Optimal estimation of physical properties of the products of an atmospheric distillation column using support vector regression

doi:10.1016/j.compchemeng.2019.106711

Computers & Chemical Engineering

Volume 134, 4 March 2020, 106711

https://doi.org/10.1016/j.compchemeng.2019.106711 Get rights and content

Abstract

Atmospheric distillation column is one of the most important units in an oil refinery where crude oil is fractioned into its more valuable constituents. Almost all of the state-of-the art online equipment has a time lag to complete the physical property analysis in real time due to complexity of the analyses. Therefore, estimation of the physical properties from online plant data with a soft sensor has significant benefits. In this paper, we estimate the physical properties of the hydrocarbon products of an atmospheric distillation column by support vector regression using Linear, Polynomial and Gaussian Radial Basis Function kernels and SVR parameters are optimized by using a variety of algorithms including genetic algorithm, grid search and non-linear programming. The optimization-based data analytics approach is shown to produce superior results compared to linear regression, the mean testing error of estimation is improved by 5% with SVR 4.01 °C to 3.8 °C.

Introduction

Crude Distillation Unit (CDU) is the heart of a petroleum refinery where crude oil is separated into its naturally occurring fractions. Atmospheric Distillation Column (ADC) is the most complex unit in the CDU which operates at atmospheric pressure (Gary et al., 2010; Leffler, 1985). The operation parameters of an ADC affect the amounts and physical properties of its products. Dynamically changing demands and prices for end products forces the planners to frequently update the optimal operating parameters and maximize the amount of a particular product while keeping the products within their specified limits. This gives rise to the need for online monitoring of the properties.

The online monitoring of temperature, pressure and flowrate within the ADC is possible with the measurement devices connected to Distributed Control Systems (DCS). However online monitoring of hydrocarbon properties is only possible by installing online analyzers which are very complex, hard-to-maintain and expensive. Therefore, chemical composition and physical properties are monitored by taking samples periodically from the ADC and analyzing the samples in a laboratory with appropriate equipment which takes up considerable amount of time.

The lack of online measurement of hydrocarbon properties can be complemented by online estimation methods. An estimation function of temperature and pressure as input variables can be fitted to laboratory analysis data of the physical properties. The ADC in İzmit Refinery of Turkish Petroleum Refineries Corporation which is the subject of this study has employed linear regression for the online estimation of the critical properties of hydrocarbon products like the 95% boiling point temperature of heavy diesel. This prediction embodies a fixed function of heavy diesel tray temperature and flash zone temperature and pressure which is then fitted to laboratory data by a gain and a bias. There have been studies on rigorous modeling of ADCs. This method is highly complex and embodies a comprehensive physical properties library, refers to the laws of thermodynamics and is hard to fit to a real operating unit. Though many of the hydrocarbon properties can be gathered from the simulation results of a rigorous model, the rigorous model still needs periodic maintenance as it sways away from real system in time.

Recent studies have successfully employed Machine Learning in modeling chemical processes in petroleum refining. Artificial Neural Network (ANN) has been a popular choice in this area but has suffered from over-fitting. ANN like any regression method that embodies Empirical Risk Minimization, minimizes the estimation error regardless of the model complexity. This in turn leads to poor performance in generalization of unseen and unvalidated testing data.

More recent studies have utilized Support Vector Machines (SVM) in regression problems. SVMs, in contrast to ANN embodies Structural Risk Minimization that aims to generate a flatter and less complex function. By implementing a new loss function, Support Vector Regression (SVR) chooses the flattest path where the error is kept within a predefined width for insensitive region and a predefined cost factor handles outlier data. Moreover, by employing kernel functions, SVR can be trained to generate a nonlinear estimation function. Lastly, the user-defined parameters of SVR can be optimized via cross validation embedded in global optimizers like Genetic Algorithm or simpler solvers like Grid Search, to maximize generalization performance.

Section snippets

Literature survey

Since Crude Distillation Units are complex and energy intensive units, safe and optimum operation of the unit has a great significance in oil refining. This leads to a need in proper monitoring of the operation for process engineers and a proper estimation of the operation dynamics for planning engineers. Both objectives prove to be troublesome as the physical properties can be measured by taking a sample from the stream periodically and then analyzing these sample in a laboratory with

Data normalization

The rate of convergence in training the support vector model is greatly affected by the eigenvalues of the Hessian. We can improve this rate by decreasing the difference between the smallest and largest eigenvalue. Therefore, this study strongly recommends and applies normalization for input data to improve the rate of convergence. Some kernel functions can only be defined in restricted intervals which already require data normalization. For unrestricted kernels, this study scales data between

The data

The atmospheric distillation column (ADC) subject to this study is an actual operating unit in İzmit Refinery of Turkish Petroleum Refineries Corporation. The ADC together with its pump-arounds, side-strippers and condenser drum is depicted in Fig. 4.

