A method and system for quickly predicting material gas adsorption performance based on artificial intelligence

Abstract

The present invention relates to the field of computational materials technology, and specifically to a method and system for quickly predicting the gas adsorption performance of materials based on artificial intelligence. The present invention innovatively adopts a descriptor that blends theory and practice, namely, a gas adsorption isotherm, combined with an innovatively proposed query sampling strategy. Only the adsorption data of a material on a gas obtained by experiment or simulation needs to be input to obtain the adsorption performance of the material on the remaining one or more gases, thereby reducing the R&D cost of the adsorption material and improving the R&D efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of materials science, and in particular to a method and system for rapidly predicting the gas adsorption performance of materials based on artificial intelligence. Background Art

[0002] The research and development of gas adsorption materials is of great significance to industries such as petrochemicals, environmental protection, and biomedicine. By innovating and improving the performance of adsorption materials, key gases can be more effectively processed and separated, such as carbon dioxide capture and storage, pollutant control, and purification of hydrogen and natural gas. These are key technologies for achieving clean energy transformation and sustainable development goals. Currently, the gas adsorption performance of materials can be evaluated through experiments and simulations. Experimental evaluation includes material synthesis, characterization, testing, etc., while simulation includes model building, force field selection, etc. Both methods have certain shortcomings, and it is necessary to establish more efficient material gas adsorption performance evaluation methods and systems to help scientific researchers reduce material research and development costs and improve research and development efficiency.

[0003] In terms of evaluating the gas adsorption performance of materials, both experimental and simulation methods have certain shortcomings. When evaluating by experimental methods, the entire process includes material synthesis, characterization, testing and other links. Especially when you want to examine the adsorption performance of multiple gases in a material, it involves multiple tests, which is lengthy and costly. When evaluating by simulation, you will face the difficulty of material modeling. Many materials have complex compositions and unclear rules, and it is impossible to parse out appropriate structural data for modeling, and thus it is impossible to carry out simulations. In addition, the accuracy of gas parameters plays a decisive role in whether the simulation results are reliable. However, currently only a small number of gas parameters are relatively accurate, which also limits the scope of application of simulation methods to evaluate the gas adsorption performance of materials. Summary of the invention

[0004] In response to the problems of long and costly experimental cycles, difficult simulation modeling, and limited application scope, the present invention innovatively adopts a descriptor that blends theory and practice, namely the gas adsorption isotherm, combined with an innovatively proposed query sampling strategy, and uses artificial intelligence technology to establish a more efficient material gas adsorption performance evaluation method and system, which effectively reduces the R&D cost of adsorption materials and improves R&D efficiency. This application solves the problems raised in the above background technology through a method and system for quickly predicting material gas adsorption performance based on artificial intelligence.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for quickly predicting the gas adsorption performance of a material based on artificial intelligence comprises the following steps:

[0007] S1: The user obtains the gas adsorption performance data of the material and determines the performance label to be predicted; then obtains the gas adsorption performance data of similar materials from the database for analysis and divides them into training set and validation set;

[0008] S2: Use the training set to build a prediction model, and use the validation set to verify it; the verified model is used to predict the gas adsorption performance of the material;

[0009] The data obtained from the database contains both data corresponding to the gas adsorption performance and data corresponding to the performance label to be predicted.

[0010] The gas adsorption performance refers to one of the adsorption amount, adsorption rate, and desorption rate for a specific gas type under specific parameter conditions; the performance label to be predicted refers to the gas adsorption performance under different parameter conditions and / or different gas types.

[0011] The different parameter conditions refer to different temperatures and / or different pressures.

[0012] The number of types of similar materials obtained from the database is greater than or equal to the number of types of materials obtained by the user.

[0013] Similar materials refer to materials of a higher concept that are classified by type among the materials obtained by users.

[0014] The gas type is selected from: methane, ethane, n-hexane, toluene, ethyl acetate, nitrogen, argon, carbon dioxide, ethylene, propylene, butene, methanol, ethanol, acetone, chloroform, benzene, xylene, carbon tetrachloride, ammonia, hydrogen sulfide, carbon monoxide, oxygen, ozone, chlorine, fluorine, bromine, methyl iodide, and styrene.

[0015] The type of the material is selected from: activated carbon, metal organic framework, covalent organic framework, molecular sieve, carbon nanotube, ceramic, porous coordination polymer.

[0016] The steps in the process of dividing the training set and the validation set include: randomly selecting a part of the data as the initial set, then picking out a part of the remaining data that has a higher similarity with the initial set, adding it to the initial set as the training set, and the rest as the validation set.

[0017] The similarity is calculated by adding the similarity of the normalized adsorption isotherms and the similarity coefficient of the saturated adsorption amount.

[0018] The similarity of the normalized adsorption isotherms is calculated from two sets of isothermal adsorption data by the following formula:

[0019]

[0020]

[0021] Among them, ρ i is the similarity of the normalized adsorption isotherms, ρ i,j is the Pearson correlation coefficient of the two sets of data, cov(i,j) is the covariance of the two sets of data, σ i and σ j are the standard deviations of the two sets of data, E is the expectation, i and j are the two sets of data, μ i and μ j is the expectation of the two sets of data, and n is the number of samples in the training set.

