点云稀疏区域存在取样不足的问题,提出了一种基于重聚类策略的精简方法。通过移动最小二乘法定义点——曲面距离函数,区分高低曲率特征,建立了原始点云和精简点云之间的距离关系,在高曲率区域重聚类,实现叶缘高曲率特征保持;并在重聚类时判断点云疏密程度,对稀疏区域进行重聚类。航空叶片、铸造模具等典型复杂曲面测量点云的精简过程中,该方法相比于均匀采样法、层次聚类法,在高曲率区域可保持较高的几何精度。

财务内控信息化方法探索

采用目前方法对企业财务信息进行控制时,未考虑降维处理财务信息,导致财务内控的信息化建设受数据维数的影响较大,存在Kappa系数和特征贡献率低的问题。为此提出财务内控信息化方法探索。采用加权距离度量测度方法,降维处理财务信息,抑制干扰样本点并充分发挥样本点的贡献。通过自适应权重聚类算法,对降维处理后的财务数据进行聚类处理,获得谱系聚类树状图,实现企业财务内控信息化。实验结果表明,所提方法财务内控的信息化建设受数据维数的影响较小,Kappa系数和特征贡献率较高。

Primary Comments on Chemotaxonomy of Paris spp. Based on Saponins Analysis

Aim The several species of the genus Paris called ＂Chonglou＂ are famous traditional Chinese herbal medicines. We established the quantitative analysis method of the steroidal saponins in some species of the genus Paris and discussed their relations. Methods We detected the contents of 11 steroidal saponins in Paris samples with a Kromasel C18 （ 150 mm× 4.6 mm ID, 5μm） column which was subjected to gradient elution with acetonitrile-water （30：70- 60：40, V/V） at a flow rate of 1 mL· min^-1 by HPLC-ELSD and established chemical cluster tree using SPSS 11 software. Results All the samples could be separated and calibration curves of 11 saponins were prepared. We successfully detected the contents of 11 steroidal saponins in 14 Paris spp. in 30 min. The recovery for the assay of saponins was between 95 % and 97 %. The RSD of precision of 11 saponins and stability of samples were below 3 %. Chemical phylogenetic tree based on saponin contents indicated that 17 samples of Paris spp. clustered separately. Conclusion The established method is accurate and convenient, and suitable for the quantitative analysis of these 11 steroidal saponins in Paris spp.. The chemical phylogenetic tree is in accordance with Takhtajian classical taxonomy.

基于引用内容与位置的共被引分析改进研究

传统的共被引研究没有考虑到共被引的两篇文献在施引文献中的位置，只要两篇文献共同被引用，其共被引关系将被等同对待。然而，共被引文献发生在施引文献的不同位置时，其关系强度也有所不同。根据共被引的两篇文献在施引文献中的位置，本文将共被引关系划分四个层次，分别是句子层次共被引、段落层次共被引、章节层次共被引和文章层次共被引，并提出一种基于引用内容相似度的共被引关系权重计算方法，将共被引位置与内容相结合，使共被引关系权重更具客观性和准确性。以3本BMC期刊作为研究对象，计算出共被引4个层次的权重分别为1、0．77、0．64和0．56。在比较加入权重的共被引聚类结果与传统共被引聚类结果后，发现加入共被引权重后的共被引聚类，不仅聚类内文献间关系更紧密，还能更好地揭示施引文献所表达的主题，从而验证了这种赋予权重的共被引分析方法的有效性。

基于改进YOLOv4-Tiny的蔗芽识别方法

为解决蔗芽识别在蔗种定向机械化种植速度不匹配问题,提出基于改进YOLOv4-Tiny的蔗芽快速识别方法。将主干网络融合SE-Resnet模块实现注意力机制以增强蔗芽的特征;颈部网络结构多增加一层预测尺度并进行锚框的K-means重聚类,利用浅层网络的细节信息来提高模型对小目标蔗芽的检测能力;设计人机交互界面实时显示蔗芽识别定位信息。该研究将改进YOLOv4-Tiny算法和NCS2加速推理部署在树莓派4B设备中测试,试验结果表明:识别蔗芽精度达到95.87%,平均精度均值mAP为92.46%,基于树莓派检测速度为0.61 s,模型大小仅为23.2 MB。实现部署在嵌入式设备中准确快速识别蔗芽,解决蔗种蔗芽识别速度慢制约蔗种机械化播种速度问题,为蔗种机械化定向种植提供解决方案。

