A method and device for predicting physical properties of electrolyte based on attention mechanism

Abstract

The application relates to a method and device for predicting physical properties of electrolyte based on an attention mechanism, applied to the field of chemical property prediction, wherein the method comprises the following steps: obtaining a to-be-trained electrolyte in a preset electrolyte formula database; the electrolyte formula database comprises formula component information of the to-be-trained electrolyte, the formula component information comprises chemical group features and site data; according to the to-be-trained electrolyte, a preset self-recurrent pre-training attention model is subjected to mask training to obtain a target electrolyte prediction model; formula component information of a to-be-predicted electrolyte is obtained, the formula component information of the to-be-predicted electrolyte is input into the target electrolyte prediction model, and target physical property prediction information of the to-be-predicted electrolyte is obtained. Through the application, the application range of electrolyte physical property prediction is improved, and the accuracy of the prediction result is higher.

Description

Technical Field

[0001] This application relates to the field of chemical property prediction, and in particular to a method and apparatus for predicting the physical properties of electrolytes based on an attention mechanism. Background Technology

[0002] Common electrolytes include lithium salts, solvents, and additives. The physical properties of electrolytes are closely related to their chemical composition and proportions. Chemically, electrolyte components include various salt groups, organic groups, and active coordinating elements. Due to the differences in various electrolytes and their components, there is a high demand for exploring the chemical space of electrolytes, thus requiring optimization of electrolyte performance.

[0003] Current physical prediction methods for electrolyte properties involve multiple aspects, including theoretical calculations, model building, and experimental verification. For example, solving Newton's equations of motion can simulate the trajectories of all particles in the system. In electrolyte research, molecular dynamics simulations can be used to predict the diffusion coefficient, conductivity, viscosity, and structural properties of ions in the solvent; or quantum chemical calculations can be used to predict important information such as energy changes during dissolution, the interaction energy between ions and the solvent, and reaction pathways.

[0004] However, different prediction methods may be applicable to different types of electrolyte systems. For example, some molecular dynamics simulations based on classical force fields may not accurately describe complex ion-solvent interactions. Similarly, quantum chemical calculations, due to computational resource limitations, can typically only handle small-scale systems (e.g., a few to tens of atoms), which limits their direct applicability to real electrolyte environments; furthermore, due to factors such as model simplification, parameter selection, and approximations, the results may contain some degree of bias. Therefore, existing methods for predicting electrolyte physical properties have a narrow scope of application, and the accuracy obtained also contains errors.

[0005] There is currently no effective solution to the problem that the methods for predicting the physical properties of electrolytes have a narrow scope of application and low accuracy. Summary of the Invention

[0006] This embodiment provides a method and apparatus for predicting the physical properties of electrolytes based on an attention mechanism, in order to solve the problems of narrow applicability and low accuracy of related technologies for predicting the physical properties of electrolytes.

[0007] Firstly, this embodiment provides a method for predicting the physical properties of electrolytes based on an attention mechanism, the method comprising:

[0008] Obtain the electrolyte to be trained from the preset electrolyte formulation database; the electrolyte formulation database includes formulation component information of the electrolyte to be trained, and the formulation component information includes chemical group characteristics and site data;

[0009] Based on the electrolyte to be trained, a pre-set autoregressive pre-trained attention model is trained with a mask to obtain a target electrolyte prediction model.

[0010] Obtain the formulation component information of the electrolyte to be predicted, input the formulation component information of the electrolyte to be predicted into the target electrolyte prediction model, and obtain the target physical property prediction information of the electrolyte to be predicted.

[0011] In some embodiments, the pre-defined autoregressive pre-trained attention model includes at least one encoding module and at least one decoding module; both the encoding module and the decoding module include a self-attention layer.

[0012] The step of performing masked training on a pre-set autoregressive pre-trained attention model based on the electrolyte to be trained, to obtain a target electrolyte prediction model, includes:

[0013] The formulation component information of the electrolyte to be trained is masked to obtain the formulation mask component information;

[0014] The formula mask component information is input into the encoding module, and the intermediate prediction features of the electrolyte to be trained are obtained based on the self-attention layer;

[0015] The formulation mask component information of the electrolyte to be trained and the intermediate prediction features are input to the decoding module, and the predicted values ​​of the physical properties of the electrolyte to be trained are obtained based on the self-attention layer.

[0016] Based on the predicted values ​​of the physical properties, and the chemical group characteristics and site data in the formulation composition information of the electrolyte to be trained, the predicted loss value is calculated.

[0017] Based on the calculated predicted loss value, the preset autoregressive pre-trained attention model is iteratively trained until the predicted loss value is lower than the preset error threshold, thus obtaining the target electrolyte prediction model.

[0018] In some embodiments, the target electrolyte prediction model includes at least one target encoding module and at least one target decoding module; both the target encoding module and the target decoding module include a target self-attention layer.

[0019] The step of inputting the formulation component information of the electrolyte to be predicted into the target electrolyte prediction model to obtain the target physical property prediction information of the electrolyte to be predicted includes:

[0020] The formulation component information of the electrolyte to be predicted is input into the target electrolyte prediction model, and the formulation component information of the electrolyte to be predicted is masked to obtain prediction mask information.

[0021] The prediction mask information is input to the target encoding module, and the intermediate arbitrary feature vector of the electrolyte to be predicted is obtained based on the target self-attention layer;

[0022] The intermediate arbitrary feature vector of the electrolyte to be predicted is input to the target decoding module, and the target physical property prediction information of the electrolyte to be predicted is obtained based on the self-attention layer.

[0023] In some embodiments, the target physical property prediction information includes target solubility, and the method further includes:

[0024] Obtain the attention weights of any intermediate feature vector;

[0025] By combining the attention weights and the preset solubility improvement strategy, the prediction mask information is input into the target electrolyte prediction model to obtain the target ratio of chemical groups and sites of the electrolyte to be predicted;

[0026] The target solubility of the electrolyte to be predicted is improved based on the target ratio to obtain the improved solubility.

[0027] In some embodiments, after obtaining the target ratio of chemical groups and sites of the electrolyte to be predicted, the method further includes:

[0028] Based on the target ratio, and combined with a preset molecular simulation method, the correlation and binding strength between the chemical groups and site data are determined;

[0029] Based on the correlation and binding strength, the target groups and target sites in the chemical group and site data are determined, and the electrolyte to be predicted is reconstructed.

