Electrolyte conductivity prediction method and system and computer readable storage medium

Abstract

The invention discloses an electrolyte conductivity prediction method and system and a computer readable storage medium, and belongs to the technical field of machine learning, and the method comprises the steps: S1, constructing a large-scale chemical molecule data set and an electrolyte conductivity data set; s2, importing a large-scale chemical molecule data set into a variational graph auto-encoder for training, and learning potential feature representation of chemical molecules through the synergistic effect of the encoder and a decoder; s3, utilizing the trained variational graph auto-encoder to extract potential feature representation of chemical molecules in the electrolyte conductivity data set, and inputting the potential feature representation into the electrolyte conductivity prediction model based on the attention mechanism for training; and S4, predicting the conductivity of the electrolyte by using the electrolyte conductivity prediction model. According to the method, conductivity prediction can be modeled under the condition of a small number of electrolyte conductivity samples, the mixing behavior of a solvent and an additive in the electrolyte is simulated in combination with an attention mechanism, and the prediction capacity of the electrolyte conductivity is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning, and in particular relates to a method and system for predicting the conductivity of an electrolyte, and a computer-readable storage medium. Background Art

[0002] With the rapid development of machine learning and artificial intelligence technology, these advanced technologies have begun to cross-combine with various traditional scientific fields, thus promoting the rapid development of various industries. Especially in the field of lithium battery materials, the application of effective machine learning technology has enabled material developers to make more intelligent and accurate decisions, significantly accelerating the efficiency of new material research and development. Especially in the electrolyte design process, ionic conductivity, as one of the most important performance indicators, must be fully considered and optimized. The traditional trial-and-error experimental method can only test the conductivity of a specific solution formula combination of a lithium salt, several fixed solvents and an additive at a time. It takes a lot of manpower and time to complete the combination and concentration screening of multiple lithium salts, multiple solvents and different additives, and the research and development cycle is very long. Therefore, it is urgent to use artificial intelligence methods to predict electrolyte conductivity in order to improve the efficiency of material research and development. Summary of the invention

[0003] In view of the deficiencies of the prior art, the present invention provides a method, system and computer-readable storage medium for predicting the conductivity of electrolytes, which overcome the problems that the sample size is insufficient when the existing machine learning technology predicts the conductivity of electrolytes, the graph convolutional neural network cannot fully learn the potential feature representation of molecules, and the linear weighted summation method cannot simulate the highly complex interaction problem of the components in the electrolyte. The variational graph autoencoder trained on a large-scale chemical molecular structure data set is used to fully extract the potential feature representation of chemical molecules in the electrolyte, and the attention mechanism is combined to simulate the mixing behavior of solvents and additives in the electrolyte to improve the ability to predict the conductivity of the electrolyte.

[0004] To achieve the above-mentioned object of the invention, the first object of the present invention is to provide a method for predicting the conductivity of an electrolyte, comprising the following steps:

[0005] S1. Build large-scale chemical molecular datasets and electrolyte conductivity datasets;

[0006] S2, importing the large-scale chemical molecule dataset into the variational graph autoencoder for training, and learning the potential feature representation of the chemical molecules through the synergy of the encoder and the decoder;

[0007] S3. Use the trained variational graph autoencoder to extract the potential feature representation of chemical molecules in the electrolyte conductivity dataset, and input it into the electrolyte conductivity prediction model based on the attention mechanism for model training;

[0008] S4. Use the trained electrolyte conductivity prediction model to predict the conductivity of the electrolyte.

