Joint Task Scheduling, Resource Allocation, and UAV Trajectory under Clustering for FANETs

This paper establishes a new layered flying ad hoc networks(FANETs) system of mobile edge computing(MEC) supported by multiple UAVs,where the first layer of user UAVs can perform tasks such as area coverage, and the second layer of MEC UAVs are deployed as flying MEC sever for user UAVs with computing-intensive tasks. In this system, we first divide the user UAVs into multiple clusters, and transmit the tasks of the cluster members(CMs) within a cluster to its cluster head(CH). Then, we need to determine whether each CH’ tasks are executed locally or offloaded to one of the MEC UAVs for remote execution(i.e., task scheduling), and how much resources should be allocated to each CH(i.e., resource allocation), as well as the trajectories of all MEC UAVs.We formulate an optimization problem with the aim of minimizing the overall energy consumption of all user UAVs, under the constraints of task completion deadline and computing resource, which is a mixed integer non-convex problem and hard to solve. We propose an iterative algorithm by applying block coordinate descent methods. To be specific, the task scheduling between CH UAVs and MEC UAVs, computing resource allocation, and MEC UAV trajectory are alternately optimized in each iteration. For the joint task scheduling and computing resource allocation subproblem and MEC UAV trajectory subproblem, we employ branch and bound method and continuous convex approximation technique to solve them,respectively. Extensive simulation results validate the superiority of our proposed approach to several benchmarks.

基于聚类与稀疏字典学习的近似消息传递

基于传统字典学习的近似消息传递(approximate message passing,AMP)算法对训练样本数量的需求较高,且运算成本较高。本文引入双稀疏模型,构建基于稀疏字典学习的AMP框架,降低迭代过程中字典学习对训练样本数量的需求,提高压缩感知图像重建的质量与效率。进一步,提出基于聚类与稀疏字典学习的AMP算法,在迭代过程中依据图像块特征进行分类,并为各类图像块分别学习稀疏字典,实现自适应去噪。与基于传统字典学习的AMP算法相比,基于聚类与稀疏字典学习的AMP算法能够将重建图像的峰值信噪比提高0.20~1.75 dB,并且能够将运算效率平均提高89%。

分布式稀疏软大间隔聚类

虽然软大间隔聚类(Soft large margin clustering,SLMC)相比其他诸如K-Means等算法具有更优的聚类性能与某种程度的可解释性,然而当面对大规模分布存储数据时,均遭遇了同样的可扩展瓶颈,其涉及的核矩阵计算需要高昂的时间代价。消减此代价的有效策略之一是采用随机Fourier特征变换逼近核函数,而逼近精度所依赖的特征维度常常过高,隐含着可能过拟合的风险。本文将稀疏性嵌入核SLMC,结合交替方向乘子法(Alternating direction method of multipliers,ADMM),给出了一个分布式稀疏软大间隔聚类算法(Distributed sparse SLMC,DS-SLMC)来克服可扩展问题,同时通过稀疏化获得更好的可解释性。

近似k-median分类属性数据聚类

数据挖掘中解决分类属性数据聚类的算法有很多种,但大多数基于划分的方法得到的聚类中心一般不是数据集中的实际数据对象,缺乏实际的物理意义,有时会导致某一聚类为空。该文研究了近似k-median的求解算法,用数据的近似中值来代替模式进行聚类,提出了分类属性数据的近似k-median聚类算法,克服了一般基于划分的可分类属性数据聚类中所遇到的问题,仿真实验证明该算法有效。

基于特征聚类的近似查询分块选择方法

大数据集群环境中,随机访问的低效性使得基于行级别抽样的近似查询处理方法在构建样本时效率低下。该文将利用集群环境中数据分块存储的特性,以分块级别来进行抽样。在基准测试数据集和真实数据集上的实验,显示此方法在降低数据读取率,提高查询响应速度的同时,保持较高的查询精度。实验中,仅需要读取少于20%的数据就可以获得低于5%的查询误差,且为数据集每个分块的预计算的特征数据所需要的存储空间小于数据集所占空间的0.04%。

