Deep chemometrics using one-dimensional convolutional neural networks for predicting crude oil properties from FTIR spectral data

The determination of physicochemical properties of crude oils is a very important and time-intensive process that needs elaborate laboratory procedures. Over the last few decades, several correlations have been developed to estimate these properties, but they have been very limited in their scope and range. In recent years, methods based on spectral data analysis have been shown to be very promising in characterizing petroleum crude. In this work, the physicochemical properties of crude oils using Fourier transform infrared (FTIR) spectrums are predicted. A total of 107 samples of FTIR spectral data consisting of 6840 wavenumbers is used. One dimensional convolutional neural networks (CNNs) were used employing FTIR spectral data as the one-dimensional input and Keras and TensorFlow were used for model building. The Root Mean Square Error decreased from 160 to around 60 for viscosity when compared to previous machine learning methods like partial least squares (PLS), principal component regression (PCR), and partial least squares regression with genetic algorithm (PLS-GA) on the same data. The important hyper-parameters of the CNN were optimized. In addition, a comparison of results obtained with different neural network architectures is presented. Some common preprocessing techniques were also tested on the spectral data to determine their impact on model performance. To increase interpretability, the intermediate neural network layers were analyzed to reveal what the convolutions represented, and sensitivity analysis was done to gather key insights about the wavenumbers that were the most important for prediction of the crude oil properties using the neural network.     

Prediction of thermodynamic properties of polyvinylpyrrolidone solutions by artificial neural networks coupled with genetic algorithm

In the present study, the artificial neural networks coupled with the genetic algorithm (ANN–GA) models were used to predict the thermodynamic properties of polyvinylpyrrolidone (PVP) solutions in water and ethanol at various temperatures, mass fractions, and molecular weights of polymer. The genetic algorithm (GA) was used to find the best weights and biases of the network and improve the performance of ANNs. The proposed model was composed of three input variables including the temperature of the solution, the mass fraction, and molecular weight of the polymer. Density, viscosity, and surface tension of PVP solutions with various molecular weights (10,000, 25,000, and 40,000) in water and ethanol have been measured in the temperature range 20–55°C and various mass fractions of polymer. The ANN–GA models were trained by the experimental datasets and the prediction of density, surface tension, and viscosity of PVP solutions was performed using these models. The predicted values were compared with the experimental ones and the mean absolute relative error was less than 0.5% for the density and surface tension and about 3% for the viscosity of solutions.     

Measurement of Whole Soybean Fatty Acids by Near Infrared Spectroscopy

Whole soybean fatty acid contents were measured by near infrared spectroscopy. Three calibration algorithms—partial least squares (PLS), artificial neural networks (ANN), and least squares support vector machines (LS-SVM)—were implemented. Three different validation strategies using independent sets and part of calibration samples as validation sets were created. There was a significant improvement of the prediction precision of all fatty acids measured on relative concentration of oil compared with previous literature using PLS (standard error of prediction of 0.85, 0.42, 1.64, 1.67, and 0.90% for palmitic, stearic, oleic, linoleic and linolenic acids respectively). ANN and LS-SVM methods performed significantly better than PLS for palmitic, oleic and linolenic acids. Calibration models developed on relative concentrations (% of oil) were compared to prediction models created on absolute fatty acid concentration (% of weight) and corrected to relative concentration by multiplying by the predicted oil content. While models were easier to develop in absolute concentration (higher coefficients of determination), the multiplication of errors with the total oil content model resulted in no net precision improvement.     

Experimental measurement and modeling of saturated reservoir oil viscosity

A novel mathematical-based approach is proposed to develop reliable models for prediction of saturated crude oil viscosity in a wide range of PVT properties. A new soft computing approach, namely least square support vector machine modeling optimized with coupled simulated annealing optimization technique, is proposed. Six models have been developed to predict saturated oil viscosity, which are designed in such a way that could predict saturated oil viscosity with every available PVT parameter. The constructed models are evaluated by carrying out extensive experimental saturated crude oil viscosity data from Iranian oil reservoirs, which were measured using a “Rolling Ball viscometer.” To evaluate the performance and accuracy of these models, statistical and graphical error analyses were used simultaneously. The obtained results demonstrated that the proposed models are more robust, reliable and efficient than existing techniques for prediction of saturated crude oil viscosity.     