[0030] In some embodiments, the method further includes:

[0031] The electrolyte formulation database is updated based on the electrolyte to be predicted and the target physical property prediction information of the electrolyte to be predicted.

[0032] In some embodiments, the preset electrolyte formulation database includes known electrolytes; obtaining the electrolyte to be trained from the preset electrolyte formulation database includes:

[0033] Obtain structural information from the known electrolyte formulation composition information;

[0034] The structural information is segmented to determine the tag data contained in the known electrolyte;

[0035] Based on the tag data, the known electrolyte is processed to obtain the chemical group characteristics of the known electrolyte; the known electrolyte is then used as the electrolyte to be trained.

[0036] Secondly, this embodiment provides a device for predicting the physical properties of electrolytes based on an attention mechanism. The device includes: a data acquisition module, a model training module, and a prediction module.

[0037] The data acquisition module is used to acquire the electrolyte to be trained from a preset electrolyte formula database; the electrolyte formula database includes formula component information of the electrolyte to be trained, and the formula component information includes chemical group characteristics and site data;

[0038] The model training module is used to perform mask training on a preset autoregressive pre-trained attention model based on the electrolyte to be trained, so as to obtain a target electrolyte prediction model.

[0039] The prediction module is used to obtain the formulation component information of the electrolyte to be predicted, and input the formulation component information into the target electrolyte prediction model to obtain the target physical property prediction information of the electrolyte to be predicted.

[0040] Thirdly, this embodiment provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the attention-based method for predicting the physical properties of electrolytes described in the first aspect.

[0041] Fourthly, this embodiment provides a storage medium storing a computer program that, when executed by a processor, implements the attention-based method for predicting the physical properties of electrolytes as described in the first aspect.

[0042] Compared with related technologies, the attention-based method and apparatus for predicting the physical properties of electrolytes provided in this embodiment utilizes the component and formulation information (including chemical group characteristics and site data) of the electrolyte to be trained in an electrolyte formulation database. This information is used to perform mask training on an autoregressive pre-trained attention model to obtain a target electrolyte prediction model. The component and formulation information of the electrolyte to be predicted are then input into the target electrolyte prediction model to obtain target physical property prediction information, including physical prediction values ​​and probability distributions. Training the target electrolyte prediction model with electrolytes of various properties from the electrolyte formulation database improves the applicability of the physical property prediction; furthermore, combining the model obtained through mask training results in higher accuracy of the physical property prediction information.

[0043] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0044] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0045] Figure 1 This is a hardware structure block diagram of the terminal for the attention mechanism-based method for predicting the physical properties of electrolytes provided in this embodiment.

[0046] Figure 2 This is a flowchart of the method for predicting the physical properties of electrolytes based on an attention mechanism, provided in an embodiment of this application.

[0047] Figure 3 This is a flowchart of the electrolyte reconstruction method provided in the embodiments of this application;

[0048] Figure 4 This is a flowchart of the electrolyte physical property prediction method based on the attention mechanism provided in this specific embodiment;

[0049] Figure 5 This is a schematic diagram of the electrolyte modification process provided in this specific embodiment;

[0050] Figure 6 This is a schematic diagram of the method for modifying chemical groups and sites provided in this specific embodiment;

[0051] Figure 7 This is a structural block diagram of the attention mechanism-based electrolyte physical property prediction device provided in this embodiment. Detailed Implementation

[0052] To better understand the purpose, technical solution, and advantages of this application, the application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0053] Unless otherwise defined, the technical or scientific terms used in this application shall have the general meaning as understood by one of ordinary skill in the art to which this application pertains. Words such as “a,” “an,” “an,” “the,” “the,” and “these,” used in this application, do not indicate quantitative limitation and may be singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or modules (units) is not limited to the listed steps or modules (units) but may include steps or modules (units) not listed, or may include other steps or modules (units) inherent to such processes, methods, products, or devices. The terms “connected,” “linked,” and “coupled,” used in this application, are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. The term “multiple” used in this application refers to two or more. The "and / or" operator describes the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: A alone, A and B simultaneously, and B alone. Typically, the character " / " indicates that the objects before and after it are in an "or" relationship. The terms "first," "second," and "third," etc., used in this application are merely for distinguishing similar objects and do not represent a specific ordering of the objects.

[0054] The method embodiments provided in this example can be executed on a terminal, computer, or similar computing device. For example, it can run on a terminal. Figure 1 This is a hardware structure block diagram of the terminal for the attention-based electrolyte physical property prediction method provided in this embodiment. (See diagram for example.) Figure 1 As shown, a terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 and a memory 104 for storing data are also included. The processor 102 may be, but is not limited to, a microprocessor (MCU) or a programmable logic device (FPGA). The terminal may also include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that… Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the terminal described above. For example, the terminal may also include components that are larger than... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown are illustrated.

[0055] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the attention-based method for predicting the physical properties of electrolytes in this embodiment. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, thereby implementing the aforementioned method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0056] The transmission device 106 is used to receive or send data via a network. This network includes a wireless network provided by the terminal's communication provider. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 can be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0057] Electrolytes are the medium used in chemical batteries, electrolytic capacitors, and other similar devices. Their applications vary considerably across different industries. There are electrolytes used in biological systems, lithium-ion electrolytes used in the battery industry, and electrolytes used in electrolytic capacitors, supercapacitors, and other similar applications.

[0058] Common lithium electrolytes include lithium salts (lithium hexafluorophosphate (LiPF6), lithium bis(fluorosulfonyl)imide (LiFSI), lithium bis(trifluoromethanesulfonyl)imide (LiTFSI), and others), solvents (cyclic carbonates, chain carbonates, carboxylic acid esters, ethers, sulfites, fluorinated solvents, nitriles, and others), and additives (film-forming additives, overcharge protection additives, flame retardant additives, electrolyte stabilizers, etc.). Aqueous or hybrid lithium batteries also contain a certain amount of water.

