稀疏卷积非负矩阵分解的语音增强算法

非平稳噪声和低信噪比条件下提高增强语音质量一直以来都是语音增强研究的难题。近年来,卷积非负矩阵分解在语音增强算法中成功应用,本文进一步考虑语音信号在时频域的稀疏性,提出了稀疏卷积非负矩阵分解(Sparse Convolutive Nonnegative Matrix Factorization, SCNMF)的语音增强算法。该算法包括训练和增强两个阶段。训练阶段通过SCNMF算法分别对纯净语音和噪声的频谱进行训练,得到纯净语音和噪声字典,并将其作为增强阶段的先验信息。增强阶段首先通过SCNMF算法对带噪语音的频谱进行分解,然后利用纯净语音和噪声联合字典对语音编码矩阵进行估计,重构增强语音。本文通过实验仿真分析了稀疏因子对增强语音质量的影响。实验结果表明,在非平稳噪声和低信噪比条件下,本文算法增强效果均优于多带谱减、非负矩阵分解、卷积非负矩阵分解等传统的算法。     

新型动磁式直线振荡执行器的动子位移自传感

针对一种新型无内定子动磁式直线振荡执行器,在建立其机电系统数学模型的基础上,提出一种基于全维状态观测器的动子位移自传感算法。通过对执行器输入电压和输出电流信号的处理和计算来估算动子位移。仿真和实验结果均表明:在变压变频控制方式下,该算法能实现不同电气驱动频率下的动子位移自传感;采用该算法进行行程估算的绝对误差最大值为0.32 mm,相对误差最大值为2.6%。此算法可以满足直线压缩机和直线泵类负载的变行程控制要求。     

模糊聚类在股票投资中的应用

聚类分析即将一组事物根据其性质上亲疏远近的程度进行分类,把性质相近的个体归为一类,使得同一类中的个体具有高度的同质性,不同类之间的个体具有高度的异质性。模糊聚类分析是现今模糊理论应用最广泛和最富成果的技术之一。阐述模糊聚类的理论,以部分石油股票为例,抽取影响石油股票收益因素的数据,利用最大最小法建立相似矩阵,用传递闭包法作出聚类分析,并进行总结。     

一种应用PCA的人脸图像集识别方法

提出一种适用于同类别的图像子集的类别判断方法。同类别的图像子集经过PCA主特征提取后,选择较大的p个特征值对应的线性无关的特征向量,组成特征矩阵,则同类别的图像子集可以转化成一个特征矩阵,图像子集与特征矩阵一一对应,进而整个图像库能够用矩阵集合来表示。定义一种矩阵间的距离及最小二乘距离,通过计算待测图像子集对应特征矩阵与图像库中不同类别对应特征矩阵之间的距离或最小二乘距离,判断待测试图像子集所属类别。     

Adaptive Linear Dynamic Matrix Control applied to an integrating process

Dynamic Matrix Control (DMC) has proven to be a powerful tool for optimal regulation of chemical processes under constrained conditions. It is based on a linear convolution model derived from step-response measurements. A model predictive control algorithm optimises closed-loop performance for a nominal operating point. However, as the process moves away from this point, control usually becomes sub-optimal due to process non-linearity. As seen in this work, the DMC algorithm can be made adaptive, to establish a new local model, by recursive estimation of the local step response parameters from normal plant variations. However, when used for control of plants containing integrating process units, steady-state offsets occur for sustained changes. Thus, a novel Adaptive Linear Dynamic Matrix Control (ALDMC) algorithm has been developed and used to control a 2-input/2-output system with an integrating behaviour. Comparisons of model control and plant control with and without these features demonstrated the importance of integral compensation for integrating processes, and model adaptation in the case of plant/model mismatch. Some cross-compensation of integration by the adaptive feature was also noted. An holistic technique is demonstrated which simultaneously recognises residual integration disturbances and matrix parameter variations, whereas previous techniques which recognise only one of these will fail in the presence of the other.     

Advancements in the design and validation of an air pollution integrated assessment model for Spain

This paper describes the design and application of the Atmospheric Evaluation and Research Integrated model for Spain (AERIS). Currently, AERIS can provide concentration profiles of NO₂, O₃, SO₂, NH₃, PM, as a response to emission variations of relevant sectors in Spain. Results are calculated using transfer matrices based on an air quality modelling system (AQMS) composed by the WRF (meteorology), SMOKE (emissions) and CMAQ (atmospheric-chemical processes) models. The AERIS outputs were statistically tested against the conventional AQMS and observations, revealing a good agreement in both cases. At the moment, integrated assessment in AERIS focuses only on the link between emissions and concentrations. The quantification of deposition, impacts (health, ecosystems) and costs will be introduced in the future. In conclusion, the main asset of AERIS is its accuracy in predicting air quality outcomes for different scenarios through a simple yet robust modelling framework, avoiding complex programming and long computing times.     

Development of soft sensor by incorporating the delayed infrequent and irregular measurements

Originated from challenging industrial applications, this paper addresses a soft sensor development problem for linear systems with two types of measurements. One is fast, regular and delay-free measurements. The other is infrequent, irregular measurements with time-varying delays. The approach to be taken is based on the Kalman filter and data fusion technique, avoiding running the full augmented state Kalman filter, and leading to a considerably lower implementation cost. Although it is suboptimal, the loss in performance is minor compared to the optimal filter. Two simulation examples demonstrate the advantages of the proposed approach. An industrial soft sensor application example is also used to demonstrate its practicality.     

基于Q学习的DDoS攻防博弈模型研究

新形势下的DDoS攻防博弈过程和以往不同,因此利用现有的方法无法有效地评估量化攻防双方的收益以及动态调整博弈策略以实现收益最大化。针对这一问题,设计了一种基于Q学习的DDoS攻防博弈模型,并在此基础上提出了模型算法。首先,通过网络熵评估量化方法计算攻防双方收益;其次,利用矩阵博弈研究单个DDoS攻击阶段的攻防博弈过程;最后,将Q学习引入博弈过程,提出了模型算法,用以根据学习效果动态调整攻防策略从而实现收益最大化。实验结果表明,采用模型算法的防御方能够获得更高的收益,从而证明了算法的可用性和有效性。     

基于改进差别矩阵和专家知识的态势指标提取算法

针对常规粗糙集约简算法在应用中的样本指标提取困难和效率低下等问题,将粗糙集理论引入到态势指标提取中,基于差别矩阵压缩和分类选择实现态势指标决策表的信息约简,结合专家知识的指标重要性度量调整态势指标的选择,提出一种基于改进差别矩阵和专家知识的态势指标提取算法.在态势指标体系实例中进行分析和验证,实验表明该算法具有较好的态势指标约简效果,提取后的指标在网络安全评估中是合理的,因此该研究为态势指标的有效提取提供一种可行的解决途径.     

The adhesion of amorphous polystyrene surfaces below Tg

The bonding of polystyrene (PS) surfaces below T_g was investigated by two different fracture tests: the lap-shear joint method and the cantilever beam method. Adhesion energy values obtained by the two methods are in agreement and develop with (time)^1/2, at temperatures as low as T_g−16 °C. Even if the double cantilever method is the most common test found in the literature for adhesions above T_g, for low adhesion values, below T_g, the lap-shear joint geometry is more appropriate. Moreover, when the glass transition temperature is used as a reference temperature, polydisperse and monodisperse PS adhesion energy curves are superposable, suggesting that the auto-adhesion is not significantly favored by the presence of numerous chain ends at the surface (due to the low molecular weight chains provided by the polydisperse PS).