Application of artificial neural network for prediction of 10 crude oil properties

This study aims to develop an industrially reliable and accurate method to estimate crude oil properties from their Fourier transform infrared spectroscopy (FTIR) spectra. We used the complete FTIR spectral data of selected crude oil samples from seven different Canadian oil fields to predict 10 important crude oil properties using artificial neural networks (ANNs). The predicted properties include specific gravity, kinematic viscosity, total acid number, micro carbon content, and production of light and heavy naphtha, Kero, and distillate in oil refineries. The 107 different (65 light oil and 42 heavy/medium oil samples) crude oil samples used in this study came from seven oil fields and reservoirs across Canada. In line with standard practice, we used 80% of the dataset for training the ANN models and used the remaining 20% of the crude oil samples to test the models. In the ANN analysis, the mean squared error (MSE) was used as the loss function in models, and the mean absolute prediction error (MAPE) was used as a reference to compare the performance of different neural networks constructed with different numbers of layers. This work demonstrates that FTIR spectroscopy is a promising technique that provides rapid and accurate estimates for the oil properties of interest to the industry. A comparison of the values predicted by the validated ANN models and their corresponding measured (actual) values showed excellent prediction with the acceptable range of error (below 15%) aimed for by our industry partner for all properties except viscosity, for which building models based on the natural logarithmic values of measured viscosities significantly improved the results.     

Determining Frying Oil Degradation by Near Infrared Spectroscopy Using Chemometric Techniques

A rapid method for the determination of some important physicochemical properties in frying oils has been developed. Partial least square regression (PLS) calibration models were applied to the physicochemical parameters and near infrared spectroscopy (NIR) spectral data. PLS regression was used to find the NIR region and the data pre-processing method that give the best prediction of the chemical parameters. Calibration and validation were appropriated by leave one out cross validation and test set validation techniques for predicting free fatty acids (FFA), total polar materials (cTPM; measured by chromatographic method and iTPM measured by an instrumental method), viscosity and smoke point of the frying oil samples. For PLS models using the cross validation techniques, the best correlations (r) between NIR predicted data and the standard method data for iTPM in oils were 93.79 and root mean square error of prediction (RMSEP) values were 5.53. For PLS models using the test set validation techniques, the best correlations (r) between NIR predicted data and standard method data for FFA, cTPM, viscosity and smoke point in oils were 92.58, 94.61, 81.95 and 84.07 and RMSEP values were 0.121, 3.96, 22.30 and 8.74, respectively. In conclusion, NIR technique with chemometric analysis was found very effective in predicting frying oil quality changes.     

Determination of Sodium Fatty Acid in Soap Formulation Using Fourier Transform Infrared (FTIR) Spectroscopy and Multivariate Calibrations

Fourier Transform Infrared (FTIR) spectroscopy using an attenuated total reflectance (ATR) accessory has been investigated as a method for the determination of sodium-fatty acid (sodium-FA) in soap formulations. Multivariate calibrations namely partial least squares regression (PLS) and principle component regression (PCR) were developed for the prediction of sodium-FA using spectral ranges on the basis of relevant IR absorption bands related to sodium-FA. The sodium-FA content in soap formulations was predicted accurately at wavenumbers of 1,570–1,550 cm⁻¹, which is specific for RCOO⁻ Na⁺ vibration. The PLS method was found to be a consistently better predictor when both PLS and principal component regression (PCR) analyses were used for quantification of sodium-FA. Furthermore, FTIR spectroscopy can be an alternative technique to American oil Chemist Society methods which use a titrimetric technique because FTIR offers rapid, easy sample preparation and is friendly to the environment.     

Quantitative-structure-effect relationship for some technical nonionic surfactants

The detergency effect has been examined for a series of technical nonionic surfactants with the use of statistical experimental designs and revealed a plateau in each of the response surfaces obtained. The surfactant concentrations and washing temperatures, needed to reach the edge of each detergency effect plateau, were also determined. These conditions, which define the edge of the plateau, could be well modeled from the physicochemical properties of the surfactants with the use of partial least squares of latent structures. It was also possible to point out the importance of the different physicochemical properties. If an experimental design has been utilized, the detergency effect of a nonionic surfactant can be modeled from multiple linear regression as a function of surfactant concentration, washing time, and washing temperature. We have shown how these regression coefficients can be modeled from the physicochemical properties of the surfactants. Partial least squares of latent structures were used to estimate these models as well. We also demonstrated how these models can be used to predict the regression coefficients of a surfactant not included in the model estimations. The resultant regression coefficients can then be used to predict the detergency effects of this surfactant at different variable settings. The detergency effects thus obtained are in good agreement with measured data acquired under corresponding conditions.     

