Sensory Modeling of Coffee with a Fuzzy Neural Network

ABSTRACT: Models were constructed to predict sensory evaluation scores from the blending ratio of coffee beans. Twenty-two blended coffees were prepared from 3 representative beans and were evaluated with respect to 10 sensory attributes by 5 coffee cup-tasters and by models constructed using the response surface method (RSM), multiple regression analysis (MRA), and a fuzzy neural network (FNN). The RSM and MRA models showed good correlations for some sensory attributes, but lacked sufficient overall accuracy. The FNN model exhibited high correlations for all attributes, clearly demonstrated the relationships between blending ratio and flavor characteristics, and was accurate enough for practical use. FNN, thus, constitutes a powerful tool for accelerating product development.     

Prognostic predictor with multiple fuzzy neural models using expression profiles from DNA microarray for metastases of breast cancer

Gene expression profiling data from DNA microarray were analyzed using the fuzzy neural network (FNN) modeling method for predicting the distant metastases of breast cancer. The best model consisting of five genes was able to predict metastases of breast cancer with 94% accuracy. Furthermore, 100% accuracy was achieved by majoritarian decision using only 25 genes from five noninferior models which were constructed independently. From the constructed model, gene expression rules, which may cause distant metastases, were explicitly extracted and 60% of the metastases cases could be explained by this rule. The FNN modeling method described in this paper enables precise extraction of significant biological markers affecting prognosis without prior knowledge.     

Selection of causal gene sets for lymphoma prognostication from expression profiling and construction of prognostic fuzzy neural network models

To assess the response of lymphomas to chemotherapy, gene expression profiling data from DNA microarrays were analyzed using the fuzzy neural network (FNN) modeling method. We used the FNN modeling method to produce 10 noninferior models. Using these models, we were able to predict diffuse large B-cell lymphoma (DLBCL) patient outcome with 93% accuracy. Of the 37 genes in the 10 models, 13 genes were repeatedly selected, indicating that these genes are important for prognostication. On Kaplan-Meier plots of overall survival, patients predicted by the FNN model to be cured survived significantly longer than those predicted to be refractory (P<0.0001), indicating that the FNN could successfully identify patients with a relatively poor prognosis among low-clinical-risk patients. The FNN modeling method presented here is able to precisely extract significant biological markers affecting prognosis.     

大豆油脱臭工艺的流程模拟及回归建模分析

利用计算机模拟技术建立大豆油脱臭工艺的流程模拟模型;以脱臭操作压力、脱臭进料温度、汽提蒸汽量为操作变量,按照中心组合设计原理制定实验方案,并利用流程模拟模型完成所有实验。根据实验结果建立大豆油脱臭工艺的油脂损失率和脱臭油中植物甾醇保留率的数学回归模型。该数学模型达到较高的拟合程度和准确性。可为计算机模拟技术和数学建模技术在油脂工艺优化领域的应用提供参考。     

Modeling of pH and acidity for industrial cheese production

A three-layer feedforward neural network was successfully used to model and predict the pH of cheese curd at various stages during the cheese-making process. An extended database, containing more than 1800 vats over 3 yr of production of Cheddar cheese with eight different starters, from a large cheese plant was used for model development and parameter estimation. Neural network models were developed with inputs selected among 33 quantitative and qualitative process variables for final pH of cheese, pH at cutting, and acidity at whey drawing-off and at pressing. In all cases, very high correlation coefficients, ranging from 0.853 to 0.926, were obtained with the validation data. A sensitivity analysis of neural network models allowed the relative importance of each input process variable to be identified. The sensitivity analysis in conjunction with a priori knowledge permitted a significant reduction in the size of the model input vector. A neural network model using only nine input process variables was able to predict the final pH of cheese with the same accuracy as for the complete model with 33 original input variables. This significant decrease in the size of neural networks is important for applications of process control in cheese manufacturing.     

Prediction of peptide binding to major histocompatibility complex class II molecules through use of boosted fuzzy classifier with SWEEP operator method

To treat autoimmune diseases, it is important to identify which peptides bind to major histocompatibility complex (MHC) class II molecules (HLA-DRs). Predicting the peptides that bind to MHC class II molecules can effectively reduce the number of experiments required for identifying helper T cell epitopes. In our previous study, we applied fuzzy neural networks (FNNs) to solve this problem. However, an FNN requires a long calculation time and a large number of peptides; this means performing several experiments. In this study, we applied a boosted fuzzy classifier with the SWEEP operator method (BFCS) to solve this problem. For comparison, two other conventional modeling methods, namely, support vector machine and FNN combined with the SWEEP operator method (FNN-SWEEP) instead of using solely an FNN, were employed. Compared with FNN, FNN-SWEEP is extremely fast and has an almost identical prediction accuracy. The model constructed by BFCS showed an accuracy approximately 5%-10% higher than that constructed by FNN-SWEEP. In addition, BFCS was 30,000-120,000 times faster than FNN-SWEEP. This result suggests that BFCS has the potential to function as a new method of predicting peptides that bind to various protein receptors.     

