基于主成分分析和小波神经网络的近红外多组分建模研究

将小麦叶片原始光谱经过预处理后,采用主成分分析(PCA)对数据进行降维,取前3个主成分输入小波神经网络,建立了基于主成分分析和小波神经网络的近红外多组分预测模型(WNN);进一步研究了小波基函数个数的选取(WNN隐层节点数)对小波神经网络模型性能的影响,并将WNN模型与偏最小二乘法(PLS)和传统的反向传播神经网络(BPNN)模型进行了比较.结果表明,所建立的WNN模型能用于同时预测小麦叶片全氮和可溶性总糖两种组分含量,其预测均方根误差(RMSEP)分别为0.101%和0.089%,预测相关系数(R)分别为0.980和0.967.另外,在收敛速度和预测精度上,WNN模型明显优于BPNN和PLS模型,从而为将小波神经网络用于近红外光谱的多组分定量分析奠定了基础.     

人工神经网络预测煤炭成浆浓度的研究

考虑煤炭的多种理化特性建立了成浆浓度的神经网络预测模型，对其数据预处理方法、学习率和中间层节点数等进行了深入讨论。水分、挥发分、分析基碳、灰分和氧等五个因子对于煤炭成浆性的预测起到主导作用。五因子、七因子和八因子神经网络模型对煤炭成浆浓度的预测误差分别为：0.53%、0.50%和0.74%，而现有回归分析方程的误差为1.15%，故神经网络模型比回归分析方程有更好的预测能力，尤以七因子模型最佳。     

水体透射光谱结合主成分分析(PCA )改进化学需氧量(COD)含量估算研究

化学需氧量（Chemical Oxygen Demand,COD）是水体有机污染的一项重要指标,化学需氧量越高,表示水污染程度越严重。 为了解决传统的COD测量方法耗时较长,不利于快速、实时地获取水体中COD的信息等问题。本文提出了基于透射光谱测量结合主成分分析(Principal Component Analysis, PCA)改进水体COD含量估算模型。具体的,采集100组COD水体光谱信息,分别使用3种不同的高光谱数据预处理方法对光谱数据进行预处理,分析不同预处理方法对模型精度的影响,并基于不同的预处理方法分别建立高斯过程回归模型(Gaussian Process Regression, GPR)和BP神经网络模型,分析不同预处理方法对模型精度的影响;并对各模型结合PCA数据降维方法进行模型的改进,通过比较模型的精度选择最优模型进行水体COD含量的检测。结果显示,相比于原始光谱数据建立的GPR模型和BP神经网络模型,数据预处理后的模型精度明显提升;且结合PCA对预处理后的数据进一步降维处理后,模型精度得到了进一步的提升。其中,基于标准正态变量变换特征结合PCA改进BP神经网络模型基于PCA改进的BP神经网络模型R^2高达0.9940,均方根误差RMSE为0.022540。证明了基于PCA改进的BP神经网络数据降维方法对预处理后的光谱数据进行降维处理,有利于去除光谱中的冗余信息,提取特征信息,可以实现高光谱检测方法可以实现COD含量估算模型的优化,从而为传统COD测量方法存在的问题提出了一种新的解决思路。     

可解释深度学习在光谱和医学影像分析中的应用

深度学习是一种基于神经网络的建模方法,通过不同功能感知层的构建获得优化模型,提取大量数据的内在规律,实现端到端的建模。数据规模的增长和计算能力的提高促进了深度学习在光谱及医学影像分析中的应用,但深度学习模型可解释性的不足是阻碍其应用的关键因素。为克服深度学习可解释性的不足,研究者提出并发展了可解释性方法。根据解释原理的不同,可解释性方法划分为可视化方法、模型蒸馏及可解释模型,其中可视化方法及模型蒸馏属于外部解释算法,在不改变模型结构的前提下解释模型,而可解释模型旨在使模型结构可解释。本文从算法角度介绍了深度学习及三类可解释性方法的原理,综述了近三年深度学习及可解释性方法在光谱及医学影像分析中的应用。多数研究聚焦于可解释性方法的建立,通过外部算法揭示模型的预测机制并解释模型,但构建可解释模型方面的研究相对较少。此外,采用大量标记数据训练模型是目前的主流研究方式,但给数据的采集带来了巨大的负担。基于小规模数据的训练策略、增强模型可解释性的方法及可解释模型的构建仍是未来的发展趋势。     