DCAD:a Dual Clustering Algorithm for Distributed Spatial Databases

Spatial objects have two types of attributes： geometrical attributes and non-geometrical attributes, which belong to two different attribute domains （geometrical and non-geometrical domains）. Although geometrically scattered in a geometrical domain, spatial objects may be similar to each other in a non-geometrical domain. Most existing clustering algorithms group spatial datasets into different compact regions in a geometrical domain without considering the aspect of a non-geometrical domain. However, many application scenarios require clustering results in which a cluster has not only high proximity in a geometrical domain, but also high similarity in a non-geometrical domain. This means constraints are imposed on the clustering goal from both geometrical and non-geometrical domains simultaneously. Such a clustering problem is called dual clustering. As distributed clustering applications become more and more popular, it is necessary to tackle the dual clustering problem in distributed databases. The DCAD algorithm is proposed to solve this problem. DCAD consists of two levels of clustering： local clustering and global clustering. First, clustering is conducted at each local site with a local clustering algorithm, and the features of local clusters are extracted clustering is obtained based on those features fective and efficient. Second, local features from each site are sent to a central site where global Experiments on both artificial and real spatial datasets show that DCAD is effective and efficient.

Research on Some Bus Transport Networks with Random Overlapping Clique Structure

On the basis of investigating the statistical data of bus transport networks of three big cities in China,wepropose that each bus route is a clique(maximal complete subgraph)and a bus transport network(BTN)consists of alot of cliques,which intensively connect and overlap with each other.We study the network properties,which includethe degree distribution,multiple edges' overlapping time distribution,distribution of the overlap size between any twooverlapping cliques,distribution of the number of cliques that a node belongs to.Naturally,the cliques also constitute anetwork,with the overlapping nodes being their multiple links.We also research its network properties such as degreedistribution,clustering,average path length,and so on.We propose that a BTN has the properties of random cliqueincrement and random overlapping clique,at the same time,a BTN is a small-world network with highly clique-clusteredand highly clique-overlapped.Finally,we introduce a BTN evolution model,whose simulation results agree well withthe statistical laws that emerge in real BTNs.

Refracturing candidate selection for MFHWs in tight oil and gas reservoirs using hybrid method with data analysis techniques and fuzzy clustering

The selection of refracturing candidate is one of the most important jobs faced by oilfield engineers. However, due to the complicated multi-parameter relationships and their comprehensive influence, the selection of refracturing candidate is often very difficult. In this paper, a novel approach combining data analysis techniques and fuzzy clustering was proposed to select refracturing candidate. First, the analysis techniques were used to quantitatively calculate the weight coefficient and determine the key factors. Then, the idealized refracturing well was established by considering the main factors. Fuzzy clustering was applied to evaluate refracturing potential. Finally, reservoirs numerical simulation was used to further evaluate reservoirs energy and material basis of the optimum refracturing candidates. The hybrid method has been successfully applied to a tight oil reservoir in China. The average steady production was 15.8 t/d after refracturing treatment, increasing significantly compared with previous status. The research results can guide the development of tight oil and gas reservoirs effectively.

Research on Distribution Pattern of Heavy Metals in Yanzhou Coalfield

Heavy metal pollution brings extensive concerns since 1940s. In order to assess the heavy metal pollution on the farmland of Yanzhou coalfield, 216 soil samples and 54 combined samples were collected. Lead, cadmium, chromium, copper, zinc, and nickel contained in both topsoil and deep soil were analyzed using atomic absorbent spectrometry analyzer (AAS). Fuzzy clustering method was used in data processing. And fuzzy synthetic assessment was applied to assess the soil contamination by heavy metals. The result shows that Yanzhou coalfield has been polluted by the heavy metals to some extent.

Synchronization in Triangled Complex Networks

Using a tunable clustering coeffcient model withoutchanging the degree distribution, we investigate the effect of clustering coefficient on synchronization of networks with both unweighted and weighted couplings. For several typical categories of complex networks, the more triangles are in the networks, the worse the synchronizability of the networks is.

