混合核支持向量机在化工软测量中的应用研究

针对核函数方法中单个核函数的局限性,提出混合核支持向量机建模方法以提高模型的泛化能力和精度.本文的混合核函数由一个局部核函数和一个全局核函数线性组合而成,并可以通过参数来调节局部和全局核函数对混合核函数的作用.通过对工业双酚A生产过程软测量建模的仿真研究表明.混合核支持向量机软测量模型适用于化工建模并具有较好的泛化能力.     

基于混合核函数和PLS的胎面尺寸预测模型

为实现对生产过程中胎面尺寸的准确预测,分析胎面生产工艺,结合混合核函数和偏最小二乘法(PLS)的特点构造了一种基于混合核函数和PLS的软测量模型。该预测模型把辅助变量通过输入非线性函数映射到高维特征空间,在高维特征空间中采用线性建模方法建立模型后经过非线性函数映射回原数据空间形成非线性模型。应用效果表明,该模型具有较强的学习和泛化能力,并且具有较高的预测精度,能较好地满足胎面实际生产要求。     

神经元网络在精馏塔控制中的应用

本文以某３００ｋｔ乙烯塔精馏控制为目的，以各种建模方法、软测量推断控制和预测控制为基础，探讨了针对具体工业过程的实际应用。具体研究了精馏过程对控制的要求和影响，以及径向基函数神经网络和ＰＬＳ－ＲＢＦ算法的建模方法，并以此建立了两个乙烯出塔浓度的软测量模型。     

基于核慢特征回归与互信息的常压塔软测量建模

针对工业过程的非线性及动态特性,提出了一种新的慢特征回归软测量方法。该方法首先通过添加时延数据构造动态数据集,利用互信息最大化准则筛选变量从而减少信息冗余的影响。同时该方法在慢特征分析的基础上引入核函数扩展,加强模型处理非线性数据的能力,并将获得的核慢特征用于回归建模。核慢特征分析通过分析样本的变化,提取具有缓慢变化特征的成分,可以有效地刻画工业过程的变化趋势,提升回归模型精度。最后该方法的有效性在常压塔常顶油干点与常一线初馏点的软测量模型中得到了验证。     

基于核慢特征回归与互信息的常压塔软测量建模

针对工业过程的非线性及动态特性,提出了一种新的慢特征回归软测量方法。该方法首先通过添加时延数据构造动态数据集,利用互信息最大化准则筛选变量从而减少信息冗余的影响。同时该方法在慢特征分析的基础上引入核函数扩展,加强模型处理非线性数据的能力,并将获得的核慢特征用于回归建模。核慢特征分析通过分析样本的变化,提取具有缓慢变化特征的成分,可以有效地刻画工业过程的变化趋势,提升回归模型精度。最后该方法的有效性在常压塔常顶油干点与常一线初馏点的软测量模型中得到了验证。     

基于分阶段的LSSVM发酵过程建模

发酵过程建模是研究微生物发酵的重要课题,基于模型可实现被测参量的软测量、系统的优化控制。鉴于引入混合核函数的最小二乘支持向量机在过程建模中具有优良表现,采用基于混合核函数的最小二乘支持向量机建模。但由于发酵过程周期较长,最小二乘支持向量机的全局模型预测精度难以保证,算法复杂度很高,因此提出一种分阶段建模方法。首先,选择表征阶段特性的辅助变量,利用模糊C均值聚类算法对样本数据聚类,将发酵过程分成不同的阶段,然后为各个阶段分别建立最优混合核最小二乘支持向量机局部模型,最后将局部模型合成构成过程的完整模型。将此方法应用于青霉素发酵过程和重组大肠杆菌发酵过程中,验证了该方法的有效性。     

