Method and system for predicting gas absorption performance of mixed absorbent by using artificial neural network, and method and system for determining optimal composition of mixed absorbent

Abstract

The present invention relates to a method and system for predicting gas absorption performance of a mixed absorbent by using an artificial neural network, and a method and system for determining the optimal composition of a mixed absorbent by using same.

Description

Method and system for predicting gas absorption performance of a mixed absorbent using an artificial neural network, and method and system for determining an optimal mixed absorbent composition

[0001] The present invention relates to a method and system for predicting the gas absorption performance of a mixed absorbent using an artificial neural network, and a method and system for determining an optimal mixed absorbent composition using the same.

[0002] According to the International Energy Agency (IEA), CO₂ emissions are steadily increasing, making carbon capture, utilization, and storage (CCUS) technology increasingly important. While wet carbon capture (PCC) processes possess high technological maturity and efficiently remove CO₂, they consume significant energy during the absorbent regeneration process, resulting in high process costs. To improve the energy efficiency of wet carbon capture processes, reducing regeneration energy is necessary. This requires predicting the vapor-liquid equilibrium (VLE) state of the amine solution, the absorbent. Existing studies predict VLE using thermodynamic models (KE, E-NRTL, UNIQUAC) and data-driven models (ANN, DNN, etc.). However, thermodynamic models require complex parameters (activity coefficients, fugacity coefficients, etc.), and data-driven models suffer from performance limitations (low predictive power, overfitting) due to a lack of mixed amine data.

[0003] The present invention provides a method and system for predicting the gas absorption performance of a mixed absorbent through machine learning based on an artificial neural network that is strong even with a small amount of learning data, in order to overcome the limitations of existing studies, and a method and system for determining an optimal mixed absorbent composition using the same.

[0004] [Prior Art Literature]

[0005] [Patent Document]

[0006] 1. Chinese Patent Publication No. 2024-10366531

[0007] [Non-patent literature]

[0008] 1. Haonan Liu et al. “A generic machine learning model for CO2 equilibrium solubility into blended amine solutions”Separation and Purification Technology 334 (2023) 126100

[0009] 2. Farid Amirkhanie et al. “Estimation of CO2 solubility in aqueous solutions of commonly used blended amines: Application to optimized greenhouse gas capture” Journal of Cleaner Production 430 (2023) 139435

[0010] 3. Chungbuk National University, “Development of an absorption equilibrium model for CO2, SOx, NOx, and mixed gases in mixed amine aqueous solutions considering thermodynamic non-ideality.”

[0011] The first purpose of the present invention is to provide a method for predicting the gas absorption performance of a mixed absorbent through machine learning based on an artificial neural network using a limited amount of learning data.

[0012] A second object of the present invention is to provide a system for predicting gas absorption performance of a mixed absorbent through machine learning based on an artificial neural network.

[0013] The third object of the present invention is to provide a method for determining the optimal composition of a mixed gas absorbent through machine learning based on an artificial neural network.

[0014] The fourth object of the present invention is to provide a system for determining the optimal composition of a mixed gas absorbent through machine learning based on an artificial neural network.

[0015] A method for predicting the gas absorption performance of a mixed absorbent according to the first object of the present invention

[0016] a) A step of creating a prediction model for a specific gas absorption performance of a single absorbent by training an artificial neural network model with absorption performance experimental data of one or more single absorbents for a specific gas to predict the specific gas absorption performance of a single absorbent;

[0017] b) A step of generating weighted average data of absorption performance of two or more mixed absorbents for the above specific gas using the concentration-weighted average formula below:

[0018]

[0019] (In the above formula,

[0020] n is an integer representing the type of single absorbent in the mixed absorbent,

[0021] c represents the concentration (wt%) of each single absorbent in the mixed absorbent,

[0022] α is the absorption performance (mol) of each single absorbent in the mixed absorbent. 기체 / mol 흡수제 ) represents); and,

[0023] c) a step of creating a prediction model for a specific gas absorption performance of a mixed absorbent by training the absorption performance prediction data of the mixed absorbent generated in step b) and the absorption performance experimental data of the mixed absorbent using an artificial neural network model to predict the specific gas absorption performance of the mixed absorbent;

[0024] It includes.

