A design method for polymer desalination membranes based on graph neural networks
By constructing a knowledge graph and generating an adjacency matrix based on a graph neural network method, and building a graph neural network, the problems of high cost and low accuracy in seawater desalination membrane design were solved, and efficient and reliable seawater desalination membrane design and prediction were achieved.
Patent Information
- Application Number
- CN202210997162.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-19
- Publication Date
- 2025-09-16
- Estimated Expiration
- 2042-08-19
AI Technical Summary
The design cost of existing seawater desalination membranes is high and traditional neural networks find it difficult to effectively utilize global molecular information, resulting in long design verification cycles and low accuracy.
A graph neural network-based method is used to construct a knowledge graph of elements-molecular groups-molecular structures, generate an adjacency matrix, and build a graph neural network. The graph neural network is trained to output a polymer desalination membrane design model, which is then used for prediction based on global node information.
It significantly reduces the cost and time of seawater desalination membrane design, improves the accuracy and reliability of predictions, can effectively retain the global information of molecular structure, and enhances the scientific nature and practical applicability of membrane performance predictions.
Smart Images
Figure CN115346619B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of seawater desalination, and in particular to a design method for a polymer seawater desalination membrane based on a graph neural network. Background Art
[0002] Seventy-one percent of Earth's surface is covered by oceans, and seawater resources are a precious gift from nature. With the advancement of economic globalization, ocean trade has flourished. Coastal areas, key hubs for this trade, are often economically developed. However, despite their prosperity, many coastal regions suffer from a severe shortage of freshwater resources, failing to meet the needs of daily life and industrial production. The high concentration of salt in seawater poses a serious health risk to human health if consumed directly, and unpurified seawater also contains numerous harmful substances. Furthermore, high salt concentrations can severely damage industrial equipment, corroding the metal components, severely shortening their service life, and posing significant safety risks. While the emergence of desalination technology has, to some extent, addressed the issue of freshwater shortages, it also has significant drawbacks.
[0003] Currently, two main methods are used to desalinate seawater: distillation and reverse osmosis. Distillation is primarily used for large-scale desalination projects and in locations with abundant thermal energy. Reverse osmosis membranes, however, are more widely used due to their wide applicability and high salt rejection rates, providing significant benefits in areas with limited freshwater supply. While a wide variety of desalination equipment is being developed, concerns remain about production and design costs. To reduce desalination energy consumption, scientists are continuously researching desalination membrane design. However, since the breakthrough of thin-film composite polyamide membranes in desalination nearly half a century ago, only marginal improvements in the water-salt selectivity of desalination membranes have been achieved. This slow progress is largely due to the high cost and high requirements for experimental personnel and equipment in traditional desalination membrane design verification, which hinders large-scale deployment. This approach requires a long validation cycle and limited understanding of the relationship between membrane synthesis, structure, and performance.
[0004] In recent years, the rise and application of neural networks has successfully driven the digitalization of traditional industries. Many methods that once relied heavily on manual feature extraction and inductive research have now been replaced by various deep learning methods. However, in the field of polymer desalination membrane design, traditional neural networks struggle to fully represent the complex data structure of polymers, making training difficult. Existing design methods fail to effectively utilize global molecular information and are unable to model the overall characteristics of data from both structural and functional perspectives. Consequently, both the accuracy and reliability of predictions need to be improved. Summary of the Invention
[0005] In response to the shortcomings of the existing technology, the present invention proposes a design method for polymer desalination membranes based on graph neural networks to solve the problems of high cost and time cost in actual desalination membrane design and the inability of existing neural networks to better describe molecular information.
[0006] In order to solve the above technical problems, the technical solution of the present invention is:
[0007] A design method for polymer desalination membranes based on graph neural networks includes the following steps:
[0008] S1. Build a knowledge graph of elements, molecular groups, and molecular structures based on chemical knowledge, and construct a molecular structure information dataset. The data in the molecular structure information dataset includes the atoms in the molecule, the related chemical bonds connecting them, and the characteristics of the molecule itself.