基于独立分量和神经网络的近红外多组分分析方法

采用小波变换对光谱数据进行压缩,用独立分量分析(ICA)方法提取近红外光谱数据矩阵的独立成分和相应的混合矩阵,再用BP神经网络对混合矩阵和实测浓度矩阵进行建模,提出了基于独立分量分析-神经网络回归(ICA-NNR)的近红外分析建模方法。进一步研究了独立分量数和网络中间隐层的神经元数对模型性能的影响,经优化后的ICA-NNR模型在相关系数与均方根误差两个指标上均优于直接用光谱矩阵作为输入所建立的模型。本方法用于玉米中水分、淀粉、蛋白质3种主要成分含量的同时测定,检验样品集的化学检测值与近红外预测值的相关系数分别达到:淀粉r=0.971,蛋白质r=0.976,水分r=0.975。     

人工神经网络模型在非ABO3型希土复氧化物电阻率特性分析中的应用

目前,人工神经网络在模式识别及其他领域的应用引起了普遍的关注.本文叙述“反向传播”神经网络模型,算法和在非ABO_3型希土复氧化物电阻率分析中的应用,实验结果表明,神经网络的判别正确率为100％,因此,该方法可望成为化合物分析,材料设计的辅助手段.     

基于神经网络的微芯片电泳信号分析

神经网络具有自学习、自组织、自适应的能力。应用神经网络将高斯函数对电泳分离曲线峰形进行拟合，提取微芯片电泳信号的特征，通过分析结果，对微芯片电泳实验的分离条件作出指导。     

分子荧光-基于主成分分析优化输入变量的径向基函数神经网络法测定尿液中的诺氟沙星

将主成分分析用于优化径向基函数神经网络的输入变量,用于提高神经网络模型的预测能力。方法能有效地解决分子荧光光谱法测定尿液中诺氟沙星过程中尿液中内源性荧光物质的干扰。在优化条件下,径向基函数神经网络模型对尿液中诺氟沙星的平均预测误差为15.34%,神经网络结构为2∶3∶1。方法已用于测定尿液中的诺氟沙星。     

常见食源性致病菌的光谱快速识别

利用拉曼光谱技术建立了4种常见食源性致病菌的分类鉴别方法。在600～1800 cm-1光谱范围内，对蜡样芽胞杆菌、副溶血性弧菌、小肠结肠炎耶尔森菌与金黄色葡萄球菌共186个单细菌进行光谱信息的采集。使用Savitzky-Golay卷积平滑算法与一阶导数对原始光谱数据进行预处理，采用主成分分析方法（PCA）并结合多层感知器（MLP）神经网络构建对4种致病菌的分类识别模型，探究了不同光谱预处理方法下的PCA-MLP模型预测效果，经平滑算法与一阶导数处理过的光谱数据训练集准确率为98.5%，测试集准确率为99.2%。本文建立的光谱技术结合PCA-MLP模型可实现对食品中常见致病菌的快速鉴别。     

用人工神经网络-近红外光谱法测定冬虫夏草中的甘露醇

提出了用近红外漫反射光谱技术快速分析发酵冬虫夏草菌粉中甘露醇含量的新方法。采用比色法测定样品中的甘露醇，其含量范围为8．082％-14．548％。在7501．7-6097．8cm^-1与5453．7-4246．5cm^-1波段，分别采用PCR、PLSR和BP神经网络方法建立了样品近红外光谱的一阶微分光谱与其甘露醇含量之间的相关模型。BP神经网络模型的内部交叉验证误差均方根为0．475，预测误差均方根为0．608，均优于PCR和PLSR的处理结果。这表明，BP神经网络法对非线性检测对象具有较好的建模效果，可用于中药近红外光谱分析的非线性校正。     

龙岩煤不同宏观煤岩组分的颗粒及其燃烧性质实验研究

用浮选法分选出3个密度档、代表不同宏观岩相的龙岩煤颗粒，进行工业分析、元素分析和煤岩分析；并进一步通过扫描电镜结合压汞实验分析颗粒的结构，通过热天平实验研究不同宏观煤岩组分的燃烧性质。结果表明，(1)龙岩煤颗粒的煤岩主要由镜质组组成。其中，亮煤富含镜质组，密度相对较低，挥发分极少、固定碳高达80％以上；暗煤富含矿物质成分，密度极高但含碳量却很少，热值很低，接近于原煤中的矸石。灰煤中惰性组的量是三种岩相中最多的，其工业分析性质与原煤接近；(2)龙岩煤颗粒结构致密，密度在1.6g/cm3以上。颗粒中绝大多数孔隙的孔径在1μm以下。其中亮煤颗粒表面光洁致密，内部孔隙以亚微孔(0.01μm<d<0.10μm)居多，大孔径较少，比表面积较大；暗煤颗粒的表面孔隙较发达，内部结构相对较蓬松，颗粒中细孔、微孔、亚微孔都存在；而灰煤颗粒中的孔隙最不发达，以超微孔(d<0.01μm)为主；(3)不同宏观煤岩组分龙岩煤的着火温度相近，约580℃；但燃尽温度相差较大：亮煤最高，大于800℃；暗煤最低，两者相差约100℃；(4) 龙岩煤不同宏观煤岩颗粒的反应性按亮煤、煤(混煤)、灰煤、暗煤依次递减，亮煤的反应性最好，暗煤的反应性最差。     