基于模糊计算法的皮带机信号辨识

为了使出料皮带机简化为悬臂梁后的激励力更贴近于合理值,先获得悬臂梁被激励时力传感器采集到的数据集,再运用模糊贴近方法计算最贴近试验数据的观测模糊向量。然后,为了确定试验信号中较大幅值的分布情况,对6个信号幅值运用模糊聚类算法进行迭代计算其较大幅值的聚类中心。最后,运用分类数据的较大幅值作为代表值,将含有的关键数据替代原信号幅值。分析表明移动破碎装置中出料皮带机的加速度振幅值能够较好地聚类,为技术人员提供了重要依据。

Approximation Algorithms on k-Correlation Clustering

In this paper,we consider the k-correlation clustering problem.Given an edge-weighted graph G(V,E)where the edges are labeled either“+”(similar)or“−”(different)with nonnegative weights,we want to partition the nodes into at most k-clusters to maximize agreements—the total weights of“+”edges within clusters and“−”edges between clusters.This problem is NP-hard.We design an approximation algorithm with the approximation ratio{a,(2-k)a+k-1/k},where a is the weighted proportion of“+”edges in all edges.As a varies from 0 to 1,the approximation ratio varies from k-1/k to 1 and the minimum value is 1/2.

New judging model of fuzzy cluster optimal dividing based on rough sets theory

To investigate the judging problem of optimal dividing matrix among several fuzzy dividing matrices in fuzzy dividing space, correspondingly, which is determined by the various choices of cluster samples in the totality sample space, two algorithms are proposed on the basis of the data analysis method in rough sets theory： information system discrete algorithm （algorithm 1） and samples representatives judging algorithm （algorithm 2）. On the principle of the farthest distance, algorithm 1 transforms continuous data into discrete form which could be transacted by rough sets theory. Taking the approximate precision as a criterion, algorithm 2 chooses the sample space with a good representative. Hence, the clustering sample set in inducing and computing optimal dividing matrix can be achieved. Several theorems are proposed to provide strict theoretic foundations for the execution of the algorithm model. An applied example based on the new algorithm model is given, whose result verifies the feasibility of this new algorithm model.

The Harmonic Approximation in Heavy-Ion Reaction Study

The derivation of the harmonic approximation of the Hamiltonian of a model of coupled three-dimensional harmonic oscillator is presented. It is shown how the splitting of the total Hamiltonian into the intrinsic and collective Hamiltonians leads to the description of the mechanism for energy dissipation in physical systems.

Quantum-Chemical Estimating Interaction of <i>sp²</i>-Carbon Nanoclusters with PE and PP Oligomers

The purpose of this work was to examine the interaction of graphene-like nanoclusters with fragments of polymers of the same nature, but somewhat different structure, for example, polyethylene (PE) and polypropylene (PP) by means of quantum chemistry. By method of density functional theory with the exchange-correlation functional B3LYP, the basis set 6 - 31 G (d, p) and the Grimme’s dispersion correction, the energy values have been calculated of interaction between nanocarbon fragments and oligomers of PE and PP, the most probable structures of their intermolecular complexes being optimized. A graphene-like plane of 40 carbon atoms and 16 atoms of hydrogen was chosen as a model for the surface of the graphene and carbon nanotubes (CNT). In order to take into account the dimensional effect of the surface of the nanotube fragment model on the interaction energy, in addition to the above described, two larger models were used, with the general formula C54H18 and C96N24. It has been found that the interaction energy of nanocarbon fragment with an oligomer of PP is greater, compared with PE, which is consistent with the experimental data on melting temperatures of pure polymers and nanotube-polymer composites. The polymer with a surface of nanocarbon fragment forms an intermolecular complex not bound covalently and retained by intermolecular dispersion forces. Oligomers of polymeric matters and carbon surfaces in formed nanocomplex are placed closer to each other than separate polymeric links between them.

Time-Dependent Density-Functional Calculations for Optical Spectra of Na2 and Na4 Clusters

With the frame of the time-dependent local density approximation, an efficient description of the optical response of clusters has been used to study the photo-absorption cross section of Na2 and Na4 clusters. It is shown that our calculated results are in good agreement with the experiment. In addition, our calculated spectrum for the Na4 cluster is in better agreement with experiment than the GW absorption spectrum.

