Electrolyte formula property prediction method and device based on graph neural network

Abstract

The present invention discloses a method and device for predicting the properties of electrolyte formulas based on a graph neural network. The method generates a target electrolyte virtual molecule based on the atomic node architecture in the original molecular graph and the molecular nodes in the target electrolyte formula. A pretrained model is constructed and trained based on an open source model. A property prediction model is constructed and trained based on the graph neural network. The target electrolyte virtual molecule is input into the trained pretrained model to obtain the characteristics of the target electrolyte virtual molecule. The characteristics of the target electrolyte virtual molecule are input into the trained property prediction model to obtain the properties of the target electrolyte formula. The present invention can obtain relatively high-precision formula properties in a very short time, improving the efficiency of electrolyte formula acquisition.

Description

Technical Field

[0001] The present invention relates to the technical field of electrolyte formula property prediction, and in particular to a method and device for predicting electrolyte formula properties based on a graph neural network. Background Art

[0002] Formula property prediction is a critical step in the development of battery electrolytes. This technology can quickly help researchers obtain the properties of the current formula and accelerate the development of battery electrolytes.

[0003] Currently, researchers mainly rely on human experience and a large number of experiments to obtain the true properties of a formula. Human experience can help screen out a large number of unreliable formulas, and then researchers can conduct experiments on the remaining electrolyte formulas. This method is time-consuming and labor-intensive.

[0004] Therefore, how to invent a method for predicting the properties of electrolyte formulas that can obtain relatively high-precision formula properties in a very short time and improve the efficiency of electrolyte formula acquisition has become an urgent problem to be solved. Summary of the Invention

[0005] To this end, the present invention provides a method and device for predicting electrolyte formula properties based on graph neural networks. It not only inherits the experience of experts in this field, but also can obtain relatively high-precision formula properties in a very short time. Then researchers only need to experiment with several better groups of formulas to obtain the formula they want.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting electrolyte formulation properties based on a graph neural network, comprising:

[0007] According to the structure of the atomic nodes in the original molecular graph, the target electrolyte virtual molecules are generated through the molecular nodes in the target electrolyte formula;

[0008] Based on the open source model, a pre-trained model is constructed; the pre-trained model is trained by setting a training data set to obtain a trained pre-trained model;

[0009] Constructing a property prediction model based on a graph neural network; training the property prediction model using the set training data set to obtain a trained property prediction model;

[0010] Inputting the target electrolyte virtual molecule into the trained pre-trained model; obtaining the characteristics of the target electrolyte virtual molecule through processing by the trained pre-trained model;

[0011] The characteristics of the target electrolyte virtual molecule are input into the trained property prediction model; and the properties of the target electrolyte formula are obtained through prediction processing by the trained property prediction model.

[0012] As a preferred solution of the electrolyte formula property prediction method based on graph neural network, in the process of training the pre-training model and the property prediction model through the set training data set, the set training data set is divided into a training set, a validation set and a test set according to a set ratio.

[0013] As a preferred solution of the electrolyte formula property prediction method based on graph neural network, the characteristics of the target electrolyte virtual molecule include low-dimensional atom-level features and Bond-level features; in the process of obtaining the low-dimensional atom-level features, the target electrolyte virtual molecule is molecularly characterized by the trained pre-trained model to obtain high-dimensional atom-level features; the high-dimensional atom-level features are processed by setting a dimensionality reduction algorithm to obtain the low-dimensional atom-level features.

[0014] As a preferred solution of the electrolyte formula property prediction method based on graph neural network, in the process of obtaining the Bond-level features, the Bond-level features include the simulated connection relationship between molecules, non-covalent interaction energy and VIP vertical ionization energy;

[0015] The calculation formula of the non-covalent interaction energy is:

[0016] ;

[0017] Where, IE is the non-covalent interaction energy; ΔE_els is the classical electrostatic interaction energy between segments, els represents electrostatic interaction; ΔE_x is the exchange energy, x represents exchange interactions; ΔE_dc is the dispersion correction energy; ΔE_orb is the orbital interaction energy;

[0018] The calculation formula of the VIP vertical ionization energy is:

[0019] ;

[0020] ;

[0021] ;

[0022] Where, is the VIP vertical ionization energy; 、 are the masses of molecules; 、 The energy of a molecule after losing an electron. 、 All are the complete energy of the molecule; is the vertical ionization energy of molecule 1; is the vertical ionization energy of molecule 2.