基于改进Kalman滤波算法的多模型融合建模方法

针对聚合物生产过程重要质量控制指标或状态变量的软测量问题,提出了一种基于改进Kalman滤波算法的多模型融合建模方法。将混合核函数主元分析(K₂PCA)与人工神经网络(ANN)相结合,建立一种基于K₂PCA-ANN的数据驱动模型;利用改进Kalman滤波算法实现K₂PCA-ANN模型与机理模型融合,构建一种并联结构的混合模型;协调二次滤波(线性滑动平滑)和方差更新对混合模型进行优化处理,使混合模型的估计性能尽可能地达到最优,使混合模型的预测稳定性得到有效改善。将该多模型融合建模方法应用于氯乙烯聚合过程聚合速率软测量中,应用研究结果表明:与单一的机理模型或K₂PCA-ANN数据驱动模型的预测性能相比,该建模方法建立的聚合速率模型具有更佳的预测性能。该建模方法的运用为进一步开展聚合物生产过程优化与控制等研究提供基础条件。     

基于辅助变量KNN分析的软测量建模方法

提出一种基于辅助变量最近邻（KNN）分析的软测量建模方法,该方法将KNN算法应用于辅助变量分类,根据分类结果,应用核主成分分析（KPCA）和支持向量回归机（SVR）相结合进行软测量建模。KNN分析独立于后继回归模型,却又直接影响模型结构,KPCA作为中间层,在KNN分类结果指导下提取不同类别包含辅助变量高阶信息的特征主元,然后使用SVR建立特征主元和主导变量之间的回归模型。用该方法建立粗汽油干点软测量模型,结果表明KNN-KPCA-SVR（KKS）模型的预测精度和泛化能力优于线性PLS、RBF核函数SVR和KPCA-SVM模型。     

基于改进扩张搜索聚类算法的多流形软测量建模

针对化工生产过程工况复杂多变，单一的软测量模型难以满足系统对估计精度的要求，提出了一种基于改进的扩张搜索聚类算法的多流形软测量建模的方法。该算法采用流形距离来代替欧氏距离，自适应地确定邻域半径，并引入局部密度用于确定聚类中心，对聚类后得到的各个子流形分别采用流形学习中的核等距映射法进行特征提取，建立基于高斯过程回归的子模型。将该方法应用于某双酚A生产装置的软测量建模，仿真结果验证了该方法的有效性。     

基于非线性偏最小二乘回归的软测量建模及应用

针对复杂工业过程存在的多变量、相关性和非线性问题,提出一种新的基于非线性偏最小二乘(partial least squares,PLS)回归的软测量建模方法。该方法利用PLS作为模型的外部框架来提取输入输出主成分变量,同时消除变量间的相关性,然后用最小二乘支持向量机(least squares support vector machine,LSSVM)作为内部函数来描述主成分变量之间的非线性关系,并引入基于误差最小化的权值更新策略,来改进模型的预测精度。以pH中和过程的Benchmark模型来验证该方法的性能,并与其他建模方法比较,结果表明该方法预测精度较高,而且具有较强的泛化能力。将该方法应用于某电站燃煤锅炉的NOx排放软测量建模之中,取得了较好的预测效果。     

Learning a data-dependent kernel function for KPCA-based nonlinear process monitoring

Kernel principal component analysis (KPCA)-based process monitoring methods have recently shown to be very effective for monitoring nonlinear processes. However, their performances largely depend on the kernel function and currently there is no general rule for kernel selection. Existing methods simply choose the kernel function empirically or experimentally from a given set of candidates. This paper proposes a kernel function learning method for KPCA to learn a kernel function tailored to specific data and explores its potential for KPCA-based process monitoring. Motivated by the manifold learning method maximum variance unfolding (MVU), we obtain the kernel function by optimizing over a family of data-dependent kernels such that the nonlinear structure in input data is unfolded in the kernel feature space and gets more likely to be linear there. Using the optimized kernel, the nonlinear principal components of KPCA which are linear principal components in the kernel feature space can effectively capture the variation in data, and thus the data under normal operating conditions can be more precisely modeled by KPCA for process monitoring. Simulation results on an simple nonlinear system and the benchmark Tennessee Eastman (TE) demonstrate that the optimized kernel functions lead to significant improvement in the performance over the popular Gaussian kernels when used in the KPCA-based process monitoring.     

