Polymeric gas separation membrane designing and screening method and device, and storage medium

Abstract

The present disclosure relates to the technical field of high-throughput molecular dynamics simulation, and provides a polymeric gas separation membrane designing and screening method and device, and a storage medium. The method comprises: acquiring monomer information of a plurality of high molecular monomers; generating a membrane search space on the basis of the monomer information; on the basis of a pre-trained membrane performance prediction model, performing performance prediction on a plurality of candidate polymeric gas separation membranes to be simulated in the search space; on the basis of performance prediction results, screening for, from the membrane search space, at least one target polymeric gas separation membrane which is to be simulated and has a desired performance; on the basis of a molecular dynamics simulation tool, performing Monte Carlo simulation and molecular dynamics simulation on the target polymeric gas separation membranes to be simulated to obtain solubility coefficients and diffusion coefficients; and on the basis of the solubility coefficients and the diffusion coefficients, determining performance indexes of said target polymeric gas separation membranes. A large number of non-existing simulated membrane materials can be formed for analysis, and the screening efficiency of a large number of polymer materials is improved.

Description

Design and screening method, device and storage medium of polymer gas separation membrane

[0001] The present application claims priority from the Chinese patent application No. 202411230657.7 filed on September 3, 2024, and entitled "Design and screening method, device and storage medium of polymer gas separation membrane", the whole content of which is incorporated herein by reference. TECHNICAL FIELD

[0002] The present disclosure relates to the technical field of high-throughput molecular dynamics simulation, in particular to a design and screening method, device and storage medium of polymer gas separation membrane. BACKGROUND

[0003] Polymer gas separation membranes are a class of thin film materials used for gas separation and purification, made of polymers, which can selectively permeate certain gas molecules according to the size, shape and chemical properties of gas molecules. The gas separation principle of polymer gas separation membrane is generally based on the solubility-diffusion mechanism, that is, gas molecules first dissolve on one side of the membrane surface, then diffuse inside the membrane, and finally release from the other side of the membrane. Polymer gas separation membranes are widely used in gas separation and purification applications such as natural gas treatment, air separation and pollution control due to their low energy consumption, simple operation and easy mass production.

[0004] Traditional polymer gas separation membranes have limitations in separation efficiency and selectivity due to their inherent physical and chemical properties. For example, traditional polymer gas separation membranes often sacrifice selectivity while improving gas permeability, which is determined by the microstructure of the material and the diffusion mechanism of gas molecules. Therefore, in order to improve the separation performance of the membrane material, the structure and properties of the traditional polymer gas separation membrane need to be optimized.

[0005] A typical design method of polymer gas separation membrane includes: designing a polymer gas separation membrane based on the performance indicated by the polymer material database and the data in the literature.

[0006] However, the high polymer material database and the data in the literature are mostly experimental values, which are limited by the diversity of experimental conditions and methods, making it difficult to perform large-scale material screening and optimization design. SUMMARY

[0007] Therefore, the present disclosure proposes a design and screening method, device and storage medium of polymer gas separation membrane, which can overcome the limitations of traditional molecular dynamics simulation in high-throughput processing, and combine machine learning technology to realize rapid screening and accurate characterization of a large number of polymer materials, thereby realizing rapid, efficient and accurate screening and characterization of a large number of polymer materials.

[0008] According to an aspect of the present disclosure, a design and screening method of a polymer gas separation membrane is provided, the method comprising:

[0009] obtaining monomer information of a plurality of polymer monomers;

[0010] generating a membrane search space based on the monomer information, the membrane search space comprising a plurality of candidate simulated polymer gas separation membranes;

[0011] performing performance prediction on the plurality of candidate simulated polymer gas separation membranes in the search space based on a pre-trained membrane performance prediction model, to obtain a performance prediction result of each candidate simulated polymer gas separation membrane; the membrane performance prediction model is trained based on a machine learning algorithm;

[0012] screening at least one target simulated polymer gas separation membrane meeting a desired performance from the membrane search space based on the performance prediction result;

[0013] performing Monte Carlo simulation on each target simulated polymer gas separation membrane based on a molecular dynamics simulation tool to determine solubility coefficients of the target simulated polymer gas separation membrane for different gases; and performing molecular dynamics simulation on each target simulated polymer gas separation membrane based on the molecular dynamics simulation tool to determine diffusion coefficients of the target simulated polymer gas separation membrane for different gases;

[0014] determining a performance index of the target simulated polymer gas separation membrane based on the solubility coefficients and the diffusion coefficients, the performance index comprising a permeability coefficient of a gas and / or a selectivity of a membrane.

[0015] The performance prediction on the plurality of candidate simulated polymer gas separation membranes in the search space based on the pre-trained membrane performance prediction model to obtain the performance prediction result of each candidate simulated polymer gas separation membrane comprises:

[0016] for each candidate simulated polymer gas separation membrane, obtaining multi-scale information of the candidate simulated polymer gas separation membrane; wherein the multi-scale information at least comprises microscopic information and mesoscopic information; wherein the microscopic information is used to indicate microscopic attributes of the candidate simulated polymer gas separation membrane, and the mesoscopic information is used to indicate mesoscopic attributes of the candidate simulated polymer gas separation membrane;

[0017] inputting the multi-scale information into the membrane performance prediction model to obtain the performance prediction result; wherein the membrane performance prediction model is established based on a hierarchical graph convolutional neural network model.

[0018] In a possible implementation, the membrane performance prediction model sequentially comprises, in an information transmission direction:

[0019] an input layer configured to input the multi-scale information;

[0020] a convolutional layer configured to perform feature extraction based on the multi-scale information;

[0021] an attention pooling layer configured to determine an attention score of feature data output by the convolutional layer, and perform weighted pooling on the feature data based on the attention score to obtain processed feature data;

[0022] an output layer configured to generate the performance prediction result based on the processed feature data.

[0023] In a possible implementation, the attention pooling layer is specifically configured to determine the attention score of the feature data based on a multi-head attention mechanism guided by a mask matrix.

[0024] In a possible implementation, the micro information includes a SMILES string of at least one polymer monomer forming the candidate simulated high polymer gas separation membrane, and / or atomic and bond information; the atomic and bond information is generated based on the SMILES string.

[0025] The meso information includes an intermonomer linkage mode of the polymer monomer.

[0026] In a possible implementation, the multi-scale information further includes macro information, and the macro information is used to indicate a macro attribute of the candidate simulated high polymer gas separation membrane.

[0027] In a possible implementation, the macro information includes a free volume fraction FFV and / or a glass transition temperature Tg; the FFV refers to a proportion of space not occupied by molecules in the candidate simulated high polymer gas separation membrane; and the glass transition temperature Tg is a temperature at which the candidate simulated high polymer gas separation membrane is converted from a glass state to a rubber state.