[0023] As a preferred solution of the electrolyte formula property prediction method based on graph neural network, in the process of training the property prediction model through the set training data set to obtain the trained property prediction model, the low-dimensional atom-level features and Bond-level features of the target electrolyte virtual molecules in the set training data set are learned through the graph neural network, and features and property labels are output; by performing label verification on the features and property labels, the parameters of the graph neural network are iteratively optimized until the set number of training times is reached to obtain the trained property prediction model.

[0024] The present invention also provides an electrolyte formula property prediction device based on a graph neural network, based on the above-mentioned electrolyte formula property prediction method based on a graph neural network, comprising:

[0025] A target electrolyte virtual molecule generation module is used to generate target electrolyte virtual molecules based on the structure of the atomic nodes in the original molecular graph and the molecular nodes in the target electrolyte formula;

[0026] A pre-training model construction and training module is used to build a pre-training model based on an open source model; the pre-training model is trained by setting a training data set to obtain a trained pre-training model;

[0027] A property prediction model construction and training module is used to construct a property prediction model based on a graph neural network; the property prediction model is trained using the set training data set to obtain a trained property prediction model;

[0028] A molecular feature acquisition module is used to input the target electrolyte virtual molecule into the trained pre-trained model; and obtain the features of the target electrolyte virtual molecule through processing by the trained pre-trained model;

[0029] The electrolyte formula property acquisition module is used to input the characteristics of the target electrolyte virtual molecule into the trained property prediction model; and obtain the properties of the target electrolyte formula through prediction processing by the trained property prediction model.

[0030] As a preferred solution of the electrolyte formula property prediction device based on graph neural network, in the pre-training model construction and training module and the property prediction model construction and training module, in the process of training the pre-training model and the property prediction model through the set training data set, the set training data set is divided into a training set, a validation set and a test set according to a set ratio.

[0031] As a preferred solution of the electrolyte formula property prediction device based on graph neural network, in the molecular feature acquisition module, the characteristics of the target electrolyte virtual molecule include low-dimensional atom-level features and Bond-level features; in the process of obtaining the low-dimensional atom-level features, the target electrolyte virtual molecule is molecularly characterized by the trained pre-trained model to obtain high-dimensional atom-level features; the high-dimensional atom-level features are processed by setting a dimensionality reduction algorithm to obtain the low-dimensional atom-level features.

[0032] As a preferred solution of the electrolyte formula property prediction device based on graph neural network, in the molecular feature acquisition module, in the process of acquiring the Bond-level features, the Bond-level features include the simulated connection relationship between molecules, non-covalent interaction energy and VIP vertical ionization energy;

[0033] The calculation formula of the non-covalent interaction energy is:

[0034] ;

[0035] Where, IE is the non-covalent interaction energy; ΔE_els is the classical electrostatic interaction energy between segments, els represents electrostatic interaction; ΔE_x is the exchange energy, x represents exchange interactions; ΔE_dc is the dispersion correction energy; ΔE_orb is the orbital interaction energy;

[0036] The calculation formula of the VIP vertical ionization energy is:

[0037] ;

[0038] ;

[0039] ;

[0040] Where, is the VIP vertical ionization energy; 、 are the masses of molecules; 、 The energy of a molecule after losing an electron. 、 All are the complete energy of the molecule; is the vertical ionization energy of molecule 1; is the vertical ionization energy of molecule 2.

[0041] As a preferred solution of the electrolyte formula property prediction device based on graph neural network, in the property prediction model construction and training module, in the process of training the property prediction model through the set training data set to obtain the trained property prediction model, the low-dimensional atom-level features and Bond-level features of the target electrolyte virtual molecules in the set training data set are learned through the graph neural network, and features and property labels are output; by performing label verification on the features and property labels, the parameters of the graph neural network are iteratively optimized until the set number of training times is reached to obtain the trained property prediction model.