Composition of the oil in palm kernel fromElaeis guineensis

The composition of oil from the outer, middle, and inner section of palm kernel had been evaluated by gas-liquid chromatography and nuclear magnetic resonance techniques. The composition was not homogenous throughout the kernel and was found to be more unsaturated in the outer kernel. The inner core of the kernel is less unsaturated, having a higher lauric content. The variation in the iodine value from the outer to inner section of the kernel suggested that composition of palm kernel oil was affected by the kernel size. This was found to be true, as small kernels tend to have oil with higher iodine value than do larger kernels.     

基于稀疏特征空间的核方法建模研究

为提高支持向量机在建模方面的拟合性能,针对核函数方法中单个核函数的局限性,尝试融合核支持向量机建模方法以提高模型的泛化能力和精度。为避免在进行核融合时,支持向量机稀疏性的缺失,提出将数据映射到稀疏特征空间进行研究。通过仿真研究表明,所建模型在保证稀疏性的前提下,能较好地提高建模精度,从而验证了算法的有效性。     

Determination of aflatoxins in individual peanuts and peanut sections

Subsamples of a given lot of peanuts may vary greatly in aflatoxin content due to extreme variability in the degree of contamination of individual kernels. A micro method, adapted from the aqueous acetone procedure recently proposed by Pons and Goldblatt for the determination of aflatoxins in cottonseed products, was developed to permit accurate determination of aflatoxins in individual kernels and kernel sections. Use of this procedure permitted the topographic distribution of aflatoxins within single kernels to be mapped and indicated that the toxins are not uniformly distributed within contaminated kernels, even when the kernel contains a high level of aflatoxins. Although wrinkling or discoloration sometimes indicated that a kernel was contaminated, this type of physical damage was not found to be a reliable indication of aflatoxin content. Also it was noted that a few apparently sound and mature kernels contained high levels of aflatoxins. Presented at the AOCS Meeting, Houston, April 1965. Honorable Mention Bond Award Competition. So. Utiliz. Res. Dev. Div., ARS, USDA.     

High shear wet granulation modelling—a mechanistic approach using population balances

A population balance approach based on splitting the coalescence kernel into two factors, the first describing the collision frequency of particles and the second describing the collision efficiency, is applied to modelling wet granulation in a high shear mixer. Four different expressions for the collision frequency are compared and discussed. The kernels are the size independent kernel, the shear kernel proposed by Smoluchowski [M. Smoluchowski, “Versuch einer matematischen Theorie der Koagulationskinetic kolloider Lösungen”, Z. Physik. Chem. 92, (1917) 129–168.] and the two kernels proposed by Hounslow [M.J. Hounslow, “The population balance as a tool for understanding particle rate processes”, Kona (1998) 179–193.], i.e. the EKE kernel and the less used kernel based on equipartition of fluctuating translational momentum (ETM kernel). Microcrystalline cellulose (mcc) is granulated under different process conditions and it is found that the ETM kernel best describes the granulation at higher impeller speeds, whereas the EKE kernel gives better agreement at lower impeller speeds. The collision efficiency is assumed to be a function of the liquid saturation. By using this assumption, it was possible to detect similar trends for the remaining part of the collision efficiency regardless of process conditions.     

Investigation of droplet breakup in liquid-liquid dispersions by CFD-PBM simulations: The influence of the surfactant type

