Elsevier

Energy

Volume 164, 1 December 2018, Pages 664-675
Energy

Characterizing CO2 capture with aqueous solutions of LysK and the mixture of MAPA + DEEA using soft computing methods

https://doi.org/10.1016/j.energy.2018.09.061Get rights and content

Highlights

  • Predicting the solubility of CO2 in the mixture of MAPA + DEEA and LysK aqueous solutions was carried out.

  • SGB and GP as two robust models were developed based on published data.

  • Graphical and statistical analyses used for showing successful performance of the models.

  • Results confirmed the superiority of developed models to published method.

Abstract

Accurate data in the field of CO2-capture using new high potential absorbents as alternatives to the traditional ones is of great interest within scientific and engineering communities. In this direction, two robust modeling strategies, viz. Stochastic Gradient Boosting (SGB) tree and Genetic Programming (GP) are used to 1) predict the solubility of CO2 in aqueous potassium lysinate (LysK) solutions as a function of temperature, partial pressure of CO2, and the mass fraction of LysK; and 2) predict the solubility of CO2 in the mixture of MAPA + DEEA aqueous solutions as a function of temperature, partial pressure of CO2, and the concentration of MAPA and DEEA based on previously published data. The efficiency and precision of the proposed models are checked graphically and statistically. Results show that both proposed models are competent in accurate and reliable predictions (R2 > 0.98 and RMSE < 0.06). However, the SGB models are superior to the GP models. Additionally, the proposed models are compared to the modified Kent-Eisenberg model for predicting the CO2 solubility in LysK solutions, and shown to have better outputs.

Introduction

Climate change has attracted much attention in the recent years. One major factor contributing to climate change is greenhouse gas emissions among which CO2 has a much more contribution. For example, CO2 emissions constitute 82% of the U.S [1]. Greenhouse gas emissions in 2015. Although CO2 is a natural part of the Earth's carbon cycle, its excessive amount generated by human has altered the cycle. The main source of CO2 emissions is the combustion of fossil fuels such as oil, natural gas, and coal which are used for heat, transportation, and various industries.

The best way to reduce CO2 emission is eliminating fossil fuel consumption which is proven to be impracticable with the current technologies [2,3]. An alternative is to use the available technologies to capture CO2 emission from the fossil fuel consumption. The captured CO2 can then be used in biological processes such as algae production [[4], [5], [6], [7], [8]], processed further using various methods such as CO2 geosequestration [7,[9], [10], [11], [12]], or converted to different chemicals or fuels [7,[13], [14], [15], [16], [17], [18], [19], [20], [21], [22]].

The main CO2 capture processes are absorption, where CO2 is absorbed selectively by a chemical solvent; adsorption, where the gas is passed through a packed bed, and CO2 is captured by the solid adsorbents; and permeation, where the gas is passed through a membrane, and CO2 is separated [[23], [24], [25], [26]]. The less common methods include cryogenics separation, where CO2 is captured by condensation; and gas hydrate crystallization, where CO2 is trapped in light hydrocarbon cage-like structures [[27], [28], [29], [30], [31], [32]]. Hybrid methods are also investigated in the literature [[33], [34], [35], [36], [37]].

CO2 capture by chemical absorption is one of the oldest of the aforementioned techniques. Many solvents such as amine solutions including monoethanolamine (MEA), diethanolamine (DEA), methylidiethanolamine (MDEA), piperazine (PZ), 2-amino-2-methyl-1-propanol (AMP), 2-(diethylamino)ethanol (DEEA), triisopropanolamine (TIPA), and 4-(diethylamino)-2-butanol (DEAB) [[38], [39], [40], [41], [42]], and ionic liquids including Imidazolium, Pyrollidinium, Pyridinium, Guanidinium, Phosphonium, Morpholinium, Piperidinium, Sulfonium, Ammonium, Hexafluorophosphate, Tetrafluoroborate, Alkylsulphate, and Triflate, Dicyanamide [[43], [44], [45], [46], [47], [48], [49], [50], [51]] are investigated theoretically and experimentally for this process among which MEA, DEA, and MDEA have been used for many decades [[27], [28], [29],[31], [32], [33], [34], [35], [36], [37],52].