A new method for determining gossypol in cottonseed oil by FTIR spectroscopy

A new method was developed to determine the gossypol content in cottonseed oil using FTIR spectroscopy with a NaCl transmission cell. The wavelengths used were selected by spiking clean cottonseed oil to gossypol concentrations of 0–5% and noting the regions of maximal absorbance. Transmittance values from the wavelength regions 3600–2520 and 1900–800 cm⁻¹ and a partial least squares (PLS) method were used to derive FTIR spectroscopic calibration models for crude cottonseed, semirefined cottonseed, and gossypol-spiked cottonseed oils. The coefficients of determination (R ²) for the models were computed by comparing the results from the FTIR spectroscopy against those obtained by AOCS method Ba 8-78. The R ² were 0.9511, 0.9116, and 0.9363 for crude cottonseed, semirefined cottonseed, and gossypol-spiked cottonseed oils, respectively. The SE of calibration were 0.042, 0.009, and 0.060, respectively. The calibration models were cross-validated within the same set of oil samples. The SD of the difference for repeatability and accuracy of the FTIR method were better than those for the chemical method. With its speed (ca. 2 min) and ease of data manipulation, FTIR spectroscopy is a useful alternative to standard wet chemical methods for rapid and routine determination of gossypol in process and/or quality control for cottonseed oil.     

An architecture of deep learning in QSPR modeling for the prediction of critical properties using molecular signatures

Deep learning rapidly promotes many fields with successful stories in natural language processing. An architecture of deep neural network (DNN) combining tree-structured long short-term memory (Tree-LSTM) network and back-propagation neural network (BPNN) is developed for predicting physical properties. Inspired by the natural language processing in artificial intelligence, we first developed a strategy for data preparation including encoding molecules with canonical molecular signatures and vectorizing bond-substrings by an embedding algorithm. Then, the dynamic neural network named Tree-LSTM is employed to depict molecular tree data-structures while the BPNN is used to correlate properties. To evaluate the performance of proposed DNN, the critical properties of nearly 1,800 compounds are employed for training and testing the DNN models. As compared with classical group contribution methods, it can be demonstrated that the learned DNN models are able to provide more accurate prediction and cover more diverse molecular structures without considering frequencies of substructures.     

Multiple input self-organizing-map ResNet model for optimization of petroleum refinery conversion units

This work introduces a deep-learning network, i.e., multi-input self-organizing-map ResNet (MISR), for modeling refining units comprised of two reactors and a separation train. The model is comprised of self-organizing-map and the neural network parts. The self-organizing-map part maps the input data into multiple two-dimensional planes and sends them to the neural network part. In the neural network part, residual blocks enhance the convergence and accuracy, ensuring that the structure will not be overfitted easily. Development of the MISR model of hydrocracking unit also benefits from the utilization of prior knowledge of the importance of the input variables for predicting properties of the products. The results show that the proposed MISR structure predicts more accurately the product yields and properties than the previously introduced self-organizing-map convolutional neural network model, thus leading to more accurate optimization of the hydrocracker operation. Moreover, the MISR model has smoother error convergence than the previous model. Optimal operating conditions have been determined via multi-round-particle-swarm and differential evolution algorithms. Numerical experiments show that the MISR model is suitable for modeling nonlinear conversion units which are often encountered in refining and petrochemical plants.     

MemDis: Predicting Disordered Regions in Transmembrane Proteins

Transmembrane proteins (TMPs) play important roles in cells, ranging from transport processes and cell adhesion to communication. Many of these functions are mediated by intrinsically disordered regions (IDRs), flexible protein segments without a well-defined structure. Although a variety of prediction methods are available for predicting IDRs, their accuracy is very limited on TMPs due to their special physico-chemical properties. We prepared a dataset containing membrane proteins exclusively, using X-ray crystallography data. MemDis is a novel prediction method, utilizing convolutional neural network and long short-term memory networks for predicting disordered regions in TMPs. In addition to attributes commonly used in IDR predictors, we defined several TMP specific features to enhance the accuracy of our method further. MemDis achieved the highest prediction accuracy on TMP-specific dataset among other popular IDR prediction methods.     

Multivariate model for the prediction of total phenolic acids in crude extracts of polyphenols from canola and rapessed meals: A preliminary study

The feasibility of using UV spectrophotometry to develop multivariate models for prediction of total phenolic acids content in crude polyphenol extracts from defatted canola and rapeseed meals was investigated. The polyphenols were extracted from the meals with methanol/acetone/water (7∶7∶6, by vol). Partial least squares regression was used to correlate the spectral data of the crude polyphenols in methanol between 320 and 355 nm with the total phenolic acid content in canola and rapeseed meals. The Folin-Denis assay was used to provide reference data for creating the model. The predictive ability of the model is good, as indicated by the RPD value (the ratio of the SD of data to the standard error of calibration) of 3.84.     

