基于光谱与空间特征结合的改进高光谱数据分类算法

针对仅利用光谱信息进行分类未充分利用高光谱数据图谱合一特性的问题,提出了基于马尔可夫随机场的改进分类模型,利用基于最大后验概率的马尔科夫随机场模型进行光谱与空间信息的融合应用,采用基于光谱信息的概率支持向量机方法提高马尔科夫随机场模型中光谱能量函数项的类条件概率估计精度,设计基于信息传播策略、信息更新策略、多尺度传播策略的多重加速策略的高效置信传播优化算法,解决了马尔科夫随机场模型中全局能量最小化优化过程中计算复杂度高、计算耗时等问题。利用航空可见-近红外成像光谱仪AVIRIS对美国印第安纳州西北部的农业示范区数据进行应用分析,并与迭代条件模型、模拟退火、置信传播等方法进行性能比较,试验结果表明：该方法能够达到总体分类精度95.78%、Kappa系数0.933 4,优于现有马尔科夫随机场分类算法,并且计算效率比置信传播优化算法提高了3倍以上。     

基于概率假设密度的无线传感器网络多目标跟踪算法

针对无线传感器网络多目标跟踪过程中杂波难以去除以及由数据关联复杂带来的计算复杂度高的问题,将概率假设密度滤波器应用于无线传感器网络,以更好地对多目标状态信息进行融合估计;首先,建立簇-树型无线传感器网络模型,并运用随机有限集理论对目标状态模型和传感器观测模型进行描述;然后,根据目标与节点之间的距离设置观测阈值,当传感器节点测量值小于观测阈值时,概率假设密度滤波器将实时对该组测量数据进行处理,从而实现传感器网络对目标状态的联合检测与跟踪;仿真结果表明,在无线传感器网络的多目标跟踪应用中,该算法比粒子滤波算法具有更高的跟踪效率和精度。     

考虑边聚类与扩散特性的信息传播网络结构优化算法

优化网络结构以促进信息在网络中传播一直是复杂网络研究的重点,网络中边的聚类特性和扩散特性对信息传播具有重要作用. K-truss分解算法是一种利用边的聚类特性识别网络关键节点的算法,然而K-truss算法会受到网络中局部聚类结果 (即相互连接的假核结构)的影响,而这些假核结构里的节点对信息扩散能力通常较弱.为此,本文提出一种衡量边扩散特性的指标,研究发现一些位于网络边缘的边具有很好的扩散性,但这类边的聚类很低,并不利于信息传播.通过同时考虑边的聚类特性和扩散特性之间的制约关系,提出一种信息传播网络结构优化算法.为了验证所提算法的有效性,使用该算法对四个真实的网络进行结构优化,并使用经典的独立级联模型来验证网络结构优化前后信息传播的有效范围.结果表明:使用提出的算法优化后的网络拓扑可以有效提高信息传播范围;并且,优化后的网络其叶子节点数目降低、聚类系数降低以及平均路径长度降低.     

基于LS的双站纯方位无源定位算法

针对基线较短情况下传统的双站三角测距精度不高的问题,利用双站方位历史信息建立了双站无源定位模型,该模型可以利用多种线性估计器来求解。采用最小二乘算法来实现对目标距离的估计。通过仿真实验说明该模型能够实现对目标距离的估计,证明该算法用于无源定位是有效的。     

一种改进的基于信息传播率的复杂网络影响力评估算法

评价网络中节点的信息传播影响力对于理解网络结构与网络功能具有重要意义.目前,许多基于最短路径的指标,如接近中心性、介数中心性以及半局部(SP)指标等相继用于评价节点传播影响力.最短路径表示节点间信息传播途径始终选择最优方式,然而实际上网络间的信息传播过程更类似于随机游走,信息的传播途径可以是节点间的任一可达路径,在集聚系数高的网络中,节点的局部高聚簇性有利于信息的有效扩散,若只考虑信息按最优传播方式即最短路径传播,则会低估节点信息传播的能力,从而降低节点影响力的排序精度.综合考虑节点与三步内邻居间的有效可达路径以及信息传播率,提出了一种SP指标的改进算法,即ASP算法.在多个经典的实际网络和人工网络上利用SIR模型对传播过程进行仿真,结果表明ASP指标与度指标、核数指标、接近中心性指标、介数中心性指标以及SP指标相比,可以更精确地对节点传播影响力进行排序.     