[0059] With the increasing demand for electrolytes, significant challenges are being placed on improving battery performance, including rate capability, coulombic efficiency, low-temperature performance, high-temperature resistance, film-forming properties, cycle life, and addressing issues like gas expansion. When designing electrolytes, it is desirable to find properties such as good solubility / compatibility, high ionic conductivity, high dielectric constant and low viscosity, high boiling point and low melting point, high lithium-ion transference number, good chemical stability, effective passivation, good electrochemical stability, and low toxicity.

[0060] These physical properties of electrolytes are closely related to their chemical composition and proportions. Chemically, electrolyte components include various salt groups (SO4-, PO3-, NO3-, -NH2), organic groups (-CH3, -CH2CH3, -CH(CH3)2), and active coordinating elements (N, O, F). Traditional electrolyte design methods struggle to address the need to explore the chemical space of electrolytes in the tens of millions of cells; and existing experimental designs, such as those based on predicting the electrochemical window of lithium-ion battery electrolytes, cannot efficiently optimize electrolyte performance.

[0061] Current methods for predicting electrolyte physical properties suffer from several limitations. For example, some molecular dynamics simulations based on classical force fields may fail to accurately describe complex ion-solvent interactions. Similarly, quantum chemical calculations, due to computational resource constraints, typically only handle small-scale systems (e.g., a few to tens of atoms), limiting their direct applicability to real-world electrolyte environments. Furthermore, factors such as model simplification, parameter selection, and approximations can introduce biases into the results. Therefore, existing methods have a narrow scope for predicting electrolyte physical properties, and their accuracy is subject to error.

[0062] Therefore, based on the above problems, this embodiment provides an attention mechanism-based method for predicting the physical properties of electrolytes, such as solubility, conductivity, viscosity, and diffusion coefficient. This method has a wider range of applications for predicting electrolyte properties and provides more accurate prediction results.

[0063] Figure 2 This is a flowchart of the method for predicting the physical properties of electrolytes based on an attention mechanism, as provided in the embodiments of this application. Figure 2 As shown, the process includes the following steps:

[0064] Step S210: Obtain the electrolyte to be trained from the preset electrolyte formulation database; the electrolyte formulation database includes formulation component information of the electrolyte to be trained, and the formulation component information includes chemical group characteristics and site data.

[0065] First, a pre-defined electrolyte formulation database is constructed. This database includes databases for organic, aqueous, and mixed electrolyte formulations. The database also includes various electrolytes to be trained, as well as data on the physical properties of these electrolytes and the chemical group characteristics and sites of the electrolytes. This includes sets of chemical groups and sites, and coordination data, corresponding to electrolytes composed of lithium salts, solvents, and additives. For example, a set of ion coordination sites (N, O, F) data.

[0066] Furthermore, the preset electrolyte formulation database includes known electrolytes; obtaining the electrolyte to be trained from the preset electrolyte formulation database includes: obtaining the structural information from the formulation component information of the known electrolyte; segmenting the structural information to determine the tag data contained in the known electrolyte; processing the known electrolyte based on the tag data to obtain the chemical group characteristics of the known electrolyte; and using the known electrolyte as the electrolyte to be trained.

[0067] This involves collecting 2D or 3D structure files corresponding to known electrolytes from existing electrolyte databases, segmenting the structure files to obtain different tag data, and preprocessing the known electrolyte data based on the tag data to obtain a training electrolyte containing chemical group features and site data, with the chemical group features and site data targeting the target physical properties.

[0068] Step S220: Based on the electrolyte to be trained, perform mask training on the preset autoregressive pre-trained attention model to obtain the target electrolyte prediction model.

[0069] The electrolyte formula database is input into the neural network model for training. Training is stopped when the error is below the threshold or tends to stabilize, resulting in a trained electrolyte prediction model. The neural network model is a general autoregressive pre-trained attention model.

[0070] Specifically, the autoregressive pre-trained attention model can be an XLNet model, a BART model, a Transformer-XL model, etc., without specific limitations. XLNet is an advanced self-supervised pre-trained language model that introduces the concept of a Permutation Language Model (PLM), allowing the model to learn dependencies in a text sequence from different perspectives. PLM predicts the word at each position by training on different permutations of the input sentence. This allows the model to learn more contextual information because it considers not only the context on the left but also the context on the right, and in a non-sequential way. XLNet is based on the Transformer-XL architecture and, by introducing a segment loop mechanism and relative position encoding, enables the model to utilize information from previous segments, thereby enhancing its ability to understand long texts.

[0071] Furthermore, the pre-set autoregressive pre-trained attention model includes at least one encoding module and at least one decoding module; both the encoding module and the decoding module include an autoattention layer; the step of masking the pre-set autoregressive pre-trained attention model with the electrolyte to be trained includes: using a masking method to mask part of the electrolyte group and coordination information of the label data with a randomly set prediction order, so as to predict the group and coordination information of the masked part.

[0072] First, the formulation composition information of the electrolyte to be trained is masked to obtain the formulation mask composition information. The formulation mask composition information is then input to the encoding module, which obtains intermediate prediction features of the electrolyte to be trained based on a self-attention layer. Next, the formulation mask composition information and intermediate prediction features of the electrolyte to be trained are input to the decoding module, which obtains predicted values ​​of the physical properties of the electrolyte to be trained based on a self-attention layer. Based on the predicted physical properties, as well as the chemical group features and site data in the formulation composition information of the electrolyte to be trained, a prediction loss value is calculated. Based on the calculated prediction loss value, the preset autoregressive pre-trained attention model is iteratively trained until the prediction loss value is lower than a preset error threshold, thus obtaining the target electrolyte prediction model.

[0073] The above-mentioned mask training method based on attention mechanism is beneficial to improving the robustness of the final target electrolyte training model to mask data, and also further improves the potential relationship between the physical properties in the electrolyte.

[0074] Step S230: Obtain the formulation component information of the electrolyte to be predicted, input the formulation component information of the electrolyte to be predicted into the target electrolyte prediction model, and obtain the target physical property prediction information of the electrolyte to be predicted.

[0075] The formulation information of the electrolyte to be predicted, namely the component and formulation information, includes one or more of the following: inorganic / organic anionic groups, lithium ion coordination sites (N, O, F), hydrophobic groups, size and length (-CH3, -CH2CH3, -CH(CH3)2), polar groups (SO4-, PO3-, NO3-, -NH2) of the electrolyte to be predicted, etc., which are not specifically limited here.

