一种基于压缩感知的邻域优化算法

非线性降维方法是目前对降维研究有着重要影响的方法,但在降维过程中经常会遇到局部邻域信息量不足、短路和噪声干扰等问题,严重影响降维效果,很难广泛应用于真实数据的处理中.对以上问题分析发现,其主要原因在于经典降维算法都是采用全局固定的邻域大小.提出了一种基于压缩感知的邻域优化算法,运用压缩感知技术对高维空间目标点近邻进行压缩采样,构建“收—放”模型,自适应得到最优子空间,同时优化邻域组成元素,使得数据的整体降维效果更加稳定.通过手工流形和真实数据集的实验,验证了算法的有效性和稳定性.     

Electrochemical investigation of LiNi0.5Mn1.5O4 thin film intercalation electrodes

This work provides kinetic and transport parameters of Li-ion during its extraction/insertion into thin film LiNi_0.5Mn_1.5O₄ free of binder and conductive additive. Thin films of LiNi_0.5Mn_1.5O₄ (0.2 μm thick) were prepared on electronically conductive gold substrate utilizing the electrostatic spray deposition technique. High purity LiNi_0.5Mn_1.5O₄ thin film electrodes were observed with cyclic voltammetry, to exhibit very sharp peaks, high reversibility, and absence of the 4 V signal related to the Mn³⁺/Mn⁴⁺ redox couple. The electrode subjected to 100 CV cycles of charge/discharge delivered a capacity of 155 mAh g⁻¹ on the first cycle and sustained a good cycling behavior while retaining 91% of the initial capacity after 50 cycles. Kinetics and mass-transport of Li-ion extraction at LiNi_0.5Mn_1.5O₄ thin film electrode were investigated by means of electrochemical impedance spectroscopy. The apparent chemical diffusion coefficient (D_app) value determined from EIS measurements changed depending on the electrode potential in the range of 10⁻¹⁰-10⁻¹² cm² s⁻¹. The D_app profile shows two minimums at the potential values close to the peak potentials of the corresponding cyclic voltammogram.     

A novel layered Li [Li0.12NizMg0.32−zMn0.56]O2 cathode material for lithium-ion batteries

Layered Li[Li_0.12Ni_zMg_0.32−zMn_0.56]O₂ oxide cathodes containing lithium atoms in the transition metal layers were synthesized and characterized using X-ray diffraction (XRD), galvanostatic cycling, and differential scanning calorimetry (DSC). The Li[Li_0.12Ni_zMg_0.32−zMn_0.56]O₂ cathodes deliver a specific discharge capacity of about 190 mAh/g at room temperature and 236 mAh/g at 55 °C when cycled between 2.7 and 4.6 V versus Li/Li⁺. Excellent capacity retention and smooth potential profiles at room and elevated temperatures over extended cycles suggest that this material does not convert into a spinel structure.     

降维方法监督分类的比较研究

对ISOMap、LDA、LLE、PCA这4种典型降维算法的主要思想和算法步骤进行了详细分析,并将它们用于有监督分类。从实验结果分析得到结论 ,其可为有监督分类提供有益的借鉴。     

Synthesis and electrochemical characteristics of Al2O3-coated LiNi1/3Co1/3Mn1/3O2 cathode materials for lithium ion batteries

LiNi_1/3Co_1/3Mn_1/3O₂ cathode materials have been coated with Al₂O₃ nano-particles using sol-gel processing to improve its electrochemical properties. The X-ray diffraction (XRD) pattern of the as-prepared Al₂O₃ nano-particles was indexed to the cubic structure of the γ-Al₂O₃ phase and had an average size of ∼4 nm. The XRD showed that the structure of LiNi_1/3Co_1/3Mn_1/3O₂ was not affected by the Al₂O₃ coating. However, the Al₂O₃ coatings on LiNi_1/3Co_1/3Mn_1/3O₂ improved the cyclic life performance and rate capability without decreasing its initial discharge capacity. These electrochemical properties were also compared with those of LiAlO₂-coated LiNi_1/3Co_1/3Mn_1/3O₂ cathode material. The electrochemical impedance spectroscopy (EIS) was studied to understand the enhanced electrochemical properties of the Al₂O₃-coated LiNi_1/3Co_1/3Mn_1/3O₂ compared to uncoated LiNi_1/3Co_1/3Mn_1/3O₂.     

基于类属属性的多标记学习算法

在多标记学习框架中,每个对象由一个示例(属性向量)描述,却同时具有多个类别标记.在已有的多标记学习算法中,一种常用的策略是将相同的属性集合应用于所有类别标记的预测中.然而,该策略并不一定是最优选择,原因在于每个标记可能具有其自身独有的特征.基于这个假设,目前已经出现了基于标记的类属属性进行建模的多标记学习算法LIFT.LIFT包含两个步骤:属属性构建与分类模型训练.LIFT首先通过在标记的正类与负类示例上进行聚类分析,构建该标记的类属属性;然后,使用每个标记的类属属性训练对应的二类分类模型.在保留LIFT分类模型训练方法的同时,考察了另外3种多标记类属属性构造机制,从而实现LIFT算法的3种变体——LIFT-MDDM,LIFT-INSDIF以及LIFT-MLF.在12个数据集上进行了两组实验,验证了类属属性对多标记学习系统性能的影响以及LIFT采用的类属属性构造方法的有效性.     