[0009] Preferably, S1 comprises:

[0010] Through web crawler technology, collect the structural information of small organic molecules and inorganic salts contained in public chemical databases, literature and other academic resources;

[0011] Use chemical informatics tools to convert raw molecular structure data into "SMILES" strings to form a large-scale chemical molecule dataset;

[0012] Obtain the conductivity value of the electrolyte formula obtained by experiment or theoretical calculation. There are no less than 20,000 types of electrolyte formulas, and the formula of each electrolyte should include the following information: the "SMILES" string of the structural data of the solvent, additives and lithium salt converted by chemical informatics tools, the weight fraction of the solvent and additive, the mass molar concentration of the lithium salt and the electrolyte temperature (K);

[0013] Convert the conductivity value of the electrolyte to a logarithmic value with a base of 10 in units of mS / cm;

[0014] Get graph data of chemical molecules. Each atom in the molecule is regarded as a node V in the graph data structure, and the bonds within the molecule are regarded as edges E connecting different nodes.

[0015] The "SMILES" string of the molecule is input into the open source chemical information software package RDKit23 to obtain the feature matrix X of node V and the adjacency matrix A of edge E. The above steps complete the construction of large-scale chemical molecule datasets and electrolyte conductivity datasets.

[0016] Preferably, the cheminformatics tool comprises Open Babel.

[0017] Preferably, node V contains the following information: atom type, number, number of connected atoms, formal charge, hybridization mode, van der Waals radius, covalent radius, number of connected hydrogen atoms, relative atomic mass, whether it is in a ring, and whether it is an aromatic hydrocarbon; edge E contains the following information: type of bond between atoms, bond direction, bond strength, whether it is conjugated, and whether it is a ring.

[0018] Preferably, S2 includes: a variational graph autoencoder is trained using a large-scale chemical molecule dataset, which generates a latent representation based on the input chemical molecule graph representation, and includes an encoding end and a decoding end, the encoder maps the input graph structure data to a latent space representation, and the decoder converts the latent space representation back to the input original graph data structure; when the output of the decoding end is close to the input of the encoding end, the latent space representation reflects the input molecular graph data structure; the variational graph autoencoder learns the latent space representation characteristics of molecules through a large-scale chemical molecule dataset, so as to better cope with the conductivity prediction task in the electrolyte conductivity dataset.

[0019] Preferably, S3 includes:

[0020] The graph data of solvents, additives and lithium salt molecules of each formula in the electrolyte conductivity dataset are input into the variational graph compiler trained using a large-scale chemical molecule dataset to obtain the potential feature representation of the molecule {Z sol ,Z add ,Z salt}; The electrolyte is composed of multiple solvents or additives and a lithium salt. The attention mechanism is first used to represent the mixing behavior of the solvent and additive. For the potential feature representation Z of the solvent and additive molecular graph data, first Q = Z·W Q , K=Z·W K , V=Z·W V , calculate the query Q, key K and value V, where W Q , W k , W v is a learnable parameter; then, the correlation score between molecules is calculated using dot product. If there is a strong interaction between molecules, they have a higher correlation score, and multiply it with V to obtain the updated representation Z′, as shown in the following formula:

[0021]

[0022] Where, d k is the length of the vector k in the key K; since the attention mechanism shares the parameter operation on each molecule representation separately, the parameter W can be learned Q , W K , W V The same arrangement is maintained, and then the weight fraction ratio, the mixture of solvent and additive is expressed as Z mix is calculated as a weighted sum:

[0023]

[0024] Where W i is the weight fraction of solvent or additive, Z i′ is the latent feature representation of the solvent or additive after being updated by the attention mechanism, and then the mixture of the solvent and the additive is represented by Z mix , the potential feature representation of lithium salt molecules, the mass molar concentration of lithium salt, and the logarithmic electrolyte temperature (K) are associated and mapped to the logarithmic conductivity of the electrolyte through a fully connected neural network:

[0025] lgσ=FNN(Z mix ,Z salt ,M salt ,1 / lgK)

[0026] In the formula, lgσ is the logarithmic conductivity, FNN is the fully connected neural network, Z salt is the potential characteristic representation of lithium salt molecules, M salt is the mass molar concentration of lithium salt, and lgK is the logarithmic electrolyte temperature.

