A prediction platform for lithium battery electrolyte formulations based on large language models

Abstract

The present invention discloses a prediction platform for lithium battery electrolyte formulations based on large language models, belonging to the technical field of automated prediction of battery materials. It includes a data acquisition module for obtaining the property parameters of the target electrolyte; a formulation prediction module for outputting the electrolyte formulation and the theoretical viscosity value based on the first large language model, and storing the electrolyte formulation in the electrolyte database; a trajectory simulation module for pre-constructing a battery model and simulating the electrochemical reaction kinetics trajectory of the electrolyte and battery materials; a formulation property calculation module for calculating the system properties of the electrolyte formulation; and a formulation output module for outputting the optimized electrolyte formulation based on the electrochemical reaction kinetics trajectory of the electrolyte and battery materials and the system properties of the electrolyte formulation, based on the second large language model. The present invention solves the problems of difficult deployment, difficult use, and inaccurate prediction results faced by the prior art in the research and development of electrolyte materials.

Description

Technical Field

[0001] The present invention relates to a lithium battery electrolyte formula prediction platform based on a large language model, belonging to the technical field of automated prediction of battery materials. Background Art

[0002] With the continuous growth of the global demand for high-performance energy storage devices, especially the rapid development of the new energy vehicle field, the research and development of lithium batteries with higher energy density has become a research hotspot. However, the research and development of high-energy-density lithium batteries face many technical challenges, among which the performance optimization of electrolyte materials is a key link. The traditional research and development of electrolyte materials mainly rely on the experimental trial-and-error method, screening and optimizing the performance of electrolytes through a large number of experiments. This method not only takes a long time and has high costs, but also is difficult to comprehensively and deeply understand the interaction mechanism between electrolytes and electrode materials. In recent years, the application of artificial intelligence technology in the field of scientific research has become increasingly widespread, providing new ideas for the research and development of electrolyte materials. Artificial intelligence technology can quickly predict and optimize the performance of electrolyte materials through the analysis and learning of a large amount of data, thus greatly shortening the research and development cycle and reducing costs. At present, some prediction models and active generation models of battery material properties based on artificial intelligence have been developed, but these models still have problems of difficult deployment and use in practical applications, and require professional technical personnel for operation and maintenance.

[0003] Although artificial intelligence technology shows great potential in the research and development of battery materials, there are still some obvious problems in the existing technology. First of all, many advanced artificial intelligence models and tools are difficult to be adopted by a wide range of scientific research teams and the industrial community due to their high complexity and technical thresholds. This results in the advantages of artificial intelligence technology not being fully utilized, restricting its application scope in the research and development of electrolyte materials. Secondly, when the existing artificial intelligence models predict the performance of electrolyte materials, they often ignore the interaction mechanism between electrolytes and electrode materials, resulting in inaccurate prediction results. Summary of the Invention

[0004] The purpose of the present invention is to provide a lithium battery electrolyte formula prediction platform based on a large language model, which can realize the rapid screening and optimization of lithium battery electrolyte materials through the trained large language model, solve the problems of difficult deployment, difficult use and inaccurate prediction results faced by the existing technology in the research and development of electrolyte materials, realize the rapid screening and optimization of electrolyte materials, reduce the research and development cost and improve the research and development efficiency.

[0005] To solve the above technical problems, the present invention is implemented by the following technical solutions.

[0006] The present invention provides a lithium battery electrolyte formula prediction platform based on a large language model, including:

[0007] A data acquisition module for acquiring property parameters of a target lithium battery electrolyte;

[0008] A formulation prediction module for predicting based on the property parameters of the target lithium battery electrolyte using a trained large language model of the first type, and outputting an electrolyte formulation and a theoretical viscosity value of the electrolyte formulation; when the difference between the theoretical viscosity value of the electrolyte formulation and the actual viscosity value of the electrolyte formulation is within a preset threshold range, storing the electrolyte formulation in an electrolyte database;

[0009] A trajectory simulation module for pre - constructing a battery model according to the electrolyte formulations in the electrolyte database that meet the viscosity threshold of the battery model, and simulating the electrochemical reaction kinetics trajectory of the electrolyte and battery materials;

[0010] A formulation property calculation module for calculating the system properties of the electrolyte formulations that meet the viscosity threshold of the battery model;

[0011] A formulation output module for predicting based on the electrochemical reaction kinetics trajectory of the electrolyte and battery materials and the system properties of the electrolyte formulations that meet the viscosity threshold of the battery model using a trained large language model of the second type, and outputting an optimized electrolyte formulation.