A new generation of mixed amine solvent system, an aqueous mixture of 2-(diethylamino) ethanol (DEEA) and 3-(methylamino)-propylamine (MAPA), has recently received a large amount of attention thanks to having characteristics of biphasic/phase-change solvents. Earlier research has revealed that the aqueous blend of DEEA + MAPA results in two liquid phases upon CO2 absorption, an upper phase very low in CO2, and a lower phase rich in CO2. If only the lower phase is regenerated, owing to low liquid circulation rate in the stripping section, the energy needed for the solvent regeneration can be decreased. Also, this will result in decreased capital and operational costs due to the decrease in the size of the desorption column [53]. DEEA is a tertiary alkanol amine whereas MAPA is diamine, which has primary and secondary amine functional groups.

Moreover, aqueous potassium lysinate (LysK), which is an amino acid salt (AAS) is investigated recently as a new absorbent and has shown to have a high potential in capturing CO2 [[54], [55], [56], [57]]. AASs in general have low volatility and have similar structure to alkanolamines. Shen et al. [54,56,57] measured the solubility of CO2 in LysK solutions at various temperatures and pressures and observed a higher CO2 loading capacity than MEA solutions. They recommended Lysk as a new absorbent with fast kinetics for CO2 Capture.

The knowledge of vapor-liquid equilibrium (VLE) of CO2 capture solvent systems is required for the design and modeling of gas treating processes. As performing experimental study is in general costly and time consuming as well as being difficult at times, mathematical models and computational methods are employed to simulate conditions that are hardly possible by experiments and to facilitate process analysis. Numerous methods are developed to model or analyze CO2 capture with chemical solvents. These methods usually fall into four categories: empirical models, activity-based models, equation of state (EOS) models, and soft computing techniques. Kent–Eisenberg [58], modified UNIQUAC [59], electrolyte NRTL [60], and Deshmukh–Mather [61] are among the thermodynamic models that are able to predict the equilibrium solubility of CO2 in different solvents. These models often suffer from a limited applicable process condition range that affects their accuracy [62]. Soft computing techniques, alternatively, are convenient and yet powerful tools that enable detailed and complicated modeling and analysis. Artificial neural network (ANN) [63,64], support vector machine (SVM) [65,66], adaptive neuro-fuzzy inference system (ANFIS) [66], genetic programming [67], and decision tree based algorithms [7,68] are among these techniques which are proven to be useful for study of VLE of sour gas absorption modeling.

In this direction, herein, a novel decision-tree-based approach named Stochastic Gradient Boosting (SGB) and genetic programming (GP) schemes are used for modeling the solubility of CO2 in systems of LysK and MAPA + DEEA blend solutions. To the best knowledge of the authors, there is no published study demonstrating the application of SGB and GP methodologies for the modeling of CO2 loading of these new capture solvents. In addition, the accuracy of GP and SGB modeling techniques are compared to the modified Kent–Eisenberg model using statistical assessment.

Section snippets

Data

The required data for developing and analyzing the models are collected from the open literature [38,57], where the CO2 loading capacity of the capture solutions is represented as a function of temperature, pressure, and the concentration of capture solvents. In other words, CO2 loading capacity is a dependent variable and temperature, pressure, and the concentration of solvents are independent variables. There, the mathematical relations can be expressed using following equations:αlysK=f1(PCO2,

Stochastic Gradient Boosting (SGB) tree

Boosting algorithm was chiefly evolved by Schapiro and Freund [69]. Friedman continued their work and provided the foundation for a recent generation of boosting algorithm called gradient boosting machine (GBM) [70,71]. In GBM, the main concept is to create additive regression models thru successively fitting a weak learner to up-to-date “pseudo”-residuals thru least squares at every step of the loop. Ergo, boosting can be defined as a way of fitting an additive spreading out in a set of

