Elsevier

Fuel

Volume 264, 15 March 2020, 116616
Fuel

Full Length Article
Estimation of CO2 equilibrium absorption in aqueous solutions of commonly used amines using different computational schemes

https://doi.org/10.1016/j.fuel.2019.116616Get rights and content

Highlights

  • Computational models are presented to prognosticate CO2 equilibrium absorption capacity of various amine solutions.

  • GA-ANFIS, PSO-ANFIS, CSA-LSSVM and RBF neural networks are employed for modeling purpose.

  • Among many different proposed models, LSSVM exhibits results that are more promising.

  • The proposed LSSVM model has better accuracy compared to the other models.

Abstract

In absorptive removal of CO2 by aqueous alkanolamine solvent, as the most prevalent CO2 capture technique, equilibrium absorption capacity of CO2 is a significant parameter for assessing the efficiency of absorption systems. In this study, unique computational models are presented to estimate CO2 solubility in commonly used amines. A series of models, including genetic algorithm-adaptive neuro fuzzy inference system (GA-ANFIS), particle swarm optimization ANFIS (PSO-ANFIS), coupled simulated annealing-least squares support vector machine (CSA-LSSVM) and radial basis function (RBF) neural networks were developed to estimate CO2 equilibrium absorption capacity in twelve aqueous amine solutions. The model inputs comprise of CO2 partial pressure, temperature, amine concentration in aqueous solution, molecular weight, hydrogen bond donor/acceptor count, rotatable bond count and complexity of the amines. The obtained results affirm that among proposed models, LSSVM exhibits more promising results with an excellent compatibility with experimental values. In detail, both mean square errors and average regression coefficient (R2) of LSSVM model are 0.02 and 0.9338, respectively. Moreover, it is confirmed that the proposed LSSVM model has better accuracy compared to the other models.

Introduction

Nowdays, reduction of CO2 emission into the atmosphere is an important issue [1], [2]. Among diverse greenhouse gases (GHGs), it is widely accepted that CO2 is the most eminent GHG which leads to the nearly 60% of the global climate change [2], [3]. Considering the before-mentioned implications of CO2 emission, mainly released by CO2 rich streams (e.g., flue gas), it is extremely important to remove CO2 by efficient separation and purification techniques.

Absorption by using aqueous amine solvent is one of the most prevalent techniques for post-combustion CO2 capture due to its privileges, such as high flexibility for industrial applications and great performance [4], [5]. Among a wide variety of influential parameters in the absorptive removal of CO2 by amine solvents, CO2 solubility or equilibrium CO2 capacity has been known as the most important parameter affecting the effectiveness and performance of amine solvents to a great extent [4].

Different separation and purification techniques have been extensively employed (e.g., physical and chemical absorption [5], cryogenic separation [6], adsorption by solid sorbents [7], membrane technology [8], [9], [10], and hybrid processes) for CO2 capture from natural gas and post-combustion CO2 capture. Although gas hydrates have attracted a great attention as a new technology [11], [12], chemical absorption of CO2 by alkanolamine solvents is the most preferred technique which has been used during past decades [13], [14]. The physical and chemical properties of commercial amine solvents, mass transfer mechanisms and kinetics of absorption can be found in the published literature [15], [16].

As well as typical measurements by experimental methods, a series of thermodynamic models have been proposed to calculate the equilibrium absorption capacity of CO2 at various operating conditions. The well known thermodynamic models developed on the basis of vapor-liquid equilibrium (VLE) theory, include Kent-Eisenberg [17], [18], electrolyte-NRTL [19], Deshmukh-Mather [20] and extended UNIQUAC [21], [22]. Nevertheless, thermodynamic models are suffering from some drawbacks, which make them inappropriate for accurate estimation of CO2 loading in a wide range of conditions and different amine solutions, and normally, physicochemical properties should be known for each amine solution. For instance, Benamor and Aroua [23] designed a number of experiments to assess the accuracy of modified Deshmukh-Mather model to estimate the CO2 loading in DEA, MDEA and DEA/MDEA solutions at different operating temperatures (30–50 °C), CO2 partial pressures (0.09–100 kPa) and amine concentrations (2–4 M). Though they reported an acceptable accuracy between experimental and model estimated values for both DEA and MDEA, the limited number of amines and narrow ranges of operating variables can seriously barricade its applicability.