[0027] Preferably, S4 includes: firstly, inputting the graph data of the solvent, additive and lithium salt molecules in the electrolyte formula into a variational graph compiler trained using a source domain to obtain a potential feature representation of the molecules, and then inputting the potential feature representation of the molecules, the weight fraction ratio of the solvent and additive, the mass molar concentration of the lithium salt and the electrolyte temperature into a trained electrolyte conductivity prediction model to perform conductivity prediction.

[0028] A second object of the present invention is to provide a conductivity prediction system for an electrolyte, comprising:

[0029] Dataset module, which builds large-scale chemical molecular datasets and electrolyte conductivity datasets;

[0030] A variational graph autoencoder module imports the large-scale chemical molecule dataset into a variational graph autoencoder for training, and learns the potential feature representation of chemical molecules through the synergy of the encoder and the decoder;

[0031] The electrolyte conductivity prediction model uses the trained variational graph autoencoder to extract the potential feature representation of chemical molecules in the electrolyte conductivity dataset and inputs it into the electrolyte conductivity prediction model based on the attention mechanism for model training;

[0032] The prediction execution module uses the trained electrolyte conductivity prediction model to predict the electrolyte conductivity.

[0033] A third object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned method for predicting the conductivity of an electrolyte.

[0034] A fourth object of the present invention is to provide a computer program product, comprising a computer program, which implements the above-mentioned method for predicting the conductivity of the electrolyte when executed by a processor.

[0035] The advantages and positive effects of this application are:

[0036] The core of the present invention is to use a variational graph autoencoder trained with a large-scale chemical molecular structure data set, which can deeply explore and fully extract the potential feature representations of chemical molecules in the electrolyte. Through this advanced technical means, we can better extract the feature representations of chemical molecules in the electrolyte. In addition, the present invention also combines an attention mechanism, which can simulate the complex mixing behavior between solvents and additives in the electrolyte, so as to accurately express their interactions. In this way, the present invention significantly improves the ability to predict the conductivity of the electrolyte, making the prediction results more accurate and reliable.

[0037] Another major advantage of the present invention is that it can effectively model conductivity prediction even when the number of electrolyte conductivity samples is limited. This is especially important for scenarios where experimental resources are limited or it is difficult to obtain a large number of samples. By combining the attention mechanism, the present invention can simulate the mixing behavior of solvents and additives in the electrolyte, further improving the predictive ability of electrolyte conductivity. This improvement in predictive ability not only helps to optimize the formulation of the electrolyte, but also provides important guidance in the battery design and manufacturing process, thereby promoting the advancement of battery technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0039] Figure 1 A flow chart provided by a preferred embodiment of the present invention is shown;

[0040] Figure 2 A schematic diagram showing the principle of a variational graph autoencoder in a preferred embodiment of the present invention is shown;

[0041] Figure 3 A model framework diagram of a preferred embodiment of the present invention is shown. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] Theoretical basis:

[0044] The variational graph autoencoder is a deep learning model that cleverly combines graph convolutional neural networks with variational autoencoders and is specifically designed to process graph-structured data, such as molecular structure data. When processing large-scale molecular data sets, the variational graph autoencoder can effectively capture the complex interactions and relationships between atoms and chemical bonds in the molecular structure through the collaborative work of its encoding and decoding ends. This model deeply learns the latent space representation of molecules, thereby enabling tasks such as accurate molecule generation, classification, and regression, providing a powerful tool for the fields of chemistry and materials science.

[0045] In the field of electrolyte mixture analysis, the use of attention mechanism is fundamentally different from the traditional linear weighted summation method. The attention mechanism can dynamically adjust the contribution weight of each molecule to the final analysis result according to its importance in the mixture. This mechanism enables the model to automatically identify and select the most relevant molecular features when aggregating information, thereby achieving a more accurate contribution assessment. In addition, the attention mechanism has the significant advantage of permutation invariance, which means that no matter how the order of molecules in the mixture changes, the model can stably output consistent and accurate results, which is of great significance for understanding and analyzing the properties of complex mixtures.