[0025] In the above method, the artificial neural network model may be a deep neural network (DNN).

[0026] In the above method, the specific gas absorption performance is determined by the specific gas absorption amount (mol) at a specific temperature, partial pressure of a specific gas, and a specific type of absorbent and concentration in a specific solvent. 기체 / mol흡수제 ) may be composed of.

[0027] In the above method, the specific gas may be CO2, and the absorbent may be an amine compound.

[0028] In the above method, the amine compound may be selected from the group consisting of MEA (Monoethanolamine), DEA (Diethanolamine), MDEA (N-methyl diethanolamine), AMP (2-amino-2-methyl-1-propanol), and PZ (Piperazine).

[0029] In the above method, the absorbent may be in the form of an aqueous solution.

[0030]

[0031] According to the second purpose of the present invention, a system for predicting the gas absorption performance of a mixed absorbent through machine learning based on an artificial neural network is provided.

[0032] i) A single absorbent's specific gas absorption performance prediction unit that generates prediction data for a single absorbent's specific gas absorption performance using an artificial neural network model that has learned absorption performance experimental data of one or more single absorbents for a specific gas;

[0033] ii) A weighted average data generation unit for a specific gas absorption performance of a mixed absorbent that generates absorption performance prediction data of two or more mixed absorbents for the specific gas using the concentration-weighted average formula below:

[0034]

[0035] (In the above formula,

[0036] n is an integer representing the type of single absorbent in the mixed absorbent,

[0037] c represents the concentration (wt%) of each single absorbent in the mixed absorbent,

[0038] α is the absorption performance (mol) of each single absorbent in the mixed absorbent. 기체 / mol 흡수제 ) represents); and,

[0039] iii) A mixed absorbent's specific gas absorption performance prediction data generation unit that generates the mixed absorbent's specific gas absorption performance prediction data using an artificial neural network model that has learned the mixed absorbent's absorption performance prediction data generated in ii) above and the mixed absorbent's absorption performance experimental data;

[0040] may include.

[0041] In the above system, the artificial neural network model may be a deep neural network (DNN).

[0042] In the above system, the specific gas absorption performance is a specific gas absorption amount (mol) at a specific temperature, partial pressure of a specific gas, and a specific type of absorbent and concentration in a specific solvent. 기체 / mol 흡수제 ) may be composed of.

[0043] In the above system, the specific gas may be CO2, and the absorbent may be an amine compound.

[0044] In the above system, the amine compound may be selected from the group consisting of MEA (Monoethanolamine), DEA (Diethanolamine), MDEA (N-methyl diethanolamine), AMP (2-amino-2-methyl-1-propanol), and PZ (Piperazine).

[0045] In the above system, the absorbent may be in the form of an aqueous solution.

[0046]

[0047] According to the third object of the present invention, a method for determining the optimal composition of a mixed gas absorbent through machine learning based on an artificial neural network is provided.

[0048] d) a step of calculating specific gas absorption performance prediction data for each composition of the mixed absorbent under each specific temperature and partial pressure condition of a specific gas for absorbing and removing a specific gas, according to a method for predicting the gas absorption performance of the mixed absorbent;

[0049] e) a step of calculating the regeneration energy of the mixed absorbent by calculating the heat of reaction, latent heat and sensible heat data based on the difference in the specific gas absorption performance data under the absorption and stripping conditions of the specific gas calculated in step d);

[0050]

[0051] f) A step may be included in which the mixed absorbent composition having the smallest value of regenerative energy among the mixed absorbent compositions calculated in step e) is selected.

[0052]

[0053] According to the fourth object of the present invention, a system for determining the optimal composition of a mixed gas absorbent through machine learning based on an artificial neural network is provided.