[0028] In a possible implementation, before the performance prediction model is used to perform performance prediction on the plurality of candidate simulated high polymer gas separation membranes in the search space to obtain the performance prediction result of each candidate simulated high polymer gas separation membrane, the method further includes:

[0029] generating a sample simulated high polymer gas separation membrane;

[0030] obtaining sample micro information and sample meso information of the sample simulated high polymer gas separation membrane;

[0031] simulate, based on the molecular dynamics simulation tool, a molecular dynamics property of each candidate simulated polymeric gas separation membrane to obtain a simulated performance parameter of the sample simulated polymeric gas separation membrane;

[0032] train a pre-created hierarchical graph convolutional neural network model based on the sample micro information, the sample meso information, and the simulated performance parameter to obtain the membrane performance prediction model.

[0033] In a possible implementation, the obtaining monomer information of a plurality of polymer monomers comprises:

[0034] receiving a SMILES string of the polymer monomer;

[0035] and / or,

[0036] receiving a polymer single-chain model; and parsing the polymer single-chain model to obtain the monomer information.

[0037] According to another aspect of the present disclosure, there is provided a device for designing and screening a polymeric gas separation membrane, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0038] According to another aspect of the present disclosure, there is provided a non-volatile computer-readable storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement the above method.

[0039] According to another aspect of the present disclosure, there is provided a computer program product comprising computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, which, when run in a processor of an electronic device, causes the processor in the electronic device to perform the above method.

[0040] The monomer information of a plurality of high molecular monomers is acquired; a membrane search space is generated based on the monomer information, the membrane search space comprising a plurality of candidate simulated high molecular gas separation membranes; performance prediction is performed on the plurality of candidate simulated high molecular gas separation membranes in the search space based on a pre-trained membrane performance prediction model, to obtain a performance prediction result of each candidate simulated high molecular gas separation membrane; at least one target simulated high molecular gas separation membrane meeting an expected performance is screened from the membrane search space based on the performance prediction result; Monte Carlo simulation is performed on each target simulated high molecular gas separation membrane based on a molecular dynamics simulation tool, to determine the solubility coefficients of the target simulated high molecular gas separation membrane for different gases; and molecular dynamics simulation is performed on each target simulated high molecular gas separation membrane based on the molecular dynamics simulation tool, to determine the diffusion coefficients of the target simulated high molecular gas separation membrane for different gases; and based on the solubility coefficients and the diffusion coefficients, a performance index of the target simulated high molecular gas separation membrane is determined, the performance index comprising a permeability coefficient of the gas and / or a selectivity of the membrane; the search space can be generated based on the monomer information, so that a large number of non-existing simulated membrane materials are analyzed, the problem that data are limited by experimental conditions and methods and are difficult to be used for large-scale material screening and optimal design is solved, and it is possible to design new membrane materials. Meanwhile, the membrane performance prediction model is used to search for feasible target simulated high molecular gas separation membranes in the search space, without manually screening the performance of a large number of simulated membrane materials in the search space, and the efficiency of screening a large number of high molecular materials can be improved.

[0041] Other features and aspects of the present disclosure will become apparent from the following detailed description of example embodiments with reference to the drawings. BRIEF DESCRIPTION OF DRAWINGS

[0042] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate example embodiments, features, and aspects of the present disclosure and serve to explain the principles of the present disclosure.

[0043] FIG. 1 shows a flowchart of a design and screening method of a high molecular gas separation membrane according to an embodiment of the present disclosure;

[0044] FIG. 2 shows a structural diagram of a membrane performance prediction model according to an embodiment of the present disclosure;

[0045] FIG. 3 shows a schematic diagram of multi-scale information according to an embodiment of the present disclosure;

[0046] FIG. 4 shows a schematic diagram of microscopic information according to an embodiment of the present disclosure;

[0047] FIG. 5 shows a block diagram of a design and screening device of a high molecular gas separation membrane according to an embodiment of the present disclosure;

[0048] FIG. 6 shows a block diagram of a design and screening apparatus for a polymer gas separation membrane according to another embodiment of the present disclosure. DETAILED DESCRIPTION

[0049] Various exemplary embodiments, features, and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numbers in the drawings represent the same elements or similar elements. Although various aspects of embodiments are illustrated in the drawings, the drawings are not necessarily drawn to scale unless specifically indicated.

[0050] The term "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any implementation described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other implementations.

[0051] In addition, for the purpose of convenience and brevity, detailed descriptions of well-known functions and structures incorporated in the disclosure will be omitted when they are considered to make the subject matter of the present disclosure unclear. It will be appreciated that the present disclosure can be practiced with the exact details as

[0052] First, several terms related to the present application are introduced.

[0053] Polymer chain: A long-chain molecule composed of many repeating monomer units (i.e., polymer monomers) connected by chemical bonds. These monomer units are connected together through polymerization reactions to form a polymer with a specific molecular weight and structure. The length of a polymer chain can range from a few thousand atoms to millions of atoms, which gives polymers unique physical and chemical properties.

[0054] Polymer monomer: The basic molecular unit that makes up a polymer chain. In polymerization reactions, polymer monomers are connected together by covalent bonds to form long-chain polymer molecules (i.e., the polymer chains described above). Each polymer monomer has one or more functional groups that can react with other monomers. For example, polyethylene is a polymer chain formed by the polymerization of ethylene monomers, each of which contains a double bond. These double bonds open up and link with the double bonds of other ethylene monomers during polymerization, forming long-chain polyethylene molecules.

[0055] Polymer single-chain model: A mathematical or computational model used to describe the physical and chemical properties of a single polymer chain. This model typically incorporates factors such as the conformation, flexibility, and interactions of the polymer chain to predict and explain the macroscopic properties of polymer materials.

[0056] Hierarchical Graph Convolutional Networks (HGCN) model: a deep learning model that combines hierarchical structure and Graph Convolutional Networks (GCN). HGCN is used to process graph data with hierarchical or layered structure, for example, to process the multi-scale simulated polymer gas separation membrane structure data in this application.

[0057] FIG. 1 shows a flowchart of a polymer gas separation membrane design and screening method according to an embodiment of the present disclosure. In this embodiment, the method is applied in an electronic device with computing capability, which can be a user terminal or a server. The user terminal can be a computer, a tablet, etc. The implementation of the electronic device is not limited in this embodiment. As shown in FIG. 1, the method includes at least the following steps:

[0058] Step 101, obtaining monomer information of a plurality of polymer monomers.