[0009] S2. Obtain relevant data on the design and production of polymer desalination membranes from literature and other databases to establish a seawater desalination membrane database. Combined with the analysis of the structural information dataset, featurization processing suitable for graph neural networks is performed to generate an adjacency matrix. The relevant data in the seawater desalination membrane database includes the test temperature, seawater desalination performance, and molecular structure of the seawater desalination membrane;
[0010] S3. Build a graph neural network, use the atoms of the desalination membrane polymer as nodes of the graph, molecular bonds as edges, add global nodes, generate a graph neural network, and initialize the graph neural network to obtain the initial input of the graph neural network, wherein the global nodes include the overall information of the desalination membrane polymer monomer;
[0011] S4. Output the polymer desalination membrane design model by training the graph neural network;
[0012] S5. Predict the chemical structure of the seawater desalination membrane that meets the performance requirements of seawater desalination through the polymer seawater desalination membrane design model.
[0013] Preferably, in step S1, the data of the knowledge graph needs to be screened and cleaned before constructing the molecular structure information data set. The data of the knowledge graph includes the electronegativity and oxidizability of molecules and groups, the synthesis difficulty and valence electrons of compounds. The method for screening and cleaning the data of the knowledge graph is: using an efficient string search and replacement algorithm flashtext to match in literature and databases, screen relevant data, and combine relevant chemical properties with comparison of similar data to clean up data with obvious errors and problems.
[0014] Preferably, in step S2, the characterization processing method for generating the adjacency matrix is: using chemdraw software to convert the obtained molecular structure formula into a simplified molecular linear input specification SMILES, and then using the GetAdjacencyMatrix method of the rdkit library to obtain the molecular adjacency matrix.
[0015] Preferably, in step S3, when an atom is used as a node of the graph, the serial number of the atom and the number of atoms adjacent to it and the serial number, and the related properties of the atom are input, so that the atom is used as a node; when a molecular bond is used as an edge, the serial number of the molecular bond and its weight, and the related properties of the molecular bond are input, so that the molecular bond is used as an edge. In addition, the number of nodes and the longest path property of the graph are added to the additional molecular information provided by the global node.
[0016] Preferably, the initialization method of the graph neural network in step S3 is:
[0017] When the main architecture of the graph neural network is initialized, the embedding layer of the node is generated according to the neighboring nodes. The embedding layer is in the form of:
[0018]
[0019] Where v represents the index of the atomic node in the graph, k represents the current convolutional layer number, N(v) represents all neighbors of node v, σ is the activation function LeakyReLU, W and B are the model parameters to be learned, edges have independent parameters and perform similar operations, and h represents the weight and bias of each node in this layer of the graph neural network;
[0020] A fully connected layer is set after the embedding layer to output the results;
[0021] Initialize the weights and biases of the graph neural network to get the initial input of the network neural network:
[0022]
[0023] Where X represents the input matrix of this layer of the graph neural network, including the relevant properties of the atoms of the desalinated membrane polymer.
[0024] Preferably, the training of the graph neural network in step S4 includes the following steps:
[0025] S4-1. Normalize and perform data augmentation on the initial molecular input of the graph neural network to facilitate convergence and improve accuracy.
[0026] S4-2. Construct an embedding layer trained with a graph neural network as the main body. For each layer, calculate the linear combination of input and weight, and then apply the activation function to the linear combination value, with different model parameters for nodes and edges;
[0027] S4-3. Each layer performs aggregation operations on nodes and edges, and aggregates the weights of neighboring nodes and global nodes into the nodes;
[0028] S4-4. Finally, the fully connected layer is used for training, the loss function is calculated, and the polymer desalination membrane design model is output.
[0029] Preferably, the standard for evaluating the desalination membrane sample by the polymer desalination membrane design model output in step S4-4, i.e., the salt rejection rate, is expressed as follows:
[0030] Salt rejection rate = 1-(sodium ions + chloride ions passing through) / (total sodium ions + chloride ions).
[0031] Preferably, the polymer desalination membrane design model uses graph neural networks to learn graph structure data, extract and explore features and patterns in molecular structure graphs, and integrates graph convolution and attention mechanisms.
[0032] The present invention also provides a computer device comprising a memory and a processor, and a program stored in the processor and runnable on the processor. When executing the program, the processor implements the above-mentioned method for designing polymer seawater desalination membranes based on graph neural networks.
