基于非线性数据变换的离群点检测算法

为了提高高维数据集合离群数据挖掘效率,在分析了传统的离群数据挖掘算法优点和缺点的基础上,提出了一种离群点检测算法,首先将非线性问题转化为高维特征空间中的线性问题,然后利用非线性数据变换进行维数约减,对所得数据对象每个投影分量逐个判断数据点是否是离群点,通过实验证明该算法不仅可用于线性可分数据集的离群点检测,而且可用于线性不可分数据集的离群点检测,表明了算法的优越性。     

基于期望核密度离群因子的离群点检测算法

针对基于密度的离群点检测方法在不同分布的数据集上检测精度低的问题,提出了一种基于期望核密度离群因子的离群点检测算法。首先,引入k近邻和反向k近邻扩展邻域空间(ENS)代替传统的k邻域范围,更加全面地考虑数据对象的邻域信息;其次,在传统核密度估计(KDE)方法的基础上引入多元高斯函数,在扩展邻域空间内估计数据对象的密度,同时借鉴自适应核带宽的思想,更好地适应不同数据集的数据分布;然后,给出期望距离的概念,进一步区分局部离群点和位于低密度区域的正常点;最后,定义了期望核密度离群因子刻画数据对象离群程度。在人工数据集和真实数据集上对所提算法进行实验验证,并与部分传统算法进行对比,验证了所提算法的有效性。     

基于多尺度关联维数和流形学习的自动机故障诊断

针对自动机振动信号非平稳、非线性的特点,提出基于多尺度关联维数和线性局部切空间排列(linear local tangent space alignment,LLTSA)相结合的自动机故障诊断方法。首先,利用局部特征尺度分解(local characteristicscale decomposition,LCD)将自动机振动信号分解为不同尺度下的内禀尺度分量(intrinsic scale component),提取出反映状态信息的主要分量并计算各分量的关联维数。然后,利用线性局部切空间排列算法挖掘出可区分度更高的特征子集。最后,将得到的低维特征输入支持向量机进行识别,自动机故障诊断实验表明,所提方法具备较高的诊断准确率。此外,将LCD与经验模态分解(empirical mode decomposition,EMD)和局部均值分解(local mean decomposition,LMD)方法的诊断结果进行比较,验证所提方法的优势。     

基于特征选择和子空间搜索的离群点检测算法

为解决高维空间下基于密度的离群点检测的低精度和高计算量的问题,提出基于特征选择和子空间搜索的离群点检测算法。该算法先通过特征选择选择数据的主要属性,然后对这些属性进行子空间搜索,从而有效的降低维数和避免全局搜索。最后通过一个数据集的实验来说明该算法的有效性。     

基于流形学习和K-最近邻分类器的旋转机械故障诊断方法

针对旋转机械故障诊断需人工干预、精度低、故障样本难以获取等问题,提出基于流形学习和K-最近邻分类器（KNNC）的故障诊断模型。提取振动信号多域信息熵以全面反映设备运行状态并构造高维特征集;利用正交邻域保持嵌入(ONPE)非线性流形学习算法的二次特征提取特性进行维数约简使特征具有更好的聚类特性;基于改进的更适用于小样本分类KNNC进行模式识别,用轴承故障诊断案例证明该模型的有效性。     

基于离群点的前沿趋势探测方法研究

针对科学前沿探测中的低频现特征,选取Web of Science 2006 -2010年的国际空间站(ISS)数据集,以高频聚类词簇外的低频现离群点为研究对象,分别构造数据集的包容指数、临近指数和等价指数矩阵,通过基于聚类的优选离群点算法,在ISS人体研究、微重力生物学、空间站技术、微重力物理学、对地观测、空间天文学等...     

基于流形正则化的批量分层编码极限学习机

针对极限学习机在处理高维数据时存在内存能耗大、分类准确率低、泛化性差等问题,提出了一种批量分层编码极限学习机算法。首先通过对数据集分批处理,以减小数据维度,降低输入复杂性;然后采用多层自动编码器结构对各批次数据进行无监督编码,以实现深层特征提取;最后利用流形正则化思想构建含有继承因子的流形分类器,以保持数据的完整性,提高算法的泛化性能。实验结果表明,该方法实现简单,在NORB,MNIST和USPS数据集上的分类准确率分别可以达到92.16%、99.35%和98.86%,与其它极限学习机算法对比,在降低计算复杂度和减少CPU内存消耗上具有较明显的优势。     

基于分形的结构损伤检测方法

将分形维数用于结构损伤检测。研究结果表明,结构不同状态下的振动信号的分形维数有明显的不同,可以将分形维数作为结构损伤检测的特征量,为结构的损伤检测技术提供了一个新的分析方法。     

基于离群分析的水位异常识别研究

面对海量的地震前兆监测数据流,传统的方式难以实现合理的分析和利用。引入时序数据挖掘的技术对地震前兆数据中的一个重要参数-地下水位进行时序分析和识别水位变化异常。从水位数据的自身特点出发,依据突变系数检测离群点,根据离群点分段,再结合基于距离的离群算法找出水位变化异常模式。实验结果表明,该方法可以比较简单直观地发现水位的异常变化。     

一种基于非线性流形学习的故障特征提取模型

流形学习作为一种挖掘高维非线性数据内在几何分布特征的有效方法,可用于故障信号的特征提取。针对机械故障诊断中的非线性、故障征兆复杂的诊断问题,提出了一种基于非线性流形学习的故障特征提取模型。该模型针对采集样本的不同处理情形,分别运用Laplacian Eigenmaps算法及其增量、监督算法,进行故障样本的特征提取与分类,由于采用非线性的维数约简方式,极大地保留了故障信号中的整体几何结构信息,增强了故障模式识别的分类性能。最后通过工程实例应用,表明了所提特征提取模型的可行性和有效性。     

基于流形覆盖的无网格法

针对无网格迦辽金法中形函数构造过于复杂以至计算时间过长的问题,基于流形覆盖思想对节点的自由度重新构造,形函数则使用形式简单的Shepard函数,避免了为得到形函数进行多个矩阵相乘的运算,从而减少了计算时间.利用罚函数法施加边界条件,不仅易于实现,而且精度降低不大.     