[0059] The monomer information is used to describe the polymer monomer. Illustratively, obtaining the monomer information of the plurality of polymer monomers includes: receiving a Simplified Molecular Input Line Entry System (SMILES) string of the polymer monomer; and / or, receiving a polymer single-chain model; and parsing the polymer single-chain model to obtain the monomer information.

[0060] Optionally, the molecular dynamics simulation tool running in the electronic device provides a human-computer interaction interface to receive the SMILES string or the polymer single-chain model describing the polymer monomer. In this embodiment, the molecular dynamics simulation tool can simulate various dynamic motion conditions of the polymer gas separation membrane during gas separation. The molecular dynamics simulation tool includes but is not limited to: Large-scale Atomic / Molecular Massively Parallel Simulator (LAMMPS), Highly Optimized Object-oriented Many-particle Dynamics-blue (HOOMD-blue), etc. open source, suitable for simulating the dynamics of polymer-gas systems molecular dynamics simulation software package, the implementation of the molecular dynamics simulation tool is not limited in this embodiment.

[0061] Optionally, the single-chain polymer model is analyzed to obtain monomer information, including: identifying at least one repeating unit in the single-chain polymer model to obtain at least one monomer; and generating a SMILES string describing each monomer to obtain the monomer information.

[0062] Optionally, in other embodiments, the electronic device can also read monomer information of the polymer monomer from an existing material library.

[0063] Step 102, generating a membrane search space based on the monomer information, the membrane search space including a plurality of candidate simulated polymer gas separation membranes.

[0064] The membrane search space is used to provide a plurality of possible simulated polymer gas separation membranes.

[0065] In one example, generating the membrane search space based on the monomer information includes: randomly selecting at least one monomer information from the obtained plurality of monomer information and randomly arranging to copolymerize to form a candidate simulated polymer gas separation membrane. Wherein, the monomer information constituting different candidate simulated polymer gas separation membranes is different, and / or the arrangement mode of the monomer information constituting different candidate simulated polymer gas separation membranes is different.

[0066] Illustratively, copolymerizing to generate a candidate simulated polymer gas separation membrane based on the monomer information includes: converting the SMILES string to a simulated single-chain polymer using the RDKit and Open Babel toolkits to obtain the candidate simulated polymer gas separation membrane.

[0067] In other embodiments, the way of generating the membrane search space based on the monomer information can also be: generating a candidate simulated polymer gas separation membrane based on user-specified monomer information and arrangement mode, and the present embodiment does not limit the way of generating the membrane search space.

[0068] Optionally, after generating a plurality of candidate simulated polymer gas separation membranes, the rationality of the candidate simulated polymer gas separation membrane can also be determined, and the candidate simulated polymer gas separation membrane whose rationality does not meet the expected target is filtered to obtain the membrane search space. Wherein, the rationality of the candidate simulated polymer gas separation membrane can be represented by a value calculated by a Q-e model or a Flory-Reaction Probability Factor model, and the present embodiment does not limit the method of determining the rationality.

[0069] Step 103, performing performance prediction on the plurality of candidate simulated polymer gas separation membranes in the search space based on a pre-trained membrane performance prediction model to obtain a performance prediction result of each candidate simulated polymer gas separation membrane.

[0070] The film performance prediction model is trained based on a machine learning algorithm. The machine learning algorithm includes but is not limited to a graph neural network model, a hierarchical graph convolutional neural network model, etc. The implementation of the machine learning algorithm is not limited in the embodiment.

[0071] In one example, the film performance prediction model is established based on a hierarchical graph convolutional neural network model. Accordingly, the performance of the plurality of candidate simulated high polymer gas separation membranes in the search space is predicted based on the pre-trained film performance prediction model, and the performance prediction result of each candidate simulated high polymer gas separation membrane is obtained, including: for each candidate simulated high polymer gas separation membrane, obtaining the multi-scale information of the candidate simulated high polymer gas separation membrane. The multi-scale information at least includes microscopic information and mesoscopic information; the multi-scale information is input into the film performance prediction model to obtain the performance prediction result.

[0072] The microscopic information is used to indicate the microscopic attribute of the candidate simulated high polymer gas separation membrane. Illustratively, the microscopic information includes the SMILES string of at least one high polymer monomer forming the candidate simulated high polymer gas separation membrane, and / or atomic and bond information. The atomic and bond information is generated based on the SMILES string, that is, in the case where the microscopic information includes the atomic and bond information, the electronic device obtains the SMILES string of at least one high polymer monomer of the candidate simulated high polymer gas separation membrane, and generates the atomic and bond information based on the SMILES string.

[0073] Illustratively, the atomic and bond information is generated based on the SMILES string, including: parsing the SMILES string to obtain atomic information and bond information, the atomic information is used to describe the atomic type, and the bond information is used to indicate the connection relationship between atoms. Optionally, the atomic information includes but is not limited to atomic symbols; the bond information includes but is not limited to bond types (such as single bond, double bond, triple bond, etc.).

[0074] The mesoscale information is used to indicate mesoscale properties of the candidate simulated polymeric gas separation membrane. Illustratively, the mesoscale information includes: an intermonomer linkage pattern of the polymeric monomers. The intermonomer linkage pattern refers to a specific way in which the polymeric monomers connect to each other during polymerization, and describes which parts (e.g., functional groups) of the polymeric monomers participate in the connection and how they are arranged relative to each other. The intermonomer linkage pattern can affect the geometry, spatial arrangement, and final physical properties of the polymer. For example, the intermonomer linkage pattern includes, but is not limited to, head-to-tail linkage, head-to-head linkage, tail-to-tail linkage, alternate linkage, branching linkage, cross-linking, cyclic linkage, and the like. The embodiments herein do not list all the intermonomer linkage patterns. Alternatively, the intermonomer linkage pattern can be randomly generated, or can be based on the human-computer interaction interface, and the embodiments herein do not limit the way of obtaining the intermonomer linkage pattern.

[0075] Optionally, the multiscale information further includes macroscopic information, and the macroscopic information is used to indicate macroscopic properties of the candidate simulated polymeric gas separation membrane. Illustratively, the macroscopic information includes: a fraction of free volume (FFV) and / or a glass transition temperature Tg. The FFV refers to a proportion of space in the candidate simulated polymeric gas separation membrane that is not occupied by molecules. The glass transition temperature Tg is a temperature at which the candidate simulated polymeric gas separation membrane changes from a glassy state to a rubbery state.

[0076] In one example, the FFV and the Tg are obtained by performing molecular dynamics simulation on the candidate simulated polymeric gas separation membrane by a molecular dynamics simulation tool.