基于电压曲线形态特征聚类的低压台区相位辨识

针对电压曲线分布集中导致低压台区用户相位难以准确辨识的问题,提出基于电压曲线形态特征聚类的低压台区相位辨识方法。首先,采用分段聚合近似(PAA)对配电变压器低压侧三相以及低压用户的电压曲线进行分段,提取反映整体分布的主要特征信息,减少冗余噪声的干扰。引入一阶导数与动态时间弯曲(DTW)增加对每个分段序列局部趋势变化的分析,并将其集成到PAA距离度量中,既弥补了均值特征缺乏对电压曲线趋势变化信息的考量,又能降低DTW的时间复杂度。然后,通过该距离度量综合判定电压特征序列的相似度,并以此改进传统k-means聚类的相似性度量,构建低压台区相位辨识模型。最后,通过实际台区算例分析验证了所提方法的有效性。

内近似约束空间整合的灵活性资源集群响应方法

针对规模化灵活性资源的管理需求,提出了一种内近似约束空间整合的灵活性资源集群响应方法。首先,建立储能、温控负荷和电动汽车充电桩的通用模型,并通过紧凑型约束空间描述其响应特性。其次,采用一种基于内近似与闵可夫斯基和的约束空间整合方法,实现对多个灵活性资源响应特性的整体描述。然后,采用一种灵活性资源集群的双层响应方法,既可利用整合后的等效模型形成集群的整体响应策略,又可根据个体特性和整体特性的映射关系,直接得到群内各个灵活性资源的响应策略。最后,通过仿真验证了所提方法的有效性。

多智能体编队控制中的迁移强化学习算法研究

针对多障碍环境下的多智能体系统协同编队避障与防撞问题,提出一种迁移学习与强化学习相结合的编队控制算法。在源任务学习阶段,利用值函数近似方法避免Q-表格求解法所需的大规模存储空间问题,有效降低对存储空间的需求,提升算法求解速度;在目标任务学习阶段,采用高斯聚类算法对源任务进行分类,根据聚类中心和目标任务之间的距离,选择最优的源任务类进行目标任务学习,有效避免了负迁移现象,进而提升了强化学习算法的泛化能力及收敛速度。仿真实验结果表明,所提方法能使多智能体系统在复杂的障碍环境下有效地形成并保持编队构型,同时实现避障与防撞。

基于概率推断的语义SLAM数据关联方法

针对传统视觉SLAM对环境语义信息理解不足的问题,语义视觉SLAM借助语义路标提高机器人的定位精度。语义路标的准确关联是实现机器人深层定位和导航的关键,错误的关联将导致机器人定位丢失。针对动态扰动和观测噪声扰动所产生的高关联模糊性问题,提出利用非参数聚类和随机近似推断结合的方法提高语义路标关联的准确性,通过正确的数据关联实现机器人的准确定位。仿真和KITTI数据集上的实验结果表明,在噪声干扰下该算法能够提高语义路标数据关联的准确性和鲁棒性,融合语义信息和几何信息优化机器人和语义路标的位姿,提高机器人的定位精度。

面向超大规模数据的自适应谱聚类算法

针对超大规模数据聚类过程中人为设定邻域参数及计算量庞大等问题,提出了一种基于近似自然近邻的自适应超大规模谱聚类算法(approximate natural nearest neighbor based self-adaptive ultra-scalable spectral clustering algorithm, AN^(3)-SUSC)。该算法首先通过混合代表选取缩小数据规模,在此基础上利用近似自然近邻自适应地确定局部邻域参数并构建相似矩阵,最后运用二部图进行迁移分割将数据空间映射到原超大规模数据空间中并完成谱聚类分析。超大规模数据集实验结果表明,该算法对超大规模数据集聚类效果有所提升,并且降低计算规模同时具有较高的鲁棒性和较强的自适应性。