[0046] See also Figures 1 to 3 ;

[0047] The first embodiment is a method for predicting the conductivity of an electrolyte, comprising the following steps:

[0048] S1: Dataset Construction

[0049] Using web crawler technology, we collected the structural information of small organic molecules and inorganic salts contained in public chemical databases, literature and other academic resources. Then, with the help of chemical informatics tools (such as Open Babel, etc.), we converted these raw molecular structure data into "SMILES" strings, and then built a large chemical molecule dataset containing no less than 30,000 records.

[0050] Collect the conductivity data of the electrolyte formula obtained through experiments or theoretical calculations to ensure that there are no less than 2,000 types of electrolyte formulas. Each electrolyte formula should record the structural data of the solvent, additives and lithium salt in detail, and these data are converted into "SMILES" strings through chemical informatics tools. At the same time, record key information such as the weight fraction of the solvent and additives, the concentration of the lithium salt and the electrolyte temperature. For the conductivity value of the electrolyte, it should be converted to a logarithmic value with a base of 10 and in units of mS / cm to facilitate the training of the conductivity prediction model.

[0051] Obtain graph data of chemical molecules, where each atom in the molecule can be regarded as a node V in the graph data structure, and the bonding within the molecule can be regarded as the edge E connecting different nodes. Node V usually contains the following information: atom type, number, number of connected atoms, formal charge, hybridization mode, van der Waals radius, covalent radius, number of connected hydrogen atoms, relative atomic mass, whether it is located in the ring, whether it is aromatic hydrocarbon, etc. Edge E usually contains the following information: type of interatomic bond (single bond, double bond, triple bond and aromatic bond), bond direction, bond strength, whether it is conjugated, and whether it forms a ring structure. Input the "SMILES" string of the molecule into the open source chemical information software package RDKit23 to obtain the feature matrix X of node V and the adjacency matrix A of edge E. The above steps complete the construction of large-scale chemical molecule datasets and electrolyte conductivity datasets.

[0052] S2: Construction of variational graph editor

[0053] The variational graph autoencoder is trained using a large-scale chemical molecule dataset. It generates a latent representation based on the input chemical molecule graph representation. It consists of two parts: the encoder and the decoder. The encoder maps the input graph structure data to a latent space representation, while the decoder converts the latent space representation back to the input original graph data structure. When the output of the decoder is as close as possible to the input of the encoder, the latent space representation can reflect the input molecular graph data structure. In this process, the variational graph autoencoder learns the latent space representation features of molecules through the source domain (large-scale chemical molecule dataset) to better cope with the conductivity prediction task in the target domain (electrolyte conductivity dataset).

[0054] Considering that the sample size of the electrolyte conductivity dataset is insufficient, the graph convolutional network cannot fully learn the latent feature representation of chemical molecules. We train an autoencoder to learn the feature representation of the source domain, and then apply the learned latent feature representation to the task in the target domain. The variational graph autoencoder is a deep learning model that combines graph convolutional networks and variational autoencoders, which is particularly suitable for processing graph data. Figure 2As shown in the figure, the variational graph autoencoder can effectively capture the complex relationship between atoms and chemical bonds in the molecular structure through the synergy of the encoder and decoder in the source domain, fully learn the potential feature representation of chemical molecules, and then achieve accurate regression tasks in the target domain. The encoder of the variational graph autoencoder is composed of a graph convolutional neural network, which takes the adjacency matrix A and the feature matrix X as input and generates the potential feature representation Z as output. A 2-layer graph convolutional neural network is defined as:

[0055]

[0056] In the formula, GCN is a graph convolutional neural network, ReLU is the activation function, is a symmetric normalized adjacency matrix, and D is the degree matrix corresponding to A. The first layer of graph convolution generates a low-dimensional feature matrix, and the second layer u = GCN u (X, A) generates the mean u of the node vector i and logσ = GCN σ (X, A) generates the variance σ of the node vector i .GCN u (X, A) and GCN σ (X, A) share the parameter W0, and their respective W l Then, we use the reparameterization trick to calculate Z:

[0057] Z=μ+σ⊙∈

[0058] Where ⊙ is the element-wise multiplication, In summary, the encoder can be expressed as:

[0059]

[0060] The decoder will use the inner product of the learned latent feature representation Z to reconstruct the adjacency matrix:

[0061]

[0062] In the formula, A ij is an element in the matrix A. The decoder outputs the reconstructed adjacency matrix Using large-scale molecular data to train and optimize the variational graph autoencoder and A as close as possible, then the latent feature representation Z can reflect the graph data of the chemical molecule.

[0063] S3: Conductivity prediction model construction

[0064] The graph data of solvents, additives and lithium salt molecules of each formula in the electrolyte conductivity dataset are input into the variational graph compiler trained using a large-scale chemical molecule dataset to obtain the potential feature representation of the molecule {Zsol ,Z add ,Z salt}; The electrolyte is composed of multiple solvents or additives and a lithium salt. The attention mechanism is first used to represent the mixing behavior of the solvent and additive. For the potential feature representation Z of the solvent and additive molecular graph data, first Q = Z·W Q , K=Z·W K , V=Z·W V , calculate the query Q, key K and value V, where W Q , W k , W v is a learnable parameter; then, the correlation score between molecules is calculated using dot product. If there is a strong interaction between molecules, they have a higher correlation score, and multiply it with V to obtain the updated representation Z′, as shown in the following formula:

[0065]

[0066] Where, d k is the length of the vector k in the key K; since the attention mechanism shares the parameter operation on each molecule representation separately, the parameter W can be learned Q , W K , W V The same arrangement is maintained, and then the weight fraction ratio, the mixture of solvent and additive is expressed as Z mix is calculated as a weighted sum:

[0067]

[0068] Where W i is the weight fraction of solvent or additive, Z i ′ is the latent feature representation of the solvent or additive after being updated by the attention mechanism, and then the mixture of the solvent and the additive is represented by Z mix , the potential feature representation of lithium salt molecules, the mass molar concentration of lithium salt, and the logarithmic electrolyte temperature (K) are associated and mapped to the logarithmic conductivity of the electrolyte through a fully connected neural network:

[0069] lgσ=FNN(Z mix ,Z salt ,M salt ,1 / lgK)

[0070] In the formula, lgσ is the logarithmic conductivity, FNN is the fully connected neural network, Z salt is the potential characteristic representation of lithium salt molecules, M salt is the mass molar concentration of lithium salt, and lgK is the logarithmic electrolyte temperature.

[0071] S4: Electrolyte conductivity prediction

[0072] By inputting the graph data of the solvent, additive, and lithium salt molecules in the electrolyte formula into the variational graph autoencoder pre-trained with source domain data, the potential feature representations of these molecules can be obtained. These potential feature representations are compressed encodings of molecular structure and chemical properties, which can capture the intrinsic properties of molecules. Subsequently, the potential feature representations of these molecules, the weight fraction of solvents and additives, the mass molar concentration of lithium salts, and the electrolyte temperature are input into a fully trained conductivity prediction model. This model is based on the attention mechanism and can predict the conductivity of the electrolyte based on the input feature data. In this way, we can effectively predict the conductivity of different electrolyte formulas, thereby providing strong data support for the design and optimization of electrolytes.

[0073] A second embodiment is a conductivity prediction system for an electrolyte, which is used to implement the method of the first embodiment. The system includes:

[0074] In the dataset module, this embodiment is dedicated to constructing a large-scale chemical molecule dataset, which contains rich chemical molecule information. At the same time, an electrolyte conductivity dataset is also constructed, which records in detail the conductivity performance of different electrolytes under different conditions.