[0022] The similarity coefficient of saturated adsorption capacity is calculated according to the following formula:

[0023]

[0024]

[0025] Among them, φ i is the similarity coefficient of saturated adsorption capacity, MAX is the maximum value, Q i and Q j are the saturated adsorption capacities of groups i and j, respectively, and ΔQ i,j is the difference in saturated adsorption capacity, and n is the number of samples in the training set.

[0026] The data obtained by the user and / or the data obtained from the database are fitted by the gas adsorption equation and the adsorption performance data under other conditions are generated.

[0027] The method also includes a step of judging the integrity of the data obtained by the user, which is to fit the adsorption data through the gas adsorption equation to determine whether the fitting degree reaches the threshold requirement. If so, it is judged to be complete.

[0028] The satisfying condition is to determine whether the adsorption performance data can match the common adsorption curve, that is, whether a relatively smooth curve can be formed. The determination method can be to fit the adsorption data with the common adsorption equation to determine whether the degree of fit meets a certain value; the gas adsorption equation is selected from the Langmuir equation, BET equation, Freundlich equation, Temkin equation, Dubinin-Radushkevich equation, Toth equation, Sips equation, and Zeta adsorption model; the degree of fit is measured by one of the coefficient of determination, mean square error, root mean square error, mean absolute error, mean absolute percentage error, residual, AIC and BIC.

[0029] A system for rapidly predicting the gas adsorption performance of materials based on artificial intelligence, comprising:

[0030] A user terminal, used for a user to obtain gas adsorption performance data of a material and determine a performance label to be predicted;

[0031] Data storage and processing module, used to extract gas adsorption performance data of similar materials and divide them into training set and verification set;

[0032] The AI ​​training and prediction module is used to build a prediction model using a training set and verify it using a validation set;

[0033] The data obtained from the user and / or the data obtained from the database contain data of the performance label to be predicted.

[0034] The gas adsorption performance refers to one of the adsorption amount, adsorption rate, and desorption rate for a specific gas type under specific parameter conditions; the performance label to be predicted refers to the gas adsorption performance under different parameter conditions and / or different gas types.

[0035] The user terminal also includes a judgment unit for judging the integrity of the data obtained by the user, by fitting the adsorption data through the gas adsorption equation to determine whether the fitting degree meets the threshold requirement, and if so, it is judged to be complete.

[0036] The data storage and processing module also includes a secondary acquisition unit for fitting the data obtained by the user and / or the data obtained from the database through the gas adsorption equation and generating adsorption performance data under other conditions.

[0037] The data storage and processing module also includes a training set-test set partitioning unit, which is used to randomly select a part of the data as the initial set, and then select a part of the remaining data that has a higher similarity with the initial set, and add it to the initial set as the training set, and the rest as the verification set.

[0038] The artificial intelligence training and prediction module also includes a model evaluation unit for comparing the performance of models trained using different artificial intelligence models.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. The present invention defines a sampling strategy based on the actual situation of material gas adsorption descriptors. Based on the sampling strategy, the training set can be optimized, thereby reducing the cost of artificial intelligence training and significantly improving the accuracy of the artificial intelligence model;

[0041] 2. Different from the descriptor types commonly used in the prior art, the present invention adopts a theoretical and practical descriptor, namely, the gas adsorption isotherm, for training artificial intelligence models. The descriptor is theoretically calculable and easy to obtain experimentally. The present invention finally realizes that only the adsorption data of a material on one gas obtained by experiment or simulation can be input, and the adsorption performance of the material on the remaining one or more gases can be obtained through the method and system, thereby reducing the R&D cost of the adsorption material and improving the R&D efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a structural schematic diagram of a system for rapidly predicting gas adsorption performance of materials based on artificial intelligence provided by one embodiment of the present invention;

[0043] Figure 2 It is a schematic diagram of a user terminal in a system for rapidly predicting gas adsorption performance of materials based on artificial intelligence provided by an embodiment of the present invention;

[0044] Figure 3 It is a schematic diagram of a data storage and processing module in a system for rapidly predicting gas adsorption performance of materials based on artificial intelligence provided by one embodiment of the present invention;

[0045] Figure 4 It is a schematic diagram of an artificial intelligence training and prediction module in a system for quickly predicting gas adsorption performance of materials based on artificial intelligence provided by one embodiment of the present invention;

[0046] Figure 5 This is a flow chart of a method for quickly predicting the gas adsorption performance of a material based on an artificial intelligence service system provided by an embodiment of the present invention;

[0047] Figure 6 1 is a comparison of the effects before and after the sampling strategy is applied to optimize the training set according to an embodiment of the present invention, wherein (a) is the true value, (b) is the prediction result of the model before the sampling strategy is applied to optimize the training set, and (c) is the prediction result of the model after the sampling strategy is applied to optimize the training set. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] The present invention provides a method and system for quickly predicting the gas adsorption performance of materials based on artificial intelligence. At the same time, combined with the actual situation of gas adsorption isotherms as descriptors, an innovative new sampling strategy is defined, based on which the training set can be optimized in one step, thereby improving the accuracy of the artificial intelligence model, and ultimately achieving the purpose of accurate prediction, reducing material R&D costs, and improving R&D efficiency.