Various methodologies of machine learning, including artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS) and support vector machine (SVM), have been presented for modeling and parametric estimation in different fields [24], [25], [26]. Similarly, many research works have investigated the application of machine learning methodologies in the area of aqueous alkanolamine solvents for CO2 capture. Koolivand Salooki et al. [27] succeeded to estimate the output parameters of stripper column, located at Hashemi Nejad Gas Refinery in Iran, by using ANN model and reported a good agreement between operational data and the result obtained from model. Adib et al. investigated the ability of SVM model to estimate output variables of regeneration column (i.e., temperature and reflux flow rate) of Hashemi Nejad Gas Refinery [28]. A simple comparison between ANN and SVM model revealed that SVM can estimate processes variables better than ANN model with minimum square correlation coefficient of 0.99. In 2014, Ghiasi and Mohammadi [29] utilized Least Squares Support Vector Machine (LSSVM) approach to model CO2 loading in different aqueous amine solvents with different concentrations, temperatures and CO2 partial pressures. In 2016, Ghiasi et al. [1] employed ANFIS technique to model a similar system and indicated that ANFIS soft computing approach substantially improves the accuracy of model to estimate the process variables, compared to the previously used LSSVM.

Saghatoleslami and his colleagues [30] performed similar experiments by genetic algorithm. Sipöcz et al. [31] used feed forward ANN to model steady-state CO2 capture process by aqueous MEA solvent in a power plant. In the published article by Zhou et al. [32], a combination of ANN with both adaptive-network-based fuzzy interface system (ANFIS) and sensitivity analysis was used to model post-combustion CO2 capture by amine solvents. Similar investigations can be found in the published research works [33], [34], [35].

ANN modeling was employed to estimate the experimental values of CO2 solubility in aqueous TIPA, TIPA/MEA and TIPA/PZ solutions by Daneshvar et al. [36]. Shahsavand and his coworkers [37] studied the capability of both radial basis function neural network and multi-layer perceptron ANN to calculate the equilibrium CO2 absorption capacity of aqueous DEA and MDEA solutions considering different concentrations and rates. A comparative study was performed by Pahlavanzadeh et al. [38] to evaluate the abilities of Deshmukh-Mather and ANN models to estimate CO2 solubility in 2-amino-2-methyl-1-propanol (AMP) at low partial pressures (7.47–69.87 kPa). ZareNezhad and Aminian used a feed-forward ANN model to estimate the H2S solubility in PZ solvent [39] and a fuzzy network model to anticipate the solubility of H2S in aqueous brine solutions [40]. In another research, Bastani et al. [41] employed feed forward ANN model to estimate CO2 absorption capacity of aqueous chemical absorbers in a wide range of operating parameters (i.e., temperature, pressure and concentration).

To the best of our knowledge, there is not any research work attempting to estimate the experimentally obtained values of CO2 absorption capacities of all kinds of aqueous amines by only a single comprehensive model.

In this work, the experimental data of CO2 solubility in three commercialized amines (including MEA, DEA, and MDEA) and nine recently developed amines (including 2-amino-2-methyl-1-propanol (AMP), piperazine (PZ), triisopropanolamine (TIPA), monoproanolamine(MPA), 1-amino-2-propanol (MIPA), 4-(diethylamino)-2-butanol(DEAB), methyl amino ethanol (MAE), 2-(Diethylamino)- ethanol(DEEA) and 3-(Methylamino)-propylamine (MAPA)) were used which can cover primary, secondary and tertiary amine classification. Hence, the accurate results of ANN models for anticipating the equilibrium CO2 absorption capacities of all before-mentioned amine solutions can ameliorate the limitations of typical theoretical models, which are only applicable for specific cases.

Almost all previously published research papers agree that data validation is the most crucial factor for developing a promising model.