Environmentally Friendly Determination of Quality Parameters of Biodiesel/Diesel Blends Using Fourier Transform Infrared Spectra

Multivariate calibration models based on data from mid‐infrared spectroscopy of biodiesel/diesel blends were obtained. The blends were prepared from diesel oil and esters of soybean oil, waste cooking oil, and hydrogenated vegetable oil in proportions ranging from 0 to 100 % biodiesel. The results showed that the multivariate regression models with interval partial least squares (iPLS), backward interval partial least squares (biPLS), and synergy interval partial least squares (siPLS) were able to determine the fractions of the infrared spectrum that contain the relevant information for estimating the values of physicochemical properties, flash point, specific gravity, and cetane number, which are used in quality control of the blends. In the best models, the values of determination coefficients were greater than 0.9500, proving their efficiency as an alternative to traditional analytical methods.     

Soft sensor for continuous product quality estimation (in crude distillation unit)

Due to the strict norm requirements of keeping products in crude refining units within specifications, laboratory testing and quality control of the products are necessary. Given this reason, virtual soft sensor for continuous quality estimation of light naphtha as the crude distillation unit (CDU) product was developed. Experimental data included available continuous measurements of CDU process streams (temperatures, pressures and flowrate) and laboratory analyses undertaken twice a day. The results are soft sensor models for light naphtha vapor pressure (RVP) estimation.Soft sensor models have been developed conducting multiple linear regression analysis and using neural network-based models such as LNN, MLP and RBF. Considering statistical and sensitivity analysis, the best results for both oils were obtained with MLP and RBF neural networks. The results show possible application of the soft sensor models for estimating light naphtha RVP as an alternative for laboratory testing.     

Multivariate model for the prediction of soluble condensed tannins in crude extracts of polyphenols from canola and rapeseed hulls

The feasibility of using ultraviolet spectrophotometry to develop multivariate models for prediction of soluble condensed tannins (SCT) content in crude polyphenols extracts from canola and rapeseed hulls was investigated. The polyphenols were extracted from hulls using 70% (vol/vol) aqueous acetone. Partial least squares regression was used to correlate the spectral data of the crude polyphenols in methanol between 265–295 nm with the SCT content in hulls. Both the proanthocyanidin (P) and the vanillin (V) assays were used to provide reference data for creating the models. The predictive ability of the models is good, as indicated by the RPD values [the ratio of the standard deviation of data to the standard error of calibration (SEC) of above 5. Additionally, the SEC values suggest that P is superior to V in predicting the SCT content of hulls using this method.     

Development of calibration models for quality control in the production of ethylene/propylene copolymers by FTIR spectroscopy,multivariate statistical tools,and artificial neural networks

Principal component regression (PCR), partial least squares (PLS), StepWise ordinary least squares regression (OLS), and back‐propagation artificial neural network (BP‐ANN) are applied here for the determination of the propylene concentration of a set of 83 production samples of ethylene–propylene copolymers from their infrared spectra. The set of available samples was split into (a) a training set, for models calculation; (b) a test set, for selecting the correct number of latent variables in PCR and PLS and the end point of the training phase of BP‐ANN; (c) a production set, for evaluating the predictive ability of the models. The predictive ability of the models is thus evaluated by genuine predictions. The model obtained by StepWise OLS turned out to be the best one, both in fitting and prediction. The study of the breakdown number of samples to be included in the training set showed that at least 52 experiments are necessary to build a reliable and predictive calibration model. It can be concluded that FTIR spectroscopy and OLS can be properly employed for monitoring the synthesis or the final product of ethylene–propylene copolymers, by predicting the concentration of propylene directly along the process line. © 2008 Wiley Periodicals, Inc. J Appl Polym Sci, 2008     

Comparison of fatty acid profiles and mid‐infrared spectral data for classification of olive oils

The composition of olive oils may vary depending on environmental and technological factors. Fatty acid profiles and Fourier‐transform infrared (FT‐IR) spectroscopy data in combination with chemometric methods were used to classify extra‐virgin olive oils according to geographical origin and harvest year. Oils were obtained from 30 different areas of northern and southern parts of the Aegean Region of Turkey for two consecutive harvest years. Fatty acid composition data analyzed with principal component analysis was more successful in distinguishing northern olive oil samples from southern samples compared to spectral data. Both methods have the ability to differentiate olive oil samples with respect to harvest year. Partial least squares (PLS) analysis was also applied to detect a correlation between fatty acid profile and spectral data. Correlation coefficients (R²) of a calibration set for stearic, oleic, linoleic, arachidic and linolenic acids were determined as 0.83, 0.97, 0.97, 0.83 and 0.69, respectively. Fatty acid profiles were very effective in classification of oils with respect to geographic origin and harvest year. On the other hand, FT‐IR spectra in combination with PLS could be a useful and rapid tool for the determination of some of the fatty acids of olive oils.     