基于三维测量的青年女性臀部体型概率神经网络识别模型构建

如何快速准确地实现体型识别是人体体型研究的热点。为满足服装臀部合体性的要求,本文结合青年女性臀部体型特征,构建了基于三维测量的青年女性臀部体型PNN识别模型。首先,运用三维人体测量仪采集数据,并提取6个典型指标,进行臀部体型细分;其次,引入概率神经网络方法,构建以典型指标作为输入层,体型类别作为输出层,径向基函数作为模式层的网络结构模型;再次,利用MATLAB R2009a软件对构建的概率神经网络模型进行仿真实验,通过训练获取精度高、结果稳定的模型;最后,测试模型识别精度。结果表明,该模型识别率高,识别性能良好,为女性臀部体型识别提供了一种新方法,同时也拓宽了概率神经网络方法的应用领域。     

烟草温光特性研究与利用：Ⅲ.基于温光效应的烟草叶片生长模拟模型建立

为准确模拟烟草叶片生长发育进程,实现烟叶精准可控生产,连续两年设置不同移栽期田间对比试验,利用Richards方程建立基于不同尺度的烟草下、中、上部叶面积变化动态模型,并分析不同模型的模拟精度。结果表明,烟草各部位叶片叶面积变化动态模型均符合典型“S”型生长曲线特征,有效积温模型对下、中部叶片生长的模拟效果优于生长时间模型,而对上部叶片生长的模拟效果较差;温光效应模型对各部位叶片不同条件下生长进程的模拟精度均高于有效积温模型与生长时间模型,具有更高的普适性;各部位叶片最终叶面积随移栽期推迟呈现先增加后降低的规律,下、中部叶片生长速率随移栽期推迟呈现加快规律,而不同移栽期上部叶的生长速率近似;推导获得各部位叶片缓增期、快增期、稳增期的温光效应值,为精准预测叶片生长提供参考。     

Application of an artificial neural network and genetic algorithm for determination of process orbits in the koji making process

In order to estimate alpha-amylase, glucoamylase, acid proteinase, and acid carboxypeptidase activities in koji from the process variables and initial conditions of the koji making process, artificial neural network (ANN) models (ANN-10, -11, -15, and -21) were constructed with 10, 11, 15, and 21 input variables, respectively. These models could estimate the enzyme activities with high accuracy. Temperature and humidity orbits were then acquired by a genetic algorithm searching in the reverse direction using ANN-10, -11, -15, and -21 (GA-10, -11, -15, and -21). The orbits acquired by GA-15 and -11 were almost identical to the actual orbits, but those acquired by GA-21 and -10 were different. Enzyme activities acquired by GA-15 had 1.3% errors compared with the target values, while those acquired by GA-11 had 9.7% errors. GA-15 was, therefore, selected as the most suitable algorithm and was used to determine temperature and humidity orbits for target enzyme activities. Test koji making was then carried out according to the orbits acquired. As a result, the enzyme activities of the koji produced were almost the same as the target values.     

Fuzzy neural network-based prediction of the motif for MHC class II binding peptides

Characterizing the interaction between major histocompatibility complex (MHC) molecules and antigenic peptides is critical for understanding immunity and developing immunotherapies for autoimmune diseases and cancer. To identify the peptide binding motif and predict peptides that bind to the human MHC classII molecule HLA-DR4(*0401), we applied a fuzzy neural network (FNN) capable of extracting the relationship between input and output. Analysis of the peptide binding motif revealed that the hydrophilicity of the position 1 residue located on the N-terminal side of the nonamer (9mer) was the most important variable and that the van der Waals volume and hydrophilicity of the position 6 residue and the hydrophilicity of the position 7 residue were also important variables. The estimation accuracy (A(ROC) value) was high and the binding motif extracted from the FNN agreed with that derived experimentally. This study demonstrates that FNN modeling allows candidate antigenic peptides to be selected without the need for further experiments.     