[0075] The role of the variational graph autoencoder module is to import the large-scale chemical molecule dataset constructed above into the variational graph autoencoder for in-depth training. In this process, the encoder and decoder work together to learn and extract the potential feature representations of chemical molecules, which are abstractions and compressions of the intrinsic properties of chemical molecules.

[0076] The electrolyte conductivity prediction model is based on the potential feature representation of chemical molecules extracted by the previously trained variational graph autoencoder. We use these feature representations as input and combine them with the prediction model based on the attention mechanism to further train the model. The attention mechanism can help the model focus more accurately on the key factors affecting the electrolyte conductivity, thereby improving the accuracy of the prediction.

[0077] The prediction execution module is the part of the whole system that is responsible for the actual prediction work. It uses a fully trained electrolyte conductivity prediction model to predict the conductivity of new electrolyte samples. The purpose of this module is to transform theoretical research into practical applications and provide a scientific basis for the design and optimization of electrolytes.

[0078] A third embodiment is a computer-readable storage medium storing a computer program, which implements the above-mentioned method for predicting the conductivity of an electrolyte when executed by a processor.

[0079] A fourth embodiment is a computer program product, comprising a computer program, wherein the computer program implements the above-mentioned method for predicting the conductivity of an electrolyte when executed by a processor.

[0080] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When the use is implemented in whole or in part in the form of a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL) or wireless (e.g., infrared, wireless, microwave, etc.) mode) to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk SolidState Disk (SSD)), etc.

[0081] The above is only a preferred embodiment of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for predicting the conductivity of an electrolyte, characterized in that: include: S1. Build large-scale chemical molecular datasets and electrolyte conductivity datasets; S2, importing the large-scale chemical molecule dataset into the variational graph autoencoder for training, and learning the potential feature representation of the chemical molecules through the synergy of the encoder and the decoder; S3. Use the trained variational graph autoencoder to extract the potential feature representation of chemical molecules in the electrolyte conductivity dataset, and input it into the electrolyte conductivity prediction model based on the attention mechanism for model training; S4. Use the trained electrolyte conductivity prediction model to predict the conductivity of the electrolyte.

2. The method for predicting the conductivity of an electrolyte according to claim 1, characterized in that: S1 includes: Through web crawler technology, collect the structural information of small organic molecules and inorganic salts contained in public chemical databases, literature and other academic resources; Use chemical informatics tools to convert raw molecular structure data into "SMILES" strings to form a large-scale chemical molecule dataset; Obtain the conductivity value of the electrolyte formula obtained by experiment or theoretical calculation. There are no less than 20,000 types of electrolyte formulas, and the formula of each electrolyte should include the following information: "SMILES" string after the structural data of solvents, additives and lithium salts are converted by chemical informatics tools, the weight fraction of solvents and additives, the mass molar concentration of lithium salts and the electrolyte temperature (K); Convert the conductivity value of the electrolyte to a logarithmic value with a base of 10 in units of mS / cm; Get graph data of chemical molecules. Each atom in the molecule is regarded as a node V in the graph data structure, and the bonds within the molecule are regarded as edges E connecting different nodes. The "SMILES" string of the molecule is input into the open source chemical information software package RDKit23 to obtain the feature matrix X of node V and the adjacency matrix A of edge E. Through the above steps, the construction of large-scale chemical molecule datasets and electrolyte conductivity datasets is completed.

3. The method for predicting the conductivity of an electrolyte according to claim 2, characterized in that: The cheminformatics tools include Open Babel.

4. The method for predicting the conductivity of an electrolyte according to claim 2, characterized in that: Node V contains the following information: atom type, number, number of connected atoms, formal charge, hybridization mode, van der Waals radius, covalent radius, number of connected hydrogen atoms, relative atomic mass, whether it is in a ring, and whether it is an aromatic hydrocarbon; edge E contains the following information: type of bond between atoms, bond direction, bond strength, whether it is conjugated, and whether it is a ring.