[0050] like Figure 1 As shown, an embodiment of the present invention provides an artificial intelligence service system, including: a user terminal 101, a data storage and processing module 102, and an artificial intelligence training and prediction module 103. The user terminal 101 is connected to the data storage and processing module 102 and the artificial intelligence training and prediction module 103 at the same time; the data storage and processing module 102 is connected to the user terminal 101 and the artificial intelligence training and prediction module 103 at the same time; the artificial intelligence training and prediction module 103 is connected to the user terminal 101 and the data storage and processing module 102 at the same time; the user terminal 101 is used for data collection, data evaluation, human-computer interaction, and sending information to the data storage and processing module 102, and receiving data output and display from the artificial intelligence training and prediction module 103; the data storage and processing module 102 is used for data storage, data extraction, data fitting, secondary data collection, training set and test set division, and passing the data to the artificial intelligence training and prediction module 103; the artificial intelligence training and prediction module 103 is used for training the artificial intelligence model, realizing the prediction of one or more adsorption properties of the target material, and transmitting the data to the user terminal 101 for display and output.

[0051] Preferably, if Figure 2 As shown, in one embodiment of the present invention, the user terminal 101 includes: an information collection unit 1011, a judgment unit 1012, a feedback unit 1013 and a display unit 1014; wherein the information collection unit 1011 is used to collect gas adsorption data of materials from users, and determine one or more gas adsorption performance labels that are desired to be predicted, where the gas adsorption performance may refer to adsorption amount, working capacity, desorption rate, etc.; the judgment unit 1012 is used to determine whether the data meets the conditions, that is, whether it can be smoothly connected into a curve; the feedback unit 1013 is used to feed back data supplement suggestions to the user or to transmit data that meets the conditions to the data storage and processing module 102; the display unit 1014 is used to receive data from the artificial intelligence training and prediction module 103, and display the data.

[0052] Preferably, the data storage and processing module 102, such as Figure 3As shown, it includes: a data storage unit 1021, a data extraction unit 1022, a data fitting unit 1023, a fitting evaluation unit 1024, a secondary acquisition unit 1025 and a training set-test set division unit 1026; wherein the data storage unit 1021 is used to store adsorption data collected from the user terminal and a large amount of existing data for artificial intelligence model training; the data extraction unit 1022 is used to extract the corresponding data information of the same type of materials in the database and the corresponding data information corresponding to the gas adsorption performance label that is expected to be predicted (for example, it can be the adsorption data and adsorption performance label of the gas under the same temperature condition). The data obtained by the user may include data of a certain adsorption performance under one parameter condition, and the corresponding adsorption performance data under another parameter condition may be predicted, or it may be another adsorption performance data under the same parameter condition (or a combination of the two). In addition to the information corresponding to the performance data obtained by the user, the database also includes data corresponding to the performance label that the user wants to predict, thereby realizing the association between data results of different parameter conditions or different performance labels. The materials in the database here can be materials of the same type (i.e., superordinate concepts) as the materials obtained by the user. For example, the user obtains data on a specific MOF material, and the similar materials in the database refer to all MOF materials, which ensures that the number of types of similar materials obtained from the database is greater than or equal to the number of types of materials obtained by the user; the data fitting unit 1023 is used to fit the extracted adsorption data using multiple methods; the fitting evaluation unit 1024 is used to compare the fitting effects of multiple fitting methods to provide a basis for the selection of the fitting method; the secondary collection unit 1025 is used to carry out secondary collection of data; the training set-test set division unit 1026 divides the training set and the test set according to the sampling strategy defined in the present invention, and transmits them to the artificial intelligence training and prediction module 103.

[0053] Preferably, the artificial intelligence training and prediction module 103, such as Figure 4 As shown, it includes: a model training unit 1031, a model evaluation unit 1032, an adsorption performance prediction unit 1033 and a result output unit 1034; wherein the model training unit 1031 is used to adopt a variety of different artificial intelligence models to perform training based on the input adsorption data; the model evaluation unit 1032 is used to evaluate the training results of a variety of different artificial intelligence models, select the optimal model, and pass it to the adsorption performance prediction unit 1033; the adsorption performance prediction unit 1033 is used to predict one or more adsorption properties of the target material; the result output unit 1034 is used to package the prediction results and transmit them to the user terminal for display and output.

[0054] In the above embodiment, the data storage unit 1021 stores more than 150,000 material structures and corresponding gas adsorption isotherm data. The gas types include CO2 (about 710,000), N2 (about 380,000), CH4 (about 850,000), H2 (about 280,000), and n / i-butane (about 24,000). The stored material types include metal organic framework materials (MOFs, about 150,000 types), covalent organic framework materials (811 types), zeolites (216 types), and amorphous porous carbon materials (614 types).