[0042] The present invention has the following advantages: according to the architecture of the atomic nodes in the original molecular graph, the target electrolyte virtual molecule is generated through the molecular nodes in the target electrolyte formula; based on the open source model, a pre-trained model is constructed; the pre-trained model is trained by setting a training data set to obtain a trained pre-trained model; based on the graph neural network, a property prediction model is constructed; the property prediction model is trained by the set training data set to obtain a trained property prediction model; the target electrolyte virtual molecule is input into the trained pre-trained model; the characteristics of the target electrolyte virtual molecule are obtained by processing the trained pre-trained model; the characteristics of the target electrolyte virtual molecule are input into the trained property prediction model; the properties of the target electrolyte formula are obtained by prediction processing through the trained property prediction model. The present invention not only inherits the experience of experts in this field, but also can obtain relatively high-precision formula properties in a very short time. Then, researchers only need to experiment with several better groups of formulas to obtain the formula they want. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and those skilled in the art can, without inventive effort, derive other implementation drawings based on the provided drawings.

[0044] The structures, proportions, sizes, etc. illustrated in this specification are intended solely to complement the contents disclosed herein and to facilitate understanding and reading by persons skilled in the art. They are not intended to limit the conditions under which the present invention may be implemented and therefore have no substantive technical significance. Any structural modifications, changes in proportions, or adjustments in sizes, without affecting the efficacy and objectives of the present invention, shall remain within the scope of the technical contents disclosed herein.

[0045] Figure 1 This is a schematic flow chart of a method for predicting electrolyte formulation properties based on a graph neural network provided in Example 1 of the present invention;

[0046] Figure 2 This is a schematic diagram of a specific implementation process of a method for predicting electrolyte formulation properties based on a graph neural network provided in Example 1 of the present invention;

[0047] Figure 3 Schematic diagram of virtual molecules of electrolyte in a method for predicting electrolyte formulation properties based on graph neural network provided in Example 1 of the present invention;

[0048] Figure 4 A schematic diagram of a process for constructing a property prediction model in a method for predicting properties of an electrolyte formulation based on a graph neural network provided in Example 1 of the present invention;

[0049] Figure 5 Schematic diagram of the architecture of an electrolyte formula property prediction device based on graph neural network provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0050] The following describes the implementation of the present invention using specific embodiments. Those skilled in the art will readily understand the other advantages and benefits of the present invention from the disclosure herein. Obviously, the embodiments described are only a portion of the present invention, not all of it. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0051] Example 1

[0052] See also Figure 1 and Figure 2 Embodiment 1 of the present invention provides a method for predicting electrolyte formulation properties based on a graph neural network, comprising the following steps:

[0053] S1. Generate a target electrolyte virtual molecule based on the atomic node structure in the original molecular graph and the molecular nodes in the target electrolyte formula;

[0054] S2. Build a pre-trained model based on the open source model; train the pre-trained model by setting a training data set to obtain a trained pre-trained model;

[0055] S3. Constructing a property prediction model based on a graph neural network; training the property prediction model using the set training data set to obtain a trained property prediction model;

[0056] S4. Inputting the target electrolyte virtual molecule into the trained pre-trained model; and obtaining the characteristics of the target electrolyte virtual molecule through processing by the trained pre-trained model;

[0057] S5. Input the characteristics of the target electrolyte virtual molecule into the trained property prediction model; and obtain the properties of the target electrolyte formula through prediction processing by the trained property prediction model.

[0058] In this embodiment, in step S1, according to the structure of the atomic nodes in the original molecular graph, the target electrolyte virtual molecules are generated through the molecular nodes in the target electrolyte formula;

[0059] Specifically, traditional molecular graphs are composed of atomic nodes, while the electrolyte formula virtual molecules are composed of electrolyte molecular nodes. Figure 3 shown.