Physicochemical properties evolution of chars from palm kernel shell pyrolysis

To clarify the effect of the pyrolysis operating conditions of the biomass on the physicochemical properties of the char and its combustion reactivity, palm kernel shell was pyrolyzed at different temperatures (400–700 °C). Analyses such as proximate and ultimate analysis, XRD, FTIR, N₂ adsorption, and SEM were used to investigate the physicochemical properties of biochar samples. The results show that an increase in pyrolysis temperature led to a development of pore structure and specific surface area of the produced biochar, which was beneficial for improving the biochar combustion reactivity. Besides, with increase in pyrolysis temperature, the carbon content exhibits a raise trend, but the oxygen and hydrogen contents exhibit the opposite behavior, and the aromaticity and graphitization degree of biochar produced at high temperature also increase. The combustion reactivity of biochar was found to be highly dependent on the pyrolysis temperature, and the aromatic structure and graphitization degree have greater effects on biochar combustion reactivity than those of the specific surface area and pore structure.     

Energy potential and thermogravimetric study of pyrolysis kinetics of biomass wastes

The use of agricultural wastes for energy conversion has been widely studied as renewable and carbon neutral energy sources. This paper aims to evaluate the energetic potential of six agricultural wastes—sugarcane bagasse, bean pods, corn stover, pineapple crown leaves, white cotton and natural coloured cotton stalks, through their characterization and pyrolysis kinetic study. The energetic potential of biomasses was evaluated by ultimate and proximate analysis, higher heating value (HHV), apparent density, and kinetic parameters of conversion and apparent activation energy (E_a) determined by Model-Free kinetics though thermogravimetric analysis data. The results indicate energetic density for dry basis biomasses, such as moisture content less than 7%, volatiles higher than 77% and moderate ash content. The HHVs were higher for the biomass with low O:C ratio. The E_a values increased with increasing O:C ratio and were also influenced by the biomass ash content. Among the studied biomasses, PCL are less explored for energy application, although the results confirm its potential for application in thermochemical processes such as pyrolysis or combustion.

     

氧化钙对流化床煤温和气化半焦性质的影响

在Φ4 8× 5 2 5mm小型流化床反应器中 ,于 4 5 0～ 75 0℃温度区间内 ,研究了氧化钙对煤温和气化产物半焦特性的影响。报道了在煤温和气化过程中添加氧化钙对所得半焦的工业分析、元素组成及热值的影响。采用等温吸附法研究了氧化钙对半焦孔结构和比表面积的影响 ,并给予了合理的解释。     

富含树皮体和半丝质体煤的液化反应性和~(13)C-核磁共振分析的研究

通过对富含树皮体的煤和半丝质体的煤进行液化研究和结构分析,探讨其液化产品的产率与结构参数之间的关系。富含树皮体的煤的氢含量和挥发分含量均高,H/C原子比也很高。13C-核磁共振分析表明,所有用煤都以芳香族碳为主。WEST样品的芳碳率最高,LP、CG和DHB样品中CH2基团碳含量高于CH3基团碳含量,且CH2基团碳含量高于其他三个样品。而P858样品中CH3基团碳含量高于CH2基团碳含量。液化结果表明,LP、CG和DHB有高的液化转化率和油收率。煤液化的油收率和煤结构中的CH2基团碳含量以及转化率和煤的芳碳率之间都存在明显的相关关系。因此,13C-核磁共振技术可以用来预测煤液化产品的收率。     

Control of stereoselective protonation of enols1,2

A decalyl framework with a siloxy enolic moiety and proximate proton transferring groups was synthesized. On enolate generation with fluoride two competitive reaction modes were possible: (a) intermolecular protonation, and (b) intramolecular proton transfer by the proximate group. Control of the protonation stereochemistry proved possible by varying the proximate group and by changing the acidity of the medium. With the groups -CH2OH, -CH=O, and -CH2OCH2OCH3 as the proximate groups, only intermolecular proton transfer was observed with no dependence on acidity. In contrast, with -COO- and COOH, only intramolecular protonation resulted but again with no dependence on acidity of the medium. In contrast, with -CH2NH2 as the proximate group, intramolecular proton transfer predominated with a dependence on the effective pH of the medium. A kinetic analysis provided a linear-log relationship of the ratio of the two stereoisomers with the medium acidity. The analysis revealed that two acetic acid molecules are involved in providing the proton to the enolate moiety. A theoretical analysis was developed paralleling the experimental results. In the ketonization transition state, the hybridization was shown to be close to sp2 hybridized at the alpha-enolate carbon.     