[0055] In a second aspect, the present invention provides a method for quickly predicting the gas adsorption performance of a material using any of the artificial intelligence service systems described in the first aspect, comprising: step 501, using the user terminal to collect material gas adsorption data and determine one or more gas adsorption performance labels to be predicted; step 502, using the judgment unit in the user terminal to determine whether the data meets the conditions, where the conditions are generally to determine whether the adsorption performance data can match the common adsorption curve, that is, whether a relatively smooth curve can be formed, and the determination method can be to fit the adsorption data with the common adsorption equation to determine whether the fit meets a certain value; the common adsorption equations here can refer to: Langmuir equation, BET equation, Freundlich equation, Temkin equation, Dubinin-Radushkevich equation, Toth equation, Sips equation, Zeta adsorption model, etc.; the fit here can use some common indicators, such as: determination coefficient (R²), mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAE), mean absolute percentage error (MAE), and so on. MAPE), residuals, AIC (Akaike Information Criterion), and BIC (Bayesian Information Criterion).Step 503, determine whether the user needs to supplement data based on the judgment result. The judgment criterion here can be achieved by setting a threshold for the degree of fit, such as R²>0.8, MAE<0.1, etc., and the threshold can be adjusted in combination with factors such as material, temperature, pressure, gas type, etc. If the condition is not met, the user is fed back through the feedback unit and is required to supplement more adsorption data under pressure (the target pressure point can be calculated by dichotomy). If the condition is met, it is transmitted to the data storage and processing module; Step 504, the adsorption data collected from the user terminal is stored using the data storage unit in the data storage and processing module; Step 505, the adsorption data and adsorption performance labels of the gas under the same temperature conditions of the same material in the database are extracted using the data extraction unit in the data storage and processing module (the same material here can refer to the upper type of the material relative to the user when collecting data, such as the same material of a specific MOF material refers to all MOF materials); Step 506, the data fitting unit in the data storage and processing module is used to fit the provided data. The adsorption data obtained are fitted with multiple methods; step 507, the fitting evaluation unit in the data storage and processing module is used to select the optimal fitting method; step 508, the secondary collection unit in the data storage and processing module is used to complete the secondary collection of data; step 509, the training set and the test set are divided by an innovatively defined sampling strategy, and the sets are passed into the artificial intelligence training and prediction module; step 510, the model training unit in the artificial intelligence training and prediction module is used to train multiple different artificial intelligence models on the adsorption data passed in; step 511, the model evaluation unit in the artificial intelligence training and prediction module is used to evaluate the training results of multiple different artificial intelligence models, select the optimal model, and pass it into the adsorption performance prediction unit; step 512, the adsorption performance prediction unit in the artificial intelligence training and prediction module is used to predict one or more adsorption properties of the target material; step 513, the result output unit in the artificial intelligence training and prediction module is used to package the prediction results and transmit them to the user terminal for display and output.

[0056] The present invention defines a new sampling method, which is deployed in the training set-test set division unit in the data storage and processing module. Based on this sampling method, the training set can be optimized in one step before the model training starts. Specifically, it includes three steps: 1. Process the incoming isotherm data one by one, extract the saturated adsorption amount of each isotherm, and normalize the isotherm at the same time; 2. Analyze the gas adsorption performance data of the material obtained from the database, divide it into a training set and a verification set, and randomly select 60% of the materials to form an initial set, and compare the adsorption data of the remaining 40% of the materials with the adsorption data of each material in the initial set one by one to calculate the similarity; 3. Sort the similarity from low to high, select the materials in the first half (i.e. 20% of the total), add their data to the initial set, form an optimized training set (accounting for 80% of the total), and the rest is used as a test set and passed to the artificial intelligence training and prediction module. It is worth noting that the proportion of material division in the process of optimizing the data set is not fixed and can be adjusted according to actual conditions.

[0057] Fourthly, the key to the innovative sampling method proposed by the present invention lies in similarity calculation. For a given material i, its adsorption data similarity with the materials in the initial set (a total of n materials) includes two parts: 1. The similarity of the normalized adsorption isotherm, i.e. i ; 2. Similarity coefficient of saturated adsorption capacity, i.e. φ i ; Then the sum of the two is taken as the comparison value of similarity. The similarity here takes into account the two aspects of the adsorption isotherm, namely the linear shape and the saturated adsorption capacity, both of which are indispensable.

[0058] The similarity 𝜌i of the normalized adsorption isotherms is calculated as follows:

[0059]

[0060]

[0061] Among them, ρ i is the similarity of the normalized adsorption isotherms, ρ i,j is the Pearson correlation coefficient of the two sets of data, cov(i,j) is the covariance of the two sets of data, σ i and σ j are the standard deviations of the two sets of data, E is the expectation, i and j are the two sets of data, μ i and μ j is the expectation of the two sets of data, and n is the number of samples in the training set.

[0062] Similarity coefficient of saturated adsorption capacity φ i The calculation method is as follows:

[0063]

[0064]

[0065] Among them, φ i is the similarity coefficient of saturated adsorption capacity, MAX is the maximum value, Q i and Q j are the i and j groups of data, ΔQ i,j is the difference in saturated adsorption capacity, and n is the number of samples in the training set.

[0066] To facilitate understanding, the following describes the methods and processes using four examples based on a variety of materials and gas types.