Electrochemical anticorrosion performance evaluation of Al2O3 coatings deposited by MOCVD on an industrial brass substrate

Alumina (Al₂O₃) coatings of different thickness were deposited on OT59 brass substrate (BS) using the metal organic chemical vapour deposition (MOCVD) technique to evaluate the corrosion performance by EIS measurements. The used precursor was dimethyl-aluminium-isopropoxide. Electrochemical characterizations of the deposited films were performed in a standard very aggressive acidic solution (aerated 1N H₂SO₄ at 25 °C up to 168 h of immersion time) by means of direct current method (Tafel curves) and electrochemical impedance spectroscopy (EIS). The Rutherford backscattering spectroscopy (RBS) indicated that the films are very pure with the correct Al₂O₃ stoichiometry, while the IR absorption spectra showed that the films did not contain any OH groups. The surface film morphology was investigated by atomic force microscopy (AFM) and displayed a globular texture. The films were very smooth, with a maximum root mean square roughness, for example, of 14 nm for a 0.96 μm thick coating. The EIS data confirmed, as expected, that a 2.40 μm Al₂O₃ layer ensures the best corrosion protection after 168 h of immersion in the very acidic environment used.     

碳化硅纳米粉体研究进展

碳化硅纳米粉体性能优于传统的碳化硅粉体，能够达到高新技术领域的严格要求，具有更为广泛的用途。综述了近年来在高新技术领域发展起来的碳化硅纳米粉体的制备方法，对一些新的制备方法进行了重点介绍，阐述了各种方法的基本原理、特点及研究现状，并对各种方法进行了比较。碳热还原法原料便宜，质量稳定，易实现工业化生产；液相法可制得纯度高的纳米级微粉；气相法是目前比较先进的方法，但不易实现大批量生产。未来碳热还原法将继续占主导地位，提出了该领域科学工作者的努力方向。     

锐钛型TiO2对尿素热解过程的影响

通过实验的方法收集了不同温度下纯尿素和尿素/TiO₂混合物热解后的固体残留物，使用红外光谱（IR）及气相色谱质谱联机（GC-MS）方法对这些热解残留物进行成分分析；使用热重-红外联机（TG-FTIR）技术研究尿素及三聚氰酸在有无催化剂TiO₂的情况下的热解特性及气体产物的生成规律；根据Coats-Renfern方法对尿素热解第一阶段的非等温热失重率曲线的数据进行动力学研究，建立动力学方程。结果表明，100~250℃的尿素热解残留物中主要为尿素和缩二脲，300~400℃的尿素热解残留物中主要为三聚氰酸等含氮杂环有机化合物；锐钛型TiO₂能促进尿素和三聚氰酸的热解反应，缩短其反应进程，HNCO与水蒸气在TiO₂表面易发生反应；尿素第一阶段热解的反应级数为2，单独热解时活化能为113.25kJ/mol，指前因子A为2.01×10¹¹min^-1，在催化剂TiO₂的作用下，活化能E为77.42kJ/mol，指前因子A为4.82×10⁷min^-1。     

Electrochemical copolymerization of aniline with m-aminophenol and novel electrical properties of the copolymer in the wide pH range

A copolymer, poly(aniline-co-m-aminophenol), has been synthesized using repeated potential cycling. The monomer concentration ratio, acid concentration and applied potential strongly affect the copolymerization rate and the properties of the copolymer. The optimum conditions for the copolymerization are that the scan potential range is controlled between −0.10 and 0.95 V (vs.SCE), and a solution consists of 0.34 M aniline, 0.012 M m-aminophenol and 2 M H₂SO₄. The IR spectra of the copolymers demonstrate that the m-aminophenol units are included in the copolymer chains. The cyclic voltammograms of the copolymers in 0.3 M Na₂SO₄ solution with various pH values were performed at the potential ranges from −0.20 to 0.80 V and at a scan rate of 60 mV s⁻¹. The results indicate that the copolymer still hold 41.7% of the electrochemical activity when the copolymer electrode was transferred from a solution of pH 5.0 to a solution of pH 11.0 in the potential range of −0.20 to 0.80 V. An impedance plot of the copolymer in a solution with pH 12.0 and at 0.40 V is constructed of a semicircle and a Warburg line with a slope of 1. This means that the electrode reaction of the copolymer at pH 12.0 is also under mass transfer control. The conductivity of the copolymer prepared under the optimum conditions is 1.42 S cm⁻¹, and slightly depends on the pH value. Thus, the pH dependence of the electrical properties of the copolymer is improved compared with poly(aniline-co-o-aminophenol), and is much better than that of the parent polyaniline.