Results and discussions

Different statistical parameters including Correlation coefficient (R2), average absolute relative deviation (AARD), Mean-square error (MSE), Root-mean-square error (RMSE), Mean absolute error (MAE), and Absolute error (AE) as well-known commonly used criteria were measured in this study to assess the performance of the SGB and GP models. These statistical parameters are as follows:R2=1i=1n(yiexp.yical.)2i=1n(yiexp.y¯exp.)2AARD(%)=(1n)i=1n|(yiexp.yical.)yiexp.|×100MSE=(1n)i=1n(yiexp.yi

Conclusion

In this work, two accurate and reliable modeling approaches namely SGB and GP are proposed for predicting the solubility of CO2 in aqueous LysK and MAPA + DEEA solutions based on 237 and 313 experimental data points collected from the literature. The input parameters of the developed models for predicting the solubility of CO2 in aqueous LysK solutions are temperature, partial pressure of CO2, the mass fraction of LysK, temperature, partial pressure of CO2, concentration of MAPA, and

References (80)

  • L. Li et al.

    Electrochemical conversion of CO2 to carbon and oxygen in LiCl–Li2O melts

    Electrochim Acta

    (2016)
  • D. Mei et al.

    Plasma-photocatalytic conversion of CO2 at low temperatures: understanding the synergistic effect of plasma-catalysis

    Appl Catal B Environ

    (2016)
  • K. Roh et al.

    A methodological framework for the development of feasible CO2 conversion processes

    Int J Greenh Gas Contr

    (2016)
  • W. Schakel et al.

    Assessing the techno-environmental performance of CO2 utilization via dry reforming of methane for the production of dimethyl ether

    J CO2 Util

    (2016)
  • Y. Wei et al.

    Fabrication of inverse opal TiO2-supported Au@ CdS core–shell nanoparticles for efficient photocatalytic CO2 conversion

    Appl Catal B Environ

    (2015)
  • N. Yang et al.

    A novel electrochemical conversion of CO2 with aryl hydrazines and paraformaldehyde into 1, 3, 4-oxadiazol-2 (3H)-one derivatives in one step

    Electrochem Commun

    (2016)
  • Y. Son et al.

    An intermittently moving bed adsorption process for carbon dioxide capture

    Int J Greenh Gas Contr

    (2016)
  • M.M. Ghiasi et al.

    Modeling of stability conditions of natural gas clathrate hydrates using least squares support vector machine approach

    J Mol Liq

    (2016)
  • P.J. Herslund et al.

    Phase equilibrium modeling of gas hydrate systems for CO2 capture

    J Chem Therm

    (2012)
  • T.E. Rufford et al.

    The removal of CO2 and N2 from natural gas: a review of conventional and emerging process technologies

    J Petrol Sci Eng

    (2012)
  • H. Yang et al.

    Progress in carbon dioxide separation and capture: a review

    J Environ Sci

    (2008)
  • B. Freeman et al.

    Hybrid membrane-absorption CO2 capture process

    Energy Procedia

    (2014)
  • P.K. Kundu et al.

    Effectiveness of membranes and hybrid membrane processes in comparison with absorption using amines for post-combustion CO2 capture

    Int J Greenh Gas Contr

    (2014)
  • M. Maroño et al.

    Experimental studies of CO2 capture by a hybrid catalyst/adsorbent system applicable to IGCC processes

    Appl Therm Eng

    (2015)
  • L. Peters et al.

    CO2 removal from natural gas by employing amine absorption and membrane technology—a technical and economical analysis

    Chem Eng J

    (2011)
  • D.-L. Zhong et al.

    Precombustion CO2 capture using a hybrid process of adsorption and gas hydrate formation

    Energy

    (2016)
  • 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

    Separ Purif Technol

    (2004)
  • S.K. Dash et al.

    Experimental and theoretical investigation of solubility of carbon dioxide in concentrated aqueous solution of 2-amino-2-methyl-1-propanol and piperazine

    J Chem Therm

    (2012)
  • S.-H. Wu et al.