[0067] The embodiments of the present invention have at least the following beneficial effects: 1. The source of input data is not limited, and it can come from experiments or simulations. Users only need to input one isotherm data of a material to obtain the other one or more gas adsorption properties, without the need for experimental testing or theoretical simulation of the material's adsorption properties for multiple gases, which greatly saves time and cost. 2. In the embodiments of the present invention, gas adsorption isotherms are used as descriptors, and such descriptors can be obtained by experimental simulations, which cleverly avoids the difficulty of material modeling. By measuring the adsorption isotherms of some small molecule gases, the adsorption properties of materials for gases with complex parameters, under extreme conditions, or even toxic gases are predicted based on artificial intelligence, which can not only avoid the limitation of the scope of simulation application, but also ensure safety. Example 1

[0068] Select MOF material, the user inputs the adsorption isotherm data of Ar at 77 K, and is required to predict the adsorption amount of n / i-butane by the material at room temperature and 3e+05 Pa pressure. First, the user inputs the Ar adsorption amount of a MOF material at 1e-02, 1e-01, 1e-00, 1e+01, 1e+02, 1e+03, 1e+04, 1e+05 Pa pressure points at 77K through the user terminal, which are 0.18, 1.64, 11.86, 43.90, 191.60, 597.22, 1127.93, 1239.15 cm 3 / g, and we hope to predict the adsorption amount of n / i-butane by the material at room temperature and 3e+05 Pa pressure. The above input Ar adsorption isotherm data are sufficient and can be connected into a smooth curve. The Langmuir equation is used for fitting, and the goodness of fit R 2The value is greater than 0.8, which meets the requirements of the subsequent process. Therefore, the data is transferred to the data storage and processing module, and the isothermal adsorption data of Ar for all MOF materials in the database at 77 K are extracted (finally including about 12,000 materials and about 100,000 data items), as well as the adsorption amount of n / i-butane at room temperature and 3e+05 Pa pressure (the number of label values ​​is about 24,000); the extracted Ar isothermal adsorption data are fitted by Langmuir, Quadratic, Brunauer-Emmett-Teller, and Dual-site Langmuir fitting methods. Finally, according to the principle of minimum determination coefficient, the Dual-site Langmuir fitting method is selected, and the secondary data collection is completed based on this. In this example, the Ar adsorption isotherms of all MOF materials in the training set are collected at 1e-03, 1e-02, 1e-01, 1e+00, 1e+01, 1e+02, 1e+03, 2e+03, The values ​​at a total of 29 pressure points (4e+03,6e+03,8e+03, 1e+04, 1.2e+04,1.4e+04, 1.6e+04, 1.8e+04, 2e+04, 2.4e+04, 2.8e+04, 3e+04, 3.2e+04, 3.6e+04,4e+04, 5e+04, 6e+04, 7e+04, 8e+04, 9e+04, 1e+05Pa) are used as descriptors for subsequent artificial intelligence training. It is worth noting that the selection of pressure points is not fixed when collecting data, but it is necessary to ensure that the pressure points of the adsorption data of all materials are consistent.

[0069] Of all the extracted MOF materials, 500 (approximately 4.17% of the extracted MOF materials) were randomly selected to form an initial set. The adsorption data of the remaining MOF materials were compared one by one with the adsorption data of each material in the initial set, and the similarity was calculated. The similarities were then sorted from low to high, and the top 200 materials were selected and added to the initial set to form an optimized training set (700 MOF materials). The remaining MOF materials (approximately 11,000) were used as a test set.

[0070] The collected data and the corresponding n / i-butane adsorption amount labels are passed into the artificial intelligence training and prediction module. A total of 7 different artificial intelligence models, including Decision Tree, Linear Regression, SVR, K-neighbors, Random Forest, XGB, and NeuralNetwork, are used for training based on the input adsorption data. The optimal model is selected based on the principle of minimum determination coefficient. In this case, the optimal model is Neural Network, and the model is passed to the adsorption performance prediction unit to predict the adsorption amount of n / i-butane on the material at room temperature and 3e+05 Pa pressure. The results are 15.15 and 14.83 mmol / g, respectively. The prediction results are packaged and transmitted to the user terminal for display and output.

[0071] Figure 6 The comparison of the effect before and after applying the sampling strategy to optimize the training set is shown. Figure 6 (a) in is the true value, Figure 6 (b) in the figure is the prediction result of the model before applying the sampling strategy to optimize the training set. Figure 6 (c) in the figure is the prediction result of the model after applying the sampling strategy to optimize the training set. It can be seen that the prediction result of the model before applying the sampling strategy to optimize the training set is inaccurate ( Figure 6 In (b) of Figure 1, the main features are as follows: (1) the n / i-butane adsorption selectivity is negative, and (2) the pore size interval where the n / i-butane adsorption selectivity is maximum is inconsistent with the actual situation. In contrast, after applying the sampling strategy to optimize the training set, the n / i-butane adsorption selectivity is positive, and the pore size interval where the adsorption selectivity is maximum is highly consistent with the actual situation, which proves the advantage of our sampling strategy.