[0076] The prediction information for the target physical properties of the electrolyte to be predicted includes the predicted values ​​or prediction probability distributions of the target physical properties. These target physical properties include the electrolyte's solubility, conductivity, viscosity, diffusion coefficient, and other physical properties. The prediction probability distribution is the probability distribution of the predicted physical property values ​​obtained by changing the value of a certain variable while keeping other physical quantities constant.

[0077] Through the above steps, the autoregressive pre-trained attention model is trained using the component and formulation information (including chemical group characteristics and site data) of the electrolyte to be trained from the electrolyte formulation database. This results in a target electrolyte prediction model. Subsequently, the component and formulation information of the electrolyte to be predicted are input into the target electrolyte prediction model to obtain target physical property prediction information, including physical prediction values ​​and probability distributions. Training the target electrolyte prediction model with electrolytes of various properties from the electrolyte formulation database improves the applicability of physical property predictions. Furthermore, combining the model obtained through mask training enhances the accuracy of the physical property prediction information.

[0078] In some embodiments, the target electrolyte prediction model includes at least one target encoding module and at least one target decoding module; both the target encoding module and the target decoding module include a target self-attention layer; inputting the formulation component information of the electrolyte to be predicted into the target electrolyte prediction model to obtain the target physical property prediction information of the electrolyte to be predicted includes: inputting the formulation component information of the electrolyte to be predicted into the target electrolyte prediction model, performing masking processing on the formulation component information of the electrolyte to be predicted to obtain prediction mask information; inputting the prediction mask information into the target encoding module, obtaining an intermediate arbitrary feature vector of the electrolyte to be predicted based on the target self-attention layer; inputting the intermediate arbitrary feature vector of the electrolyte to be predicted into the target decoding module, obtaining the target physical property prediction information of the electrolyte to be predicted based on the self-attention layer.

[0079] The input to the target self-attention layer is determined by tag data with a randomly set prediction order, while the output is determined by all tag data. Tag data refers to the entire electrolyte molecular structure, including chemical functional groups, auxiliary groups, polar groups, nonpolar groups, and the size and coordination atoms. Furthermore, it can be understood that the input to the target self-attention layer includes multiple different groups labeled with custom digital tags, and the output is the target physical property information determined by all tag data of these multiple input groups.

[0080] In some embodiments, the target physical property prediction information includes the target solubility, and the method further includes: obtaining the attention weight of an intermediate arbitrary feature vector; combining the attention weight and a preset solubility improvement strategy, inputting the prediction mask information into the target electrolyte prediction model to obtain the target ratio of chemical groups and site data of the electrolyte to be predicted; and improving the target solubility of the electrolyte to be predicted based on the target ratio to obtain the improved solubility.

[0081] When it is necessary to improve the target solubility physical properties of the electrolyte under test, the formulation composition information of the electrolyte to be predicted must first be input into any layer of the target electrolyte prediction model to obtain an intermediate arbitrary feature vector. Then, the attention weight of the intermediate arbitrary feature vector is determined. The attention weight of the intermediate result is used as one of the criteria for selecting improved chemical groups and sites. Combined with a preset solubility improvement strategy, the chemical groups and sites to be improved in the formulation to be improved are masked and then input into the trained target electrolyte prediction model to obtain the optimal ratio of the types of components to be improved in the electrolyte under test, i.e., the target ratio. Finally, the target solubility of the electrolyte to be predicted is improved according to the target ratio to obtain the improved solubility. For example, the preset solubility improvement strategy includes strategies based on chemical physics models, such as molecular or atomic dynamics simulation methods, to obtain solubility at the molecular or atomic level, which is beneficial to further improve the accuracy of solubility improvement.

[0082] In some of these embodiments, an electrolyte reconstruction method is also provided after obtaining the target ratio of chemical groups and sites of the electrolyte to be predicted. Figure 3 This is a flowchart of the electrolyte reconstruction method provided in the embodiments of this application, such as... Figure 3 As shown, the process includes the following steps S310 to S320.

[0083] Step S310: Based on the target ratio and using a preset molecular simulation method, determine the correlation and binding strength between chemical groups and site data.

[0084] In this process, after masking the chemical groups and sites to be improved in the formulation of the electrolyte to be predicted, the data are input into the trained target electrolyte prediction model to obtain the target ratio of the types of components to be improved in the electrolyte to be tested. Then, based on the target ratio, the correlation and binding strength between the chemical groups and sites in the electrolyte to be predicted are calculated using a preset molecular simulation method based on deep learning prediction improvement.

[0085] Step S320: Based on correlation and binding strength, determine the target groups and target sites in the chemical group and site data, and reconstruct the electrolyte to be predicted.

[0086] After calculating the correlation and binding strength between chemical groups and sites in the electrolyte to be predicted based on the target ratio, the optimal target chemical groups and target sites are determined. Then, the electrolyte to be predicted is redesigned from scratch based on the target chemical groups and target sites.

[0087] In some embodiments, the method further includes updating the electrolyte formulation database based on the electrolyte to be predicted and the target physical property prediction information of the electrolyte to be predicted.

[0088] Furthermore, after masking the chemical groups and sites to be improved in the electrolyte to be predicted and inputting them into the trained target electrolyte prediction model, the target ratio of the types of component groups to be improved in the electrolyte to be tested is obtained, and the electrolyte to be trained in the electrolyte formulation database is screened according to the target ratio.

[0089] The present embodiment will be described and explained below through specific examples.

[0090] This specific embodiment provides a device for predicting the physical properties of electrolytes based on an attention mechanism. An electrolyte formulation database is constructed for organic, aqueous, and mixed electrolyte compositions. This database includes chemical group and site data, and coordination data sets for lithium salts, solvents, and additives. Electrolytes from the database are input into a neural network model for training until a well-trained prediction model for the target electrolyte's physical properties (solubility, conductivity, viscosity, and diffusion coefficient) is obtained. The neural network model is a general autoregressive pre-trained attention model. The component and formulation information of the electrolyte to be predicted are input into the target electrolyte physical property prediction model to obtain predicted values ​​or probability distributions of the electrolyte's solubility, conductivity, viscosity, and diffusion coefficient. This embodiment achieves model training for predicting the molecular chemical composition and physical properties of electrolytes. Furthermore, the trained model can be applied to electrolyte formulation modification, and intermediate model results can be used as feature vectors in various electrolyte property prediction and modification tasks, offering advantages such as wide applicability, flexibility, and simplicity.