[0077] The molecular dynamics simulation on the candidate simulated polymeric gas separation membrane by the molecular dynamics simulation tool includes: performing molecular dynamics simulation on the candidate simulated polymeric gas separation membrane by an optimized potentials for liquid simulations-all atom (OPLS-AA) force field for all-atom molecular dynamics (MD) simulation.

[0078] Since the OPLS-AA is more effective and robust in simulating various molecular systems, its use in the polymer-gas simulation system in the embodiments can improve the simulation accuracy.

[0079] Specifically, taking the LAMMPS software package as an example of a molecular dynamics simulation tool, the simulation process includes the following steps 31 and 32:

[0080] Step 31, generating a periodic molecular box suitable for LAMMPS simulation;

[0081] Before starting the simulation by LAMMPS, a three-dimensional space region is defined, which is the simulation box (or simulation cell), and the simulation molecules corresponding to the candidate simulation high polymer gas separation membrane to be simulated are placed in this region for subsequent simulation calculation to obtain the molecular box.

[0082] When creating the simulation box, the boundary conditions of the simulation cell need to be specified. Common boundary conditions include periodic boundary conditions and non-periodic boundary conditions. Periodic boundary conditions indicate that when a particle leaves one boundary of the simulation box, it will re-enter the simulation cell from the opposite boundary, which can simulate an infinite system. Non-periodic boundary conditions do not have such restrictions. Non-periodic boundary conditions indicate that the boundaries of the simulation box are not considered part of the continuous space, i.e., particles in the simulation box do not jump to the other side of the box when they reach the boundary, but are affected by the boundary, such as being reflected or absorbed. In this embodiment, the boundary condition of the periodic molecular box is an infinite periodic molecular box, i.e., the boundaries of the simulation box are considered to be infinitely extended, and the particles will immediately appear at the opposite boundary when they reach one side boundary, thereby simulating an infinite system.

[0083] Optionally, before step 31, the simulation system can also be subjected to an energy minimization and annealing process to eliminate any unnatural atomic overlap or high-energy state through the energy minimization process, and to ensure that the simulation system reaches a thermodynamic equilibrium state by slowly reducing the system temperature through the annealing process.

[0084] Step 32, executing a run file for performing molecular dynamics simulation, simulating the temperature variation process through the heating sequence and the cooling sequence in the run file; determining the glass transition temperature Tg of the candidate simulation high polymer gas separation membrane based on the density of the candidate simulation high polymer gas separation membrane recorded in the temperature variation process through a customized analysis script; and / or obtaining the FFV by processing the ratio of the volume occupied by the atoms in the candidate simulation high polymer gas separation membrane to the volume of the pores in the temperature variation process through a customized analysis script.

[0085] In this embodiment, the run file includes a heating sequence and a cooling sequence for simulating the temperature variation process in the molecular dynamics simulation process.

[0086] Illustratively, the temperature change process simulated by the heating sequence and the cooling sequence comprises: starting from a preset initial temperature, gradually increasing to a preset maximum temperature according to a preset temperature interval. The maximum temperature is greater than the initial temperature, and the initial temperature can be 200 Kelvin (K), and the maximum temperature can be 1000 K. In other embodiments, the values of the initial temperature and the maximum temperature can also be other values, and the present embodiment does not limit the values of the initial temperature and the maximum temperature.

[0087] During the glass transition, the volume of the candidate simulated high polymer gas separation membrane will change significantly, and then the density of the candidate simulated high polymer gas separation membrane will change significantly. Based on this, after the temperature climbing process is completed, the density recorded by the script is processed by the analysis script, the discrete derivative (or slope) of the density as a function of temperature is calculated by the script, and the temperature at which the slope is maximum (or the rate of change of density is maximum) is determined as the glass transition temperature Tg. In addition, by analyzing the volume of atomic occupation and the volume of pores recorded by the script, then calculating the ratio of the volume of atomic occupation and the volume of pores, the FFV is obtained.

[0088] Optionally, the running file further includes other files required by the molecular dynamics simulation process, such as: a topology file, a coordinate file and a parameter file of each polymer, etc., to indicate how to construct the force field model of the molecule, determine the position and direction of the molecule, define the topology information and parameters of a single molecule type, etc. The present embodiment does not limit the contents included in the running file.

[0089] In the present embodiment, referring to FIG. 2, the pre-trained membrane performance prediction model comprises, in sequence according to the information transmission direction: an input layer 210, a convolution layer 220, an attention pooling layer 230 and an output layer 240.

[0090] The input layer 210 is configured to input multi-scale information. In the present embodiment, the electronic device labels the attributes of different scale information in the multi-scale information to record the source and level of each piece of information. The attributes of different scale information remain unchanged in the prediction process.

[0091] Taking the multi-scale information including micro information, meso information and macro information as an example, referring to the schematic diagram of the 3-layer input configuration of the input layer 210 shown in FIG. 3, according to FIG. 3, the information input by the input layer 210 includes 3 levels, the scale information of the third level is micro information, and accordingly the attribute of the third level indicates that the micro information belongs to the third level. The micro information includes the SMILES string of at least one polymer monomer forming the candidate simulated polymer gas separation membrane; then, the input layer 210 generates the atomic and bond information of each polymer monomer based on the SMILES string. Referring to FIG. 4, the atomic and bond information includes node features, node labels, edge features and edge weights. Among them, the node features include: atomic charge number, type of atom (such as hydrogen mixture, aromaticity) and the like; the node label includes: atomic number. The edge features include: bond type and form of conjugated system; the edge weight: steric effect.

[0092] When constructing the atomic and bond information, if the candidate simulated polymer gas separation membrane is generated by at least two different polymer monomer units, an additional edge can be added to indicate the connection mode of different kinds of polymer monomer units, to represent the possibility of other monomers connecting each polymer monomer unit. In this way, when analyzing one of the polymer monomer units, the membrane performance prediction model can consider the existence of other monomers according to the additional edge, so that the end atom can understand the local chemical environment. Among them, the polymer monomer unit refers to a unit composed of multiple repeating polymer monomers, and the performance of this representation is better than using a virtual atom to represent the end of the monomer. For example: the polymer chain corresponding to the candidate simulated polymer gas separation membrane is: AAABBABCCCBBAAA, where A, B and C represent different polymer monomer units, at this time, an edge can be connected between each two of the polymer monomer units A, B and C to define the corresponding connection mode, such as: connecting an edge between the polymer monomer units A and B to define that the polymer monomer units A and B are connected by a C-C bond; connecting an edge between the polymer monomer units A and B to define that the polymer monomer units A and B are connected by a C-N bond; connecting an edge between the polymer monomer units B and C to define that the polymer monomer units B and C are connected by a C-N bond.