[0072] It is worth noting that there are various fitting methods and artificial intelligence models, and they are not limited to those listed in the embodiments; the evaluation criteria for the fitting effect and the training effect of the artificial intelligence model are also not fixed, and are not limited to the principle of the minimum determination coefficient in the demonstration example. Example 2

[0073] Anion column hybrid MOF material is a special type of MOF material. The anion column in its structure can be used as a selective adsorption site, which makes the material have great potential in the field of gas separation. In this embodiment, anion column hybrid MOF material is selected. The user inputs the adsorption isotherm data of CH4 at room temperature, and at the same time requires the prediction of the working ability of the material to adsorb CO2 in the pressure range of 1e+05 to 1e+07 Pa at room temperature. First, the user inputs the CH4 adsorption amount of an anion column hybrid MOF material at 1e+03, 1e+04, 5e+04, 1e+05, 5e+05, 1e+06, 5e+06, 1e+07 Pa pressure points at 298K through the user terminal, which are 0.048, 0.44, 1.59, 2.39, 4.15, 4.69, 5.46, 5.65 mmol / g, respectively, hoping to predict the working ability of the material to adsorb CO2 in the pressure range of 1e+05 to 1e+07 Pa at room temperature. The CH4 adsorption isotherm data input above are sufficient and can be connected into a smooth curve. The Langmuir equation is used for fitting. The goodness of fit R 2 It is greater than 0.8, which meets the requirements of the subsequent process. Therefore, the data is transferred to the data storage and processing module, and the isothermal adsorption data of CH4 of all anion column hybrid MOF materials in the database at 298K temperature conditions are extracted (the sample contains about 1000 materials and the data volume is about 8000), as well as the working capacity of materials for CO2 adsorption in the pressure range of 1e+05 to 1e+07 Pa at room temperature (the sample label data volume is about 1000); the extracted CH4 isothermal adsorption data are fitted by Langmuir, Quadratic, Brunauer-Emmett-Teller, and Dual-site Langmuir. Finally, according to the principle of minimum determination coefficient, Dual-site isothermal adsorption ... Langmuir fitting method is used as the basis for secondary data collection. In this example, the values ​​of CH4 adsorption isotherms of all materials at 18 pressure points, including 1e+00, 1e+01, 1e+02, 1e+03, 5e+03, 1e+04, 3e+04, 5e+04, 7e+04, 1e+05, 3e+05, 5e+05, 7e+05, 1e+06, 3e+06, 5e+06, 7e+06, and 1e+07 Pa, are collected as descriptors for subsequent artificial intelligence training. It is worth noting that the selection of pressure points is not fixed when collecting data, but it is necessary to ensure that the pressure points of the adsorption data of all materials are consistent.

[0074] 60% of the anion column hybrid MOF materials were randomly selected to form an initial set, and the adsorption data of the remaining 40% of the anion column hybrid MOF materials were compared with the adsorption data of each material in the initial set one by one to calculate the similarity; then the similarities were sorted from low to high, and the first half of the materials, i.e. 20% of the total, were selected, and their data were added to the initial set to form an optimized training set, accounting for 80% of the total, and the rest were used as test sets.

[0075] The collected data and the corresponding CO2 adsorption capacity labels are passed into the artificial intelligence training and prediction module. Seven different artificial intelligence models, including Decision Tree, Linear Regression, SVR, K-neighbors, Random Forest, XGB, and NeuralNetwork, are used for training based on the input adsorption data. The optimal model is selected based on the principle of minimum determination coefficient. In this case, the optimal model is Neural Network, and is passed to the adsorption performance prediction unit to predict the CO2 adsorption capacity of the material in the pressure range of 1e+05 to 1e+07 Pa at 298 K. The result is 1.40 mmol / g. The prediction results are packaged and transmitted to the user terminal for display and output. Example 3

[0076] Taking zeolite as an example, the user inputs the adsorption isotherm data of H2 at 77 K, and requires to predict the adsorption amount of H2 by the material at 273 K and 4e+07 Pa. First, the user inputs the H2 adsorption amount of a zeolite material at 1e+05, 2.71e+05, 7.39e+05, 2e+06, 3e+06, 5.5e+06, and 1.5e+07 Pa pressure points at 77 K through the user terminal, which are 15.52, 21.55, 25.76, 27.99, 28.44, 28.81, and 29.26 g / l, respectively. The user hopes to predict the adsorption capacity of H2 by the material at 273 K and 4e+07 Pa. The above-mentioned H2 adsorption isotherm data are sufficient and can be connected into a smooth curve. The dual-site Langmuir equation is used for fitting, and the goodness of fit R 2It is greater than 0.8, which meets the requirements of the subsequent process. Therefore, the data is transmitted to the data storage and processing module. At the same time, the isothermal adsorption data of H2 of all zeolite materials in the database at 77 K are extracted (the sample contains about 200 materials and the data volume is about 1500), as well as the adsorption amount of H2 of materials at 273 K and 4e+07 Pa (the sample label data volume is about 200); the extracted H2 isothermal adsorption data are fitted by Langmuir, Quadratic, Brunauer-Emmett-Teller, and Dual-site Langmuir. Finally, according to the principle of minimum determination coefficient, the Langmuir fitting method is selected, and the secondary data collection is completed on this basis. In this example, the H2 adsorption isotherms of all materials are collected at 1e+05, 3e+05, 5e+05, 7e+05, 1e+06, 3e+06, 5e+06, 7e+06, and 1e+07 Pa, a total of 9 pressure points, as descriptors for subsequent artificial intelligence training.

[0077] 60% of the zeolite materials were randomly selected to form an initial set, and the adsorption data of the remaining 40% of the zeolite materials were compared with the adsorption data of each material in the initial set one by one to calculate the similarity; then the similarities were sorted from low to high, and the first half of the materials, i.e. 20% of the total, were selected, and their data were added to the initial set to form an optimized training set, accounting for 80% of the total, and the rest were used as test sets.