[0060] In this embodiment, in step S2, a pre-trained model is constructed based on the open source model; the pre-trained model is trained by setting a training data set to obtain a trained pre-trained model;

[0061] Specifically, the pre-trained model can adopt an open source model, or it can be constructed and trained based on an open source model. Open source models include Uni-mol, chemformer, etc. The training data set is set to include the low-dimensional atom-level features and Bond-level features of the target electrolyte virtual molecules; the training data set is set to be divided into a training set, a validation set, and a test set in a ratio of 8:1:1. The pre-trained model is trained by using the low-dimensional atom-level features and Bond-level features of the target electrolyte virtual molecules, so that the trained pre-trained model can extract the low-dimensional atom-level features and Bond-level features of the target electrolyte virtual molecules.

[0062] In this embodiment, in step S3, a property prediction model is constructed based on a graph neural network; the property prediction model is trained using the set training data set to obtain a trained property prediction model;

[0063] Specifically, such as Figure 4As shown, a property prediction model is constructed based on a graph neural network; the low-dimensional atom-level features and Bond-level features of the target electrolyte virtual molecules are input into traditional graph neural networks such as GNN or MPNN in the form of molecular graphs, and the low-dimensional atom-level features and Bond-level features of the target electrolyte virtual molecules are learned through the graph neural network, and features and property labels are output; by performing label verification on the features and property labels, the parameters of the graph neural network are iteratively optimized until the set number of training times is reached, and the trained property prediction model is obtained.

[0064] In this embodiment, in step S4, the target electrolyte virtual molecule is input into the trained pre-trained model; the trained pre-trained model is used to process the target electrolyte virtual molecule to obtain the characteristics of the target electrolyte virtual molecule;

[0065] Specifically, the characteristics of the target electrolyte virtual molecule include low-dimensional atom-level features and Bond-level features; in the process of obtaining the low-dimensional atom-level features, the target electrolyte virtual molecule is molecularly characterized by the trained pre-trained model to obtain high-dimensional atom-level features; the high-dimensional atom-level features are processed by setting a dimensionality reduction algorithm to obtain the low-dimensional atom-level features.

[0066] In the process of obtaining the Bond-level features, the Bond-level features include simulated connection relationships between molecules, non-covalent interaction energy, and VIP vertical ionization energy;

[0067] The calculation formula of the non-covalent interaction energy is:

[0068] ;

[0069] Where, IE is the non-covalent interaction energy; ΔE_els is the classical electrostatic interaction energy between segments, els represents electrostatic interaction; ΔE_x is the exchange energy, x represents exchange interactions; ΔE_dc is the dispersion correction energy; ΔE_orb is the orbital interaction energy;

[0070] The calculation formula of the VIP vertical ionization energy is:

[0071] ;

[0072] ;

[0073] ;

[0074] Where, is the VIP vertical ionization energy; 、 are the masses of molecules; 、 The energy of a molecule after losing an electron. 、 All are the complete energy of the molecule; is the vertical ionization energy of molecule 1; is the vertical ionization energy of molecule 2.

[0075] In this embodiment, in step S5, the characteristics of the target electrolyte virtual molecule are input into the trained property prediction model; and the properties of the target electrolyte formula are obtained through prediction processing by the trained property prediction model.

[0076] Specifically, the characteristics of the target electrolyte virtual molecules are input into the trained property prediction model for prediction to obtain the properties of the target electrolyte formula.

[0077] In a possible embodiment, a specific recipe property prediction example is provided as follows:

[0078] The researchers need to be given a formula A containing 6 molecules. Through the transformation of step S1 of the present invention, a formula virtual molecule A containing 6 nodes is generated.

[0079] First, each of the six nodes is fed into the pre-trained Uni-mol model to obtain atom-level and bond-level features. The atom features obtained by Uni-mol have a relatively high dimensionality. To reduce the difficulty of model training, these features are reduced using the PCA algorithm. The combination of these two features forms the molecular graph features of virtual molecule A.

[0080] Then, the atom features and bond features of each node after dimensionality reduction are used as input into the property prediction model, and finally the properties of formula A can be obtained.

