人工神经网络-近红外光谱法用于粉末药品扑热息痛的非破坏定量分析

研究了应用人工神经网络进行粉末药品的非破坏定量分析,使用扑热息痛粉末药品的近红外漫反射光谱数据建立人工神经网络模型,预测未知样品,讨论了影响网络的各参数,采用逼近度作为网络新的评价标准,由于人工神经网络好的非线性的多变量校正特点,预测结果是准确的。     

Nondestructive Quantitative Analysis of Cofrel Medicines by Double ANN-NIR Spectroscopy

It has been proved that near-infrared (NIR) spectroscopy is a powerful analytical tool in the pharmaceutical industries1, especially in the quantitative analysis of the pharmaceu- tical tests during the last decades2-4. Currently, the quantitative analyti…     

Modeling of synergistic solvent extraction of Zr(IV) / HCl / TOA+Cyanex 921 system using statistical analysis and neural network

Multi-linear regression analysis (MLR), radial basis function (RBF) and multilayer perceptron (MLP) of artificial neural network (ANN) with five inputs (temperature, concentrations of HCl, TOA, Cyanex 921, Zr (IV) and percentage of extraction (%E)) as only output were employed for the construction of models. It was observed that ANN (RBF and MLP) performed better as compared to the MLR model. Based on the models proposed, the extraction of Zr(IV) could be predicted under variable experimental conditions of concentrations of HCl, TOA (Tri-n-octylamine), Cyanex 921 (Tri-n-octyl phosphineoxide), Zr(IV) and temperature. The nonlinear and complex relation between the percentage of extraction and operating variables have been determined using two and three layered feed forward neural network with back-propagation of error learning algorithm. Uncertainties in data have been determined in terms of statistical parameters such as root mean-squared error and R-squared values to check the efficiency of the model for prediction.     

Estimation of Polymerization Conditions Needed to Make Ethylene/1‐olefin Copolymers with Specific Microstructures Using Artificial Neural Networks

Two artificial neural network models (forward and inverse) are developed to describe ethylene/1‐olefin copolymerization with a catalyst having two site types using training and testing datasets obtained from a polymerization kinetic model. The forward model is applied to predict the molecular weight and chemical composition distributions of the polymer from a set of polymerization conditions, such as ethylene concentration, 1‐olefin concentration, cocatalyst concentration, hydrogen concentration, and polymerization temperature. The results of the forward model agree well with those from the kinetic model. The inverse model is applied to determine the polymerization conditions to produce polymers with desired microstructures. Although the inverse model generates multiple solutions for the general case, unique solutions are obtained when one of the three key process parameters (ethylene concentration, 1‐olefin concentration, and polymerization temperature) is kept constant. The proposed model can be used as an efficient tool to design materials from a set of polymerization conditions.

     

The prediction of total anthocyanin concentration in red-grape homogenates using visible-near-infrared spectroscopy and artificial neural networks

This study compares the performance of partial least squares (PLS) regression analysis and artificial neural networks (ANN) for the prediction of total anthocyanin concentration in red-grape homogenates from their visible-near-infrared (Vis-NIR) spectra. The PLS prediction of anthocyanin concentrations for new-season samples from Vis-NIR spectra was characterised by regression non-linearity and prediction bias. In practice, this usually requires the inclusion of some samples from the new vintage to improve the prediction. The use of WinISI LOCAL partly alleviated these problems but still resulted in increased error at high and low extremes of the anthocyanin concentration range. Artificial neural networks regression was investigated as an alternative method to PLS, due to the inherent advantages of ANN for modelling non-linear systems. The method proposed here combines the advantages of the data reduction capabilities of PLS regression with the non-linear modelling capabilities of ANN. With the use of PLS scores as inputs for ANN regression, the model was shown to be quicker and easier to train than using raw full-spectrum data. The ANN calibration for prediction of new vintage grape data, using PLS scores as inputs, was more linear and accurate than global and LOCAL PLS models and appears to reduce the need for refreshing the calibration with new-season samples. ANN with PLS scores required fewer inputs and was less prone to overfitting than using PCA scores. A variation of the ANN method, using carefully selected spectral frequencies as inputs, resulted in prediction accuracy comparable to those using PLS scores but, as for PCA inputs, was also prone to overfitting with redundant wavelengths.     