5. The method for predicting the conductivity of an electrolyte according to claim 1, characterized in that: S2 includes: the variational graph autoencoder is trained with a large-scale chemical molecule dataset, which generates a latent representation based on the input chemical molecule graph representation, including an encoding end and a decoding end. The encoder maps the input graph structure data to a latent space representation, while the decoder converts the latent space representation back to the input original graph data structure; when the output of the decoding end is close to the input of the encoding end, the latent space representation reflects the input molecular graph data structure; the variational graph autoencoder learns the latent space representation characteristics of molecules through large-scale chemical molecular structures, so as to better cope with the conductivity prediction task in the electrolyte conductivity dataset.

6. The method for predicting the conductivity of an electrolyte according to claim 1, characterized in that: S3 includes: The graph data of solvents, additives and lithium salt molecules of each formula in the electrolyte conductivity dataset are input into the variational graph compiler trained using a large-scale chemical molecule dataset to obtain the potential feature representation of the molecule { sol ,Z add ,Z salt }; The electrolyte is composed of multiple solvents or additives and a lithium salt. The attention mechanism is first used to represent the mixing behavior of the solvent and additive. For the potential feature representation Z of the solvent and additive molecular graph data, first Q = Z·W Q , K=Z·W K , V=Z·W V , calculate the query Q, key K and value V, where W Q , W k , W v is a learnable parameter; then, the correlation score between molecules is calculated using dot product. If there is a strong interaction between molecules, they have a higher correlation score, and multiply it with V to obtain the updated representation Z′, as shown in the following formula: Where, d k is the length of the vector k in the key K; since the attention mechanism shares the parameter operation on each molecule representation separately, the parameter W can be learned Q , W K , W V The same arrangement is maintained, and then the weight fraction ratio, the mixture of solvent and additive is expressed as Z mix is calculated as a weighted sum: Where W i is the weight fraction of solvent or additive, Z i ′ is the latent feature representation of the solvent or additive after being updated by the attention mechanism, and then the mixture of the solvent and the additive is represented by Z mix , the potential feature representation of lithium salt molecules, the mass molar concentration of lithium salt, and the logarithmic electrolyte temperature (K) are associated and mapped to the logarithmic conductivity of the electrolyte through a fully connected neural network: lgσ=FNN(Z mix ,Z salt ,M salt ,1 / lgK) In the formula, lgσ is the logarithmic conductivity, FNN is the fully connected neural network, Z salt is the potential characteristic representation of lithium salt molecules, M salt is the mass molar concentration of lithium salt, and lgK is the logarithmic electrolyte temperature.

7. The method for predicting the conductivity of an electrolyte according to claim 1, characterized in that: S4 includes: First, the graph data of the solvent, additive and lithium salt molecules in the electrolyte formula are input into the variational graph compiler trained using the source domain to obtain the potential feature representation of the molecules. Then, the potential feature representation of the molecules, the weight fraction of the solvent and additive, the mass molar concentration of the lithium salt and the electrolyte temperature are input into the trained electrolyte conductivity prediction model to perform conductivity prediction.

8. A conductivity prediction system for an electrolyte, characterized in that: include: Dataset module, which builds large-scale chemical molecular datasets and electrolyte conductivity datasets; A variational graph autoencoder module imports the large-scale chemical molecule dataset into a variational graph autoencoder for training, and learns the potential feature representation of chemical molecules through the synergy of the encoder and the decoder; The electrolyte conductivity prediction model uses the trained variational graph autoencoder to extract the potential feature representation of chemical molecules in the electrolyte conductivity dataset and inputs it into the electrolyte conductivity prediction model based on the attention mechanism for model training; The prediction execution module uses the trained electrolyte conductivity prediction model to predict the electrolyte conductivity.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for predicting the conductivity of the electrolyte according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the program is executed by a processor, the method for predicting the conductivity of the electrolyte as described in any one of claims 1 to 7 is implemented.

Images