[0091] Figure 4 This is a flowchart of the attention mechanism-based method for predicting the physical properties of electrolytes provided in this specific embodiment. (Reference) Figure 4 The method includes steps A, B, and C.

[0092] Step A involves collecting and preprocessing data on electrolyte formulations (lithium salts, solvents, and additive electrolytes) to obtain a chemical group dataset, which is essentially preprocessing the data into a dataset containing chemical group characteristics.

[0093] This involves constructing an electrolyte database, which is a collection of electrolyte chemical groups targeting specific physical properties. Specifically, this collection of electrolyte chemical groups can be one or more of the following: inorganic / organic anionic groups, lithium ion coordination sites (N, O, F), hydrophobic groups of different lengths, and polar groups (SO4-, PO3-, NO3-, -NH2). It can also be other data sets tailored to the problem at hand.

[0094] For example, training can be performed using only auxiliary groups as model inputs. This can be done using lithium-ion coordination sites (N, O, F), the hydrophobic groups, sizes, and lengths (-CH3, -CH2CH3, -CH(CH3)2) of the electrolyte to be predicted, and the polar groups (SO4-, PO3-, NO3-, -NH2) of the electrolyte to be predicted, all within a general autoregressive pre-trained attention model (XLNet model). Given the diversity of auxiliary groups, training specifically for these auxiliary groups is particularly important.

[0095] The trained target electrolyte prediction model effectively captures the connections between major functional groups and auxiliary groups, thus accurately predicting the optimal group and site composition for a given portion of major functional groups. Because its training method is independent of specific order and considers global information, this model outperforms other methods such as generative pre-trained attention models (GPT) and convolutional neural networks. In tasks involving modifying the affinity of auxiliary groups for solvent molecules, this model can preserve specific sites, thereby providing the optimal combination of remaining groups. For example, if certain sites are known to be crucial for lithium-ion coordination, these sites can be retained to predict the optimal modification combination for other sites. Furthermore, in de novo electrolyte design, several sites can be determined using a physical energy model, combined with a geometric feature-based structural model, to generate the optimal combination of remaining sites, or a combination of this model and a physical model can be used to design molecules.

[0096] Step B involves inputting the chemical group dataset into a generalized autoregressive pre-trained model for training (using the XLNet model as an example) until the error is below the threshold and stabilizes, at which point training is stopped, thus obtaining a trained prediction model for the physical properties of the electrolyte.

[0097] Specifically, the content can be as follows: Obtain and construct relevant electrolyte molecule data from an organic small molecule database, including data on inorganic or organic lithium salts, solvents, and additive molecules. The training output includes the physical properties (solubility, conductivity, viscosity, diffusion coefficient) of existing electrolyte components and formulations, as well as the physical properties (solubility, conductivity, viscosity, diffusion coefficient) estimated from molecular dynamics-based tools (Lammps, Gromacs, Amber). The data is input into a neural network, which obtains predicted values ​​of relevant electrolyte functional groups and the target physical properties of the electrolyte. The neural network is trained based on the predicted values ​​of relevant electrolyte functional groups and the target physical properties of the electrolyte, as well as the true values ​​of relevant electrolyte functional group connections and the target physical properties of the electrolyte. In this implementation, the relevant electrolyte functional group connections and the true values ​​of the target physical properties of the electrolyte are used as supervisory information to train the neural network. This allows the neural network to learn the autocorrelation of the internal physical properties of the electrolyte, thereby enabling the neural network to learn the ability to predict the target physical properties of the electrolyte to be predicted. This approach eliminates the need for prior knowledge of ion cluster structures, enabling the acquisition of relatively accurate predictions using readily available electrolyte data, thus offering the advantages of simplicity and ease of implementation. If physical properties (solubility, conductivity, viscosity, diffusion coefficient) obtained through experiments and computational simulations are available, the electrolyte functional groups and sites can be designed in conjunction with the energy of the physical model. The optimal functional groups and site types for the predicted sites are further ranked using the energy and geometric characteristics of the physical model, thereby further improving the accuracy of the prediction results.

[0098] Step C: Input the electrolyte formulation component information into the electrolyte prediction model to obtain the predicted solubility, diffusion coefficient, conductivity and viscosity of the electrolyte.

[0099] The trained target electrolyte prediction model can include an encoding model and a decoding model. The chemical groups and site information from the composition and formulation information of the electrolyte to be predicted are input into the encoding model, which then obtains the intermediate results corresponding to the electrolyte to be predicted. The composition and formulation information of the electrolyte to be predicted, along with the intermediate results, are input into the decoding model, which then obtains the predicted values ​​of the target physical properties of the electrolyte to be predicted. In this implementation, the encoding model is used to obtain information on the main functional groups and coordination sites, while the decoding model is used to predict the information on auxiliary groups. Using this implementation to predict the structural design and modification of the electrolyte to be predicted can improve the accuracy of the prediction results. This implementation can further consider the main functional groups and coordination sites, using the encoding model to obtain information on the entire polar group region and the decoding model to predict the information on non-polar groups. Using this implementation to predict the target physical properties of the electrolyte to be predicted can improve the accuracy of the prediction results.

[0100] The encoding and decoding models can each include at least one attention module. The input to the attention module is determined by the label data with a randomly set prediction order, and the output of the attention module is determined by all the label data. Specifically, the label data with a randomly set prediction order uses a masking method to mask information about some functional groups in order to predict the information of the masked functional groups. This attention module is the self-attention layer in the aforementioned embodiments.

[0101] Further, the specific process of step C above is as follows: at least one of the data of the electrolyte groups and sites to be predicted is input into the self-attention module in the encoding model, and the intermediate result corresponding to the electrolyte to be predicted is obtained through the self-attention module. Then, at least one of the data of the electrolyte groups and sites to be predicted and the intermediate result are input into the self-attention module in the encoding and decoding model, and the predicted value of the target physical property corresponding to the electrolyte to be predicted is obtained through the self-attention module.