[0054] iv) A specific gas absorption performance calculation unit of the mixed absorbent that calculates specific gas absorption performance data for each composition of the mixed absorbent under each specific temperature and partial pressure condition of a specific gas for absorbing and removing a specific gas, according to a system for predicting the gas absorption performance of the mixed absorbent;

[0055] v) A regeneration energy calculation unit that calculates the regeneration energy of the mixed absorbent by calculating the heat of reaction, latent heat, and sensible heat data based on the difference in the specific gas absorption performance data under the absorption and stripping conditions of the specific gas calculated in iv) using the mathematical formula below;

[0056]

[0057] vi) It may include an optimal mixed amine absorbent composition determination unit that selects a mixed absorbent composition having the smallest value of regeneration energy among the mixed absorbent compositions calculated in v).

[0058] According to the present invention,

[0059] - The first artificial intelligence machine learning-based framework is presented that addresses data-driven modeling and optimization to predict specific gas absorption performance of mixed absorbents.

[0060] - A data-driven model is developed that can accurately predict the VLE of mixed absorbents without thermodynamic parameters.

[0061] - Solve the problem of lack of absorption performance data of mixed absorbents by utilizing absorption performance data of single absorbents,

[0062] - Optimization of the regenerative energy of the absorbent will enable the selection of efficient mixed absorbent candidates.

[0063] This will contribute to increasing the efficiency of the wet carbon capture process.

[0064] Figure 1 is a schematic diagram showing the structure of a mixed absorbent performance prediction model (ABN Model) according to the present invention.

[0065] Figure 2 is a parity plot showing the consistency between actual VLE data between CO2 and MEA and MDEA, which are types of CO2 absorbents, and VLE data generated by a performance prediction model of a single absorbent of the present invention.

[0066] Figure 3 is a box plot comparing the performance of a conventional DNN model and a mixed absorbent performance prediction model of the present invention.

[0067] Figure 4 is a schematic diagram showing the structure of a conventional model for predicting the performance of a general mixed absorbent.

[0068] FIG. 5 is a graph showing the consistency between actual VLE data of MEA / MDEA mixed amines and CO2 at various mixing ratios at various temperatures and VLE data generated by a performance prediction model of the mixed absorbent of the present invention.

[0069] Figure 6 is a schematic diagram of a typical wet carbon capture process for feed gas, consisting of an absorber and a stripper.

[0070] Figure 7 is a graph showing the determination of the optimal composition by calculating the regeneration energy at various MEA / MDEA mixing ratios under specific CO2 absorption and stripping conditions.

[0071] Figure 8 is a graph showing the calculated regeneration energy in terms of heat of reaction (Q-rxn), sensible heat (Q-sen), and latent heat (Q-latent) at various MEA / MDEA mixing ratios under specific CO2 absorption and stripping conditions.

[0072] Figure 9 is a graph comparing the optimal regeneration energy divided into heat of reaction, latent heat, and sensible heat of various single amines and mixed amines as CO2 wet absorbents.

[0073]

[0074] According to the first object of the present invention, the artificial neural network model used to create a prediction model for a specific gas absorption performance of a gas absorbent in the method for predicting the gas absorption performance of a mixed absorbent may be one of the known machine learning models.

[0075] These machine learning algorithms, such as deep neural networks (DNN), convolutional neural networks (CNN), recurrent neural networks (RNN), and deep belief networks (DBN), are widely used in various fields.

[0076] Among these, deep neural networks (DNNs) are artificial neural networks (ANNs) comprised of multiple hidden layers between the input and output layers. Like conventional ANNs, DNNs can model complex non-linear relationships. However, like conventional ANNs, DNNs can also suffer from numerous problems, such as overfitting and high time complexity, if trained in a naive manner.

[0077] This artificial neural network machine learning method can be used to create a prediction model for specific gas absorption performance of single and mixed absorbents.

[0078] In the present invention, among the artificial neural network models, a deep neural network (DNN) may be particularly suitable.

[0079] In the present invention, the specific gas is CO2, and the absorbent may be an amine compound selected from the group consisting of MEA (Monoethanolamine), DEA (Diethanolamine), MDEA (N-methyl diethanolamine), AMP (2-amino-2-methyl-1-propanol), and PZ (Piperazine), and the absorbent may be in the form of an aqueous solution.

[0080] The above specific gas absorption performance is the specific gas absorption amount (mol) at a specific temperature, partial pressure of a specific gas, and a specific type of absorbent and concentration in a specific solvent. 기체 / mol 흡수제 ) may be composed of.

