神经网络在波谱分析中的应用：用亚图估计和预测烷烃的^13C NMR

系统研究了神经网络在波谱分析中的应用，采用多层／三层前馈神经网络以误差反传及改进算法估计和预测Ｃ１－Ｃ１０的６０余种烷烃的化学位移。烷烃中碳原子由十余种对应于所谓根亚树的相嵌频率描述子所决定。这些描述子等于由１至６个碳原子组成的更小结构骨加组成。     

用亚图参数与回归技术估计和预测烷烃的核磁共振碳谱

系统研究了分子建模在波谱分析中的应用.采用多元线性回归算法(MLR)估计和预测了60余种烷烃的碳谱化学位移.烷烃中碳原子由十余种对应于所谓根亚树的相嵌频率描述子所决定.这些描述子等于由2~5个碳原子组成的更小结构骨架组成.说明了所用描述子作为很有用的工具可适当地描述烷烃中碳所处微观环境.同时还比较了与神经网络的计算结果.     

Nitronyl Nitroxides as a Spin Probe in EPR Tomography In Vivo

Recently, new water- and blood-soluble nitronyl nitroxides, 2-(5-methyl-1H-imidazole-4-yl)-4,4,5,5-tetramethyl-4,5-dihydroimidazole-3-oxide-1-oxyl (NN1) and 2-(1H-imidazole-4-yl)-4,4,5,5-tetramethyl-4,5-dihydroimidazole-3-oxide-1-oxyl (NN2), (Fig. 1), were synthesized and used as contrast agents for MRI (Savelov et al. Dokl Academ Nauk 416(4): 493–495, 2007). Taking into account the high rate constants of NN’s reduction by ascorbic acid and other biologically relevant reductants, it is not clear which factors helped with the use of these nitroxides in vivo as a contrast reagent. Moreover, due to high solubility in an aqueous solution and low toxicity (Ovcharenko et al. in Dokl Academ Nauk 404(2):198–200, 2005, Eriksson et al. in Drug Metab Dispos 15(2):155–160, 1987, Afzal et al. in Polyhedron 22(14):1957–1964, 2003) of NNs, it seems possible to use them as a spin probe for NO in vivo with EPR tomography. In this paper, we studied reduction of NN1 and NN2 in model conditions (by ascorbic acid) and in vitro. In addition, the possibility of NN1 and NN2 to be used as paramagnetic probes for L-band EPR imaging in vivo was investigated. Nitric oxide (NO) expression in vivo leads to the decrease in concentrations of NN1, 2 upon the injection in a mouse body, that can be explained by the reaction of studied radicals with NO and fast transformation of the reaction products to diamagnetic species. Pharmacokinetics of NN1, 2 and limitations of their application as contrast agents in MRI are discussed also. Finally, the results of EPR tomography were compared with MRI data. It is shown that the fast reduction of the reaction product of NN with NO—imino nitroxides—is the main obstacle to use NN as a spin probe in vivo.

[M(SS)(NN)](M=Zn2+,Cd2+)配合物分子内跃迁与结构的关系

报道了配合物[M(SS)(NN)](M=Zn2 ,Cd2 )(SS=mnt2-,1,2-二氰基乙烯-1,2-二硫醇离子,NN=5-NO2-phen,5-硝基-1,10-邻菲咯啉)的合成,探讨了Zn(SS)(NN)和Cd(SS)(NN)在二甲基亚砜(DM-SO)、二甲基甲酰胺(DMF)、丙酮(acetone)、吡啶(Py)等溶剂中的电子吸收光谱。研究发现配合物的紫外区的吸收带270～280nm,320～350nm,350～390nm本质上属于配体mnt2-,5-NO2-phen内部的πb→π*跃迁,可见光区400～500nm本质上属于配体mnt2-到配体5-NO2-phen的荷移跃迁(LL′CT)。确认了标题配合物的荷移跃迁光谱在相关分子轨道能级图中的对应关系。标题配合物的LL′CT吸收带与组成相似配合物M(SS)(NN)(M=Co2 ,Fe2 ,Ni2 ,Cu2 )的对应吸收带相比较弱一些,这是由于[M(SS)(NN)](M=Zn2 ,Cd2 )的LL′CT谱带(7b2→7b1)是部分解除轨道禁阻的跃迁。     

Van Vleck second moments and hydrogen diffusion in YH2.1--measurements and simulations