The accurate prediction of the droplet size distribution (DSD) in liquid-liquid turbulent dispersions is of fundamental importance in many industrial applications and it requires suitable kernels in the population balance model.When a surfactant is included in liquid-liquid dispersions,the droplet breakup behavior will change as an effect of the reduction of the interracial tension.Moreover,also the dynamic interfacial tension may be different with respect to the static,due to the fact that the surfactant may be easily desorbed from the droplet surface,generating additional disruptive stresses.In this work,the performance of five breakup kernels from the literature is assessed,to investigate their ability to predict the time evolution of the DSD and of the mean Sauter diameter,when different surfactants are employed.Simulations are performed with the Quadrature Method of Moments for the solution of the population balance model coupled with the two-fluid model implemented in the compressibleTwoPhaseEulerFoam solver of the open-source computational fluid dynamics (CFD) code OpenFOAM v.2.2.x.The time evolution of the mean Sauter diameter predicted by these kernels is validated against experimental data for six test cases referring to a stirred tank with different types of surfactants (Tween 20 and PVA 88％) at different concentrations operating under different stirrer rates.Our results show that for the dispersion containing Tween 20 additional stress is generated,the multifractal breakup kernel properly predicts the DSD evolution,whereas two other kernels predict too fast breakup of droplets covered by adsorbed PVA.Kernels derived originally for bubbles completely fail.     

Enzyme-assisted aqueous extraction of fat from kernels of the shea tree,Butyrospermum parkii

Chemical characteristics of kernels of the shea tree (Butyrospermum parkii) from Ghana were determined in order to design and evaluate studies on a traditional enzyme-assisted fat extraction of the kernels. The effectiveness of a number of cellular degrading enzymes in assisting the shea fat extraction were also tested by treating meals of the kernels with one or more of these enzymes before extraction and comparing the yield with control extractions. Proximate composition of the kernel on dry-matter basis was: total lipids, 59.04%; crude fat, 54.85; protein, 7.81%; total carbohydrates, 34.77%; ash, 2.57%. Starch content was 7.59%; hemicellulose, 10.84%; cellulose, 5.95%; and pectic substances, 2.93%. Total fiber content was 20.35%. The fat extracted by the Soxhlet method was pale-yellow in color and solid at room temperatures. Its physicochemical characteristics were: melting range, 34–36°C; iodine value, 58.53; saponification value, 180.37; and unsaponifiable matter content, 7.48%. The predominant fatty acids were: palmitic (3.55%), stearic (44.44%), oleic (42.41%), linoleic (5.88%) and linolenic (1.66%) acids. The enzyme-assisted extraction tests showed increases in extraction yield when the shea kernel meals were treated with the enzyme(s) before extraction. An increase of about 20% was realized when a protease and an enzyme with both cellulase and hemicellulase activities were used together. These observations confirmed the fact that the shea kernel is a rich source of fat. They also indicate the possibility of improving shea kernel extraction processes by pretreating the kernel meal with cell structure-degrading enzymes.     

Bitter Almond,Plum and Mango Kernels as Sources of Lipids

The chemical composition of bitter almond, plum and mango kernels and the physico-chemical characteristics of their lipids were studied. Bitter almond and plum kernels contained higher amounts of lipids in comparison to mango kernels. All kernel lipids were found free from hydrocyanic acid. The predominant lipid class of the studied kernel lipids was triglycerides. Oleic acid was the major fatty acid present in bitter almond and plum kernel lipids, while mango kernel lipids were rich in stearic and oleic acid.     

Bicomponent aggregation with composition-dependent rates and the approach to well-mixed state

We consider the problem of aggregative mixing of components from a theoretical standpoint. We formulate equations for the evolution of the bivariate size distribution, the classical size distribution, and the compositional distribution, and introduce a segregation index to quantify the degree segregation and blending in the population. We consider composition-dependent kernels and examine their effect on the degree of blending of components by aggregation. To systematically study the effect of the kernel, we introduce a new kernel that allows us to adjust the strength of cross-aggregation over that of self-aggregation. Kernels that promote cross-aggregation lead to more efficient blending of components, as judged by the rapid decrease of the segregation index, while kernels that promote self-aggregation inhibit blending and may even result in segregation during the earlier stages of aggregation. In all cases, however, the segregation index ultimately scales inversely with the mean particle mass. Thus, regardless of kernel details, the segregation index at long times becomes arbitrarily small and follows the same scaling as with kernels that are independent of composition.