[0033] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned method for designing a polymer seawater desalination membrane based on a graph neural network.
[0034] The present invention has the following characteristics and beneficial effects:
[0035] Adopting the above technical solution, with the goal of maximizing the salt rejection rate and water-salt ratio of the desalination membrane, experimental and literature data on different desalination membranes were obtained. Combined with existing databases, a specialized database for desalination membranes was established. The molecular structures of the desalination membranes were graphically transformed to obtain a data structure suitable for graph neural network training. This data was trained using an improved graph neural network algorithm to obtain the most suitable molecular structure for the polymer membrane material. With the goal of optimizing desalination efficiency, a reasonable membrane structure was selected, taking into account the cost of raw materials, to enhance the practical usability of the model. Taking into account the polymer membrane's material, shape, area, and other conditions, a graph neural network model was trained to identify a set of materials with superior theoretical performance, which was then experimentally verified. This method utilizes a novel graph construction method for the material molecular structure, effectively reducing information loss. Furthermore, the inclusion of global nodes effectively preserves global information about the molecular structure, improving prediction accuracy. Furthermore, the method is highly reliable, accurate, and scientifically effective. The present invention significantly lowers the threshold for designing and manufacturing seawater desalination membranes, accelerates the design and manufacturing process, and the database can be used for the development of similar products with good reusability, thus having broad application prospects in the field of seawater desalination membrane design. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0037] Figure 1 It is the main module of the design method of polymer seawater desalination membrane based on graph neural network provided by an embodiment of the present invention.
[0038] Figure 2 This is the structure of a polymer seawater desalination membrane design prediction model training module based on a graph neural network provided in an embodiment of the present invention.
[0039] Figure 3 It is a schematic diagram of the seawater desalination process provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0040] In order to clearly explain the advantages of the present invention, the embodiments of the present invention are described in detail below with reference to examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0041] It should be understood by those skilled in the art that unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art in the art to which this invention belongs.
[0042] Those skilled in the art will understand that, unless otherwise specified, the terms "first," "second," and the like in the description and claims of the present invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the numbers used in this manner are interchangeable where appropriate so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having," as well as any variations thereof, are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatus.
[0043] To facilitate understanding of the present invention, the present invention is described in detail below with reference to the accompanying drawings.
[0044] The present invention provides a design method for a polymer seawater desalination membrane based on a graph neural network, comprising the following steps:
[0045] S1. Construct a knowledge graph of elements-molecular groups-molecular structures based on chemical knowledge, and construct a molecular structure information data set. The data in the molecular structure information data set include the atoms in the molecules and the related chemical bonds connected, as well as the characteristics of the molecules themselves. Before constructing the molecular structure information data set, the data of the knowledge graph needs to be screened and cleaned. The data of the knowledge graph includes the electronegativity and oxidizability of molecules and groups, the synthesis difficulty and valence electrons of compounds. The method for screening and cleaning the data of the knowledge graph is: use the efficient string search and replacement algorithm flashtext to match in literature and databases, screen relevant data, and combine the relevant chemical properties with the comparison of the same type of data to clean up the data with obvious errors and problems.
[0046] S2. Obtain relevant data on the design and production of polymer desalination membranes from literature and other databases, establish a seawater desalination membrane database, and combine the analysis of the structural information dataset to perform featurization processing suitable for graph neural networks to generate an adjacency matrix. The relevant data of the seawater desalination membrane database includes the test temperature of the seawater desalination membrane, the seawater desalination performance, and the molecular structure of the seawater desalination membrane.
[0047] Furthermore, the featurization processing method for generating the adjacency matrix is: using the rdkit library supported by Python to first convert the chemical structure SMILES into a Mol object, then using the WeaveFeaturizer and MolGraphConvFeaturizer converters of the deepchem library to generate a ConvMol object, and then extracting information such as the adjacency matrix from the ConvMol object, using the chemdraw software to convert the obtained molecular structure into a simplified molecular linear input specification SMILES, and then using the GetAdjacencyMatrix method of the rdkit library to obtain the adjacency matrix of the molecule.