Effect of structural parameters on the flow field and power consumption of in-line high shear mixer

The effects of three structural parameters on flow field and power consumption of in-line high shear mixer (HSM) were investigated by large eddy simulation (LES). In addition, an artificial neural network (ANN) is used to predict the relationship between the structural parameters and the power consumption, and the effect of dimensionless structural parameters on the power number constant Poz and k1 is studied. The results show that the stator tooth thickness and the tooth tip-base distance have a slight effect on the flow field, and the shear gap width is a key parameter affecting the flow field. As the stator teeth thickness, the tooth tip-base distance and the shearing gap width increases, the power number Po decreases. There is a linear relationship between the constant k₁ and the dimensionless structural parameters. With the increase of the dimensionless parameter Ts/Ds-o of the stator tooth thickness, the dimensionless parameter St/H of the tooth tip-base distance, and the dimensionless parameter Sg/DR-o of the shear gap width, the constant k₁ decreases. With the increase of the parameter St/H, Sg/DR-o and Ts/Ds-o, the constant Po_z first increases and then decreases. There is a linear relationship between the constant Po_z and the parameter Ts/h. With the increase of the parameter Ts/h, the constant Po_z increases.     

Ultrasound-assisted extraction of phenolic acids,flavonols, and flavan-3-ols from muscadine grape skins and seeds using natural deep eutectic solvents and predictive modelling by artificial neural networking

The objective of this study was to investigate the extraction efficiency of 9 natural deep eutectic solvents (NDES) with the assistance of ultrasound for phenolic acids, flavonols, and flavan-3-ols in muscadine grape (Carlos) skins and seeds in comparison to 75% ethanol. Artificial neural networking (ANN) was applied to optimize NDES water content, ultrasonication time, solid-to-solvent ratio, and extraction temperature to achieve the highest extraction yields for ellagic acid, catechin and epicatechin. A newly formulated NDES (#1) consists of choline chloride: levulinic acid: ethylene glycol 1:1:2 and 20% water extracted the highest amount of ellagic acid in the skin at 22.1 mg/g. This yield was 1.73-fold of that by 75% ethanol. A modified NDES (#3) consisting of choline chloride: proline: malic acid 1:1:1 and 30% water extracted the highest amount of catechin (0.61 mg/g) and epicatechin (0.89 mg/g) in the skin, and 2.77 mg/g and 0.37 mg/g in the seed, respectively. The optimal yield of ellagic acid in the skin using NDES #1 was 25.3 mg/g (observed) and 25.3 mg/g (predicted). The optimal yield of (catechin + epicatechin) in seed using NDES #3 was 9.8 mg/g (observed) and 9.6 mg/g (predicted). This study showed the high extraction efficiency of selected NDES for polyphenols under optimized conditions.     

Melamine detection by mid- and near-infrared (MIR/NIR) spectroscopy: a quick and sensitive method for dairy products analysis including liquid milk, infant formula, and milk powder

Melamine (2,4,6-triamino-1,3,5-triazine) is a nitrogen-rich chemical implicated in the pet and human food recalls and in the global food safety scares involving milk products. Due to the serious health concerns associated with melamine consumption and the extensive scope of affected products, rapid and sensitive methods to detect melamine's presence are essential. We propose the use of spectroscopy data-produced by near-infrared (near-IR/NIR) and mid-infrared (mid-IR/MIR) spectroscopies, in particular—for melamine detection in complex dairy matrixes. None of the up-to-date reported IR-based methods for melamine detection has unambiguously shown its wide applicability to different dairy products as well as limit of detection (LOD) below 1 ppm on independent sample set. It was found that infrared spectroscopy is an effective tool to detect melamine in dairy products, such as infant formula, milk powder, or liquid milk. ALOD below 1 ppm (0.76 ± 0.11 ppm) can be reached if a correct spectrum preprocessing (pretreatment) technique and a correct multivariate (MDA) algorithm—partial least squares regression (PLS), polynomial PLS (Poly-PLS), artificial neural network (ANN), support vector regression (SVR), or least squares support vector machine (LS-SVM)—are used for spectrum analysis. The relationship between MIR/NIR spectrum of milk products and melamine content is nonlinear. Thus, nonlinear regression methods are needed to correctly predict the triazine-derivative content of milk products. It can be concluded that mid- and near-infrared spectroscopy can be regarded as a quick, sensitive, robust, and low-cost method for liquid milk, infant formula, and milk powder analysis.     