Reconstructing bubble profiles from gas-liquid two-phase flow data using agglomerative hierarchical clustering method

The knowledge of bubble profiles in gas-liquid two-phase flows is crucial for analyzing the kinetic processes such as heat and mass transfer, and this knowledge is contained in field data obtained by surface-resolved computational fluid dynamics (CFD) simulations. To obtain this information, an efficient bubble profile reconstruction method based on an improved agglomerative hierarchical clustering (AHC) algorithm is proposed in this paper. The reconstruction method is featured by the implementations of a binary space division preprocessing, which aims to reduce the computational complexity, an adaptive linkage criterion, which guarantees the applicability of the AHC algorithm when dealing with datasets involving either non-uniform or distorted grids, and a stepwise execution strategy, which enables the separation of attached bubbles. To illustrate and verify this method, it was applied to dealing with 3 datasets, 2 of them with pre-specified spherical bubbles and the other obtained by a surface-resolved CFD simulation. Application results indicate that the proposed method is effective even when the data include some non-uniform and distortion.

Research on Novel Natural Image Reconstruction and Representation Algorithm based on Clustering and Modified Neural Network

In this paper, we conduct research on the novel natural image reconstruction and representation algorithm based on clustenng and modified neural network. Image resolution enhancement is one of the earliest researches of single image interpolation. Although the traditional interpolation and method for single image amplification is effect, but did not provide more useful information. Our method combines the neural network and the clustering approach. The experiment shows that our method performs well and satisfactory.

A multi-view K-multiple-means clustering method

The K-multiple-means(KMM)retains the simple and efficient advantages of the K-means algorithm by setting multiple subclasses,and improves its effect on non-convex data sets.And aiming at the problem that it cannot be applied to the Internet on a multi-view data set,a multi-view K-multiple-means(MKMM)clustering method is proposed in this paper.The new algorithm introduces view weight parameter,reserves the design of setting multiple subclasses,makes the number of clusters as constraint and obtains clusters by solving optimization problem.The new algorithm is compared with some popular multi-view clustering algorithms.The effectiveness of the new algorithm is proved through the analysis of the experimental results.

Heavy minerals clustering analysis in application of provenance analysis of Kong 2 Member in Kongnan area

The main task of provenance analysis is to determine the source of sediments and the position of parent rocks.Provenance analysis may find out the relationship between erosion districts and sediment zone,between the uplift and the depression in the process of basin development.The authors use the method of heavy mineral clustering analysis and estimate the provenance direction of Huanghua Depression in the Paleogene Kong 2 Member.Research shows that there were five provenance areas of Kong 2 Member in Kongnan area.They are western(Shenusi),northwestern(Cangzhou),eastern(Ganhuatun),northeastern and southeastern.The main provenance areas were northwestern and western,while the southern provenance could not be ruled out.And these areas are consistent with the known provenance areas.

真实世界医院集中监测中药不良反应的类不平衡高维预测及其风险因素分类识别

目的:基于真实世界医院集中监测的中药上市后安全性数据,实现中药不良反应的类不平衡高维预测并分类识别影响药品不良反应(ADR)发生的风险因素。方法:采用集成聚类重采样结合正则化Group Lasso回归,对ADR类不平衡的数据进行高维平衡处理,进而集成平衡数据实现ADR预测及其风险因素分类识别。结果:对中药安全性监测数据的示例研究结果显示,建立的ADR预测模型在测试集上的预测正确率、ADR发生的预测灵敏度、ADR未发生的预测特异度及受试者工作特征曲线下面积(AUC)4个指标均达到0.8以上。同时,该方法从600个高维变量中筛选出40个影响ADR发生的保护因素或危险因素,并分类加权识别各类因素对ADR发生的影响,重要风险因素类别依次为既往史、给药方案、合并用药名称、病症情况、合并用药数量、个人资料。结论:在ADR罕见且存在大量临床变量的真实世界数据中,本文实现了精准ADR类不平衡高维预测,并分类识别出关键风险因素及其所属类别的临床重要性,为临床合理用药及联合用药提供风险预警,也为中药上市后安全性再评价提供科学依据。

基于YOLOv3的军事目标检测算法改进

YOLOv3目标检测模型预设的anchorboxes是在VOC数据集上采用K-means聚类算法得到,并不适用于本文研究的巡飞弹对地目标侦察的使用场景。同时,K-means聚类算法在训练数据集上无法得到稳定的最优解。本文通过改进聚类算法,并对军事目标数据集重聚类后,更新anchor boxes数据。训练模型后的avg-IOU和loss曲线表明,改进算法使模型更快速地收敛,同时能够获得更高质量的目标检测结果。