[0048] S3. Build a graph neural network, use the atoms of the desalination membrane polymer as nodes of the graph, and the molecular bonds as edges, add global nodes, generate a graph neural network, and initialize the graph neural network to obtain the initial input of the graph neural network. The global nodes include the overall information of the desalination membrane polymer monomer, specifically the overall information of the desalination membrane polymer monomer refers to information such as aromaticity, number of atoms, and longest chain length;
[0049] Furthermore, when atoms are used as nodes of the graph, the serial number of the atom and the number and number of adjacent atoms, as well as the related properties of the atom, are input, so that the atom is used as a node; when molecular bonds are used as edges, the serial number of the molecular bonds and their weights, as well as the related properties of the molecular bonds, are input, so that the molecular bonds are used as edges. In addition, the number of nodes, the longest path and other properties of the graph are added to the additional molecular information provided by the global node, so that adding global information helps the training of graph neural networks.
[0050] It can be understood that the overall structure of the graph neural network includes a multi-layer network structure of input layer, embedding layer, hidden layer and output layer. Each layer considers multiple influencing factors, and each layer is composed of multiple neurons.
[0051] The hidden layer is a two-layer MLP fully connected layer, and the output layer is a single neuron. The aggregation function can also be LSTM or max-pooling to aggregate neighbor node information.
[0052] Specifically, when the main architecture of the graph neural network is initialized, the embedding layer of the node is generated according to the neighboring nodes. The form of the embedding layer is:
[0053]
[0054] Where v represents the index of the atomic node in the graph, k represents the current convolutional layer number, N(v) represents all neighbors of node v, σ is the activation function LeakyReLU, W and B are the model parameters to be learned, edges have independent parameters and perform similar operations, and h represents the weight and bias of each node in this layer of the graph neural network;
[0055] A fully connected layer is set after the embedding layer to output the results;
[0056] Initialize the weights and biases of the graph neural network to get the initial input of the network neural network:
[0057]
[0058] Where X represents the input matrix of this layer of the graph neural network, including the relevant properties of the atoms of the desalinated membrane polymer.
[0059] S4. Train the graph neural network to train the weights of the polymer desalination membrane variables to minimize the loss function and output the polymer desalination membrane design model.
[0060] In this embodiment, Figure 2 As shown, it mainly has five units: a training condition confirmation unit, which is used to set the objective function and constraint conditions, determine the training end conditions and the target requirements of model training;
[0061] The graph neural network training epoch is set to 500, the loss function is the mean square error, and the training is stopped when the correlation coefficient of the test set is greater than 0.95 or the epoch is completed.
[0062] Furthermore, the training of the graph neural network in step S4 includes the following steps:
[0063] S4-1. The data normalization and enhancement unit normalizes and enhances the initial molecular input of the graph neural network to facilitate convergence and improve accuracy. The input matrix consists of the structural characteristics of different polymers and the desalination performance labels.
[0064] Specifically, the input desalination membrane database is randomly divided into 80% training set and 20% test set, and the mean of the sample parameters is calculated. and standard deviation S k
[0065]
[0066] Change the mean and standard deviation according to the following rules
[0067]
[0068] The influence parameters are normalized to the range [0, 1] by translation transformation
[0069]
[0070] Add random perturbations to the training set to enhance the model's ability to resist interference
[0071] S4-2. Construct an embedding layer trained with a graph neural network as the main body. For each layer, calculate the linear combination value of the input and weight, and then apply the activation function to the linear combination value, with different model parameters for nodes and edges.
[0072] It can be understood that the embedding layer transforms the sparse matrix into a dense matrix through some linear transformations and convolutions.
[0073] S4-3, through the feature aggregation unit, is used to aggregate the parameter features of adjacent nodes or edges. Each layer performs aggregation operations on nodes and edges, and aggregates the weights of neighboring nodes and global nodes into the node. The core aggregation method is:
[0074]
[0075] In this embodiment, graph convolution and attention mechanisms are integrated. When representing each atomic node in the computation graph, different weights are assigned to it based on the characteristics of its neighboring atomic nodes. The importance is:
[0076] α ij =softmax j (e ij )
[0077] Where a is implemented using a single-layer feedforward neural network, and is calculated as follows:
[0078]
[0079] After calculating the attention, we can get a new representation of a node aggregating its neighbor node information
[0080]
[0081] Merge the results of K independent calculations to improve the model's fitting ability
[0082]
[0083] S4-4. Finally, the training is performed through the fully connected layer to calculate the loss function. The training is terminated when the loss function of the training result meets the training target requirements through the prediction model output unit. A five-fold cross-validation is performed to output the polymer seawater desalination membrane design model.