基于LSD聚类拟合与KF的轨道检测算法

针对传统无人机巡检中视觉轨道识别的识别效率低、精度差的问题,提出了一种基于LSD的约束聚类拟合与卡尔曼滤波相结合的轨道线检测算法。首先针对由于镜头视角造成的视角畸变,采用IPM算法矫正,并通过LSD算法检测出轨道轮廓,在轨道间距约束将LSD检测结果进行聚类并进行最小二乘拟合得到轨道直线。然后根据轨道几何特征和无人机动力学特性建立数学模型,并结合卡尔曼滤波器对轨道坐标信息进行跟踪估计,以保证算法的稳定性和鲁棒性。采用无人机采集多个场景的轨道图像作为测试样本,对检测算法与其他算法进行对比实验。实验结果表明,本文轨道识别算法优于其他算法,其轨道准确识别率达到92.49%,识别速率达到23frame/s,满足轨道检测的稳定性和实时性要求。

融合局部结构学习的大规模子空间聚类算法

常规的大规模子空间聚类算法在计算锚点亲和矩阵时忽略了数据之间普遍存在的局部结构,且在计算拉普拉斯(Laplacian)矩阵的近似特征向量时存在较大误差,不利于数据聚类。针对上述问题,提出一种融合局部结构学习的大规模子空间聚类算法(LLSC)。所提算法将局部结构学习嵌入锚点亲和矩阵的学习,从而能够综合利用全局和局部信息挖掘数据的子空间结构;此外,受非负矩阵分解(NMF)的启发,设计一种迭代优化方法以简化锚点亲和矩阵的求解过程;其次,根据Nystr?m近似方法建立锚点亲和矩阵与Laplacian矩阵的数学联系,并改进Laplacian矩阵特征向量的计算方法以提升聚类性能。相较于LMVSC(Large-scale Multi-View Subspace Clustering)、SLSR(Scalable Least Square Regression)、LSC-k(Landmark-based Spectral Clustering using k-means)和k-FSC(k-Factorization Subspace Clustering),LLSC在4个广泛使用的大规模数据集上显示出明显的提升,其中,在Pokerhand数据集上,LLSC的准确率比k-FSC高28.18个百分点,验证了LLSC的有效性。

基于离群值检测的铅酸电池故障诊断方法

单体电池故障是导致铅酸电池组在运行过程中突发失效的主要原因,传统的识别方法需要依靠高精度的测试设备和复杂的电池机理模型,部署成本高、使用范围有限。考虑到故障电池和正常电池之间的等效电阻、等效电容等内部参数的差异,会由于浮充电流在时间尺度上通过电压凸显出来,文中设计了一种基于离群值检测的故障诊断方法,该方法采用时间序列聚类分析技术,对各个单体电池运行过程中产生的电压时间序列进行相似性分析,通过判断相异度较高离群值对故障电池进行定位。为了减小长跨度的时间序列造成的计算爆炸风险,采用分段聚合近似表示方法对时间序列进行降维处理,加快了计算速度。该方法可直接应用于微控制器,有较强的实用性。

基于Tsallis熵的近似差分隐私K-means算法

利用K-means算法对用户信息进行聚类时,存在隐私泄露的风险。差分隐私保护技术可提供严格的隐私保护,但目前大多数满足差分隐私的K-means算法在处理多维数据时,存在随机选择质心和噪声添加不均衡的问题,因而导致聚类结果不理想。为此,本文提出一种基于Tsallis熵的近似差分隐私K-means算法。针对质心选择的随机性问题,提出Tsallis熵对属性赋权的策略来优化对象间的欧氏距离,然后对比各对象到唯一随机初始质心的赋权欧式距离来确定其余初始质心,使算法在减少随机选择初始质心的同时,提高模型准确率;在此基础上,针对噪声添加不均衡的问题,提出一种能够平衡信噪比的隐私预算分配策略,然后对迭代质心加入高斯扰动,使算法在不增加计算复杂度的情况下满足(ε,δ)-差分隐私保护,同时提升扰动结果的准确性;最后在四个真实数据集上对算法进行有效性评价。实验结果表明,所提出的算法能够在保证用户隐私安全的同时实现高效用的聚类。