[0081] In summary, the present invention generates target electrolyte virtual molecules based on the architecture of atomic nodes in the original molecular graph and through the molecular nodes in the target electrolyte formula; constructs a pre-trained model based on an open source model; trains the pre-trained model by setting a training data set to obtain a trained pre-trained model; constructs a property prediction model based on a graph neural network; trains the property prediction model by setting a training data set to obtain a trained property prediction model; inputs the target electrolyte virtual molecules into the trained pre-trained model; obtains the characteristics of the target electrolyte virtual molecules through processing by the trained pre-trained model; inputs the characteristics of the target electrolyte virtual molecules into the trained property prediction model; obtains the properties of the target electrolyte formula through prediction processing by the trained property prediction model. The present invention not only inherits the experience of experts in this field, but also can obtain relatively high-precision formula properties in a very short time, and then researchers only need to experiment with several better groups of formulas to obtain the formula they want.

[0082] It should be noted that the method of the embodiments of the present disclosure can be performed by a single device, such as a computer or server. The method of the embodiments of the present disclosure can also be applied in a distributed scenario, where multiple devices cooperate to perform the method. In such a distributed scenario, one of the multiple devices may only perform one or more steps of the method of the embodiments of the present disclosure, and the multiple devices will interact with each other to complete the method.

[0083] It should be noted that the above description is limited to some embodiments of the present disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in an order different from that described in the above embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0084] Example 2

[0085] See also Figure 5 , Embodiment 2 of the present invention further provides an electrolyte formula property prediction device based on graph neural network, comprising:

[0086] The target electrolyte virtual molecule generation module 001 is used to generate the target electrolyte virtual molecules according to the structure of the atomic nodes in the original molecular graph and the molecular nodes in the target electrolyte formula;

[0087] The pre-training model construction and training module 002 is used to construct a pre-training model based on the open source model; train the pre-training model by setting a training data set to obtain a trained pre-training model;

[0088] The property prediction model construction and training module 003 is used to construct a property prediction model based on a graph neural network; the property prediction model is trained using the set training data set to obtain a trained property prediction model;

[0089] The molecular feature acquisition module 004 is used to input the target electrolyte virtual molecule into the trained pre-trained model; and obtain the features of the target electrolyte virtual molecule through processing by the trained pre-trained model;

[0090] The electrolyte formula property acquisition module 005 is used to input the characteristics of the target electrolyte virtual molecule into the trained property prediction model; and obtain the properties of the target electrolyte formula through prediction processing by the trained property prediction model.

[0091] In this embodiment, in the pre-training model construction and training module 002 and the property prediction model construction and training module 003, in the process of training the pre-training model and the property prediction model through the set training data set, the set training data set is divided into a training set, a validation set and a test set according to a set ratio.

[0092] In this embodiment, in the molecular feature acquisition module 004, the features of the target electrolyte virtual molecule include low-dimensional atom-level features and Bond-level features; in the process of obtaining the low-dimensional atom-level features, the target electrolyte virtual molecule is molecularly characterized by the trained pre-trained model to obtain high-dimensional atom-level features; the high-dimensional atom-level features are processed by setting a dimensionality reduction algorithm to obtain the low-dimensional atom-level features.

[0093] In this embodiment, in the molecular feature acquisition module 004, in the process of acquiring the Bond-level features, the Bond-level features include the simulated connection relationship between molecules, non-covalent interaction energy, and VIP vertical ionization energy;

[0094] The calculation formula of the non-covalent interaction energy is:

[0095] ;

[0096] Where, IE is the non-covalent interaction energy; ΔE_els is the classical electrostatic interaction energy between segments, els represents electrostatic interaction; ΔE_x is the exchange energy, x represents exchange interactions; ΔE_dc is the dispersion correction energy; ΔE_orb is the orbital interaction energy;

[0097] The calculation formula of the VIP vertical ionization energy is:

[0098] ;

[0099] ;

[0100] ;

[0101] Where, is the VIP vertical ionization energy; 、 are the masses of molecules; 、 The energy of a molecule after losing an electron. 、 All are the complete energy of the molecule; is the vertical ionization energy of molecule 1; is the vertical ionization energy of molecule 2.

[0102] In this embodiment, in the property prediction model construction and training module 003, in the process of training the property prediction model through the set training data set to obtain the trained property prediction model, the low-dimensional atom-level features and Bond-level features of the target electrolyte virtual molecules in the set training data set are learned through the graph neural network, and features and property labels are output; by performing label verification on the features and property labels, the parameters of the graph neural network are iteratively optimized until the set number of training times is reached to obtain the trained property prediction model.