Modeling pedigree accuracy and uncertain parentage in single-step genomic evaluations of simulated and US Holstein datasets

The objective of this study was to model differences in pedigree accuracy caused by selective genotyping. As genotypes are used to correct pedigree errors, some pedigree relationships are more accurate than others. These accuracy differences can be modeled with uncertain parentage models that distribute the paternal (maternal) contribution across multiple sires (dams). In our case, the parents were the parent on record and an unknown parent group to account for pedigree relationships that were not confirmed through genotypes. Pedigree accuracy was addressed through simulation and through North American Holstein data. Data were simulated to be representative of the dairy industry with heterogeneous pedigree depth, pedigree accuracy, and genotyping. Holstein data were obtained from the official evaluation for milk, fat, and protein. Two models were compared: the traditional approach, assuming accurate pedigrees, and uncertain parentage, assuming variable pedigree accuracy. The uncertain parentage model was used to add pedigree relationships for alternative parents when pedigree relationships were not certain. The uncertain parentage model included 2 possible sires (dams) when the sire (dam) could not be confirmed with genotypes. The 2 sires (dams) were the sire (dam) on record with probability 0.90 (0.95) and the unknown parent group for the birth year of the sire (dam) with probability 0.10 (0.05). An additional set of assumptions was tested in simulation to mimic an extensive dairy production system by using a sire probability of 0.75, a dam probability of 0.85, and the remainder attributed to the unknown parent groups. In the simulation, small bias differences occurred between models based on pedigree accuracy and genotype status. Rank correlations were strong between traditional and uncertain parentage models in simulation (≥0.99) and in Holstein (≥0.99). For Holsteins, the estimated breeding value differences between models were small for most animals. Thus, traditional models can continue to be used for dairy genomic prediction despite using genotypes to improve pedigree accuracy. Those genotypes can also be used to discover maternal parentage, specifically maternal grandsires and great grandsires when the dam is not known. More research is needed to understand how to use discovered maternal pedigrees in genetic prediction.     

Application of Fourier transform near-infrared spectroscopy to optimization of green tea steaming process conditions

In this study, we constructed prediction models by metabolic fingerprinting of fresh green tea leaves using Fourier transform near-infrared (FT-NIR) spectroscopy and partial least squares (PLS) regression analysis to objectively optimize of the steaming process conditions in green tea manufacture. The steaming process is the most important step for manufacturing high quality green tea products. However, the parameter setting of the steamer is currently determined subjectively by the manufacturer. Therefore, a simple and robust system that can be used to objectively set the steaming process parameters is necessary. We focused on FT-NIR spectroscopy because of its simple operation, quick measurement, and low running costs. After removal of noise in the spectral data by principal component analysis (PCA), PLS regression analysis was performed using spectral information as independent variables, and the steaming parameters set by experienced manufacturers as dependent variables. The prediction models were successfully constructed with satisfactory accuracy. Moreover, the results of the demonstrated experiment suggested that the green tea steaming process parameters could be predicted on a larger manufacturing scale. This technique will contribute to improvement of the quality and productivity of green tea because it can objectively optimize the complicated green tea steaming process and will be suitable for practical use in green tea manufacture.     

基于MIV-BP神经网络的成品烟丝质量预测模型构建

目的：构建卷烟制丝过程成品烟丝质量模拟预测模型。方法：使用平均影响值法（the Mean Impact Value, MIV）对制丝加工过程工艺参数进行筛选,然后通过反向传播（Back-Propagation,BP）神经系统构建起制丝关键工艺参数和成品烟丝质量的模拟模型。结果：通过模拟数据与实测数据比较,填充值的模拟预测平均相对误差为3.16%;整丝率的模拟预测平均相对误差为0.67%;碎丝率的模拟预测平均相对误差为5.33%。结论：该模型预测值与实测值之间相对误差较小,精确性高,该模型适用于卷烟制丝生产过程工艺参数仿真优化。     

遗传算法联合LS-SVM的苹果原醋成分定量分析

利用近红外光谱技术对苹果原醋中的重要指标进行定量分析,并进行模型优化以提高性能。采用遗传偏最小二乘法（GA-PLS）提取的特征波长作为最小二乘支持向量机（LS-SVM）的输入变量,先后建立苹果原醋中总酸、可溶性固形物的近红外定量模型,并与建立的偏最小二乘（PLS）模型结果进行比较。用决定系数（R2）、预测均方根误差（RMSEP）以及相对分析误差（RPD）对模型进行评价,确定最佳建模方法。结果表明,相比于PLS模型,总酸及可溶性固形物指标的LS-SVM定量模型的R2、RMSEP以及RPD值均有更好的表现,且在进行独立测试集验证时,LS-SVM模型的预测精度也明显优于PLS模型。说明遗传算法联合LS-SVM建立的定量模型有很高的准确度及稳定性,可以应用于苹果原醋总酸和可溶性固形物含量的快速检测。     

Two‐way prediction of cotton yarn properties and fiber properties using multivariate multiple regression

This paper explains the feasibility of two‐way prediction by developing direct models relating fiber to yarn and reverse models relating yarn to fiber using multivariate methods simultaneously. These models evaluate the dependencies of cotton yarn properties on fiber properties and vice versa with minimum random errors and maximum accuracy. To this end, cotton fiber properties were measured from rovings carefully untwisted. An HVI system and an evenness tester of premier were used to measure the various properties. The samples of cotton yarns (108 samples) produced yarn counts ranging from 16 to 32 Ne with optimum twist factor. In this study, effective variables were selected by multivariate statistical test (m‐test). Then, multivariate analysis of variance (MANOVA) was used for evaluating the significance of obtained models. Next, the optimal separate equations were determined through multivariate multiple regression. After solving the linear equation system, a reverse model was achieved. By selecting fiber properties and machine factors as appropriate variables, the relative importance of these factors was also investigated. The results showed that the obtained equations were significant at the significance level α = 0.01.     