Prediction of Bromate Formation Using Multi-Linear Regression and Artificial Neural Networks

The main objective of this study was to develop simple models for the prediction of bromate formation in ozonated bottled waters, using rapidly and practically measurable raw water quality and/or operational parameters. A total of 6 multi-linear regression (MLR) with or without principal component analysis (PCA) and 2 artificial neural networks (ANN) models with multilayer perceptron architecture were developed for the prediction of bromate formation. PCA was employed to better identify relations between variables and reduce the number of variables. Experimental data used in modeling was provided from the ozonation of samples from 5 groundwater sources at various applied ozone dose and contact time. MLR models#1 and #2 well-predicted bromate formation although correlations (i.e., the signs of regression constants) among pH (as input variable) and bromate concentrations did not agree with the chemistry. MLR model#6, containing practical input parameters that are measured on-line in full-scale treatment plants, adequately predicted bromate formation and agreed with the chemistry, although fewer input parameters were used compared to MLR#1 and #2. Although both of the ANN models exhibited high regression coefficients (R²) (0.97 for both) ANN#1 was found to provide better prediction of bromate formation based on mean square error (MSE) values. However, since ANN#2 included easily measurable input parameters it may be practically used by water companies employing ozonation. Results overall indicated that ANN models have stronger prediction capabilities of bromate formation than MLR models. ANN modeling appears to be a strong tool in situations where the relations between variables are non-linear, interactive and complex, as in the bromate formation by ozonation.     

Neuro–fuzzy modelling of spectroscopic data. Part B – Dye concentration prediction

An adaptive network–based fuzzy inference system, ANFIS, has been used for predicting dye concentrations using spectroscopic absorbance data in the visible region. The samples were two–component red/yellow dye solutions with a concentration range of 0–900 mg/l for the one component (red) while the concentration of the other component (yellow) was kept constant. The modelled system had two inputs (wavelength and absorbance) with the concentration values as output. Generalised bell–shaped membership functions were used for the inputs. The inference system used was a first–order Sugeno fuzzy model. The ANFIS models gave concentration prediction results with approximately the same standard error of prediction as artificial neural network (ANN) models. However, the ANFIS model building runs faster than in the case of ANN.     

Response Surface Methodology and Artificial Neural Networks for Optimization of Catalytic Esterification of Lactic Acid

Response surface methodology (RSM) and artificial neural network (ANN) models were employed to study the esterification of lactic acid and isoamyl alcohol. A carbon-based solid acid catalyst prepared by wet impregnation was used in the esterification reaction. Experimental characterization revealed its potential to serve as catalyst for the esterification reaction. The experiments were performed based on the design of experiments provided by RSM and ANN models. Both models were compared on the basis of prediction efficacies and deviation from actual data. The prediction data results demonstrated that the ANN model gave better prediction efficiency and lower prediction deviation than the RSM model. The ANN model provided a higher coefficient of determination and lower error values than the RSM model. Moreover, the catalyst exhibited a good stability and recyclability up to four reaction cycles.     

Viscosity prediction by computational method and artificial neural network approach: The case of six refrigerants

There are some computational models for fluids viscosity calculation. However, each of these models is reliable in confined density. In this comparative study two methods are evaluated for viscosity prediction in all range of density. We determine the effectiveness of each of the models and we demonstrate the strengths and weaknesses of them. Viscosity of the six refrigerants is calculated by some computational models based on Chapman⿿Enskog and Rainwater⿿Friend theories. Then a feed forward artificial neural network (ANN) with multilayer perceptrons is used to viscosity prediction and finally two methods (computational models and artificial neural network) are comparing. It is concluded that there is no opinion by computational methods to calculate viscosity from low to high density. The results show that prediction accuracy of computational models in low and moderate densities is good as ANN method. However artificial neural network has very good accuracy in high densities while computational method is defeated when the density is more than 8.     

A comparative study of a CRT colorimetric prediction model by neural networks and the models by conventional method

In order to produce desired colors on CRT screens, much work has been done on the problem of the CRT colorimetric prediction. However, it would take great pains to overcome the troubles such as the constant channel chromaticity, the gun or channel independence, and the screen background effect, etc., with the conventional prediction methods such as PLCC and PLVC models, etc. To solve such problems, we propose a completely different CRT colorimetric prediction model by using a set of Artificial Neural Networks (ANN), where a set of back‐propagation (BP) neural networks is used to perform a nonlinear conversion between RGB values and XYZ values. By comparing some typical conventional CRT colorimetric prediction models with our neural‐networks‐based model theoretically, the article indicates that our new model can overcome the troubles faced by the conventional models, and by experiment the article shows that our new model can yield a satisfactory prediction result. © 1999 John Wiley & Sons, Inc. Col Res Appl, 24, 45–51, 1999     

Estimating Oil and Protein Content of Sesame Seeds Using Image Processing and Artificial Neural Network

Image processing has many applications in different fields of agriculture. The present study aimed to use image processing techniques and artificial neural networks (ANN) to estimate oil and protein contents of sesame genotypes without the use of time-consuming and costly laboratory methods. The proposed method accurately estimates the parameters in sesame seeds without destructing the genetically valuable material. In this study, a set of 138 morphological features were extracted from the digital image of 125 sesame seed genotypes. A multilayer perceptron (MLP) ANN was then employed to estimate oil and protein contents and determine the relationship between estimated values and laboratory-measured values. The efficiency of this model was compared to radial bases function (RBF), extended RBF (ERBF), GRNN, M5-Rule, M5-Tree, support vector machine regression, and linear regression models. Results showed that MLP performed better in estimating qualitative parameters of seeds in the sesame germplasm. The model estimated oil content with an root mean square error (RMSE) of 2.13% (the accuracy of 97.87%) and an R² of 0.93. Protein content was estimated by an RMSE of 0.378% (the accuracy of 99.62%) and an R² of 0.96.     