[0103] It should be noted that the information interaction, execution process, etc. between the modules of the above-mentioned system are based on the same concept as the method embodiment in Example 1 of the present application, and the technical effects they bring are the same as those of the method embodiment of the present application. For specific contents, please refer to the description in the method embodiment shown above in the present application, and no further details will be given here.

[0104] Example 3

[0105] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium, in which a program code for a method for predicting the properties of an electrolyte formula based on a graph neural network is stored. The program code includes instructions for executing embodiment 1 or any possible implementation thereof. A method for predicting the properties of an electrolyte formula based on a graph neural network.

[0106] Computer-readable storage media can be any available medium that can be accessed by a computer, or a data storage device such as a server or data center that includes one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0107] Example 4

[0108] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0109] The processor and the memory communicate with each other through a bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute an electrolyte formula property prediction method based on a graph neural network according to Example 1 or any possible implementation thereof.

[0110] Specifically, the processor can be implemented by hardware or by software. When implemented by hardware, the processor can be a logic circuit, an integrated circuit, etc.; when implemented by software, the processor can be a general-purpose processor, which is implemented by reading software code stored in a memory. The memory can be integrated into the processor or located outside the processor and exist independently.

[0111] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be 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 and 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 can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable systems. The computer instructions can 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 can be transmitted from a website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode.

[0112] Obviously, those skilled in the art will appreciate that the various modules or steps of the present invention described above can be implemented using a general-purpose computing system. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Alternatively, they can be implemented using program code executable by a computing system, and thus, they can be stored in a storage system and executed by the computing system. In some cases, the steps shown or described herein can be performed in a different order than that shown, or they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0113] Although the present invention has been described in detail above using general descriptions and specific embodiments, it will be apparent to those skilled in the art that modifications and improvements may be made thereto. Therefore, such modifications and improvements, without departing from the spirit of the present invention, are intended to be within the scope of protection claimed herein.

Claims

1. A method for predicting electrolyte formulation properties based on graph neural network, characterized in that: include: According to the structure of the atomic nodes in the original molecular graph, the target electrolyte virtual molecules are generated through the molecular nodes in the target electrolyte formula; Build pre-trained models based on open source models; The pre-training model is trained by setting a training data set to obtain a trained pre-training model; Constructing a property prediction model based on a graph neural network; training the property prediction model using the set training data set to obtain a trained property prediction model; Inputting the target electrolyte virtual molecule into the trained pre-trained model; obtaining the characteristics of the target electrolyte virtual molecule through processing by the trained pre-trained model; Inputting the characteristics of the target electrolyte virtual molecule into the trained property prediction model; obtaining the properties of the target electrolyte formula through prediction processing by the trained property prediction model; The characteristics of the target electrolyte virtual molecule include low-dimensional atom-level characteristics and Bond-level characteristics; in the process of obtaining the low-dimensional atom-level characteristics, the target electrolyte virtual molecule is subjected to molecular characterization processing by the trained pre-trained model to obtain high-dimensional atom-level characteristics; the high-dimensional atom-level characteristics are processed by a set dimensionality reduction algorithm to obtain the low-dimensional atom-level characteristics; In the process of obtaining the Bond-level features, the Bond-level features include simulated connection relationships between molecules, non-covalent interaction energy, and VIP vertical ionization energy; The calculation formula of the non-covalent interaction energy is: IE=ΔE_els+ΔE_x+ΔE_dc+ΔE_orb Where IE is the non-covalent interaction energy; ΔE_els is the classical electrostatic interaction energy between segments, els represents the electrostatic interaction; ΔE_x is the exchange interaction energy, x represents the exchange interaction; ΔE_dc is the dispersion correction energy; ΔE_orb is the orbital interaction energy; The calculation formula of the VIP vertical ionization energy is: In the formula, VIP bond is the VIP vertical ionization energy; m1 and m2 are the masses of the molecule; The energy of a molecule after losing an electron. Both are the complete energy of molecules; VIP1 is the vertical ionization energy of molecule 1; VIP2 is the vertical ionization energy of molecule 2.