[0078] The collected data and the corresponding H2 adsorption labels are passed into the artificial intelligence training and prediction module. Seven different artificial intelligence models, including Decision Tree, Linear Regression, SVR, K-neighbors, Random Forest, XGB, and NeuralNetwork, are used for training based on the input adsorption data. The optimal model is selected based on the principle of minimum determination coefficient. In this case, the optimal model is Random Forest, and the model is passed into the adsorption performance prediction unit to predict the H2 adsorption of the material at a temperature of 273 K and a pressure of 4e+07 Pa. The result is 6.19 g / l. The prediction results are packaged and transmitted to the user terminal for display and output. Example 4

[0079] Taking amorphous materials as an example, the user inputs the adsorption isotherm data of N2 at 77 K, and requires to predict the adsorption amount of Ar on the material at 77 K and 1e+05 Pa. First, the user inputs the N2 adsorption amount of an amorphous material at 77 K at 1e-1, 1e+00, 1e+01, 5e+01, 1e+02, 5e+02, 1e+03, 2e+03, 5e+03, 1e+04, 2e+04, 5e+04, and 1e+05 Pa pressure points through the user terminal, which are 12.16, 16.71, 21.47, 25.61, 27.07, 31.76, 34.21, 35.53, 40.60, 42.29, 46.25, 47.97, and 49.58 cm 3 / g, hoping to predict the adsorption capacity of the material for Ar at 77 K and 1e+05Pa. The above input N2 adsorption isotherm data are sufficient and can be connected into a smooth curve. The dual-site Langmuir equation is used for fitting. The goodness of fit R 2 It is greater than 0.8, which meets the requirements of the subsequent process. Therefore, the data is transferred to the data storage and processing module. At the same time, the isothermal adsorption data of N2 of all amorphous materials in the database at 77K temperature conditions are extracted (the sample contains about 600 materials and the data volume is about 8000), as well as the adsorption amount of Ar of materials at 77K temperature and 1e+05 Pa pressure (the sample label data volume is about 600); the extracted N2 isothermal adsorption data are fitted by Langmuir, Quadratic, Brunauer-Emmett-Teller, and Dual-site Langmuir fitting methods. Finally, according to the principle of minimum determination coefficient, the Dual-site Langmuir fitting method is selected, and the secondary data collection is completed on this basis. In this example, the N2 adsorption isotherms of all materials are collected at 1e+00, 1e+01, 3e+01, 1e+02, 3e+02, 1e+03, 3e+03, 1e+04, 3e+04, 1e+05 Pa, the values ​​at a total of 10 pressure points are used as descriptors for subsequent artificial intelligence training.

[0080] 60% of amorphous materials were randomly selected to form an initial set, and the adsorption data of the remaining 40% amorphous materials were compared with the adsorption data of each material in the initial set one by one to calculate the similarity; then the similarities were sorted from low to high, and the first half of the materials, i.e. 20% of the total, were selected, and their data were added to the initial set to form an optimized training set, accounting for 80% of the total, and the rest were used as test sets.

[0081] The collected data and the corresponding Ar adsorption labels were passed into the artificial intelligence training and prediction module. Seven different artificial intelligence models, including Decision Tree, Linear Regression, SVR, K-neighbors, Random Forest, XGB, and NeuralNetwork, were used for training based on the input adsorption data. The optimal model was selected based on the principle of minimum determination coefficient. In this case, the optimal model was Random Forest, and passed to the adsorption performance prediction unit to predict the Ar adsorption of the material at 77 K and 1e+05 Pa. The result was 60.86 cm 3 / g, packages the prediction results and transmits them to the user terminal for display and output.

[0082] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting from any point of view, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention, and any reference numerals in the claims should not be regarded as limiting the claims involved.

Claims

1. A method for quickly predicting the gas adsorption performance of materials based on artificial intelligence, characterized in that: The following steps are involved: S1: The user obtains the gas adsorption performance data of the material and determines the performance label to be predicted; then obtains the gas adsorption performance data of similar materials from the database and divides it into a training set and a validation set; S2: Use the training set to build a prediction model, and use the validation set to verify it; the verified model is used to predict the gas adsorption performance of the material; The data obtained from the database contains both the data corresponding to the gas adsorption performance and the data corresponding to the performance label to be predicted; In the process of dividing the training set and the validation set, the steps include: randomly selecting a part of the data as the initial set, then selecting a part of the remaining data that has a higher similarity with the initial set and adding it to the initial set as the training set, and the rest as the validation set; The similarity is calculated by adding the similarity of the normalized adsorption isotherms and the similarity coefficient of the saturated adsorption amount; The similarity of the normalized adsorption isotherms is calculated from two sets of isothermal adsorption data by the following formula: ； ； Among them, ρ i is the similarity of the normalized adsorption isotherms, ρ i,j is the Pearson correlation coefficient of the two sets of data, cov(i,j) is the covariance of the two sets of data, σ i and σ j are the standard deviations of the two sets of data, E is the expectation, i and j are the two sets of data, μ i and μ j is the expectation of the two sets of data, and n is the number of samples in the training set; The similarity coefficient of saturated adsorption capacity is calculated according to the following formula: ； ； φ i is the similarity coefficient of saturated adsorption capacity, MAX is the maximum value, Q i and Q j are the saturated adsorption capacities of groups i and j, respectively, and ΔQ i,j is the difference in saturated adsorption capacity, and n is the number of samples in the training set.