[0093] The input layer 210 of the embodiment supports inputting multiple SMILES strings, such as: "monomer SMILES1.monomer SMILES2.monomer SMILES3.monomer SMILES4", while the traditional neural network model only allows input of a single SMILES string, at this time, only the performance of the candidate simulated polymer gas separation membrane polymerized by a single polymer monomer can be predicted, while the embodiment supports predicting the performance of the candidate simulated polymer gas separation membrane polymerized by multiple polymer monomers, expanding the application scenario of the model.

[0094] The attribute of the second level indicates that the mesoscopic information belongs to the second level, and the mesoscopic information includes the inter-monomer linkage mode of the monomer corresponding to the atoms and bonds indicated by the microscopic information.

[0095] The attribute of the first level indicates that the macroscopic information belongs to the first level, and the macroscopic information includes the Tg and FFV of the candidate simulated polymer gas separation membrane composed of the monomer corresponding to the atoms and bonds indicated by the microscopic information.

[0096] Based on this, the input layer 210 associates the multi-scale information based on the input information, so that the hierarchical graph convolutional neural network can predict the performance of the membrane material in combination with the different scale characteristics of the candidate simulated polymer gas separation membrane, thereby improving the prediction accuracy.

[0097] Optionally, in order to process attributes of different scales, the input layer 210 includes a multilayer perceptron (MLP) to pre-process the multi-scale information, so that each attribute is subjected to a nonlinear transformation before being input to the convolutional layer, and a processed feature representation is obtained, which is uniformly input to the same convolutional layer to ensure the integrity and continuity of the information.

[0098] Optionally, the input layer 210 is also connected to an embedding layer to map the input information received by the input layer 210 to a continuous vector space (or embedding space), so as to improve the understanding ability of the model to the input information and improve the performance of the model.

[0099] The convolutional layer 220 is configured to perform feature extraction based on the multi-scale information. In the embodiment, the convolutional layer is a graph convolutional network (GCN) layer. The GCN layer updates the feature representation of a node by performing specific operations on the node features and the adjacency matrix.

[0100] The attention pooling layer 230 is configured to determine an attention score of the feature data output by the convolutional layer, and perform weighted pooling on the feature data based on the attention score to obtain processed feature data.

[0101] Optionally, the attention pooling layer is specifically configured to determine the attention score of the feature data based on a multi-head attention mechanism guided by a mask matrix. That is, for each head in the multi-head attention mechanism, after the attention score corresponding to the head is determined, the mask matrix is used to set the attention score between the preset feature position and other feature positions to a preset smaller value (such as negative infinity), to obtain an updated attention score; and the updated attention score is used to perform weighted pooling on the feature data to obtain the processed feature data.

[0102] An output layer 240 is configured to generate a performance prediction result based on the processed feature data.

[0103] Optionally, the performance prediction result is determined based on a performance prediction requirement. For example, if the performance prediction requirement is to predict the Tg of a candidate simulated polymeric gas separation membrane, the membrane performance prediction model is trained at least with Tg as a label, and the performance prediction result includes a parameter indicating Tg. For another example, if the performance prediction requirement is to predict the solubility of a candidate simulated polymeric gas separation membrane, the membrane performance prediction model is trained at least with solubility as a label, and the performance prediction result includes a parameter indicating solubility. In addition, the performance prediction result can be one or more. In the case of multiple performance prediction results, the membrane performance prediction model is a multi-task learning model.

[0104] The training process of the membrane performance prediction model is described in the following embodiments, which will not be repeated here.

[0105] At step 104, at least one target simulated polymeric gas separation membrane meeting the desired performance is selected from the membrane search space based on the performance prediction result.

[0106] The desired performance is preset in the electronic device. The electronic device compares the performance prediction result with the desired performance. If the performance prediction result matches the desired performance, the candidate simulated polymeric gas separation membrane corresponding to the performance prediction result is determined as the target simulated polymeric gas separation membrane.

[0107] In this embodiment, the membrane performance prediction model is used to screen a large number of candidate simulated polymeric gas separation membranes in the membrane search space, and the target simulated polymeric gas separation membrane is designed, which can filter out some candidate simulated polymeric gas separation membranes that cannot be designed, and improve the design efficiency of the polymeric gas separation membrane.

[0108] At step 105, the Monte Carlo simulation is performed on each target simulated polymeric gas separation membrane based on the molecular dynamics simulation tool to determine the solubility coefficient of the target simulated polymeric gas separation membrane for different gases; and the molecular dynamics simulation is performed on each target simulated polymeric gas separation membrane based on the molecular dynamics simulation tool to determine the diffusion coefficient of the target simulated polymeric gas separation membrane for different gases.

[0109] The molecular dynamics simulation tool is described in the above embodiments, which will not be repeated here.

[0110] In one example, the solubility coefficient of the target simulated polymeric gas separation membrane is determined first, and then the diffusion coefficient of the target simulated polymeric gas separation membrane is determined.

[0111] Monte Carlo simulation is performed on each target simulated high polymer gas separation membrane to obtain solubility coefficients of the target simulated high polymer gas separation membrane for different gases, including the following steps 51-58:

[0112] Step 51, energy minimization and annealing process are performed on the simulated system to obtain a stable system;

[0113] The related description of the energy minimization and annealing process is described above, and will not be repeated here.

[0114] Step 52, constant number of atoms, volume and temperature (NVT) simulation is performed on the stable system;

[0115] The NVT simulation refers to molecular dynamics simulation under constant number of atoms (N), volume (V) and temperature (T) conditions.

[0116] For example, the electronic device performs 0.5 nanosecond NVT simulation at 500K on the stable system. In other embodiments, the simulation time and temperature can also be other values, and the simulation time and temperature are not limited in this embodiment.

[0117] Step 53, constant number of atoms, pressure and temperature (NPT) simulation is performed on the system after NVT simulation;

[0118] The NPT simulation refers to molecular dynamics simulation under constant number of atoms (N), pressure (P) and temperature (T) conditions.

[0119] For example, the electronic device performs 0.5 nanosecond NPT simulation at 500K and 1 bar (bar) on the system after NVT simulation. In other embodiments, the simulation time, temperature and pressure can also be other values, and the simulation time, temperature and pressure are not limited in this embodiment.