Comparative Modeling and Analysis of the Mass Transfer Coefficient in a Turbulent Bed Contactor using Artificial Neural Network and Adaptive Neuro-Fuzzy Inference Systems

This study deals with predicting the gas film volumetric mass transfer coefficient (K_oga) in a turbulent bed contactor (TBC), using both artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) techniques. The networks have been trained and evaluated with the experimental data available in the literature. Input variables to the networks are process variables such as gas and liquid phase concentration, gas and liquid superficial velocities and also specific area of packings. The results obtained the ability of developed ANN and ANFIS for prediction of K_oga. Although it was observed that both ANN and ANFIS models provided a good statistical prediction in terms of coefficient of determination (R²), mean relative error (MRE) and root mean square error (RMSE), the accuracies of ANFIS predictions were better than those of ANN predictions.     

SOFT COMPUTING METHOD FOR PREDICTION OF CO2 CORROSION IN FLOW LINES BASED ON NEURAL NETWORK APPROACH

An important aspect of corrosion prediction for oil/gas wells and pipelines is to obtain a realistic estimate of the corrosion rate. Corrosion rate prediction involves developing a predictive model that utilizes commonly available operational parameters, existing lab/field data, and theoretical models to obtain realistic assessments of corrosion rates. This study presents a new model to predict corrosion rates by using artificial neural network (ANN) systems. The values of pH, velocity, temperature, and partial pressure of the CO₂ are input variables of the network and the rate of corrosion has been set as the network output. Among the 718 data sets, 503 of the data were implemented to find the best ANN structure, and 108 of the data that were not used in the development of the model were used to examine the reliability of this method. Statistical error analysis was used to evaluate the performance and the accuracy of the ANN system for predicting the rate of corrosion. It is shown that the predictions of this method are in acceptable agreement with experimental data, indicating the capability of the ANN for prediction of CO₂ corrosion rate in production flow lines.     

基于多传感器技术的原油含水率预测模型研究

通过多传感器技术对原油含水率测量有影响的多个参量进行测定,提出基于多元非线性回归和神经网络融合处理两种方法建立原油含水率预测模型,并采用分段建模的方法分别进行改进.评价结果表明:神经网络模型预测效果优于多元非线性回归模型,原油含水率分段预测模型效果优于统一模型.尤其是改进的神经网络分段预测模型具有网络结构简化、收敛速度快,泛化能力强的特点,取得很好的拟合精度和预测效果.     

ANN方法分析预测聚丙烯材料的力学性能

利用Ｂ－Ｐ人工神经网络（ＡＪＮＮ）对聚丙烯（ＰＰ）的力学性能进行了分析和预测。首先将ＰＰ材料接纯ＰＰ、共混和增韧及填充和增强ＰＰ等进行分类,并根据这些数据的特点建立Ｂ－Ｐ网络,然后用各类ＰＰ材料的组成和力学性能数据对网络进行学习训练,最后用“未知样品”的数据对网络进行验证。结果表明,所建立的网络能反映ＰＰ的力学性能特性,预测有一定的准确性,但不同类别的材料预测准确性不同。     

Prediction of air pollutants by using an artificial neural network

The purpose of this study is to predict the amount of primary air pollution substances in Seoul, Korea. An artificial neural network (ANN) was used as a prediction method. The ANN with three layers is learned with past data, and the concentrations of air pollutants are predicted based on the pre-learned weights. The error back propagation method that has a powerful application to various fields was adopted as the learning rule. The concentrations of air pollutants from one to six hours in the future were predicted with the ANN. To verify the performance of the prediction method used in the present study, the predicted concentrations of air pollutants were compared with the measured data. From the comparison, it was found that the prediction method based on the ANN gives an acceptable accuracy for the limited prediction horizon.     