[0084] Among them, the output polymer desalination membrane design model evaluates the standard of desalination membrane samples, namely the salt rejection rate, and its expression is as follows:
[0085] Salt rejection rate = 1-(sodium ions + chloride ions passing through) / (total sodium ions + chloride ions).
[0086] It should be noted that step S4 also includes step S4-5, which is used to adjust the parameter variables and retrain through the retraining unit if the result does not meet the training target requirements or exceeds the set rounds until the training target requirements are met. If the requirements cannot be met after multiple times, return to step S3 for redesign.
[0087] Understandably, the polymer desalination membrane design model uses neural networks to learn graph structure data, extract and discover features and patterns in molecular structure graphs, and integrates graph convolution and attention mechanisms.
[0088] S5. Predict the chemical structure of the seawater desalination membrane that meets the performance requirements of seawater desalination through the polymer seawater desalination membrane design model.
[0089] like Figure 3 As shown, this is a schematic diagram of the seawater desalination process provided by an embodiment of the present invention.
[0090] Embodiment 2 of the invention provides an information data processing terminal for implementing the design of a polymer seawater desalination membrane based on a graph neural network, comprising a memory and a processor that can communicate with each other.
[0091] Embodiment 3 of the present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program including instructions that, when executed on a computer, enable the computer to execute a method for designing a polymer desalination membrane based on a graph neural network. The present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program including instructions that, when executed on a computer, enable the computer to execute a method for designing a polymer desalination membrane based on a graph neural network.
[0092] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an all-hardware embodiment, an all-software embodiment, or a combination of hardware and software embodiments. Furthermore, the present invention may also take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0093] According to embodiments of the application, the application is described with reference to methods, flowcharts and / or block diagrams, and computer program products of the device (system). Computer program instructions should be understood as each process in the flowchart and / or block diagram and / or block in the block diagram and / or each process in the flowchart and / or block diagram and / or block diagram. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce such machine instructions for execution by the processor of the computer or other programmable data processing device to produce a device that implements the function specified in the flowchart or multiple processes and / or block or block diagram.
[0094] These computer program instructions may also be stored in a computer-readable memory that directs a computer or other programmable data processing device to operate in a specific manner, such that the instructions stored in the computer-readable memory produce a product of manufacture including an instruction device that implements the functions specified in the flowchart process or processes and / or one or more block diagrams.
[0095] These computer program instructions may also be loaded onto a computer or other programmable data processing device so as to execute a series of operational steps on the computer or other programmable device to produce a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in the flowchart process or multiple processes and / or one or more block diagrams.
[0096] The specific embodiments described herein are intended only to illustrate the present invention and are not intended to define the present invention. Various modifications and variations are possible for those skilled in the art. Any modifications or variations made by those skilled in the art, without requiring creative effort, in accordance with the spirit and principles of the invention, fall within the scope of the invention.
Claims
1. A design method for polymer desalination membrane based on graph neural network, characterized in that: The steps include: S1. Build a knowledge graph of elements, molecular groups, and molecular structures based on chemical knowledge, and construct a molecular structure information dataset. The data in the molecular structure information dataset includes the atoms in the molecule, the related chemical bonds connecting them, and the characteristics of the molecule itself. S2. Obtain relevant data on the design and production of polymer desalination membranes from literature and other databases to establish a seawater desalination membrane database. Combined with the analysis of the structural information dataset, featurization processing suitable for graph neural networks is performed to generate an adjacency matrix. The relevant data in the seawater desalination membrane database includes the test temperature, seawater desalination performance, and molecular structure of the seawater desalination membrane; S3. Build a graph neural network, use the atoms of the desalination membrane polymer as nodes of the graph, molecular bonds as edges, add global nodes, generate a graph neural network, and initialize the graph neural network to obtain the initial input of the graph neural network, wherein the global nodes include the overall information of the desalination membrane polymer monomer; S4. Train the graph neural network to output a polymer desalination membrane design model; S5. Predict the chemical structure of the seawater desalination membrane that meets the performance requirements of seawater desalination through the polymer seawater desalination membrane design model.