基于粒子群算法优化BP神经网络漏钢预报的研究

针对标准BP神经网络中收敛速度慢以及易陷入局部最优解等问题,利用粒子群算法的全局搜索性,将粒子群算法应用到BP神经网络训练中建立了PSO-BP神经网络模型,结果表明改进模型不仅可以克服传统 BP 网络收敛速度慢和易陷入局部权值的局限问题,而且很大程度地提高了结果精度和 BP 网络学习能力,将此模型应用到结晶器漏钢预报系统中,并用某钢厂采集到的历史数据对该模型进行训练与测试,与标准BP神经网络测试结果进行分析与比较,实验表明PSO-BP网络模型预报更加实时、准确,具有很好的应用前景。     

浅海环境下数据同化声层析方法研究

讨论了适用于浅海的基于声速局部测量模型、声传播模型及声压场局部测量模型的声学数据同化方法,并给出了具体的执行算法。该算法根据最优化准则把局部声速和声压场的先验测量信息,以及声传播模型进行了有效地融合,提高了海洋环境参数估计的精度,为浅海声层析提供了新思路。同时,利用实验浅海声速测量数据,通过经验正交函数对实际海洋声速剖面进行了估计,并分析了各类噪声对水体声速场及海底声学参数估计精度的影响,验证了该执行算法的有效性。      

基于有向渗流理论的关联微博转发网络信息传播研究

微博网络的快速性、爆发性和时效性, 以及用户复杂的行为模式, 使得研究其信息传播模型及影响因素成为网络舆情的热点方向. 利用压缩映射定理, 分析不动点迭代过程的收敛条件, 得到有向网络信息传播过程的渗流阈值和巨出向分支的数值解法; 通过可变同配系数生成模型, 分析关联特征对信息传播的影响; 最后利用微博转发网络数据进行仿真对比实验. 结果表明: 虽然四类关联特征同时体现出同配、异配特征, 但信息传播结果更多受入度-入度相关性、入度-出度相关性影响; 通过删除少量节点的方法, 提取边同配比例, 验证大部分节点的四类关联特征呈现一致性.     

有限步传播范围期望指标判别节点传播影响力

在线社交网络逐渐成为人们不可或缺的重要工具,识别网络中具有高影响力的节点作为初始传播源,在社会感知与谣言控制等方面具有重要意义.本文基于独立级联模型,给出了一个描述有限步传播范围期望的指标-传播度,并设计了一种高效的递推算法.该指标在局部拓扑结构信息的基础上融合了传播概率对影响力进行刻画,能够较好地反映单个节点的传播影响力.对于多传播源影响力极大化问题,本文提出了一种基于传播度的启发式算法-传播度折扣算法,使得多个传播源的联合影响力最大.最后,将上述方法应用到三个真实网络中,与经典指标和方法相比,该方法不需要知道网络的全局结构信息,而是充分了利用网络的局部结构信息,可以较快地筛选出高传播影响力的传播源.     

直升机光电系统网络协同目标瞄准方法北大核心CSCD

针对直升机网络化和信息化的探测需求,提出了利用数据链信息快速引导机载光电系统实现目标瞄准的方法。分析网络信息引导过程中光电系统的目标瞄准指示误差,给出目标信息坐标转换和延迟处理算法以及光电系统瞄准线偏差估计方法,最后通过仿真实验重点研究了数据精度、数据率、延迟时间等对直升机光电系统网络协同目标瞄准的影响。仿真结果表明:减小数据延迟时间,提高数据率和数据精度是减小网络协同目标瞄准误差的有效途径,采用合理的目标信息延迟处理算法,对提高目标状态估计精度,减小目标瞄准线估计误差具有一定的作用。     

稀磁体系相变的相界关系

本文将纯键和纯座逾渗的普适关系作了推广,建立了键座混合逾渗的相界关系,并讨论了稀磁体系相变问题,给出了稀磁体系相变的相界关系,与以往文献中所给出的级数展开、有限集团近似,以及数值模拟等方法相比较,本文所给出的关系式简单、直观、物理概念清晰,且较为精确。 关键词：     

Multifragmentation of excited nuclei within a schematic hydrodynamical and percolation picture

We present a simple model to describe the evolution of hot and compressed nuclei. It is based on irrotational hydrodynamics and a 3-dimensional site-bond percolation model. Hydrodynamics is used if the evolution of the system is dominated by the mean field. This situation occurs when the fluctuations of the mean field are small. These latter are evaluated using the percolation model. In some cases it turns out that these fluctuations become very large and the system breaks up into several pieces (multifragmentation). The results of this process are described by the percolation model. We have obtained analytical or fitted expressions for all the results which compare well with those obtained in a previous and more involved model based on the same physical ideas. In particular we found that a noncompressed nucleus undergoes multifragmentation if the thermal excitation energy is larger than 70% of its binding energy. If the nucleus is compressed this value is notably decreased.     