Elman网络温度预测气密性检测研究

分析温度在差压法气密性检测中对检测准确度的影响及影响温度变化的因素,提出建立Elman网络,利用气密性检测平台在不同实验参数条件下得到的试验样本来训练网络。将训练好的神经网络用来预测气密性检测中待测密闭容器内的温度,并将预测得到的温度应用到气密性检测判断中。通过与应用较为广泛的BP网络和实际泄漏流量值比较可得,Elman网络进行41次训练后就达到预设目标,而且预测准确度更高。经过温度补偿后的泄漏流量值为12.67 m L/min(实际值为13 m L/min),相对误差为2.52%,极大地提高气密性检测的准确度。     

The research of unsupervised change detection method based on the clustering characteristic of 3D histogram

Abstract

One of the main problems related to unsupervised change detection methods based on the ‘difference image’ lies in the lack of efficient automatic techniques for discriminating between changed and unchanged pixels in the difference image. Such discrimination is usually performed by using empirical strategies or manual trial-and-error procedures, which affect both the accuracy and the reliability of the change detection process. To overcome such drawbacks, in this paper, we propose an automatic techniques (based on the clustering characteristic of 3D histogram) for the analysis of the difference image. The 3D histogram is formed by pixel grey levels, contiguous average grey levels and local average grey levels of the difference image. First, the optimal plane threshold and plane direction are searched by using maximal entropy principle based on the clustering characteristic of 3D histogram. Then, by using the optimal plane threshold and plane direction, a plane is established to segment the 3D histogram into changed clustering and unchanged clustering. Finally, the changed pixels in the difference image are discriminated according to the segmentation of 3D histogram. The theoretical analysis and experiment results confirm the effectiveness of the proposed method.     

Numerical manifold method based on the method of weighted residuals

Usually, the governing equations of the numerical manifold method (NMM) are derived from the minimum potential energy principle. For many applied problems it is difficult to derive in general outset the functional forms of the governing equations. This obviously strongly restricts the implementation of the minimum potential energy principle or other variational principles in NMM. In fact, the governing equations of NMM can be derived from a more general method of weighted residuals. By choosing suitable weight functions, the derivation of the governing equations of the NMM from the weighted residual method leads to the same result as that derived from the minimum potential energy principle. This is demonstrated in the paper by deriving the governing equations of the NMM for linear elasticity problems, and also for Laplaces equation for which the governing equations of the NMM cannot be derived from the minimum potential energy principle. The performance of the method is illustrated by three numerical examples.     

A process monitoring scheme based on independent component analysis and adjusted outliers

Recently, the independent component analysis (ICA) has been widely used for non-Gaussian multivariate process monitoring. An elliptical type measure is traditionally used for ICA-based process monitoring. However, it will not work appropriately since the extracted ICA components exhibit skewed distribution. Thus, this study aims to develop a novel process monitoring scheme for ICA. The basic idea of the proposed method is to first screen out outliers in order to describe well majority for training dataset. Hereafter, a rectangular type measure is applied to monitor the process. The efficiency of proposed monitoring scheme will be implemented via a five variables simulation example and a case study of Tennessee Eastman process. Results indicate that the proposed method cannot only deal with the contaminated training dataset but also shows superior fault detection ability when compared with alternative methods.     

基于面阵CCD的医用X光机焦点检测研究

针对星卡测试法存在的不足,采用电荷耦合器件(charge-coupled device,CCD)数字成像测量X射线管焦点。根据狭缝成像的原理,重点讨论面阵CCD的焦点成像,运用Matlab二值化图像处理的方法。通过实验测量,测得的焦点尺寸符合医用X射线辐射源检定规程要求。     

Dynamic high order numerical manifold method based on weighted residual method

High Order Numerical Manifold Method (HONMM) is a powerful method to solve static problems. A development of HONMM to achieve a dynamic solution with high accuracy and less computational cost is addressed in the current paper. In the developed method, the global approximation is obtained through increasing the order of local approximation functions without any Linear Dependence (LD) of the unknowns. The weighted residual formulations are modified to be used in dynamic high order simulation. Moreover, a modified Newmark method formulation is adjusted for time integration of high order equations. The superiority of the proposed method over the conventional NMM is demonstrated through a special beam example. The dynamic free fall block example is used to exhibit the removal of mass matrix singularity. As cases of dynamic analysis, beam free and forced vibrations are illustrated which include a moving load. Finally, a non‐uniform cross‐section beam under dynamic variable loads with accelerated motion is solved while demonstrating the capability of the new method such as simplicity, accuracy and time efficiency for simulation of complex dynamic problems. Copyright © 2014 John Wiley & Sons, Ltd.