Rapid honey characterization and botanical classification by an electronic tongue

In this paper a commercial electronic tongue (αAstree, Alpha M.O.S.) was applied for botanical classification and physicochemical characterization of honey samples. The electronic tongue was comprised of seven potentiometric sensors coupled with an Ag/AgCl reference electrode. Botanical classification was performed by PCA, CCA and ANN modeling on 12 samples of acacia, chestnut and honeydew honey. The physicochemical characterization of honey was obtained by ANN modeling and the parameters included were electrical conductivity, acidity, water content, invert sugar and total sugar. The initial reference values for the physicochemical parameters observed were determined by traditional methods. Botanical classification of honey samples obtained by ANN was 100% accurate while the highest correlation between observed and predicted values was obtained for electrical conductivity (0.999), followed by acidity (0.997), water content (0.994), invert sugar content (0.988) and total sugar content (0.979).All developed ANN models for rapid honey characterization and botanical classification performed excellently showing the potential of the electronic tongue as a tool in rapid honey analysis and characterization. The advantage of using such a technique is a simple sample preparation procedure, there are no chemicals involved and there are no additional costs except the initial measurements required for ANN model development.     

基于BP-ANN和PLS的近红外光谱无损检测李果实品质的研究

可溶性固形物（SSC）和可滴定总酸（TA）含量是影响李果实品质的重要指标,经典的破坏性检测方法不适用于果实按品质分级,近红外光谱(NIRS)检测方法具有速度快、操作简便、可无损检测果实品质。为实现NIRS无损快速检测安哥诺李果实可溶性固形物和可滴定总酸含量,利用NIRS采集李果实的漫反射光谱,同时采用糖度计测定安哥诺李果实的SSC,采用滴定法测定了李果实TA含量,使用杠杆值和F概率值剔除异常样品,采用软件优化结合人工筛选光谱波段,使用了消除常数偏移量、减去一条直线、矢量归一化（SNV）、最大-最小归一化、多元散射校正（MSC）、一阶和二阶导数结合平滑处理、一阶导数结合减去一条直线和平滑处理、以及一阶导数结合SNV或MSC校正等光谱预处理方法,分别采用偏最小二乘法（PLS）和主成分分析结合反向传播人工神经网络（BP-ANN）建立李果实SSC、TA的定量分析模型。结果表明,李果实SSC和TA的最佳PLS建模效果波段范围分别为4 000～8 852和4 605～6 523 cm-1。SSC的PLS模型的最佳光谱预处理方法为MSC校正,最佳模型校正相关系数（R_c）为0.914 4,预测相关系数（R_p）为0.878 5,校正均方根误差（RMSEC）为0.91,预测均方根误差（RMSEP）为1.00。经一阶微分结合SNV和9点平滑的方法预处理后,TA的PLS模型效果最佳,R_c,R_p,RMSEC,RMSEP分别为0.860 3,0.819 6,0.80和0.86。提取了李果实SSC和TA光谱数据的主成分,并基于前10个主成分得分建立了李果实SSC和TA最佳BP-ANN定量分析模型,其R_c,R_p,RMSEC和RMSEP分别为0.976 7,0.889 7,0.75和0.99;TA的BP-ANN模型的相应参数值依次为0.974 3,0.897 7,0.62和0.83,与采用PLS算法建立的定量模型相比较,BP-ANN模型具有较高的R_c,R_p和较低的RMSEC,RMSEP,因此BP-ANN模型对SSC和TA指标的定量分析结果更佳。