Section snippets

Radial basis function neural network (RBF-NN)

RBF-ANN is constructed from three layers; namely input, output and hidden layers [42]. Among different ANNs, RBF-ANN can be categorized into the single hidden layer group which is more beneficial than multiple interconnected hidden layers, such as the multilayer perception neural network (MLP-NN) [42]. In detail, activation function and the maximum number of neurons are two important parts of RBF-NN, which can substantially affect the processing conditions. Gaussian function is normally used as

General step

In order to employ machine learning technique for modeling CO2 solubility in diverse aqueous amine solutions (e.g., MEA, DEA, and MDEA, AMP, PZ, TIPA, MPA, MIPA, DEAB, MAE, DEEA and MAPA), a comperhensive databank of experimental data was extracted from literature. The CO2 solubility is defined as the moles of CO2 per moles of amine and is a function of CO2 partial pressure (PCO2, kPa), temperature (T, K), amine concentration (Camine, mol/L), molecular weight, hydrogen bond donor/acceptor

Model development

The CSA approach was utilized to find the optimal values of the LSSVM parameters, including γ and σ2. Values of these parameters are 101.6330 and 0.389, respectively. The structure of RBF consists of couple of tuning parameters that are maximum number of neurons and spread. The trial and error were employed to find the optimum values, which were 95 and 260, respectively. The MSE values at different neurons are depicted in Fig. 1. PSO and GA methods were employed to determine optimum values of

Conclusion

In this work, different models (i.e. GA-ANFIS, PSO-ANFIS, RBF-ANN and LSSVM) were successfully developed to anticipate the equilibrium absorption capacity of CO2 in 12 aqueous alkanolamine solutions. The results indicate that the models are capable to acceptably estimate experimental values of CO2 loading over a wide range of independent variables (i.e. CO2 partial pressure, temperature and amine concentration). Furthermore, it is declared that the results of the developed model (in this study)

References (112)

  • R. Deshmukh et al.

    A mathematical model for equilibrium solubility of hydrogen sulfide and carbon dioxide in aqueous alkanolamine solutions

    Chem Eng Sci

    (1981)
  • U.E. Aronu et al.

    Solubility of CO2 in 15, 30, 45 and 60 mass% MEA from 40 to 120 C and model representation using the extended UNIQUAC framework

    Chem Eng Sci

    (2011)
  • A. Benamor et al.

    Modeling of CO2 solubility and carbamate concentration in DEA, MDEA and their mixtures using the Deshmukh-Mather model

    Fluid Phase Equilib

    (2005)
  • H. Saghafi et al.

    Modeling of CO2 solubility in MEA, DEA, TEA, and MDEA aqueous solutions using AdaBoost-decision tree and artificial neural network

    Int J Greenhouse Gas Control

    (2017)
  • M. Rezakazemi et al.

    Development of hybrid models for prediction of gas permeation through FS/POSS/PDMS nanocomposite membranes

    Int J Hydrogen Energy

    (2018)
  • A. Dashti et al.

    Estimating CH4 and CO2 solubilities in ionic liquids using computational intelligence approaches

    J Mol Liq

    (2018)
  • H. Adib et al.

    Support Vector Machine based modeling of an industrial natural gas sweetening plant

    J Nat Gas Sci Eng

    (2013)
  • M.M. Ghiasi et al.

    Rigorous modeling of CO2 equilibrium absorption in MEA, DEA, and TEA aqueous solutions

    J Nat Gas Sci Eng

    (2014)
  • N. Sipöcz et al.

    The use of Artificial Neural Network models for CO2 capture plants

    Appl Energy

    (2011)
  • Q. Zhou et al.

    From neural network to neuro-fuzzy modeling: applications to the carbon dioxide capture process

    Energy Proc

    (2011)
  • Q. Zhou et al.

    Application of neuro-fuzzy modeling technique for operational problem solving in a CO2 capture process system

    Int J Greenhouse Gas Control

    (2013)
  • Q. Zhou et al.

    Modeling of the carbon dioxide capture process system using machine intelligence approaches

    Eng Appl Artif Intell

    (2011)
  • Y. Wu et al.

    Analysis of data for the carbon dioxide capture domain

    Eng Appl Artif Intell

    (2011)
  • N. Daneshvar et al.

    Carbon dioxide equilibrium absorption in the multi-component systems of CO2+ TIPA+ MEA+ H2O, CO2+ TIPA+ Pz+ H2O and CO2+ TIPA+ H2O at low CO2 partial pressures: experimental solubility data, corrosion study and modeling with artificial neural network

    Sep Purif Technol

    (2004)
  • A. Shahsavand et al.

    Application of artificial neural networks for simulation of experimental CO2 absorption data in a packed column

    J Nat Gas Sci Eng

    (2011)
  • H. Pahlavanzadeh et al.

    Experimental analysis and modeling of CO2 solubility in AMP (2-amino-2-methyl-1-propanol) at low CO2 partial pressure using the models of Deshmukh-Mather and the artificial neural network

    J Chem Thermodyn

    (2011)
  • D. Bastani et al.