[0102] The group information of a portion of the electrolyte to be predicted is input into the self-attention module to obtain a predicted value at a random location in the unpredicted region of the electrolyte. This predicted value is then combined with the group information of the portion of the electrolyte input in the previous round as the input for the new round, and this process is repeated until predicted values ​​for all regions of the electrolyte to be predicted are obtained. In this implementation, the self-attention module can predict the target physical properties of the electrolyte in any order, resulting in more accurate prediction results.

[0103] Figure 5 This is a schematic diagram of the electrolyte modification process provided in this specific embodiment. (Refer to...) Figure 5 The process involves acquiring electrolyte formulations from a database for training, including inorganic and organic groups, polar and nonpolar groups, and ionic coordination sites of lithium salt anions. These chemical groups and sites are then trained using a multilayer attention mechanism (Trm) to obtain a trained target electrolyte physical property prediction model. Based on this model, the physical properties of the electrolyte to be predicted are then calculated, yielding predictions for solubility, conductivity, viscosity, or diffusion coefficient. Once the predicted physical properties are obtained, the electrolyte is modified to obtain a modified electrolyte formulation.

[0104] The above-mentioned method for predicting the physical properties of electrolytes based on the attention mechanism can be applied to the following specific application scenarios.

[0105] Application Scenario 1: Prediction and modification of electrolyte solubility.

[0106] According to the attention-based electrolyte physical property prediction method in this embodiment, a target electrolyte physical property prediction model trained using a corresponding electrolyte database is used. The electrolyte components and formulation information to be predicted are input into the target electrolyte physical property prediction model. The intermediate result feature vectors of any layer in the neural network of the target electrolyte physical property prediction model are used as input, and then input into any machine learning model to perform the prediction task. Furthermore, the attention weights of the intermediate result feature vectors are used as one of the criteria for improving the selection of chemical groups and sites. Combined with other solubility improvement methods, the chemical groups and sites to be improved in the formulation are masked before being input into the trained model to obtain the optimal ratio of the types of components to be improved.

[0107] Application Scenario 2: Electrolyte conductivity prediction and modification.

[0108] Based on the aforementioned attention-based method for predicting electrolyte physical properties, training is performed using only datasets of important chemical groups and sites in the electrolyte. These important chemical groups and sites can be active coordinating elements (N, O, F). The composition and formulation information of the electrolyte to be predicted are input into the target electrolyte physical property prediction model. The intermediate result feature vector of any layer of the neural network for predicting the target electrolyte physical properties is used as input and then input into any machine learning model to perform coordination prediction. Furthermore, the attention weights of the intermediate results are used as one of the criteria for improving site selection. The groups and sites to be improved corresponding to the electrolyte to be predicted are masked and then input into the trained model to obtain the optimal ratio of coordination sites and group types to be improved.

[0109] Application Scenario 3: Electrolyte viscosity prediction and modification.

[0110] Based on the above-mentioned method for predicting the physical properties of electrolytes using an attention mechanism, a target electrolyte physical property prediction model trained using a corresponding electrolyte database is used. The electrolyte components and formulation information to be predicted are input into the target electrolyte physical property prediction model. The intermediate result feature vector of any layer in the neural network of the target electrolyte physical property prediction model is used as input and then input into any machine learning model to perform the prediction task. Furthermore, the attention weight of the intermediate result is used as one of the criteria for improving the selection of chemical groups and sites. Combined with other expression level improvement methods, the sites or groups to be improved are masked and then input into the trained target electrolyte physical property prediction model to obtain the optimal ratio of the types of coordination sites and groups to be improved.

[0111] Application Scenario 4: Prediction of electrolyte diffusion coefficient.

[0112] Based on the above-mentioned method for predicting the physical properties of electrolytes using an attention mechanism, a target electrolyte physical property prediction model trained using a corresponding electrolyte database is used. The electrolyte components and formulation to be predicted are input into the target electrolyte physical property prediction model. The intermediate result feature vector of any layer of the neural network in the target electrolyte physical property prediction is used as input, and then input into any machine learning model to perform the prediction task, thereby obtaining the prediction of the correlation and binding strength of the interaction between electrolyte anions, additive molecules, solvent molecules and cations.

[0113] Application Scenario 5: Construction of a general electrolyte modification library.

[0114] Based on the above-mentioned method for predicting the physical properties of electrolytes based on the attention mechanism, the target electrolyte physical property prediction model is trained using the corresponding organic, aqueous, and mixed electrolyte databases. The chemical groups and sites of the electrolyte to be improved are masked and then input into the trained target electrolyte physical property prediction model to obtain the optimal ratio of coordination sites and group types to be improved. Based on this, a library is built for electrolyte formulation screening.

[0115] Application Scenario 6: Electrolyte design from scratch.

[0116] Based on the above-mentioned attention-based method for predicting electrolyte physical properties, a target electrolyte physical property prediction model is trained using the corresponding electrolyte database. The chemical groups and sites to be designed in the electrolyte are then masked and input into the trained target electrolyte physical property prediction model to obtain the optimal ratio of coordination sites and group types to be improved. At the same time, the physical model is combined with tools such as Lammps, Gromacs, Amber, or deep learning to predict the improvement effect to calculate the interaction correlation and binding strength of the corresponding design, thereby comprehensively selecting the best sites and groups for de novo electrolyte design.

[0117] In this specific embodiment, after the important groups and sites are determined, the other sites are filled in a random order. Figure 6 This is a schematic diagram of the method for modifying chemical groups and sites provided in this specific embodiment. For example... Figure 6 As shown, the target physical property prediction model obtained based on the above-mentioned attention mechanism-based electrolyte physical property prediction method, namely the model in the figure, determines the physical prediction properties of the electrolyte to be predicted, and stores the electrolyte to be predicted and its physical prediction properties in the general electrolyte modification demonstration library, namely the electrolyte formula database in the aforementioned embodiment, for subsequent storage and training.

[0118] Furthermore, using existing molecules as templates, targeted and random modifications are made to key sites and auxiliary groups. A robust attention mechanism and encoding / decoding model are also introduced to further enhance the model's expressive power, for example... Figure 6 As shown, the polar groups -SO2- and -O- in the electrolyte LiTSI can be input into the encoding model and the enhanced decoding model can be used to predict their auxiliary groups -SO2-F and -SO2-CF3 to obtain the electrolyte LiTFSI. Alternatively, each auxiliary group -OCH3 and -OCH2CH3 can be predicted separately.