    Vapor pressures of aqueous blended-amine solutions containing (TEA/AMP/MDEA)+(DEA/MEA/PZ) at temperatures (303.15–343.15) K

    Exp Therm Fluid Sci

    (2013)
  • M.W. Arshad et al.

    Heat of absorption of CO2 in aqueous solutions of DEEA, MAPA and their mixture

    Energy Procedia

    (2013)
  • S. Shen et al.

    CO 2 absorption into aqueous potassium salts of lysine and proline: density, viscosity and solubility of CO 2

    Fluid Phase Equil

    (2015)
  • S. Shen et al.

    CO2 absorption using aqueous potassium lysinate solutions: vapor–liquid equilibrium data and modelling

    J Chem Therm

    (2017)
  • 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)
  • R. Deshmukh et al.

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

    Chem Eng Sci

    (1981)
  • G. Chen et al.

    Artificial neural network models for the prediction of CO2 solubility in aqueous amine solutions

    Int J Greenh Gas Contr

    (2015)
  • M. Hamzehie et al.

    Developing a feed forward multilayer neural network model for prediction of CO2 solubility in blended aqueous amine solutions

    J Nat Gas Sci Eng

    (2014)
  • A. Shafiei et al.

    Estimating hydrogen sulfide solubility in ionic liquids using a machine learning approach

    J Supercrit Fluids

    (2014)
  • M.-A. Ahmadi et al.

    Connectionist technique estimates H2S solubility in ionic liquids through a low parameter approach

    J Supercrit Fluids

    (2015)
  • A. Tatar et al.

    Comparison of two soft computing approaches for predicting CO2 solubility in aqueous solution of piperazine

    Int J Greenh Gas Contr

    (2016)
  • M.A. Ahmadi et al.

    Estimation of H2S solubility in ionic liquids using a rigorous method

    J Supercrit Fluids

    (2014)
  • Cited by (13)

    • Screening of important parameters in optimal design of compressed air energy storage system using an ensemble learning method

      2022, Journal of Energy Storage
      Citation Excerpt :

      The stochastic nature of the algorithm producing an incremental improvement in the model accuracy and reduces overfitting. In recent years very few applications of SGBTree model have been reported for both classification and regression problems such as carbon dioxide‐oil minimum miscibility pressure prediction [22], carbon dioxide solubility in polymers forecasting [23], estimation of interfacial tension for geological carbon dioxide storage [24], predicting carbon dioxide solubility in aqueous amine solutions [25,26], global and local wind energy prediction as well as to a solar radiation problem [27]. The use of SGBTree model as a refinement of classification tree analysis in a remote sensing issue was shown by Lawrence et al. [28].

    • Research on integrated CO<inf>2</inf> absorption-mineralization and regeneration of absorbent process

      2021, Energy
      Citation Excerpt :

      Chemical absorption is one of the most mature CO2 separation methods post-combustion and demonstrates excellent performance in the separation of CO2 from flue gas [1–4].

    • Toward smart carbon capture with machine learning

      2021, Cell Reports Physical Science
      Citation Excerpt :

      In addition to amines and ILs, the equilibrium solubility of CO2 in other absorbents has been predicted using ML algorithms. In this context, the solubility values in solutions of various amino acids, such as sodium glycinate51 and potassium lysinate,52,53 and in inorganic solvents, such as trisodium phosphate solutions,54 have been modeled using various ML algorithms. The CO2 partial pressure, temperature, and absorbent concentration have been considered as input features to construct the model; however, as mentioned earlier, this approach may result in a lack of deep insights because these models do not consider the chemical nature (and related parameters) of the absorbent.

    • Characterization of physico-chemical properties of biodiesel components using smart data mining approaches

      2020, Fuel
      Citation Excerpt :

      This modification process results a new population (i.e. next generation or children), because of changing the mathematical functions due to the tree structure changes. The generation is repeated several times and the model gets optimized more in each iteration until the last generation is implemented which includes the most-optimized functions with specific arrangement of genes to solve the problem [34,35]. For modeling purposes of GP, the regression between dependent variable and independent ones is usually known as “multi-gene symbolic regression”.

    View all citing articles on Scopus
    View full text