[0081] The data required to create the absorption performance prediction model of the mixed absorbent of the present invention may not be sufficient actual absorption performance data. In the present invention, the absorption performance weighted average data of two or more mixed absorbents for a specific gas, which is obtained from the single absorbent data using the following weighted average formula, can be used as learning data for an artificial neural network model to create a prediction model for the absorption performance of a specific gas of the mixed absorbent:

[0082] (In the above formula,

[0083] n is an integer representing the type of single absorbent in the mixed absorbent,

[0084] c represents the concentration (wt%) of each single absorbent in the mixed absorbent,

[0085] α is the absorption performance (mol) of each single absorbent in the mixed absorbent. 기체 / mol 흡수제 ) represents.);

[0086] According to the second purpose of the present invention, a system for predicting the gas absorption performance of a mixed absorbent through machine learning based on an artificial neural network is provided.

[0087] i) Prediction of specific gas absorption performance of a single absorbent;

[0088] ii) Generation section of weighted average data on specific gas absorption performance of mixed absorbents: and,

[0089] iii) Generation unit for predicting specific gas absorption performance of mixed absorbent;

[0090] The artificial neural network for machine learning, the weighted average formula, and other technical configurations specified in the first purpose of the present invention used in the implementation of such a system may be equally applied to technical configurations according to the second purpose and subsequent purposes of the present invention.

[0091]

[0092] Hereinafter, the present invention will be described in more detail through practical examples and, if necessary, with reference to the attached drawings. In the drawings, the absolute and relative sizes of systems, components, layers, and regions may be exaggerated for clarity.

[0093]

[0094] The terminology used to describe specific examples is not intended to limit the invention. When a particular step of a method is referred to as subsequent to another step, it will be further understood that, unless otherwise specified, it may directly follow the other step, or that another step or one or more intermediate steps may be performed prior to performing the particular step. Similarly, when a connection between structures or components is described, it will be understood that, unless otherwise specified, the connection may be achieved directly or through intermediate structures or components.

[0095]

[0096] A method for predicting the gas absorption performance of a mixed absorbent according to the purpose of the present invention was developed based on data on CO2 absorption using a mixed amine aqueous solution as an absorbent, and the following practical examples are also described within this scope.

[0097]

[0098] 1. Data acquisition

[0099] 1) Data collection

[0100] For a total of five amines, including Monoethanolamine (MEA), Diethanolamine (DEA), N-methyl diethanolamine (MDEA), 2-amino-2-methyl-1-propanol (AMP), and Piperazine (PZ), 2,328 VLE (vapor-liquid equilibrium) data of single amine absorbent-CO2 and 2,187 VLE data of mixed amine absorbent-CO2 for eight mixed amine combinations, including MEA / MDEA and DEA / MDEA, were collected from research papers and other literature.

[0101]

[0102] 2) Data preprocessing

[0103] Using Exploratory Data Analysis (EDA), which visualizes the distribution and patterns of VLE data, outliers were removed, resulting in 2,073 VLE data sets for single amine absorbent-CO2 and 1,548 VLE data sets for mixed amine absorbent-CO2.

[0104]

[0105] 2. Model Development

[0106] 1) Development of a single amine model

[0107] As shown in the upper yellow boxes of Fig. 1, using the standard DNN algorithm structure, temperature (T), CO₂ partial pressure (P) among the VLE data of a single amine CO₂ ), amine concentration (w amine = c amine ) as input variables and the CO2 loading of the single amine absorbent (α = absorbed CO₂ mol / total amine mol) as an output variable, a model for predicting the CO2 absorption performance of a single amine was developed.

[0108] The standard DNN algorithm consists of three hidden layers (each with 16 nodes) between the input and output variables, and uses the Log-sigmoid function to learn nonlinear relationships. The loss function uses MSE (Mean squared error), the learning rate is 0.001, the batch size is 8, and the optimizer uses Adam (Adaptive Moment Estimation).