2. The method for predicting electrolyte formulation properties based on graph neural network according to claim 1, characterized in that: In the process of training the pre-training model and the property prediction model using the set training data set, the set training data set is divided into a training set, a validation set and a test set according to a set ratio.

3. The method for predicting electrolyte formulation properties based on graph neural network according to claim 1, characterized in that: In the process of training the property prediction model through the set training data set to obtain the trained property prediction model, the low-dimensional atom-level features and Bond-level features of the target electrolyte virtual molecules in the set training data set are learned through a graph neural network, and features and property labels are output; by performing label verification on the features and property labels, the parameters of the graph neural network are iteratively optimized until the set number of training times is reached to obtain the trained property prediction model.

4. An electrolyte formula property prediction device based on a graph neural network, using the electrolyte formula property prediction method based on a graph neural network according to any one of claims 1 to 3, characterized in that: include: A target electrolyte virtual molecule generation module is used to generate target electrolyte virtual molecules based on the structure of the atomic nodes in the original molecular graph and the molecular nodes in the target electrolyte formula; Pre-trained model construction and training module, used to build pre-trained models based on open source models; The pre-training model is trained by setting a training data set to obtain a trained pre-training model; A property prediction model construction and training module is used to construct a property prediction model based on a graph neural network; the property prediction model is trained using the set training data set to obtain a trained property prediction model; A molecular feature acquisition module is used to input the target electrolyte virtual molecule into the trained pre-trained model; and obtain the features of the target electrolyte virtual molecule through processing by the trained pre-trained model; The electrolyte formula property acquisition module is used to input the characteristics of the target electrolyte virtual molecule into the trained property prediction model; and obtain the properties of the target electrolyte formula through prediction processing by the trained property prediction model.

5. The electrolyte formula property prediction device based on graph neural network according to claim 4 is characterized in that: In the pre-training model construction and training module and the property prediction model construction and training module, in the process of training the pre-training model and the property prediction model through the set training data set, the set training data set is divided into a training set, a validation set and a test set according to a set ratio.

6. The electrolyte formula property prediction device based on graph neural network according to claim 5, characterized in that: In the molecular feature acquisition module, the features of the target electrolyte virtual molecule include low-dimensional atom-level features and Bond-level features; in the process of obtaining the low-dimensional atom-level features, the target electrolyte virtual molecule is molecularly characterized by the trained pre-trained model to obtain high-dimensional atom-level features; the high-dimensional atom-level features are processed by a set dimensionality reduction algorithm to obtain the low-dimensional atom-level features.

7. The electrolyte formula property prediction device based on graph neural network according to claim 6, characterized in that: In the molecular feature acquisition module, in the process of acquiring the Bond-level features, the Bond-level features include simulated connection relationships between molecules, non-covalent interaction energy, and VIP vertical ionization energy; The calculation formula of the non-covalent interaction energy is: IE=ΔE_els+ΔE_x+ΔE_dc+ΔE_orb Where IE is the non-covalent interaction energy; ΔE_els is the classical electrostatic interaction energy between segments, els represents the electrostatic interaction; ΔE_x is the exchange interaction energy, x represents the exchange interaction; ΔE_dc is the dispersion correction energy; ΔE_orb is the orbital interaction energy; The calculation formula of the VIP vertical ionization energy is: In the formula, VIP bond is the VIP vertical ionization energy; m1 and m2 are the masses of the molecule; The energy of a molecule after losing an electron. Both are the complete energy of molecules; VIP1 is the vertical ionization energy of molecule 1; VIP2 is the vertical ionization energy of molecule 2.

8. The electrolyte formulation property prediction device based on graph neural network according to claim 7, characterized in that: In the property prediction model construction and training module, in the process of training the property prediction model through the set training data set to obtain the trained property prediction model, the low-dimensional atom-level features and Bond-level features of the target electrolyte virtual molecules in the set training data set are learned through the graph neural network, and features and property labels are output; by performing label verification on the features and property labels, the parameters of the graph neural network are iteratively optimized until the set number of training times is reached to obtain the trained property prediction model.

Images