MD-1200YJ码垛机器人腰部支架的多目标结构优化设计

建立了MD-1200YJ型码垛机器人腰部支架的有限元模型,通过模态分析与模态试验验证有限元模型的准确性,并基于5次非均匀有理B样条运动规律进行考虑动力学因素的静力学分析,分析结果表明该零件具有轻量化设计的潜力;确定以质量最小、最大位移最小、最大应力最小为优化目标,以结构参数为设计变量,以第一、二阶固有频率不降低以及设计变量的边界条件为约束条件,利用Box-Behnken和RSM方法建立目标函数和约束函数的RSM模型,并验证了模型的准确性;利用NSGA-Ⅱ算法求得最优解,进而得到腰部支架的优化模型;通过对优化模型进行静力学分析、模态分析,并与初始模型进行对比,在保证前两阶固有频率不降低以及最大应力和位移均在允许范围内的情况下,质量减轻了8.2%,验证了该优化设计方法的有效性。     

中红外光谱技术结合竞争性自适应重加权算法快速分析白酒风味组分

采用中红外光谱分析技术结合竞争性自适应重加权算法(CARS)对浓香型白酒基酒中的乳酸乙酯和乙酸乙酯的特征波长变量进行筛选后,建立偏最小二乘法(PLS)模型,并对其进行验证。结果表明,采用中红外光谱分析技术剔除明显噪声区域建立的PLS模型效果较好,而经CARS法进行特征波长选择后建立的CARS-PLS模型效果优于PLS模型,乙酸乙酯和乳酸乙酯的CARS-PLS模型相关系数R²分别为0.995、0.989,预测均方根误差(RMSEP)分别为12.80、4.54,相对分析误差(RPD)分别为8.78及8.60,模型经独立验证均取得了较高的预测精度,验证数据相关系数R²分别为0.994及0.992,RMSEP分别为13.55及4.86。该模型有较高的准确度及稳定性,能够用于白酒基酒中的乳酸乙酯和乙酸乙酯的快速分析,可为白酒酿造过程的质量把控提供技术方法。     

Comparison of models to identify lame cows based on gait and lesion scores, and limb movement variables

Bovine lameness results in pain and suffering in cattle and economic loss for producers. A system for automatically detecting lame cows was developed recently that measures vertical force components attributable to individual limbs. These measurements can be used to calculate a number of limb movement variables. The objective of this investigation was to explore whether gait scores, lesion scores, or combined gait and lesion scores were more effectively captured by a set of 5 limb movement variables. A set of 700 hind limb examinations was used to create gait-based, lesion-based, and combined (gait- and lesion-based) models. Logistic regression models were constructed using 1, 2, or 3 d of measurements. Resulting models were tested on cows not used in modeling. The accuracy of lesion-score models was superior to that of gait-score models; lesion-based models generated greater values of areas under the receiving operating characteristic curves (range 0.75 to 0.84) and lower mean-squared errors (0.13 to 0.16) compared with corresponding values for the gait-based models (0.63 to 0.73 and 0.26 to 0.31 for receiving operating characteristic and mean-squared errors, respectively). These results indicate that further model development and investigation could generate automated and objective methods of lameness detection in dairy cattle.     

黑曲霉固态发酵麻竹笋壳动力学研究

利用黑曲霉对麻竹笋壳进行固态发酵,每12 h取样,测定菌落总数,纤维素酶活,并基于Logistic和Luedeking-Piret方程,研究了黑曲霉固态发酵麻竹笋壳过程的菌体生长动力学,产物合成动力学。研究结果表明,模型模拟计算结果与试验结果吻合良好,该模型基本能反映黑曲霉固态发酵麻竹笋壳的过程。     

基于FLUENT的菊花热风干燥流场特性仿真分析

在菊花干燥过程中,干燥机内部的流场分布特点是影响菊花干燥品质的主要因素。为进一步研究在菊花干燥过程中干燥机内部的流场分布及变化规律,针对负压式干燥机,建立了干燥机的物理模型。通过对模型进行网格划分、参数确定和求解计算,得到了菊花干燥过程中速度场、压强场、温度场在干燥机内部的分布情况,实现了对菊花干燥过程中流场分布的数值模拟。为验证模拟结果的准确性,着重将温度场的模拟结果和实测结果进行了对比分析,发现相对误差在8%以内,充分证实了模拟结果的可靠性。