[0120] Step 54, x thermal annealing cycles are performed on the system after NPT simulation to make the temperature of the simulated system reach a first preset temperature, the pressure is maintained at a preset pressure value during the annealing process, and NVT simulation and NPT simulation are performed on the simulated system after annealing again;

[0121] The first preset temperature can be 250K, and in other embodiments, the first preset temperature can also be other values, and the first preset temperature is not limited in this embodiment. x is the number of cycles for reducing the temperature of the simulated system to 250K, for example, if the initial temperature of the simulated system is 500K and each thermal annealing cycle reduces 50K, then the value of x is 5. The preset pressure value can be 1 bar.

[0122] The duration of the NVT simulation and the NPT simulation can be the same as the above-mentioned step, such as 0.5 nanoseconds, and in other embodiments, the duration can also be other values, and the duration of the NVT simulation and the NPT simulation is not limited in the embodiment.

[0123] In step 55, the solubility coefficients of the target simulated polymer gas separation membrane to different gases are determined.

[0124] Under infinite dilution, the solubility coefficient can be represented by the Henry constant of the target simulated polymer gas separation membrane. The Henry constant of the polymer gas separation membrane refers to the proportional constant between the solubility of the gas in the polymer membrane material and the partial pressure of the gas on the surface of the membrane under specific temperature and pressure conditions. The solubility coefficient is the limit value of the ratio of the concentration of the gas in the target simulated polymer gas separation membrane to the partial pressure of the gas on the surface of the membrane. The solubility coefficient and the Henry constant are inversely related in value, the larger the Henry constant, the smaller the solubility coefficient, indicating that the solubility of the gas in the target simulated polymer gas separation membrane is lower.

[0125] Illustratively, the solubility coefficient S i is represented by the following formula:

[0126] wherein c i represents the concentration of the i-th gas in the target simulated polymer gas separation membrane, f i represents the partial pressure of the i-th gas on the surface of the membrane, and the limit c i → 0 represents the case when the gas concentration approaches 0, i.e. under the condition of infinite dilution. i is a positive integer.

[0127] In step 56, after the solubility calculation is completed, gas molecules are introduced into the molecular box containing the target simulated polymer gas separation membrane; and the energy minimization and annealing process of the simulated system filled with gas are carried out, so that the molecules in the simulated system reach a stable or equilibrium state.

[0128] The gas molecules can be CO2 and / or N2, etc., or can be the gas in the actual working environment of the target simulated polymer gas separation membrane, and the type of the gas molecules is not limited in the embodiment.

[0129] In step 57, NVT simulation and NPT simulation are carried out at a second preset temperature.

[0130] The second preset temperature can be 300K, and in other embodiments, the second preset temperature can also be other values, and the value of the second preset temperature is not limited in the embodiment.

[0131] For example, the electronic device first performs an NVT simulation and maintains for 1 nanosecond; then, performs an NPT simulation at 300K and 1 atmosphere and maintains for 2 nanoseconds. In other embodiments, the NVT simulation and the NPT simulation in this step can also be maintained for other time lengths, which can enable the system to reach stability.

[0132] Step 58, the simulation system is run for a preset maintenance time length, and data collection and analysis are performed within the preset maintenance time length to obtain a mean squared displacement (MSD), and the diffusion coefficient of the target simulated high polymer gas separation membrane for different gases is determined according to the mean squared displacement.

[0133] The preset maintenance time length is used to enable the simulation system to reach equilibrium and complete data collection and analysis. For example, the preset maintenance time length is 50 nanoseconds, the first 2 nanoseconds are used for equilibrium, and the remaining 48 nanoseconds are used for data collection and analysis. In actual implementation, the preset maintenance time length can also be other numerical values, and the present embodiment does not limit the value of the preset maintenance time length.

[0134] The diffusion coefficient is used to describe the rate of random motion of the gas in the target simulated high polymer gas separation membrane.

[0135] The diffusion coefficient D of the i-th gas is i which can be expressed by the following formula:

[0136] where t represents time, r(t) represents the position vector of the particle of the i-th gas at time t; r(0) represents the position vector of the particle of the i-th gas at the initial moment (t=0); |r(t)-r(0)| represents the square of the distance between the position of the particle at time t and the initial moment; and represents statistical average, i.e., the average value of all particles or multiple experimental results. 2

[0137] According to the above formula, the diffusion coefficient is one sixth of the average squared displacement of the particle in an infinite time. In actual application, since it is impossible to measure an infinite time, the MSD is measured on a time scale long enough, and the diffusion coefficient is estimated by linear fitting. In an ideal case, the MSD is linearly related to time, i.e., <|r(t)-r(0)| 2 >=6D i t, i.e., in three-dimensional space, the MSD of the particle increases linearly with time, and the slope of the MSD is the diffusion coefficient D i .

[0138] Step 106, based on the solubility coefficient and the diffusion coefficient, the performance index of the target simulated high polymer gas separation membrane is determined, and the performance index includes the permeability coefficient of the gas and / or the selectivity of the membrane.​

[0139] The permeability coefficient is used to describe the rate of gas permeation through a target model polymer gas separation membrane. The permeability coefficient combines the solubility and diffusion capabilities of a gas in the target model polymer gas separation membrane. For the i-th gas, the permeability coefficient P i may be expressed by the following equation: P i = S i · D i ;

[0140] where S i represents the solubility coefficient corresponding to the i-th gas, and D i represents the diffusion coefficient corresponding to the i-th gas.

[0141] The selectivity is used to describe the separation capability of the target model polymer gas separation membrane for different gases. Generally, the selectivity is expressed by the ratio of the permeability coefficients of two gases.

[0142] Illustratively, for the selectivity of gas i with respect to gas j, the selectivity a i / j may be expressed by the following equation: a i / j = P i / P j ;

[0143] where P i represents the permeability coefficient corresponding to the i-th gas, and P j represents the permeability coefficient corresponding to the j-th gas, where j is a positive integer different from the value of i.

[0144] In summary, the design and screening method of the polymer gas separation membrane provided in the embodiment is as follows: monomer information of a plurality of polymer monomers is obtained; a membrane search space is generated based on the monomer information, the membrane search space including a plurality of candidate simulated polymer gas separation membranes; performance prediction is performed on the plurality of candidate simulated polymer gas separation membranes in the search space based on a pre-trained membrane performance prediction model to obtain a performance prediction result of each candidate simulated polymer gas separation membrane; at least one target simulated polymer gas separation membrane meeting an expected performance is screened from the membrane search space based on the performance prediction result; Monte Carlo simulation is performed on each target simulated polymer gas separation membrane based on a molecular dynamics simulation tool to determine solubility coefficients of the target simulated polymer gas separation membrane for different gases; and molecular dynamics simulation is performed on each target simulated polymer gas separation membrane based on the molecular dynamics simulation tool to determine diffusion coefficients of the target simulated polymer gas separation membrane for different gases; and a performance index of the target simulated polymer gas separation membrane is determined based on the solubility coefficients and the diffusion coefficients, the performance index including a permeability coefficient of the gas and / or a selectivity of the membrane; the search space can be generated based on the monomer information, a large number of non-existing simulated membrane materials are analyzed, and the problem that data is limited by experimental conditions and methods and it is difficult to perform large-scale material screening and optimal design is solved, which provides a possibility for designing new membrane materials. Meanwhile, the membrane performance prediction model is used to search for a feasible target simulated polymer gas separation membrane in the search space, the performance of a large number of simulated membrane materials in the search space does not need to be manually screened, and the efficiency of a large number of polymer material screening can be improved.