The crude distillation unit as a total was designed for processing crude oil from Kirkuk, Iraq, but currently processes different kinds of crude oils. A pre-flash column separates LPG and Light Naphtha and sends the remaining stream to the ADC.

Optimization results

The use of $β_{i} = α_{i} - α_{i}^{*}$ , and $β_{i}^{*} = α_{i} + α_{i}^{*}$ as decision variables in QP greatly improved the rate of convergence. The CPU time of SVR training with α_i and $α_{i}^{*}$ as decision variables is approximately 1.9 seconds while the CPU time of SVR training with the simplified variables was reduced by 80% to a range between 350 and 400 ms even though the problem size is the same. This improvement is achieved by halving the number of decision variables in the second-degree part of the cost function, which creates a

Discussion of optimization results

The results show that different optimization solvers have found different local optimal points with similar mean absolute errors and standard deviations for the same kernels, meaning that there are many similar local optima. Genetic method has triumphed in most of the criteria. SVR with Gaussian RBF kernel optimized by Grid Search has best performance in testing. Grid Search method has highest computational.

It should be noted that any solution linear regression also showed quite bit good

Conclusions

In this paper, Support Vector Regression (SVR) was employed for estimating 95% Boiling Point Temperature property of Heavy Diesel product of a real operating Atmospheric Distillation Column (ADC) in İzmit Refinery of Turkish Petroleum Refineries Corporation. Linear, Polynomial and Gaussian Radial Basis Function (Gaussian RBF) kernels were tested and the SVR parameters were optimized by embedding k-fold cross validation in an optimizer such as Genetic Algorithm (GA) and Grid Search (GS). The

CRediT authorship contribution statement

Ahmet Can Serfidan: Investigation, Data curation, Software, Validation, Visualization, Writing - review & editing. Firat Uzman: Conceptualization, Data curation, Formal analysis, Validation, Writing - original draft. Metin Türkay: Conceptualization, Formal analysis, Investigation, Methodology, Supervision, Writing - review & editing.

Declaration of Competing Interest

None.

References (35)

S.V. Inamdar et al.
Multi-objective optimization of an industrial crude distillation unit using the elitist non-dominated sorting genetic algorithm
Chem. Eng. Res. Des.
(2004)
V. Kumar
A crude distillation unit model suitable for online applications
Fuel Process. Technol.
(2001)
L. Li et al.
Modeling of isomerization of C8 aromatics by online least squares support vector machine
Chin. J. Chem. Eng.
(2009)
R.K. More
Optimization of crude distillation system using aspen plus: effect of binary feed selection on grass-root design
Chem. Eng. Res. Des.
(2010)
G. Soave
Equilibrium constants from a modified Redlich–Kwong equation of state
Chem. Eng. Sci.
(1972)
W. Yan et al.
Soft sensing modeling based on support vector machine and Bayesian model selection
Comput. Chem. Eng.
(2004)
H. Yao et al.
Operational optimization of a simulated atmospheric distillation column using support vector regression models and information analysis
Chem. Eng. Res. Des.
(2012)
A. Aizerman et al.
Theoretical foundations of the potential function method in pattern recognition learning
Autom. Remote Control
(1964)
(2008)
K. Basak
On-line optimization of a crude distillation unit with constraints on product properties
Ind. Eng. Chem. Res.
(2002)

K.P. Bennett et al.

Robust linear programming discrimination of two linearly inseparable sets

Optimiz. Methods Softw.

(1992)

B.E. Boser et al.

A training algorithm for optimal margin classifiers

J.F. Boston et al.

A new class of solution methods for multicomponent, multistage separation processes

Can. J. Chem. Eng.

(1974)

C.-C. Chang et al.

LIBSVM: a library for support vector machines

ACM Trans. Intell. Syst. Technol.

(2011)

S.B. Chitralekha et al.

Application of support vector regression for developing soft sensors for nonlinear processes

Can. J. Chem. Eng.

(2010)

C. Cortes et al.

Support-vector networks

Mach. Learn.