[0109]

[0110] 2) Development of a mixed amine absorption performance prediction model (ABN: Amine Blending Network)

[0111] The mixed amine absorption performance prediction model (ABN) of the present invention was developed as a model to predict the VLE of mixed amines by integrating single amine models. That is, as shown in Fig. 1, the new ABN model is a mole-weighted average (α) of the predicted values ​​of the CO2 loading prediction model of the single amine absorbent. 12, avg ) was also added as input values, and a new DNN model structure was developed based on the VLE data of mixed amine absorbent-CO2. The ABN model was also configured as a model with two hidden layers (16 nodes) based on DNN.

[0112] Temperature (T), CO₂ partial pressure (P) among VLE data of mixed amine absorbent-CO2 CO₂ ), concentration of each amine (w amine1 , w amine2 = c amine1 , c amine2 ) was used as an input variable, and CO2 loading was used as an output variable.

[0113]

[0114] 3) Evaluation of the ABN model

[0115] The ABN model of the present invention uses the molar-weighted average CO2 loading value of a single amine solution as an input variable, so a single amine model with high accuracy is required. The performance of the single amine model is evaluated by MSE (mean square error) and R 2 It was evaluated by score and tested based on optimal hyperparameters.

[0116] Figure 2 shows a parity plot comparing the predicted and actual values ​​of the single-amine model for MEA (a) and MDEA (b). The x-axis represents the actual values, the y-axis represents the predicted values, and the red line indicates a perfect match between the predicted and actual values. The blue dots represent data points, and the green dotted line represents a 10% error. The high prediction accuracy of the single-amine model can be confirmed by the close distribution of the blue dots to the red line.

[0117] All single amine models had MSE values ​​less than 0.01 and R 2 The score was over 0.98, demonstrating high prediction accuracy. This demonstrates that this model can be reliably used in the ABN framework.

[0118]

[0119] The performance of the ABN according to the present invention was evaluated in two ways.

[0120] In the first evaluation, the model was iteratively developed and evaluated based on the initial weights and training dataset size. Typically, when data is scarce, the performance of data-driven models is significantly affected by initial weights. However, the model of the present invention, which includes the molar-weighted average CO2 loading value of a single amine solution as a guide feature, demonstrated stable prediction performance even in limited data environments.

[0121] Figure 3 shows the performance of a conventional model (as shown in Figure 4) and an ABN model (Hybrid model) according to the present invention, divided into seven cases in which the number of learning samples was sequentially increased by 10 from 20 to 80, and the results of 50 repeated experiments with different initial weights for each case are compared in a box plot. In each case, 30% of the data was used as validation data, and the experiment was terminated early when the validation loss no longer improved.

[0122] For reference, the general artificial neural network modeling method shown in Fig. 4 is a model developed using only mixed amine information without single amine information, unlike the ABN of the present invention, and is composed of two hidden layers (each with 16 nodes) between input variables and output variables, and uses a Log-sigmoid function for nonlinear relationship learning, uses MSE (Mean squared error) as the loss function, uses a learning rate of 0.001, a batch size of 2, and uses Adam (Adaptive Moment Estimation) as the optimizer.

[0123] Both models improved prediction accuracy as the training data size increased, but performance improvement plateaued after 60 training samples. This suggests that a minimum of 60 samples is required for robust model training. Specifically, when the dataset is very small, the conventional model exhibited a high standard deviation of the R2 score and an increased frequency of incorrect predictions, while the ABN model exhibited a significantly lower standard deviation, demonstrating greater robustness and consistent accuracy.

[0124] The second evaluation used K-fold cross-validation to compare model performance across training data distributions. K-fold cross-validation divides data into K folds, using one fold as test data and the remaining folds as training data. In this invention, we divided the dataset into six folds, using only one fold as training data, and the remaining folds as test data. We trained both the general model and the ABN model with the same initial weights across all folds to evaluate robustness, and the results are shown in the table below.