[0145] In addition, the membrane performance prediction model is established based on the hierarchical graph convolutional neural network model, so that the multi-scale information of the simulated membrane material is input into the membrane performance prediction model for performance prediction, the membrane performance prediction model can perform performance prediction based on the comprehensive characteristics of the simulated membrane material, and the accuracy of performance prediction is improved.

[0146] In addition, the attention pooling layer is arranged in the membrane performance prediction model, so that the membrane performance prediction model can dynamically weight according to the importance of the characteristics, effectively retain more key information, filter out redundant information, improve the accuracy of performance prediction, and improve the calculation efficiency.

[0147] In addition, the attention pooling layer is arranged based on the multi-head attention mechanism guided by the mask matrix, the attention score of the feature data is determined, the influence and importance of the change of different hierarchical attributes on performance prediction can be analyzed, the model can dynamically adjust the importance of the characteristics, so that the key characteristics can be more accurately captured, and the model precision is improved.

[0148] In addition, the monomer information can be input by a non-professional person through manually constructing a polymer single-chain model or a SMILES string description of the polymer monomer, and the applicability of the method is improved.

[0149] Based on the above embodiment, before step 103, that is, before the performance prediction of each candidate simulated high polymer gas separation membrane is obtained based on the pre-trained membrane performance prediction model, the membrane performance prediction model is also trained.

[0150] The training process of the membrane performance prediction model includes the following steps:

[0151] Step 1, generating a sample simulated high polymer gas separation membrane.

[0152] The sample simulated high polymer gas separation membrane is generated in the same way as the candidate simulated high polymer gas separation membrane, that is, based on the obtained monomer information; or the sample simulated high polymer gas separation membrane is a simulation model corresponding to an existing high polymer gas separation membrane, and the embodiment does not limit the way of obtaining the sample simulated high polymer gas separation membrane.

[0153] Step 2, obtaining sample micro-information and sample meso-information of the sample simulated high polymer gas separation membrane.

[0154] The sample micro-information and sample meso-information are obtained in the same way as the micro-information and meso-information of the candidate simulated high polymer gas separation membrane, and the information types are consistent, which will not be repeated here.

[0155] Step 3, simulating the molecular dynamics properties of each sample simulated high polymer gas separation membrane based on a molecular dynamics simulation tool to obtain the simulation performance parameters of the candidate simulated high polymer gas separation membrane.

[0156] In the embodiment, the simulation performance parameters can be obtained by simulating the molecular dynamics properties of the sample simulated high polymer gas separation membrane through the molecular dynamics simulation tool, and the simulation performance parameters are obtained by collecting data during the simulation process. The simulation performance parameters are the same as the prediction target of the membrane performance prediction model, for example: the membrane performance prediction model is used to predict the Tg, solubility coefficient and diffusion coefficient of the candidate simulated high polymer gas separation membrane, and the simulation performance parameters include the Tg, solubility coefficient and diffusion coefficient of the sample simulated high polymer gas separation membrane.

[0157] Step 4, training the pre-created hierarchical graph convolutional neural network model based on the sample micro-information, sample meso-information and simulation performance parameters to obtain the membrane performance prediction model.

[0158] Specifically, after inputting the sample microscopic information and the sample mesoscopic information into a pre-created hierarchical graph convolutional neural network model, a prediction result of the model is obtained, and the prediction result is compared with a simulation performance parameter; based on a comparison result, model parameters of the hierarchical graph convolutional neural network model are iteratively optimized to obtain a membrane performance prediction model.

[0159] FIG. 5 shows a block diagram of a design and screening device of a polymer gas separation membrane according to an embodiment of the present disclosure. As shown in FIG. 5, the device includes the following modules: an information acquisition module 510, a space generation module 520, a performance prediction module 530, a membrane screening module 540, a motion simulation module 550, and a performance determination module 560.

[0160] The information acquisition module 510 is configured to acquire monomer information of a plurality of polymer monomers.

[0161] The space generation module 520 is configured to generate a membrane search space based on the monomer information, the membrane search space including a plurality of candidate simulated polymer gas separation membranes.

[0162] The performance prediction module 530 is configured to perform performance prediction on the plurality of candidate simulated polymer gas separation membranes in the search space based on a pre-trained membrane performance prediction model to obtain a performance prediction result of each candidate simulated polymer gas separation membrane; the membrane performance prediction model is trained based on a machine learning algorithm.

[0163] The membrane screening module 540 is configured to screen at least one target simulated polymer gas separation membrane meeting a desired performance from the membrane search space based on the performance prediction result.

[0164] The motion simulation module 550 is configured to perform Monte Carlo simulation on each target simulated polymer gas separation membrane based on a molecular dynamics simulation tool to determine solubility coefficients of different gases for the target simulated polymer gas separation membrane, and perform molecular dynamics simulation on each target simulated polymer gas separation membrane based on the molecular dynamics simulation tool to determine diffusion coefficients of different gases for the target simulated polymer gas separation membrane.

[0165] The performance determination module 560 is configured to determine a performance index of the target simulated polymer gas separation membrane based on the solubility coefficients and the diffusion coefficients, the performance index including a permeability coefficient of a gas and / or a selectivity of the membrane.

[0166] In some embodiments, the device provided by the embodiments of the present disclosure has functions or includes modules that can be used to perform the methods described in the above method embodiment descriptions, and specific implementations can refer to the descriptions of the above method embodiments. For brevity, they will not be described here again.

[0167] The embodiments of the present disclosure further provide a computer readable storage medium, having stored thereon computer program instructions, which, when executed by a processor, implement the method described above. The computer readable storage medium can be a volatile or non-volatile computer readable storage medium.

[0168] The embodiments of the present disclosure further provide an electronic device, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to implement the method described above when executing the instructions stored in the memory.

[0169] The embodiments of the present disclosure further provide a computer program product, comprising computer readable code, or a non-volatile computer readable storage medium carrying computer readable code, when the computer readable code is run in the processor of an electronic device, the processor in the electronic device executes the method described above.