Dynamic crude oil fouling prediction in industrial preheaters using optimized ANN based moving window technique

The objective of this paper is to develop and validate a reliable, efficient and robust artificial neural network (ANN) model for online monitoring and prediction of crude oil fouling behavior for industrial shell and tube heat exchangers. To explore the complex dynamics of fouling, a new modeling strategy based on moving-window neural network approach is proposed. The essential character of this modeling approach is online updating of the ANN model whenever a new data block is available, so that it can effectively capture the slowly changing of process dynamics. The results of these models have been compared with appropriate sets of experimental data. The mean relative errors (MRE) of training and prediction subsets were about 6.61% and 8.06%, respectively. Since the data extraction in the refinery was performed every 2 h, the modeling approach led to an MRE of about 8% for fouling rate prediction of the next 50 h.     

Process Modelling of Combined Degumming and Bleaching in Palm Oil Refining Using Artificial Neural Network

Combined degumming and bleaching is the first stage of processing in a modern physical refining plant. In the current practice, the amount of phosphoric acid (degumming agent) and bleaching earth (bleaching agent) added during this process is usually fixed within a certain range. There is no system that can estimate the right amount of chemicals to be added in accordance with the quality of crude palm oil (CPO) used. The use of an Artificial Neural Network (ANN) for an improved operating procedure was explored in this process. A feed forward neural network was designed using a back-propagation training algorithm. The optimum network for the response factor of phosphoric acid and bleaching earth dosages prediction were selected from topologies with the smallest validation error. Comparisons of ANN predicted results with industrial practice were made. It is proven in this study that ANN can be effectively used to determine the phosphoric acid and bleaching earth dosages for the combined degumming and bleaching process. In fact, ANN gives much more precise required dosages depending on the quality of the CPO used as feedstock. Therefore, the combined degumming and bleaching process can be further optimised with savings in cost and time through the use of ANN.     

Optimization of supercritical carbon dioxide extraction of Passiflora seed oil

This study investigates extraction of Passiflora seed oil by using supercritical carbon dioxide. Artificial neural network (ANN) and response surface methodology (RSM) were applied for modeling and the prediction of the oil extraction yield. Moreover, process optimization were carried out by using both methods to predict the best operating conditions, which resulted in the maximum extraction yield of the Passiflora seed oil. The maximum extraction yield of Passiflora seed oil was estimated by ANN to be 26.55% under the operational conditions of temperature 56.5 °C, pressure 23.3 MPa, and the extraction time 3.72 h; whereas the optimum oil extraction yield was 25.76% applying the operational circumstances of temperature 55.9 °C, pressure 25.8 MPa, and the extraction time 3.95 h by RSM method. In addition, mean-squared-error (MSE) and relative error methods were utilized to compare the predicted values of the oil extraction yield obtained from both models with the experimental data. The results of the comparison reveal the superiority of ANN model compared to RSM model.     

Artificial and Hybrid Fuzzy Linear Neural Network‐Based Estimation of Seed Oil Content of Safflower

Inexpensive and rapid methods for measurement of seed oil content by near infrared reflectance spectroscopy (NIRS) are useful for developing new oil seed cultivars. Adopting default multiple linear regression (MLR), the predictions of safflower oil content were made by 20–140 samples using a Perten Inframatic 8620 NIR spectrometer. Although the obtained interpolation results of MLR had desired accuracy, the extrapolation was extremely poor. The extrapolation determination coefficient (R²) and standard error (SE) of cross validation for MLR models were 0.63–0.78 and 3.71–4.44, respectively. In order to overcome the accuracy limitation of linear MLR models, a common suggestion is to use a nonlinear artificial neural network (ANN); however, it needs a large number of data to yield significant accurate results. We developed a novel robust hybrid fuzzy linear neural (HFLN) network to capture simultaneously linear and nonlinear patterns of data with a limited number of safflower samples. Empirical extrapolation results showed that the HFLN had higher R² (=0.85) and lower SE (=1.83) compared to those obtained by MLR and ANN models. It is concluded that hybrid methodologies could be used to construct efficient and appropriate models for estimation of seed oil content set up on NIR system.     

Multivariate determination of cloud point in palm oil using partial least squares and principal component regression based on FTIR spectroscopy

A rapid FTIR spectroscopic method was developed for quantitative determination of the cloud point (CP) in palm oil samples. Calibration samples were prepared by blending randomized amounts of palm olein and palm stearin to produce a wide range of CP values ranging between 8.3 and 47.9°C. Both partial least squares (PLS) and principal component regression (PCR) calibration models for predicting CP were developed by using the FTIR spectral regions from 3000 to 2800 and 1800 to 1600 cm⁻¹. The prediction capabilities of these calibration models were evaluated by comparing their standard errors of prediction (SEP) in an independent prediction set consisting of 14 palm oil samples. The optimal model based on PLS in the spectral range 1800-1600 cm⁻¹ produced lower SEP values (2.03°C) than those found with the PCR (2.31°C) method. FTIR in conjunction with PLS and PCR models was found to be a useful analytical tool for simple and rapid quantitative determination of CP in palm oil.