[0125] Cross-validation results of the general model and the ABN model. Evaluation Matrics (R) 2 )Conventional modelABN modelFold 1Fold 2Fold 3Fold 4Fold 5Fold 60.96830.97340.96130.06560.94280.87660.97430.97170.90280.96190.97270.9774Avg±Std0.7980±0.32910.9601±0.0261

[0126] R of two models in each fold 2 The scores calculated showed that both models showed high predictive performance in most folds, but the ABN model showed better results in terms of mean and standard deviation. In particular, in fold 4, the general model showed R 2The score was 0.0656, which showed a significantly low prediction performance. This is because the model produced a constant output in a state of insufficient training due to the vanishing gradient problem of the activation function (Log-sigmoid) when the input value was extreme. On the other hand, the ABN model did not suffer from this problem, and the vanishing gradient problem was alleviated and the robustness of the model was secured thanks to the guidance function using the molar-weighted average CO2 loading value. Meanwhile, the VLE data of mixed amines was predicted using the ABN model of the present invention, which was verified as described above. Figure 5 shows the VLE curve of the MEA / MDEA mixture, i.e., the relationship between the CO2 partial pressure and the CO2 loading. Here, the dots represent experimental data reported in the literature under the same conditions, and the solid line represents the predicted value using the ABN model of the present invention. The predicted curve was in good agreement with the literature data, demonstrating the high accuracy of the ABN model of the present invention.

[0127]

[0128] 3. Application of the ABN Model - Determination of the Optimal Mixed Amine Absorbent Composition

[0129] 1) Calculation of renewable energy

[0130] Using VLE data under simulated operating conditions of a stripping column of single and mixed amine absorbents absorbing CO2, the regeneration energy for stripping and separating CO2 was calculated for all amine blending ratios, with a blending ratio of one amine absorbent up to 40 wt% (Grid Search). The regeneration energy consists of heat of reaction, latent heat, and sensible heat, and the regeneration energy was calculated for a typical post-combustion carbon capture process consisting of an absorber and a stripper (see Fig. 6), using the heat energy balance equation of the stripper section and the Gibbs-Helmholtz equation, utilizing VLE data. The optimization objective function is to minimize the regeneration energy, and the optimal ratio of mixed amines was searched for assuming basic process conditions.

[0131] The specific renewable energy calculation formula is as follows.

[0132] [Formula 1]

[0133]

[0134]

[0135] Here,

[0136]

[0137] m am is the molar amount of absorbent (e.g., amine) per ton CO2 (mol / tonCO2),

[0138] m w is the molar amount of water per ton CO2 (mol / tonCO2),

[0139] r w is the molar ratio of water to the absorbent (e.g., amine),

[0140] α is the amount of CO2 loading per mole of absorbent (e.g., amine) (mol CO2 / mol amine )as,

[0141] α rich , α leanEach represents the value when the absorbent absorbs CO2 (rich) and removes it (lean),

[0142] The upper limits of the integral formulas in the H1 and H2 calculation formulas, 'rich' and 'lean', represent the values ​​of the variable α when the absorbent absorbs (rich) and removes (lean) CO2, respectively.

[0143] M is the amount of CO2 (2.2727 x 10 4 mol / tonCO2),

[0144] c am , c w , c CO2 are the heat capacities (kJ / mol K) of the absorbent (e.g., amine), water, and CO2, respectively,

[0145] Δh rxn is the heat of reaction absorption (kJ / molCO2),

[0146] λ is the latent heat of water (kJ / mol),

[0147] T1 is the temperature (K) of the absorbent flowing out of the CO2 absorption tower,

[0148] T2 is the temperature (K) of the absorbent flowing into the CO2 absorption tower,

[0149] T4 is the temperature (K) of the absorbent that enters the absorption tower again from the CO2 stripping tower condenser.

[0150] T R is the CO2 removal tower reboiler operating temperature (K),

[0151] T top is the temperature (K) at the top of the CO2 removal tower.

[0152]

[0153] The above temperature separator refers to the stream number of the CO2 absorption and removal system in Fig. 6. In this embodiment, T R Silver is 393K, T top was set to 373K.

[0154]

[0155] 2) Determination of the optimal mixed amine absorbent composition

[0156] By comparing the regeneration energy values ​​for stripping CO2 from a single amine absorbent and a mixed amine absorbent that absorbed CO2, the amine composition that has the minimum regeneration energy value was determined as the optimal amine mixing ratio, i.e., the optimal mixed amine composition ratio.

[0157] The regenerative energies of single amines and mixed amines in various ratios were calculated using a grid search method. The total amine mass ratio was limited to a maximum of 40% to ensure industrial applicability.