[0170] FIG. 6 is a block diagram of an apparatus 1900 for designing and screening a polymer gas separation membrane according to an exemplary embodiment. For example, the apparatus 1900 can be provided as a server or a terminal device. Referring to FIG. 6, the apparatus 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932 for storing instructions, such as an application program, executable by the processing component 1922. The application program stored in the memory 1932 can include one or more than one module each corresponding to a set of instructions. In addition, the processing component 1922 is configured to execute the instructions to perform the method described above.

[0171] The apparatus 1900 can further include a power supply component 1926 configured to perform power management of the apparatus 1900, a wired or wireless network interface 1950 configured to connect the apparatus 1900 to a network, and an input / output interface 1958 (I / O interface). The apparatus 1900 can operate based on an operating system stored in the memory 1932, such as Windows Server TM , Mac OS X TM , Unix TM , Linux TM , FreeBSD TM or the like.

[0172] In exemplary embodiments, a non-volatile computer readable storage medium, such as the memory 1932 including computer program instructions executable by the processing component 1922 of the apparatus 1900 to complete the method described above, is also provided.

[0173] Having described above several embodiments of the disclosure, any modifications and variations that fall within the scope of the described embodiments are also intended to be within the scope of the disclosure. As will be apparent to those skilled in the art, some modifications and variations to the embodiments described above can be practiced while staying within the scope and spirit of the described embodiments. The foregoing description of the described embodiments has been presented for the purposes of illustration and description. It is not intended to be exhaustive or to limit the described embodiments to the precise form disclosed. Many modifications and variations are possible in light of the above teachings. It is intended that the disclosed embodiments be limited only by the claims.

Claims

1. A method for designing and screening a polymer gas separation membrane, characterized by, The method comprises: obtaining monomer information of a plurality of high molecular monomers; generating a membrane search space based on the monomer information, the membrane search space comprising a plurality of candidate simulated high molecular gas separation membranes; performing performance prediction on the plurality of candidate simulated high molecular gas separation membranes in the search space based on a pre-trained membrane performance prediction model, to obtain a performance prediction result of each candidate simulated high molecular gas separation membrane; the membrane performance prediction model is trained based on a machine learning algorithm; screening at least one target simulated high molecular gas separation membrane meeting the expected performance from the membrane search space based on the performance prediction result; performing Monte Carlo simulation on each target simulated high molecular gas separation membrane based on a molecular dynamics simulation tool to determine the solubility coefficients of the target simulated high molecular gas separation membrane for different gases; and performing molecular dynamics simulation on each target simulated high molecular gas separation membrane based on the molecular dynamics simulation tool to determine the diffusion coefficients of the target simulated high molecular gas separation membrane for different gases; determining a performance index of the target simulated high molecular gas separation membrane based on the solubility coefficients and the diffusion coefficients, the performance index comprising a permeability coefficient of the gas and / or a selectivity of the membrane.

2. The method of claim 1, wherein, The performance prediction on the plurality of candidate simulated high molecular gas separation membranes in the search space based on the pre-trained membrane performance prediction model to obtain the performance prediction result of each candidate simulated high molecular gas separation membrane comprises: for each candidate simulated high molecular gas separation membrane, obtaining multi-scale information of the candidate simulated high molecular gas separation membrane; wherein the multi-scale information at least comprises microscopic information and mesoscopic information; wherein the microscopic information is used to indicate the microscopic attribute of the candidate simulated high molecular gas separation membrane, and the mesoscopic information is used to indicate the mesoscopic attribute of the candidate simulated high molecular gas separation membrane; inputting the multi-scale information into the membrane performance prediction model to obtain the performance prediction result; wherein the membrane performance prediction model is established based on a hierarchical graph convolutional neural network model.

3. The method of claim 2, wherein, The membrane performance prediction model comprises in sequence according to the information transmission direction: an input layer for inputting the multi-scale information; a convolutional layer for feature extraction based on the multi-scale information; an attention pooling layer for determining an attention score of the feature data output by the convolutional layer; performing weighted pooling on the feature data based on the attention score to obtain processed feature data; an output layer for generating the performance prediction result based on the processed feature data.

4. The method of claim 3, wherein, The attention pooling layer is specifically configured to determine the attention score of the feature data based on a multi-head attention mechanism guided by a mask matrix.

5. The method of claim 2, wherein: the microscopic information comprises a SMILES string of at least one high molecular monomer forming the candidate simulated high molecular gas separation membrane, and / or atomic and bond information; wherein the atomic and bond information is generated based on the SMILES string; the mesoscopic information comprises an inter-monomer linking mode of the high molecular monomer.

6. The method of claim 2, wherein, The multi-scale information further comprises macro information, the macro information being used to indicate a macro attribute of the candidate simulated high polymer gas separation membrane.

7. The method of claim 6, wherein, The macro information comprises a free volume fraction FFV and / or a glass transition temperature Tg, the FFV being a proportion of space not occupied by molecules in the candidate simulated high polymer gas separation membrane, and the glass transition temperature Tg being a temperature at which the candidate simulated high polymer gas separation membrane is transformed from a glass state to a rubber state.

8. The method of claim 2, wherein, Before the performance prediction model is used to predict the performance of the plurality of candidate simulated high polymer gas separation membranes in the search space, the method further comprises: generating a sample simulated high polymer gas separation membrane; obtaining sample micro information and sample meso information of the sample simulated high polymer gas separation membrane; simulating a molecular dynamics property of each candidate simulated high polymer gas separation membrane based on the molecular dynamics simulation tool to obtain a simulated performance parameter of the sample simulated high polymer gas separation membrane; training a pre-created hierarchical graph convolutional neural network model based on the sample micro information, the sample meso information and the simulated performance parameter to obtain the membrane performance prediction model.

9. The method according to any one of claims 1 to 8, characterized in that, The obtaining of the monomer information of the plurality of high polymer monomers comprises: receiving a SMILES string of the high polymer monomer; and / or, receiving a high polymer single-chain model, and parsing the high polymer single-chain model to obtain the monomer information.

10. A device for designing and screening a polymer gas separation membrane, characterized by comprising: a device for designing a polymer gas separation membrane; and a device for screening a polymer gas separation membrane. comprise: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to implement the method of any one of claims 1 to 9 when executing the instructions stored in the memory.

11. A non-transitory computer readable storage medium having stored thereon computer program instructions, wherein, The computer program instructions, when executed by the processor, implement the method of any one of claims 1 to 9. The computer program instructions, when executed by the processor, implement the method of any one of claims 1 to 9.

Images