Pyrolysis kinetics of elephant grass pretreated biomasses

The Elephant Grass (Pennisetum purpureum Schum) was pretreated by two independent processes, through washing with hot water (W-EG) and acid solution (AW-EG) to improve its energy properties to apply it in a thermochemical process conversion into fuel. The biomasses were analyzed by proximate and ultimate analysis; and the pyrolysis kinetics, before and after pretreatments, were evaluated by the apparent activation energy (E _a) for decomposition in the temperature range of greater volatile matter through the Model-free kinetics using thermogravimetric analysis data. The kinetics of the microcrystalline cellulose Avicel PH-101 was performed to evaluate the E _a result of pure cellulose. The pretreatments were efficient in increasing the volatile matter and heating value, decreasing moisture and ash content, and improving its energetic power to the application in fast pyrolysis process for bio-oil production. The TG results have shown that the reduction in ash content facilitates the pyrolysis process, increasing the volatile matter and decreasing the apparent activation energy required to biomasses degradation, due to less diffusional resistances to heat and mass transfer of W-EG and AW-EG. The Avicel PH-101 showed the highest value of apparent activated energy (E _a = 276.2 kJ mol^?1) which could be explained by its crystallinity, suggesting that crystalline cellulose regions are less accessible to heat diffusion than amorphous regions, requiring more energy to its degradation.     

Influence of gel content on the physical properties of unfilled and carbon black filled natural rubber vulcanizates

Recently, gel content has been considered as a standard property for evaluating commercial grade natural rubber (NR). In this study, NR containing various amounts of gel was prepared by accelerated storage hardening as a model to clarify the influence of gel content on the physical properties of both unfilled and carbon black filled vulcanizates. Furthermore, the NR samples were investigated to determine the effect of gel fraction on Mooney viscosity and the structure of the gel after mastication. The results revealed that Mooney viscosity was related to the percentage of gel fraction that has been proven to be the result of interactions between proteins and phospholipids at chain ends. After mastication, although the gel fraction of NR can be decomposed to ∼0% w/w, the interactions of proteins and phospholipids at the chain ends still existed, corresponding to the gel content of the raw rubber. In the case of unfilled vulcanizates, the gel content showed no effect on cure characteristics, crosslink density and ultimate tensile strength, whereas the upturn of stress occurred at a smaller strain when the gel content increased. However, in the case of carbon black filled vulcanizates, the gel content played a dominant role in the carbon black dispersion, which was poorer when gel content increased, contributing to a decrease of crosslink density and ultimate tensile strength.     

Evaluating of the exergy efficiency of the silicon production process using artificial neural networks

Abstract

In this study, the extent and effect of fluctuations in the proximate composition of carbon materials on specific electrical energy consumption and exergy efficiency in the production process of silicon furnaces was evaluated using a novel artificial neural network approach. The neural network architecture 4-5-2 with hyperbolic sigmoid in the hidden layer and linear function in the output layer was used. The proposed model successfully predicts the values of the specific electrical energy consumption and the exergy efficiency through the correlation coefficient (0.93 and 0.87) of the actual and predicted values for the various proximate compositions of carbon materials. The interactive effects of components in the mixtures of carbon materials were investigated via contour diagrams. Fixed carbon had the biggest impact on electrical energy consumption and exergy efficiency in the mixtures of carbon materials, followed by moisture. Compared with volatile matter from woodchips without any pretreatment, volatile matter from petcoke, charcoal, and coal had a significant effect on electrical energy consumption and exergy efficiency.     

结焦催化剂上焦炭氢碳比的测定方法

提出了一种测定裂化结焦催化剂上焦碳H/C的方法--酸溶-元素分析法。该方法主要利用HF和HCl将焦碳从催化剂上进行剥离，然后利用元素分析仪测定焦碳中的碳含量和氢含量，从而可以得到结焦催化剂上焦碳的H/C质量比。这种测定方法避免了在测氢过程中催化剂结晶水和吸附水对测定结果的影响，大大提高了催化剂上焦碳H/C的测定精度。