2. The design method of polymer seawater desalination membrane based on graph neural network according to claim 1 is characterized in that: In step S1, the data of the knowledge graph needs to be screened and cleaned before constructing the molecular structure information dataset. The data of the knowledge graph includes the electronegativity and oxidizability of molecules and groups, the synthesis difficulty and valence electrons of compounds. The method for screening and cleaning the data of the knowledge graph is: using an efficient string search and replacement algorithm flashtext to match in literature and databases, screen relevant data, and combine relevant chemical properties with comparison of similar data to clean up data with obvious errors and problems.
3. The design method of polymer desalination membrane based on graph neural network according to claim 1 is characterized in that: In step S2, the characterization processing method for generating the adjacency matrix is: using chemdraw software to convert the obtained molecular structure into a simplified molecular linear input specification SMILES, and then using the GetAdjacencyMatrix method of the rdkit library to obtain the molecular adjacency matrix.
4. The design method of polymer desalination membrane based on graph neural network according to claim 1 is characterized in that: In step S3, when an atom is used as a node of the graph, the sequence number of the atom and the number of atoms adjacent to it and the related properties of the atom are input, so that the atom is used as a node; when a molecular bond is used as an edge, the sequence number of the molecular bond and its weight, the related properties of the molecular bond are input, so that the molecular bond is used as an edge. In addition, the number of nodes and the longest path property of the graph are added to the additional molecular information provided by the global node.
5. The design method of polymer seawater desalination membrane based on graph neural network according to claim 4 is characterized in that: The initialization method of the graph neural network in step S3 is: When the main architecture of the graph neural network is initialized, the embedding layer of the node is generated according to the neighboring nodes. The embedding layer is in the form of: Where v represents the index of the atomic node in the graph, k represents the current convolutional layer number, N(v) represents all neighbors of node v, σ is the activation function LeakyReLU, W and B are the model parameters to be learned, edges have independent parameters and perform similar operations, and h represents the weight and bias of each node in this layer of the graph neural network; A fully connected layer is set after the embedding layer to output the results; Initialize the weights and biases of the graph neural network to get the initial input of the network neural network: Where X represents the input matrix of this layer of the graph neural network, including the relevant properties of the atoms of the desalinated membrane polymer.
6. The method for designing a polymer desalination membrane based on graph neural network according to claim 5, characterized in that: The training of the graph neural network in step S4 includes the following steps: S4-1. Normalize and perform data augmentation on the initial molecular input of the graph neural network to facilitate convergence and improve accuracy. S4-2. Construct an embedding layer trained with a graph neural network as the main body. For each layer, calculate the linear combination of input and weight, and then apply the activation function to the linear combination value, with different model parameters for nodes and edges; S4-3. Each layer performs aggregation operations on nodes and edges, and aggregates the weights of neighboring nodes and global nodes into the nodes; S4-4. Finally, the fully connected layer is used for training, the loss function is calculated, and the polymer desalination membrane design model is output.
7. The method for designing a polymer desalination membrane based on graph neural network according to claim 6, characterized in that: The standard for evaluating the desalination membrane samples using the polymer desalination membrane design model output in step S4-4, i.e., the salt rejection rate, is expressed as follows: Salt rejection rate = 1-(sodium ions + chloride ions passing through) / (total sodium ions + chloride ions).
8. The method for designing a polymer desalination membrane based on graph neural network according to claim 6, characterized in that: The polymer desalination membrane design model uses graph neural networks to learn graph structure data, extract and discover features and patterns in molecular structure graphs, and integrates graph convolution and attention mechanisms.
9. A computer device comprising a memory and a processor, and a program stored in the processor and executable on the processor, wherein: When executing the program, the processor implements the design method of polymer seawater desalination membrane based on graph neural network as described in 1-8.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the design method of the polymer seawater desalination membrane based on the graph neural network as described in 1-8 is implemented.