[0119] The above methods achieve universality for all cases of electrolyte group design and have universal application potential for downstream scenarios, such as de novo electrolyte design, construction of electrolyte modification libraries, and construction of libraries for predicting specific physical properties based on existing information.

[0120] When the plan is to fine-tune the solubility or improve the conductivity by modifying only a limited number of functional groups and sites, a trained neural network model, i.e., a target electrolyte prediction model, can be used directly for prediction. The highest-ranking functional groups and sites are selected for direct modification testing. Using the above method, functional group prediction is performed on the target sites, and a library of high-ranking functional groups and sites is built for further electrolyte formulation design. This results in an electrolyte formulation exhibiting high conductivity, low viscosity, good compatibility, and good diffusion performance in terms of anion-cation interactions, solvent-anion interactions, and additive-cation interactions, supplementing an electrolyte library with overall good electrolyte properties. Furthermore, this method can be combined with other information to comprehensively design electrolyte functional groups, such as coordination-based structure-physical energy models and coordination-based geometric feature-based structural models, thereby rationally designing a library or group of functional groups with excellent electrolyte performance. Further, this model can also be combined with physics-based modeling methods to design electrolyte functional groups from scratch through computation. Each functional group or site is determined by simultaneously considering the prediction results of the target electrolyte prediction model and the energy values ​​calculated by the physics model.

[0121] Furthermore, by utilizing the training methods described above to obtain a deep learning model, it can predict the connection sites of the main functional groups of inorganic anions, organic anions, solvents, and additive molecules in a given electrolyte, including polar and non-polar group regions. This allows for recommendations on the optimal combination of groups and sites for any desired modification position in various electrolyte modification tasks, particularly polar group region modification. The advantage of this training method is that it thoroughly describes the relationships between arbitrary positions from scratch, without relying on a fixed order or only locally adjacent groups. It also comprehensively considers the correlation between auxiliary groups and polar group regions, ensuring that focusing on polar group region modification has less impact on the interaction between auxiliary groups and solution molecules, thus preserving compatibility.

[0122] Specifically, in the modification of fluorinated electrolytes targeting polar group regions, a model trained solely on fluorinated groups can be used to make the predicted group and site composition more consistent with the characteristics of fluorinated groups. In modifications to solubility, conductivity, viscosity, and diffusion coefficient, the model will provide the group and site combinations that best match the characteristics of the electrolyte itself. This significantly improves modification efficiency.

[0123] The attention-based method for predicting the physical properties of electrolytes provided in this specific embodiment, based on a neural network model, enables the prediction of electrolyte components and formulation information, and has the following beneficial effects and advantages:

[0124] (1) It can predict different inorganic / organic anionic groups, lithium coordination sites, different numbers and sizes of hydrophobic groups, and different polar groups (SO4-, PO3-, NO3-, -NH2), with a wide coverage.

[0125] (2) The influence of the interaction correlation and binding strength between electrolyte anions, additive molecules, solvent molecules and cations is considered, reflecting the chemical and physical nature and physical properties of similar and compatible chemical components of electrolyte.

[0126] (3) It can predict both electrolyte group and site information and electrolyte physical properties, which helps to screen efficiently.

[0127] (4) It can fix any group and site to predict the optimal group and site composition distribution of the remaining groups and sites, which helps to flexibly design electrolyte components and formulations.

[0128] (5) It takes into account anionic groups, lithium coordination sites, hydrophobic groups and polar groups, and will perform better for different target tasks.

[0129] (6) By using the trained target electrolyte physical property prediction model, a better high-dimensional space embedding for the electrolyte chemical composition can be obtained, thus obtaining a better high-dimensional representation vector from the intermediate results, which can be used to improve the accuracy of prediction and modification of various electrolyte properties.

[0130] This embodiment also provides a device for predicting the physical properties of electrolytes based on an attention mechanism. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. The terms "module," "unit," "subunit," etc., used below can refer to combinations of software and / or hardware that implement a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0131] Figure 7 This is a structural block diagram of the attention-based electrolyte physical property prediction device provided in this embodiment, as shown below. Figure 7 As shown, the device includes: a data acquisition module 10, a model training module 20, and a prediction module 30.

[0132] The data acquisition module 10 is used to acquire the electrolyte to be trained from the preset electrolyte formula database; the electrolyte formula database includes the formula component information of the electrolyte to be trained, and the formula component information includes chemical group characteristics and site data.

[0133] The model training module 20 is used to perform mask training on the preset autoregressive pre-trained attention model based on the electrolyte to be trained, so as to obtain the target electrolyte prediction model.

[0134] The prediction module 30 is used to obtain the formulation component information of the electrolyte to be predicted, and input the formulation component information into the target electrolyte prediction model to obtain the target physical property prediction information of the electrolyte to be predicted.

[0135] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0136] This embodiment also provides an electronic device including a memory and a processor, the memory storing a computer program and the processor being configured to run the computer program to perform the steps in any of the above method embodiments.

[0137] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0138] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0139] S1, Obtain the electrolyte to be trained from the preset electrolyte formula database; the electrolyte formula database includes the formula component information of the electrolyte to be trained, and the formula component information includes chemical group characteristics and site data.

[0140] S2, based on the electrolyte to be trained, perform mask training on the preset autoregressive pre-trained attention model to obtain the target electrolyte prediction model.

[0141] S3, obtain the formulation component information of the electrolyte to be predicted, input the formulation component information of the electrolyte to be predicted into the target electrolyte prediction model, and obtain the target physical property prediction information of the electrolyte to be predicted.

[0142] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated in this embodiment.

[0143] Furthermore, in conjunction with the attention-based electrolyte physical property prediction method provided in the above embodiments, this embodiment can also provide a storage medium for implementation. This storage medium stores a computer program; when executed by a processor, the computer program implements any of the attention-based electrolyte physical property prediction methods described in the above embodiments.