2. The method for rapidly predicting the gas adsorption performance of materials based on artificial intelligence according to claim 1 is characterized in that: The gas adsorption performance refers to one of the adsorption amount, adsorption rate, and desorption rate of a set gas type under set parameter conditions; the performance label to be predicted refers to the gas adsorption performance under different parameter conditions and / or different gas types; The different parameter conditions refer to different temperatures and / or different pressures; The number of types of similar materials obtained from the database is greater than or equal to the number of types of materials obtained by the user.

3. The method for rapidly predicting the gas adsorption performance of materials based on artificial intelligence according to claim 1 is characterized in that: The gas type is selected from: methane, ethane, n-hexane, toluene, ethyl acetate, nitrogen, argon, carbon dioxide, ethylene, propylene, butene, methanol, ethanol, acetone, chloroform, benzene, xylene, carbon tetrachloride, ammonia, hydrogen sulfide, carbon monoxide, oxygen, ozone, chlorine, fluorine, bromine, methyl iodide, styrene; The type of the material is selected from: activated carbon, metal organic framework, covalent organic framework, molecular sieve, carbon nanotube, ceramic, porous coordination polymer.

4. The method for rapidly predicting material gas adsorption performance based on artificial intelligence according to claim 1, characterized in that: The data obtained by the user and / or the data obtained from the database are fitted by the gas adsorption equation and the adsorption performance data under other conditions are generated.

5. The method for rapidly predicting the gas adsorption performance of materials based on artificial intelligence according to claim 1, characterized in that: The method also includes a step of judging the integrity of the data obtained by the user, which is to fit the adsorption data through the gas adsorption equation to determine whether the fitting degree reaches the threshold requirement. If so, it is judged to be complete.

6. The method for rapidly predicting material gas adsorption performance based on artificial intelligence according to claim 5, characterized in that: The gas adsorption equation is selected from one of the Langmuir equation, BET equation, Freundlich equation, Temkin equation, Dubinin-Radushkevich equation, Toth equation, Sips equation or Zeta adsorption model; the fit is measured by one of the coefficient of determination, mean square error, root mean square error, mean absolute error, mean absolute percentage error, residual, AIC or BIC.

7. A system for rapidly predicting the gas adsorption performance of materials based on artificial intelligence, characterized in that: include: A user terminal (101), used for a user to obtain gas adsorption performance data of a material and determine a performance label to be predicted; A data storage and processing module (102) is used to extract gas adsorption performance data of similar materials and divide them into a training set and a verification set; An artificial intelligence training and prediction module (103), used to construct a prediction model using a training set and to verify it using a validation set; The data obtained from the database contains data corresponding to the performance label to be predicted; In the process of dividing the training set and the validation set, the steps include: randomly selecting a part of the data as the initial set, then selecting a part of the remaining data that has a higher similarity with the initial set and adding it to the initial set as the training set, and the rest as the validation set; The similarity is calculated by adding the similarity of the normalized adsorption isotherms and the similarity coefficient of the saturated adsorption amount; The similarity of the normalized adsorption isotherms is calculated from two sets of isothermal adsorption data by the following formula: ； ； Among them, ρ i is the similarity of the normalized adsorption isotherms, ρ i,j is the Pearson correlation coefficient of the two sets of data, cov(i,j) is the covariance of the two sets of data, σ i and σ j are the standard deviations of the two sets of data, E is the expectation, i and j are the two sets of data, μ i and μ j is the expectation of the two sets of data, and n is the number of samples in the training set; The similarity coefficient of saturated adsorption capacity is calculated according to the following formula: ； ； φ i is the similarity coefficient of saturated adsorption capacity, MAX is the maximum value, Q i and Q j are the saturated adsorption capacities of groups i and j, respectively, and ΔQ i,j is the difference in saturated adsorption capacity, and n is the number of samples in the training set.

8. The system for rapidly predicting material gas adsorption performance based on artificial intelligence according to claim 7 is characterized in that: The gas adsorption performance refers to one of the adsorption amount, adsorption rate, and desorption rate of a specific gas type under specific parameter conditions; the performance tag to be predicted refers to the gas adsorption performance under different parameter conditions and / or different gas types; The user terminal (101) also includes a judgment unit (1012) for performing a step of judging the integrity of the data obtained by the user, by fitting the adsorption data through a gas adsorption equation to determine whether the degree of fit meets a threshold requirement, and if so, determining that the data is complete.

9. The system for rapidly predicting material gas adsorption performance based on artificial intelligence according to claim 8 is characterized in that: The data storage and processing module (102) also includes a secondary acquisition unit (1025) for fitting the data obtained by the user and / or the data obtained from the database through the gas adsorption equation and generating adsorption performance data under other conditions; The data storage and processing module (102) also includes a training set-test set partitioning unit (1026) for randomly selecting a portion of the data as an initial set, and then selecting a portion of the remaining data that has a higher similarity to the initial set and adding it to the initial set as a training set, and the rest as a verification set; The artificial intelligence training and prediction module (103) also includes a model evaluation unit (1032) for comparing the performance of models trained using different artificial intelligence models.

Images