(1995)

R. Fletcher et al.

Nonlinear programming and nonsmooth optimization by successive linear programming

Math. Program.

(1989)

Cited by (16)

Machine learning-based prediction and optimization of green hydrogen production technologies from water industries for a circular economy
2023, Desalination
Currently, there exists a significant number of green hydrogen production (GHP) technologies based on scaling-up issues (SCUI). Optimal prediction and process optimization could be one of the most substantial SCUI of GHP. Machine learning (ML)-based prediction and optimization of GHP technologies from water industries for a circular economy (CRE) could be a plausible solution for these SCUI. We studied a detailed techno-economic and environmental feasibility study, which recommended proton exchange membrane (PEM) and dark fermentation (DF) as the most promising and environment-friendly technologies for GHP. Thus, the present investigation aims to apply different ML models to predict and optimize the GHP of DF and PEM technologies to solve the SCUI. The results revealed K-nearest neighbor and random forest are the best-fitted models to predict GHP for DF and PEM, correspondingly based on the regression co-efficient (R²), root mean squared error (RMSE) and mean absolute error (MEA). The permutation variable index (PVI) recommended that chemical oxygen demand (COD), butyrate, temperature, pH and acetate/butyrate ratio are the most influential process parameters in decreasing order for DF, while temperature, cell areas, cell pressure, cell voltage and catalysts loadings are the most effective process parameters for PEM in reducing order. The partial dependency analysis (PDA) demonstrated GHP increases with increasing COD values up to 10 mg/L, and the optimal temperature range in the DF process is between 25 and 30 °C. On the other hand, cell temperature up to 35 °C should be considered optimum for PEM, and 40–70 cm² cell areas could produce a significant GHP. In summary, the present study underscores the potential of machine learning (ML) and artificial intelligence (AI) as promising techniques for optimizing GHP, ultimately addressing scaling-up challenges in large-scale industrial GHP production and ensuring a sustainable hydrogen economy (HE).
Modeling the magnetic cooling efficiency of spinel ferrite magnetocaloric compounds for magnetic refrigeration application using hybrid intelligent computational methods
2022, Materials Today Communications
Spinel ferrite exhibits large magnetocaloric effect that promotes its candidature for solid state magnetic refrigeration technology. This method of cooling is promising and can conveniently replace the gas-compression method of refrigeration due to its energy saving capacity and ecological cleanliness. Among the available magnetocaloric materials, spinel ferrite metal oxide enjoys low production cost, non-toxicity, ease of preparation and non-oxidative. However, effective utilization of this magnetocaloric material for cooling technology requires determination of its magnetic cooling efficiency which is subjected to time consuming, intensive and laborious experimental procedures. This work employs the applied magnetic field, ionic radii and the concentrations of spinel ferrite constituents to develop a genetically hybridized support vector regression model (G-SVR) and extreme learning machine (ELM) intelligent model for determining magnetic cooling efficiency of spinel ferrite compounds. The developed G-SVR model outperforms ELM model for the testing set of spinel ferrite samples with percentage improvement of 60.57%, 19.69% and 64.73% using root mean square error, correlation coefficient and mean absolute error respectively, as performance measuring parameters. Influence of metal ions such as zinc, nickel and cadmium on magnetic cooling efficiency of various kinds of spinel ferrite compounds was investigated using the developed models. The presented intelligent approaches for spinel ferrite magnetic cooling efficiency determination in this work, with characteristic lower cost, quickness, effectiveness and precision would definitely facilitate and strengthen practical implementation of magnetic refrigeration technology for energy conservation and environmental friendliness.
A study on corrosion failure of the circulation line with valve openings in atmospheric tower based on CFD
2022, Engineering Failure Analysis
Citation Excerpt :
Atmospheric distillation tower is the main device for crude oil processing in petrochemical enterprises and plays a vital role in the refining system. The safety of atmospheric distillation towers has always been the focus of people's attention [1–4]. The water dissolved of corrosive components will damage the tower and other important components.
The perforation failure occurs frequently in the pipeline system of the atmospheric distillation tower. In this paper, the methods of technological process and computational fluid dynamic (CFD) were used to reveal the failure reason of the reducing pipe and provide theoretical guidance. The results of X-ray diffraction analysis shows that the corrosion products of circulation line reducing pipe is FeS and other corrosive substances. The corrosion mechanism was studied and the process was described. Besides, it was found that the valve in the circulation line was not fully opened in the early maintenance. The CFD result shows that the increase of the valve opening will aggravate the corrosion risk of the reducing pipe bottom wall.
Pruning Pareto optimal solutions for multi-objective portfolio asset management
2022, European Journal of Operational Research
Budget allocation problems in portfolio management are inherently multi-objective as they entail different types of assets of which performance metrics are not directly comparable. Existing asset management methods that either consolidate multiple goals to form a single objective (a priori) or populate a Pareto optimal set (a posteriori) may not be sufficient because a decision maker (DM) may not possess comprehensive knowledge of the problem domain. Moreover, current techniques often present a Pareto optimal set with too many options, making it counter-productive. In order to provide the DM with a diverse yet compact solution set, this paper proposes a three-step approach. In the first step, we employ different approximation functions to capture investment-performance relationships at the asset-type level. These simplified relationships are then used as inputs for the multi-objective optimisation model in the second step. In the final step, Pareto optimal solutions generated by a selected evolutionary algorithm are pruned by a clustering method. To measure the spread of representative solutions over the Pareto front, we present two novel indicators based on average Euclidean distance and cosine similarity between original Pareto solutions and representative solutions. Through numerical examples, we demonstrate that this approach can provide a set of representative solutions that maintain high integrity of the original Pareto front. We also put forward suggestions on choosing appropriate approximation functions, pruning methods, and indicators.
Machine learning modeling and analysis of biohydrogen production from wastewater by dark fermentation process
2022, Bioresource Technology
Dark fermentation process for simultaneous wastewater treatment and H₂ production is gaining attention. This study aimed to use machine learning (ML) procedures to model and analyze H₂ production from wastewater during dark fermentation. Different ML procedures were assessed based on the mean squared error (MSE) and determination coefficient (R²) to select the most robust models for modeling the process. The research showed that gradient boosting machine (GBM), support vector machine (SVM), random forest (RF) and AdaBoost were the most appropriate models, which were optimized by grid search and deeply analyzed by permutation variable importance (PVI) to identify the relative importance of process variables. All four models demonstrated promising performances in predicting H₂ production with high R² values (0.893, 0.885, 0.902 and 0.889) and small MSE values (0.015, 0.015, 0.016 and 0.015). Moreover, RF-PVI demonstrated that acetate, butyrate, acetate/butyrate, ethanol, Fe and Ni were of high importance in decreasing order.
A framework for transportation infrastructure cost prediction: a support vector regression approach
2022, Transportation Letters
Highway agencies and Departments of Transportation (DOTs) often face a challenge in estimating annual maintenance expenditures (AMEXs) to allocate available funds to various divisions in their jurisdictions. Few AMEX models were developed in the past, which are based on traditional statistical methods. This study introduces a support vector regression (SVR) approach to predict AMEX using data on the U.S. rural interstate highway pavement performance. Results show that the best SVR model is based on the Gaussian kernel and sequential minimal optimization algorithm. The developed SVR model can reliably predict the AMEX at the state level with a root mean squared error of 1.28. Entropy analysis results indicate the existence of mutual information (MI) between all the considered variables and the AMEX, with the highest MI value of about 79% between the AMEX and the number of lane miles with IRI values greater than 1.5 m/km.