Percolation and cluster distribution. III. Algorithms for the site-bond problem

Algorithms for estimating the percolation probabilities and cluster size distribution are given in the framework of a Monte Carlo simulation for disordered lattices for the generalized site-bond problem. The site-bond approach is useful when a percolation process cannot be exclusively described in the context of pure site or pure bond percolation. An extended multiple labeling technique (ECMLT) is introduced for the generalized problem. The ECMLT is applied to the site-bond percolation problem for square and triangular lattices. Numerical data are given for lattices containing up to 16 million sites. An application to polymer gelation is suggested.Supported by NSF Grant DMR76-07832 and NIH Grant NS08116-9.     

Double site-bond percolation model for biomaterial implants

We present a double site-bond percolation model to account, on the one hand, for the vascularization and/or resorption of biomaterial implant in bones, and on the other hand, for its mechanical continuity. The transformation of the implant into osseous material, and the dynamical formation/destruction of this osseous material is accounted for by creation and destruction of links and sites in two, entangled, networks. We identify the relevant parameters to describe the implant and its evolution, and separate their biological or chemical origin from their physical one. We classify the various phenomena in the two regimes, percolating or nonpercolating, of the networks. We present first numerical results in two dimensions.     

Multifragmentation of hot and compressed nuclei within a time dependent thomas fermi and percolation model

We have used a spherical time dependent Thomas Fermi model to study the expansion of hot and/or compressed nuclei. This approach was coupled to a site-bond percolation model to study the disassembly of the nucleus into many pieces (multifragmentation). We find that a non compressed nucleus undergoes multifragmentation if the thermal excitation energy is larger than 70% of its binding energy. If the nucleus is compressed this value is notably decreased.     

Limited universality at the percolation threshold in 2 to 6 dimensions

The universality of the spanning fraction R(p) of percolation is confirmed by comparing bond percolation with site-bond percolation in four to six dimensions. However, different boundary conditions change the universality class, as shown also for site percolation in two dimensions.     

Fragmentation of nuclei at intermediate and high energies in modified cascade model

The process of nuclear multifragmentation has been implemented, together with evaporation and fission channels of the disintegration of excited remnants, in nucleus-nucleus collisions using percolation theory and the intranuclear cascade model. Colliding nuclei are treated as face-centered-cubic lattices with nucleons occupying the nodes of the lattice. The site-bond percolation model is used. The code can be applied for calculation of the fragmentation of nuclei in spallation and multifragmentation reactions.     

Extension of the Arrowsmith–Essam Formula to the Domany–Kinzel Model

Arrowsmith and Essam gave an expansion formula for point-to-point connectedness functions of the mixed site-bond percolation model on oriented lattices, in which each term is characterized by a graph. We extend this formula to general k-point correlation functions, which are point-to-set (with k points) connectivities in the context of percolation, of the two-neighbor discrete-time Markov process (stochastic cellular automata with two parameters) in one dimension called the Domany–Kinzel model, which includes the mixed site-bond oriented percolation model on a square lattice as a special case. Our proof of the formula is elementary and based on induction with respect to time-step, which is different from the original graph-theoretical one given by Arrowsmith and Essam. We introduce a system of m interacting random walkers called m friendly walkers (m FW) with two parameters. Following the argument of Cardy and Colaiori, it is shown that our formula is useful to derive a theorem that the correlation functions of the Domany–Kinzel model are obtained as an m0 limit of the generating functions of the m FW.     

Monte Carlo simulation of the percolation process caused by the random sequential adsorption of k-mers on heterogeneous triangular lattices

Mixed site-bond percolation is studied for a random sequential adsorption process of k-mers on heterogeneous lattices with variable connectivity by means of Monte Carlo simulation. The percolation phase diagrams are built to render evidence about complex structures. Critical exponents are also calculated to show that the universality class corresponding to ordinary percolation in two dimensions is preserved.     

Impurity diffusion in a harmonic potential and nonhomogeneous temperature

Classical blockmodel is known as the simplest among models of networks with community structure. The model can be also seen as an extremely simply example of interconnected networks. For this reason, it is surprising that the percolation transition in the classical blockmodel has not been examined so far, although the phenomenon has been studied in a variety of much more complicated models of interconnected and multiplex networks. In this paper we derive the self-consistent equation for the size the global percolation cluster in the classical blockmodel. We also find the condition for percolation threshold which characterizes the emergence of the giant component. We show that the discussed percolation phenomenon may cause unexpected problems in a simple optimization process of the multilevel network construction. Numerical simulations confirm the correctness of our theoretical derivations.