[0158] Figure 7 shows the results of calculating the regeneration energy according to the amine ratio of the MEA / MDEA absorbent. A higher amine mass ratio tended to result in a decrease in regeneration energy, and the amount of water in the absorbent significantly affected the regeneration energy. While reducing the water content can minimize the regeneration energy, this can pose a potential risk to process stability. Therefore, the regeneration energy of single and mixed amine absorbents was analyzed under conditions of a minimum water content of 60 wt% and, conversely, a maximum amine content of 40 wt%, to determine the optimal composition.

[0159] Figure 8 shows the regeneration energy and each heat contribution factor according to the amine mixing ratio when the total amine mass ratio is 40%. Reducing the MEA ratio and increasing the MDEA ratio significantly reduced the reaction heat, but the latent and sensible heat changes were relatively minimal.

[0160] Figure 9 is a graph comparing the minimum regeneration energies of single amines and mixed amines. While no synergistic effect was observed in the MEA / MDEA mixture, mixed amines generally required lower regeneration energies than single amines, and a synergistic effect was confirmed.

[0161] Specifically, comparing the graphs of MEA 40%, MDEA 40%, and MEA 4% + MDEA 36% in Fig. 9, the regeneration energy (3.16) of MEA 4% + MDEA 36% is calculated to be greater than that of MDEA 40% (3.13), but the regeneration energies of the remaining optimal mixed amines are all calculated to be smaller than the regeneration energies of a single amine. In particular, the MDEA / AMP mixture showed the lowest regeneration energy at 2.829 GJ / tCO2, which is about 31.5% lower than the required regeneration energy of the reference case, the MEA 30 wt% absorbent. This has confirmed the possibility of reducing the total energy consumption of the wet carbon capture process (PCC) by about 25% and significantly reducing the operating cost of the CO2 capture process.

[0162] The present invention utilizes a machine learning model to accurately predict the carbon capture performance of a mixed absorbent for wet carbon capture, and provides a framework for finding the optimal blending ratio that minimizes renewable energy. Specifically, the prediction performance of a mixed absorbent, for which measured data is insufficient, was achieved by using the molar-weighted average CO2 loading value of a single absorbent as training data. This model was proven to be 13% more accurate and robust than existing deep learning models.

[0163] This framework presents a novel approach to effectively utilize the scarce VLE data of mixed absorbents, which will shorten the time required to analyze the performance of gas absorbents, accelerate the development of cost-effective carbon capture processes, and contribute to global carbon emission reduction and technology commercialization.

Claims

1. a) A step of creating a prediction model for a specific gas absorption performance of a single absorbent by training an artificial neural network model with absorption performance experimental data of one or more single absorbents for a specific gas to predict the specific gas absorption performance of the single absorbent; b) A step of generating absorption performance prediction data of two or more mixed absorbents for the above specific gas using the concentration-weighted average formula below: (In the above formula, n is an integer representing the type of single absorbent in the mixed absorbent, c represents the concentration (wt%) of each single absorbent in the mixed absorbent, α is the absorption performance (mol) of each single absorbent in the mixed absorbent. 기체 / mol 흡수제 ) represents); and, c) a step of creating a prediction model for a specific gas absorption performance of a mixed absorbent by training the absorption performance prediction data of the mixed absorbent generated in step b) and the absorption performance experimental data of the mixed absorbent using an artificial neural network model to predict the specific gas absorption performance of the mixed absorbent; A method for predicting specific gas absorption performance of a mixed absorbent, comprising:

2. In paragraph 1, A method for predicting specific gas absorption performance of a mixed absorbent, characterized in that the artificial neural network model is a deep neural network (DNN).

3. In paragraph 1, The above specific gas absorption performance is the specific gas absorption amount (mol) at a specific temperature, partial pressure of a specific gas, and a specific type of absorbent and concentration in a specific solvent. 기체 / mol 흡수제 ) is characterized by comprising a method for predicting a specific gas absorption performance of a mixed absorbent.

4. In paragraph 1, A method for predicting the specific gas absorption performance of a mixed absorbent, characterized in that the specific gas is CO2 and the absorbent is an amine compound.