死端微滤酵母悬浮液比阻的预测

本文首先设计了三因素四水平的正交实验表作为建模样本,其次利用人工神经网络方法和多元线性回归方法分别建立了基于操作条件（压力△P=0.04-0.12 MPa,浓度C = 0.3-2.0 g.L^-1,温度T = 20-40℃）的比阻预测模型,以期用于死端微滤过程操作条件的优化,最后以检验样本的相对误差作为衡量指标,分别采用BP人工神经网络方法和多元线性回归方法对死端微滤过滤酵母悬浮液时的比阻进行了预测。研究结果表明：(1) 在本实验范围内,BP人工神经网络模型的最佳拓朴结构为3-7-1,隐层神经元个数为7,学习速率为0.05,学习函数为traingdx, 传递函数为Logsig;用多元线性回归方法得到的比阻与操作条件之间的数学关系式为1.639883+44.2 +0.86217 -0.0607 ; (2)利用BP人工神经网络和多元线性回归方法预测死端微滤比阻的平均相对误差分别为3.55％和5.16％.由此可见,这两种方法都可用于死端微滤比阻预测,并且前者优于后者。     

The Use of FTIR Spectroscopy and Chemometrics for Rapid Authentication of Extra Virgin Olive Oil

The authenticity of high value edible fats and oils including extra virgin olive oil (EVOO) is an emerging issue, currently. The potential employment of Fourier transform infrared (FTIR) spectroscopy in combination with chemometrics of multivariate calibration and discriminant analysis has been exploited for rapid authentication of EVOO from canola oil (Ca‐O). The optimization of two calibration models of partial least square (PLS) and principle component regression was performed in order to quantify the level of Ca‐O in EVOO. The chemometrics of discriminant analysis (DA) was used for making the classification between pure EVOO and EVOO adulterated with Ca‐O. The individual oils and their blends were scanned on good contact with ZnSe crystals in horizontal attenuated total reflectance, as a sampling technique. The wavenumbers of 3,028–2,985 and 1,200–987 cm^?1 were used for quantification and classification of EVOO adulterated with Ca‐O. The results showed that PLS with normal FTIR spectra was well suited for quantitative analysis of Ca‐O with a value of the coefficient of determination (R²) > 0.99. The error, expressed as root mean square error of calibration obtained was relatively low, i.e. 0.108 % (v/v). DA can make the classification between pure EVOO and that adulterated with Ca‐O with one misclassified reported.     

FTIR Spectroscopy and Chemometric Class Modeling Techniques for Authentication of Chinese Sesame Oil

This investigation was aimed at developing a rapid analysis method for authentication of Chinese sesame oils by FTIR spectrometry and chemometrics. Ninety-five sesame oil samples were collected from the six main producing areas of China to include most if not all of the significant spectral variations likely to be encountered in future authentic materials. Two class modeling techniques, the soft independent modeling of class analogy (SIMCA) and the partial least squares class model (PLSCM) were investigated and the data preprocessing techniques including smoothing, derivative and standard normal variate (SNV) tests were performed to improve the classification performance. It was demonstrated that SIMCA and PLSCM can detect various adulterated sesame oils doped with 3% or more (w/w) of other cheaper oils, including rapeseed, soybean, palm and peanut oils. First derivative, second derivative and SNV tests significantly enhanced the class models by reducing baseline and background shifts. Smoothing of raw spectra led to inferior identification performance and proved itself to be unsuitable because some of the detailed frequency details were lost during smoothing. The best model performance was obtained with second derivative spectra by SIMCA (sensitivity 0.905 and specificity 0.944) and PLSCM (sensitivity 0.952 and specificity 0.937). Although it is difficult to perform an exhaustive sampling of all types of pure sesame oils and potential adulterations, PLS and SIMCA combined with FTIR spectrometry can detect most of current adulterations of sesame oils on the Chinese market.     

Maximum solid concentrations of coal water slurries predicted by neural network models

The nonlinear back-propagation (BP) neural network models were developed to predict the maximum solid concentration of coal water slurry (CWS) which is a substitute for oil fuel, based on physicochemical properties of 37 typical Chinese coals. The Levenberg-Marquardt algorithm was used to train five BP neural network models with different input factors. The data pretreatment method, learning rate and hidden neuron number were optimized by training models. It is found that the Hardgrove grindability index (HGI), moisture and coalification degree of parent coal are 3 indispensable factors for the prediction of CWS maximum solid concentration. Each BP neural network model gives a more accurate prediction result than the traditional polynomial regression equation. The BP neural network model with 3 input factors of HGI, moisture and oxygen/carbon ratio gives the smallest mean absolute error of 0.40%, which is much lower than that of 1.15% given by the traditional polynomial regression equation.