[0144] It should be understood that the specific embodiments described herein are merely illustrative of the application and not intended to limit it. All other embodiments derived by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0145] Obviously, the accompanying drawings are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar situations based on these drawings without any creative effort. Furthermore, it is understood that although the work done in this development process may be complex and lengthy, for those skilled in the art, certain design, manufacturing, or production modifications made based on the technical content disclosed in this application are merely conventional technical means and should not be considered as insufficient disclosure of this application.

[0146] The term "embodiment" in this application refers to a specific feature, structure, or characteristic described in connection with an embodiment that may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily imply the same embodiment, nor does it imply that it is mutually exclusive with or independent of other embodiments. It will be clearly or implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0147] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of patent protection. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the appended claims.

Claims

1. A method for predicting the physical properties of electrolytes based on an attention mechanism, characterized in that, The method includes: Obtain the electrolyte to be trained from the preset electrolyte formulation database; the electrolyte formulation database includes formulation component information of the electrolyte to be trained, and the formulation component information includes chemical group characteristics and site data; Based on the electrolyte to be trained, a pre-set autoregressive pre-trained attention model is trained using a mask to obtain a target electrolyte prediction model; the target electrolyte prediction model includes at least one target encoding module and at least one target decoding module; both the target encoding module and the target decoding module contain a target self-attention layer; The formulation component information of the electrolyte to be predicted is obtained, and the formulation component information of the electrolyte to be predicted is input into the target electrolyte prediction model. The formulation component information of the electrolyte to be predicted is masked to obtain prediction mask information. The prediction mask information is input into the target encoding module, and the intermediate arbitrary feature vector of the electrolyte to be predicted is obtained based on the target self-attention layer. Input any intermediate feature vector of the electrolyte to be predicted to the target decoding module, and obtain the target physical property prediction information of the electrolyte based on the self-attention layer; The target physical property prediction information includes target solubility. The method further includes: obtaining the attention weight of the intermediate arbitrary feature vector; combining the attention weight and a preset solubility improvement strategy, inputting the prediction mask information into the target electrolyte prediction model to obtain the target ratio of chemical groups and site data of the electrolyte to be predicted; and improving the target solubility of the electrolyte to be predicted based on the target ratio to obtain the improved solubility.

2. The method for predicting the physical properties of electrolytes based on an attention mechanism according to claim 1, characterized in that, The pre-trained autoregressive attention model includes at least one encoding module and at least one decoding module; both the encoding module and the decoding module include a self-attention layer. The step of performing masked training on a pre-set autoregressive pre-trained attention model based on the electrolyte to be trained, to obtain a target electrolyte prediction model, includes: The formulation component information of the electrolyte to be trained is masked to obtain the formulation mask component information; The formula mask component information is input into the encoding module, and the intermediate prediction features of the electrolyte to be trained are obtained based on the self-attention layer; The formulation mask component information of the electrolyte to be trained and the intermediate prediction features are input to the decoding module, and the predicted values ​​of the physical properties of the electrolyte to be trained are obtained based on the self-attention layer. Based on the predicted values ​​of the physical properties, and the chemical group characteristics and site data in the formulation composition information of the electrolyte to be trained, the predicted loss value is calculated. Based on the calculated predicted loss value, the preset autoregressive pre-trained attention model is iteratively trained until the predicted loss value is lower than the preset error threshold, thus obtaining the target electrolyte prediction model.

3. The method for predicting the physical properties of electrolytes based on an attention mechanism according to claim 1, characterized in that, After obtaining the target ratio of chemical groups and sites of the electrolyte to be predicted, the method further includes: Based on the target ratio, and combined with a preset molecular simulation method, the correlation and binding strength between the chemical groups and site data are determined; Based on the correlation and binding strength, the target groups and target sites in the chemical group and site data are determined, and the electrolyte to be predicted is reconstructed.

4. The method for predicting the physical properties of electrolytes based on an attention mechanism according to any one of claims 1 to 3, characterized in that, The method further includes: The electrolyte formulation database is updated based on the electrolyte to be predicted and the target physical property prediction information of the electrolyte to be predicted.

5. The method for predicting the physical properties of electrolytes based on an attention mechanism as described in claim 4, characterized in that, The preset electrolyte formula database includes known electrolytes; obtaining the electrolyte to be trained from the preset electrolyte formula database includes: Obtain structural information from the known electrolyte formulation composition information; The structural information is segmented to determine the tag data contained in the known electrolyte; Based on the tag data, the known electrolyte is processed to obtain the chemical group characteristics of the known electrolyte; the known electrolyte is then used as the electrolyte to be trained.

6. A device for predicting the physical properties of electrolytes based on an attention mechanism, characterized in that, The device includes: a data acquisition module, a model training module, and a prediction module; The data acquisition module is used to acquire the electrolyte to be trained from a preset electrolyte formula database; the electrolyte formula database includes formula component information of the electrolyte to be trained, and the formula component information includes chemical group characteristics and site data; The model training module is used to perform mask training on a preset autoregressive pre-trained attention model based on the electrolyte to be trained, so as to obtain a target electrolyte prediction model; the target electrolyte prediction model includes at least one target encoding module and at least one target decoding module; both the target encoding module and the target decoding module contain a target self-attention layer; The prediction module is used to acquire the formulation component information of the electrolyte to be predicted, input the formulation component information into the target electrolyte prediction model, perform masking processing on the formulation component information of the electrolyte to be predicted to obtain prediction mask information; input the prediction mask information into the target encoding module, obtain an intermediate arbitrary feature vector of the electrolyte to be predicted based on the target self-attention layer; input the intermediate arbitrary feature vector of the electrolyte to be predicted into the target decoding module, obtain the target physical property prediction information of the electrolyte to be predicted based on the self-attention layer; the target physical property prediction information includes target solubility; and also to acquire the attention weight of the intermediate arbitrary feature vector; combine the attention weight and a preset solubility improvement strategy, input the prediction mask information into the target electrolyte prediction model to obtain the target ratio of chemical groups and site data of the electrolyte to be predicted; and improve the target solubility of the electrolyte to be predicted based on the target ratio to obtain the improved solubility.

7. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method for predicting the physical properties of electrolytes based on an attention mechanism, as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for predicting the physical properties of electrolytes based on the attention mechanism as described in any one of claims 1 to 5.

Images