5. In paragraph 4, A method for predicting specific gas absorption performance of a mixed absorbent, characterized in that the amine compound is selected from the group consisting of MEA (Monoethanolamine), DEA (Diethanolamine), MDEA (N-methyl diethanolamine), AMP (2-amino-2-methyl-1-propanol), and PZ (Piperazine).

6. In paragraph 1, A method for predicting a specific gas absorption performance of a mixed absorbent, characterized in that the absorbent is in the form of an aqueous solution. 7.i) A single absorbent's specific gas absorption performance prediction unit that generates prediction data for a single absorbent's specific gas absorption performance using an artificial neural network model that has learned absorption performance experimental data of one or more single absorbents for a specific gas; ii) A weighted average data generation unit for a specific gas absorption performance of a mixed absorbent that generates absorption performance prediction data of two or more mixed absorbents for the specific gas using the concentration-weighted average formula below: (In the above formula, n is an integer representing the type of single absorbent in the mixed absorbent, c represents the concentration (wt%) of each single absorbent in the mixed absorbent, α is the absorption performance (mol) of each single absorbent in the mixed absorbent. 기체 / mol 흡수제 ) represents); and, iii) A mixed absorbent's specific gas absorption performance prediction data generation unit that generates the mixed absorbent's specific gas absorption performance prediction data using an artificial neural network model that has learned the mixed absorbent's absorption performance prediction data generated in ii) above and the mixed absorbent's absorption performance experimental data; A system for predicting specific gas absorption performance of a mixed absorbent, comprising:

8. In paragraph 7, A system for predicting specific gas absorption performance of a mixed absorbent, characterized in that the artificial neural network model is a deep neural network (DNN).

9. In paragraph 7, The above specific gas absorption performance is the specific gas absorption amount (mol) at a specific temperature, partial pressure of a specific gas, and a specific type of absorbent and concentration in a specific solvent. 기체 / mol 흡수제 ) is characterized by comprising a system for predicting the specific gas absorption performance of a mixed absorbent.

10. In paragraph 7, A system for predicting the specific gas absorption performance of a mixed absorbent, characterized in that the specific gas is CO2 and the absorbent is an amine compound.

11. In paragraph 10, A system for predicting specific gas absorption performance of a mixed absorbent, characterized in that the above amine compound is selected from the group consisting of MEA (Monoethanolamine), DEA (Diethanolamine), MDEA (N-methyl diethanolamine), AMP (2-amino-2-methyl-1-propanol), and PZ (Piperazine).

12. In paragraph 7, A system for predicting specific gas absorption performance of a mixed absorbent, characterized in that the absorbent is in the form of an aqueous solution. 13.d) A step of calculating prediction data for specific gas absorption performance of each composition of the mixed absorbent under each specific temperature and partial pressure condition of a specific gas for absorbing and removing a specific gas, according to any one of clauses 1 to 6; e) a step of calculating the regeneration energy of the mixed absorbent by calculating the heat of reaction, latent heat and sensible heat data based on the difference in the prediction data of the absorption performance of a specific gas under the absorption and stripping conditions of a specific gas calculated in step d) using the following mathematical formula; f) A method for determining an optimal mixed amine absorbent composition, comprising a step of selecting a mixed absorbent composition having the smallest value of regeneration energy among the mixed absorbent compositions calculated in step e). 14.iv) A specific gas absorption performance calculation unit for a mixed absorbent that calculates specific gas absorption performance data for each composition of the mixed absorbent under each specific temperature and partial pressure condition of a specific gas for absorbing and removing a specific gas according to any one of the 7th to 12th clauses; v) A regeneration energy calculation unit that calculates the regeneration energy of the mixed absorbent by calculating the heat of reaction, latent heat and sensible heat data based on the difference in the specific gas absorption performance data under the absorption and stripping conditions of the specific gas calculated in iv) using the following mathematical formula; vi) An optimal mixed amine absorbent composition determination system including an optimal mixed amine absorbent composition determination unit for selecting a mixed absorbent composition having the smallest value of regenerative energy among the mixed absorbent compositions calculated in v).

Images