基于多特征的PSO-MSVM动态过程质量异常模式识别

为了提高动态过程质量异常模式识别效率,将动态过程质量模式的均值特征与小波包分解特征作为分类特征,并构建两层多支持向量机识别模型进行分类.利用均值特征,在第一层MSVM中把动态过程变化趋势划分为正常与周期、上升与向上阶跃、下降与向下阶跃三大类别;采用小波包分解特征,在第二层MSVM中对这三大类别进行再分类.仿真结果表明提出的识别模型的识别精度相比采用单一特征的识别模型有明显提高.     

多通信半径的无线传感器网络DV-Hop定位算法

为了实现无线传感器网络中基于无需测距DV-Hop定位算法的高精度定位,改进基于2、3通信半径的DV-Hop节点定位算法,并利用最小均方差准则改进计算平均每跳距离的公式,在最佳指数值下,修正锚节点平均每跳距离。MATLAB仿真结果表明,改进的算法在随机分布环境能提高节点定位精度;在最佳指数值下,进一步提升节点定位精度;在锚节点人工部署环境下,大大提高节点定位精度。改进的算法实现高精度定位同时,不需要增加网络泛洪次数和硬件成本。     

Distributed consensus tracking for multi‐agent systems under two types of attacks

This paper studies a distributed coordinated control problem for a class of linear multi‐agent systems subject to two types of attacks. The problem boils down to how to achieve secure consensus tracking for multi‐agent systems with connected and disconnected (paralyzed) directed switching topologies caused by two types of attacks. The attacks on the edges instead of nodes lead to the loss of security performance. Two cases are studied in this paper. First, under only a class of connectivity‐maintained attacks, sufficient conditions are derived to achieve secure consensus tracking in mean‐square. Second, when the multi‐agent systems are further subject to a class of connectivity‐broken attacks, novel sufficient conditions are further obtained to ensure secure consensus tracking with a specified convergence rate by virtue of the idea of average dwell time switching between some stable and unstable subsystems. Three numerical simulations are finally given to illustrate the theoretical analysis. Copyright © 2015 John Wiley & Sons, Ltd.     

一种基于项目属性评分的协同过滤推荐算法

协同过滤是电子商务推荐系统中应用最成功的推荐技术之一,但面临着严峻的用户评分数据稀疏性和推荐精度低等问题。针对数据稀疏性高和单一评分导致的推荐精度低等问题,提出一种基于项目属性评分的协同过滤推荐算法。首先通过均值法或缩放法构造用户-项目属性评分矩阵将单一评分转化为多评分;其次基于每个属性评分矩阵,计算用户间的偏好相似度,得到目标用户的偏好最近邻居集;然后针对每个最近邻居集,在用户-项目评分矩阵上完成对目标用户的初步评分预测;最后,将多个初步预测评分加权求和作为综合评分,完成推荐。在Movie Lens扩展数据集上的实验结果表明,该算法能有效提高推荐精度。     

An analysis of the exponential stability of linear stochastic neutral delay systems

This paper is concerned with the analysis of the mean square exponential stability and the almost sure exponential stability of linear stochastic neutral delay systems. A general stability result on the mean square and almost sure exponential stability of such systems is established. Based on this stability result, the delay partitioning technique is adopted to obtain a delay‐dependent stability condition in terms of linear matrix inequalities (LMIs). In obtaining these LMIs, some basic rules of the Ito calculus are also utilized to introduce slack matrices so as to further reduce conservatism. Some numerical examples borrowed from the literature are used to show that, as the number of the partitioning intervals increases, the allowable delay determined by the proposed LMI condition approaches h_max, the maximal allowable delay for the stability of the considered system, indicating the effectiveness of the proposed stability analysis. Copyright © 2013 John Wiley & Sons, Ltd.     

On stability delay bounds of simple input-delayed linear and non-linear systems: Computational results

This paper deals with the problem of delay size stability analysis of single input-delayed linear and nonlinear systems. Conventional reduction, reduction linked by sliding mode, and linear memoryless control approaches are used for simple input-delayed systems to obtain the stability conditions. Several first order examples are investigated systematically to demonstrate the capabilities and limitations of the advanced stability analysis techniques including Lyapunov-Krasovskii functionals, Newton-Leibniz formula, and a newly addressed Lagrange mean value theorem. Numerical comparative results show the usefulness and effectiveness of the advanced delay size analysis techniques proposed in this paper.     

基于特征参数归一化的鲁棒语音识别方法综述

目前,自动语音识别系统往往会因为环境中复杂因素的影响,造成训练环境和测试环境存在不匹配现象,使得识别系统性能大幅度下降,极大地限制了语音识别技术的应用范围。近年来,很多鲁棒语音识别技术成功地被提出,这些技术的目标都是相同的,主要是提高系统的鲁棒性,进而提高识别率。其中,基于特征的归一化技术简单而有效,常常被作为鲁棒语音识别的首选方法,它主要是通过对特征向量的统计属性、累积密度函数或功率谱的归一化来补偿环境不匹配产生的影响。该文主要对目前主流的归一化方法进行介绍,其中包括倒谱矩归一化方法、直方图均衡化方法以及调频谱归一化方法等。     

GPU并行加速的均值偏移算法

为克服mean shift算法计算复杂度高、运行速度慢的缺点,提出一种基于GPU的快速mean shift算法.首先使用k-means算法对图像像素进行预分类,之后在预分类、下采样后缩小的数据集上进行mean shift聚类,以有效地降低算法复杂度.此外,借助GPU的通用计算功能对k-means和mean shift分别进行并行了处理.实验结果表明,通过对图像进行预处理,有效地提高了几何模板查找在强噪声、低信噪比图像中的识别率;同时,改进后的mean shift算法的运行速度提高了近40倍,满足了高速机器视觉检测的实时性要求.     

基于曲面曲率和RLDA的3D掌纹识别方法

提出一种新的3D掌纹识别方法,利用掌纹曲面的3D曲率信息提高当前2D掌纹识别算法的精确度和鲁棒性。首先因曲率是3维物体的重要表征,能够与视点无关地表现曲面的局部形状,不管手掌发生旋转或者平移,曲率信息都是稳定的,因此提取3D掌纹的均值曲面曲率特征来刻画3D掌纹的曲面特征;继而获得3D掌纹映射到2D空间上的灰度图像——均值曲率图像(MCI);然后在获得的MCI上采用正则化的LDA(RLDA)方法来进行二次特征抽取,以消除传统线性判别分析(LDA)应用于识别时存在的小样本问题和优化准则函数并不直接与识别率相关等问题。实验结果表明,所提出的方法相比与传统的LDA、PCA、ICA、LPP等方法有更高的精度和速度。