Proton magnetic resonance absorption spectra of yttrium dihydride (YH(2+x)), with x = 0.10, were recorded in the temperature ranges 4.2-310 K at 36.01 MHz and 150-400 K at 299.8 MHz. The evidence of proton self-diffusion follows from the changes of linewidth with temperature. The second moment of the resonance lines was determined from the experimental spectra and was compared with values calculated from the crystallographic data. The averaging effect of diffusion on the second moment was taken into account through Monte Carlo simulations of the diffusion process. The simulation was performed in a block of unit cells 5 x 5 x 5 with periodic boundary conditions. They compensated the effect of finite dimensions of the block. The calculated temperature dependence of the proton second moment values was fitted to the experimental ones. The fitting parameters were: the attempt frequency v0 and the activation energy Ea for hydrogen diffusion, assuming Arrhenius behavior of the jump frequencies vc = v0 exp(-Ea/k(B)T). In these preliminary studies, the Monte Carlo simulations were performed for tetrahedral-octahedral exchanges while direct tetrahedral-tetrahedral jumps were neglected for simplicity. Three models of hydrogen diffusion, differing in the maximum jump lengths allowed for a given model, were considered. These lengths were taken as the distances from the hydrogen attempting to jump to the first (1NN), second (2NN), and third (3NN) nearest neighbor position able to accept the jumping atom. Assuming the same attempt frequency v0 = 6.0 x 10(12)s(-1) for all three models, the activation energies giving the best fit to experimental data were 0.5, 0.54, and 0.55 eV for 1NN, 2NN, and 3NN models, respectively.     

Saturation of azimuthal anisotropy in Au + Au collisions at (square root)s(NN) = 62-200 GeV

New measurements are presented for charged hadron azimuthal correlations at midrapidity in Au+Au collisions at (square root)s(NN) = 62.4 and 200 GeV. They are compared to earlier measurements obtained at (square root)s(NN) = 130 GeV and in Pb + Pb collisions at (square root)s(NN) = 17.2 GeV. Sizeable anisotropies are observed with centrality and transverse momentum (pT) dependence characteristic of elliptic flow (upsilon2). For a broad range of centralities, the observed magnitudes and trends of the differential anisotropy, upsilon2(pT), change very little over the collision energy range (square root)s(NN) = 62-200 GeV, indicating saturation of the excitation function for upsilon2 at these energies. Such a saturation may be indicative of the dominance of a very soft equation of state for (square root)s(NN) approximately 60-200 GeV.     

大豆叶面积指数的高光谱估算方法比较

叶面积指数(leaf area index,LAI)是重要的生物物理参数,亦是各种生态模型、生产力模型以及碳循环研究等的重要生物物理参量,因此具有重要的研究意义。通过分析大量实测数据,选用归一化植被指数(normalized difference vegetation index,NDVI)和比值植被指数(ratio vegetation index,RVI)、主成分分析(principcal component analysis, PCA)、神经网络(neural network NN)三种方法对大豆使LAI进行了估算,比较分析了三种方法的估算结果。研究结果表明,植被指数法(NDVI,RVI),主成分分析,神经网络方法LAI都取得了较为理想的结果,验证模型的确定性系数分别达0.758和0.753, 0.954, 0.899,其中主成分分析方法和神经网络方法精度较高,主成分分析方法验证模型的稳定性更好,其验证模型的RMSE为0.267,明显低于两个植被指数(NDVI和RVI的RMSE分别为0.594和0.616)和神经网络(RMSE=0.413)。当叶面积指数较小时,植被指数能够较好地去除土壤、大气等条件影响,并精确估算LAI;当叶面积指数较大时,主成分分析能够弥补植被指数饱和的缺陷,得到很好的LAI估算效果。神经网络受LAI大小的影响效果居中,其对叶面积指数较小和较大时具有一致的估算效果,具有较好的应用潜力。     