View all citing articles on Scopus

View full text

Optimal estimation of physical properties of the products of an atmospheric distillation column using support vector regression

Abstract

Introduction

Section snippets

Literature survey

Data normalization

The data

Optimization results

Discussion of optimization results

Conclusions

CRediT authorship contribution statement

Declaration of Competing Interest

Chem. Eng. Res. Des.

Fuel Process. Technol.

Chin. J. Chem. Eng.

Chem. Eng. Res. Des.

Chem. Eng. Sci.

Comput. Chem. Eng.

Chem. Eng. Res. Des.

Theoretical foundations of the potential function method in pattern recognition learning

Autom. Remote Control

On-line optimization of a crude distillation unit with constraints on product properties

Ind. Eng. Chem. Res.

Robust linear programming discrimination of two linearly inseparable sets

Optimiz. Methods Softw.

A training algorithm for optimal margin classifiers

A new class of solution methods for multicomponent, multistage separation processes

Can. J. Chem. Eng.

LIBSVM: a library for support vector machines

ACM Trans. Intell. Syst. Technol.

Application of support vector regression for developing soft sensors for nonlinear processes

Can. J. Chem. Eng.

Support-vector networks

Mach. Learn.

Nonlinear programming and nonsmooth optimization by successive linear programming

Math. Program.