    Prediction of CO2 loading capacity of chemical absorbents using a multi-layer perceptron neural network

    Fluid Phase Equilib

    (2013)
  • A. Khajeh et al.

    Prediction of solubility of gases in polystyrene by adaptive neuro-fuzzy inference system and radial basis function neural network

    Expert Syst Appl

    (2010)
  • M.-Y. Chen

    A hybrid ANFIS model for business failure prediction utilizing particle swarm optimization and subtractive clustering

    Inf Sci

    (2013)
  • C.-J. Lin et al.

    The design of neuro-fuzzy networks using particle swarm optimization and recursive singular value decomposition

    Neurocomputing

    (2007)
  • J.-S. Chiou et al.

    A PSO-based adaptive fuzzy PID-controllers

    Simul Model Pract Theory

    (2012)
  • A. Karkevandi-Talkhooncheh et al.

    Application of adaptive neuro fuzzy interface system optimized with evolutionary algorithms for modeling CO2-crude oil minimum miscibility pressure

    Fuel

    (2017)
  • S.R. Amendolia et al.

    A comparative study of k-nearest neighbour, support vector machine and multi-layer perceptron for thalassemia screening

    Chemometr Intell Lab Syst

    (2003)
  • A. Shokrollahi et al.

    On accurate determination of PVT properties in crude oil systems: committee machine intelligent system modeling approach

    J Taiwan Inst Chem Eng

    (2015)
  • S. Rafiee-Taghanaki et al.

    Implementation of SVM framework to estimate PVT properties of reservoir oil

    Fluid Phase Equilib

    (2013)
  • A. Baylar et al.

    Application of least square support vector machines in the prediction of aeration performance of plunging overfall jets from weirs

    Expert Syst Appl

    (2009)
  • A. Hemmati-Sarapardeh et al.

    On the evaluation of the viscosity of nanofluid systems: modeling and data assessment

    Renew Sustain Energy Rev

    (2018)
  • S.K. Dash et al.

    Vapour liquid equilibria of carbon dioxide in dilute and concentrated aqueous solutions of piperazine at low to high pressure

    Fluid Phase Equilib

    (2011)
  • S. Bishnoi et al.

    Absorption of carbon dioxide into aqueous piperazine: reaction kinetics, mass transfer and solubility

    Chem Eng Sci

    (2000)
  • J.-Y. Park et al.

    Solubility of carbon dioxide in aqueous solutions of 2-amino-2-ethyl-1, 3-propanediol

    Fluid Phase Equilib

    (2002)
  • J.I. Lee et al.

    Equilibrium between carbon dioxide and aqueous monoethanolamine solutions

    J Appl Chem Biotech

    (1976)
  • B. Lemoine et al.

    Partial vapor pressure of CO2 and H2S over aqueous methyldiethanolamine solutions

    Fluid Phase Equilib

    (2000)
  • R. Sidi-Boumedine et al.

    Experimental determination of carbon dioxide solubility data in aqueous alkanolamine solutions

    Fluid Phase Equilib

    (2004)
  • P.J.G. Huttenhuis et al.

    Gas solubility of H2S and CO2 in aqueous solutions of N-methyldiethanolamine

    J Petrol Sci Eng

    (2007)
  • M. Haji-Sulaiman et al.

    Analysis of equilibrium data of CO2 in aqueous solutions of diethanolamine (DEA), methyldiethanolamine (MDEA) and their mixtures using the modified Kent Eisenberg model

    Chem Eng Res Des

    (1998)
  • M. Aroua et al.

    Modelling of carbon dioxide absorption in aqueous solutions of AMP and MDEA and their blends using Aspenplus

    Sep Purif Technol

    (2002)
  • A. Dashti et al.

    Efficient hybrid modeling of CO2 absorption in aqueous solution of piperazine: applications to energy and environment

    Chem Eng Res Des

    (2019)
  • G.T. Rochelle

    Amine scrubbing for CO2 capture

    Science

    (2009)
  • S. Zhang et al.

    Surfactants facilitating carbonic anhydrase enzyme-mediated CO2 absorption into a carbonate solution

    Environ Sci Technol

    (2017)
  • A. Samanta et al.

    Post-combustion CO2 capture using solid sorbents: a review

    Ind Eng Chem Res

    (2011)
  • Cited by (0)

    View full text