基于Adaboost及谱回归判别分析的近红外光谱固态发酵过程状态识别

为了实现固态发酵过程状态的快速监测,以饲料蛋白固态发酵为实验对象,开展了基于近红外光谱分析技术的饲料蛋白固态发酵过程状态定性识别研究。首先利用Antaris Ⅱ型傅里叶变换近红外光谱仪采集140个固态发酵物样本的近红外光谱,并采用标准正态变换(SNV)光谱预处理方法对获得的原始光谱进行预处理;其次,采用谱回归判别分析(SRDA)法对预处理后的近红外光谱进行特征提取;最后,采用最近邻(NN)分类算法作为弱分类器建立固态发酵过程状态识别模型,并对测试集样本进行识别。结果显示,与利用主成分分析(PCA)法和线性判别分析(LDA)法提取的光谱特征建立的识别模型结果相比较,SRDA-NN识别模型获得的结果最佳,在测试集中的正确识别率达到94.28%;为了进一步提高识别模型的准确率,将自适应提升法(Adaboost)与SRDA-NN方法结合,提出了Adaboost-SRDA-NN集成学习算法来建立饲料蛋白固态发酵过程状态的在线监测模型。通过Adaboost算法提升后的SRDA-NN模型预测性能得到了进一步增强,Adaboost-SRDA-NN模型在测试集中的正确识别率达到100%。试验结果表明：在近红外光谱定性分析模型校正过程中,SRDA方法能有效地对近红外光谱数据进行特征提取,以实现维数约简;另外,Adaboost算法能很好地提升最终分类模型的预测精度。     

蓝光锆（Ⅳ）配合物的合成及光致发光性能研究

合成了一系列蓝光锆（Ⅳ）配合物 Cl[Zr（NN）（DBM）₃]（DBM=二苯甲酰基甲烷,NN=1,10-邻菲罗啉 1;2,9-二甲基-1,10-邻菲罗啉 2;4,4'-二甲基-2,2'-联吡啶 3;1-乙基-2-（1-萘基）-1H-咪唑-4,5-f-1,10-邻菲罗啉4）,以波长为355 nm的紫外光激发,4个配合物的最大发射峰均位于445 nm左右,半峰宽只有36~45 nm。在相同的测试条件下,配合物 1 的蓝光发射强度最大,量子效率最高,其原因是结构刚性强,能够有效地减少非辐射跃迁造成的能量损失。而且,配合物 1~3 的热稳定性很好,分解温度均在200℃以上,可以用于制作有机发光器件,不会在器件的制作过程中发生分解。     

Pseudorapidity distributions of charged particles from Au + Au collisions at the maximum RHIC energy,square root[s(NN)] = 200 GeV

We present charged-particle multiplicities as a function of pseudorapidity and collision centrality for the 197Au+197Au reaction at square root[s(NN)] = 200 GeV. For the 5% most central events we obtain dN(ch)/deta/(eta = 0) = 625+/-55 and N(ch)/(-4.7< or =eta < or =4.7) = 4630 +/- 370, i.e., 14% and 21% increases, respectively, relative to square root[s(NN)] = 130 GeV collisions. Charged-particle production per pair of participant nucleons is found to increase from peripheral to central collisions around midrapidity. These results constrain current models of particle production at the highest RHIC energy.     

基于自组织映射神经网络的位相差异波前传感新方法

针对位相差异（Phase Diversity,PD）技术存在计算量大和神经网络训练样本集规模过大等缺点,提出一种结合人工神经网络的PD波前传感新方法－基于自组织映射（Self-Organizing Featwe Map,SOFM）网络的PD波前传感方法.通过计算机仿真实验对新方法的性能进行了测试和分析.仿真实验结果表明,该方法在保证传感准确度的前提下可以有效提高PD波前传感的速度和缩减网络训练样本集的规模.     

A comparison of genetic programming and neural networks; new formulations for electrical resistivity of Zn–Fe alloys

It is difficult to automatically solve a problem in a systematic method without using computers. In this study, a comparison between Neural Network (NN) and genetic programming (GEP) soft computing techniques as alternative tools for the formulation of electrical resistivity of zinc–iron (Zn–Fe) alloys for various compositions is proposed. Different formulations are supplied to control the verity and robustness of NN and GEP for the formulation to design composition and electrolyte conditions in certain ranges. The input parameters of the NN and GEP models are weight percentages of zinc and iron in the film and in the electrolyte, measurement temperature, and corrosion voltage of the films. The NN- and GEP-based formulation results are compared with experimental results and found to be quite reliable with a very high correlation (R ²=0.998 for GEP and 0.999 for NN).     

The πNN vertex function

The πNN vertex function is calculated with a dispersion theoretic approach which reduces to standard field theoretic results in the limit of zero-width σ- and ρ-mesons. Results are shown to be sensitive to the choice of various on-shell-equivalent couplings of the J = 1 ππ system to the NN? system.     

激子转移对GaP:N发光热猝灭过程的影响

本文研究了20K-120K范围内GaP:N束缚激子发光与温度的关系,从实验结果发现,各发光中心之间存在着激子转移.初步分析表明,采用带带激发方式,激子转移会导致深束缚中心的热激活能拟合值偏高,其发光热猝灭是由于空穴的热离化所造成的.