[0012] Further, the training method of the large language model of the first type includes:

[0013] Pre - training the large language model of the first type, and generating a molecular structure and molecular ratio as an electrolyte formulation according to the element composition category and additive type of a given electrolyte;

[0014] Introducing chemical rule constraints to perform domain adaptation training on the pre - trained large language model of the first type;

[0015] Establishing an electrolyte formulation - viscosity correlation database according to the electrolyte formulation and the viscosity range of the electrolyte as a supervision signal to guide the training process;

[0016] During the training process, using a reinforcement learning strategy cross - entropy loss function to calculate the sequence generation loss and property prediction loss, and performing viscosity value reward feedback on the electrolyte formulation according to the supervision signal to guide the large language model of the first type to consider the viscosity range of the electrolyte when generating the molecular structure and molecular ratio, and outputting the electrolyte formulation and the theoretical viscosity value of the electrolyte formulation;

[0017] The sequence generation loss is used to measure the rationality of generating the molecular structure and molecular ratio;

[0018] The property prediction loss is used to measure the difference between the theoretical viscosity value of the electrolyte formulation and the actual viscosity value of the electrolyte formulation.

[0019] Further, pre-train the first large language model to generate a molecular structure and molecular ratio as an electrolyte formula according to the elemental composition category and additive type of the given electrolyte, including:

[0020] Obtain the SMILES string corresponding to the electrolyte and the molecular structure and molecular ratio corresponding to the SMILES string from a professional database in the electrolyte field;

[0021] Annotate the SMILES string corresponding to the electrolyte according to the molecular structure and molecular ratio corresponding to the SMILES string to construct a Prompt dataset;

[0022] Use the Prompt dataset to pre-train the first large language model to generate a molecular structure and molecular ratio as an electrolyte formula according to the elemental composition category and additive type of the given electrolyte.

[0023] Further, calculate the actual viscosity value of the electrolyte formula through molecular dynamics simulation, including:

[0024] Construct an initial simulation box according to the electrolyte formula and the theoretical viscosity value of the electrolyte formula, and add solvents, cations, and anions according to the molecular structure and molecular ratio of the electrolyte formula and the theoretical viscosity value of the electrolyte formula;

[0025] Optimize the molecular structure and interactions in the initial simulation box using the energy minimization method;

[0026] Set NVT (constant number of particles, volume, and temperature) simulation parameters to pre-equilibrate the initial simulation box to generate an initial conformation;

[0027] Judge whether the initial simulation box reaches a stable state according to the energy, temperature, pressure, and density of the initial simulation box;

[0028] If the initial simulation box does not reach a stable state, adjust the NVT simulation parameters or extend the equilibration time until the initial simulation box is stable, record and output the conformation in the stable state as the simulated electrolyte formula;

[0029] Use the liquid shear viscosity formula to calculate the actual viscosity value of the electrolyte formula according to the acceleration amplitude of the conformation in the stable state, the fitting velocity, the density of the initial simulation box, and the length in the Z-axis direction.

[0030] Further, pre-construct a battery model according to the electrolyte formula in the electrolyte database that meets the viscosity threshold of the battery model, including:

[0031] Construct an electrolyte according to the electrolyte formula that meets the viscosity threshold of the battery model;

[0032] A multi-scale model that defines the initial parameters of the battery model, the lattice structure of the electrode material, the electrolyte infiltration layer, and the SEI (Solid Electrolyte Interface) film. The initial parameters include the molecular structure and molecular ratio of the electrolyte, the type of electrode, and the size of the initial simulation box;

[0033] For the known molecular structures in the electrolyte, obtain the stable conformations of the known molecular structures from the professional database of electrolyte molecules;

[0034] For the unknown molecular structures in the electrolyte, calculate the stable conformations of the unknown molecular structures through density functional theory software;

[0035] According to the stable conformations of the known molecular structures and the stable conformations of the unknown molecular structures, calculate the filling amounts of each molecular structure in the initial simulation box;

[0036] According to the filling amounts of each molecular structure in the initial simulation box, distribute them in the battery model space according to the molecular structure and ratio;

[0037] Add an electrode layer in the reserved electrode space to obtain a pre-constructed battery model.

[0038] Furthermore, simulate the electrochemical reaction kinetic trajectory of the electrolyte and the battery material, including:

[0039] Map the molecular distribution of the electrolyte in the space of the pre-constructed battery model;

[0040] Simulate the charge transfer process at the electrolyte-electrode interface through reactive molecular dynamics, calculate the ion migration energy barrier and the interface impedance parameters, and output the voltage-time curve and the interface structure evolution trajectory as the electrochemical reaction kinetic trajectory of the electrolyte and the battery material.

[0041] Furthermore, calculate the system properties of the electrolyte formulation that meets the viscosity threshold of the battery model, including:

[0042] Convert the electrolyte formulation that meets the viscosity threshold of the battery model into the SMILES representation of the Lewis base target molecule, and perform prediction based on the trained DN value prediction model to obtain the DN value of the electrolyte formulation.

[0043] Furthermore, convert the electrolyte formulation that meets the viscosity threshold of the battery model into the SMILES representation of the Lewis base target molecule, including:

[0044] Extract the molecular structures in the electrolyte formulation that meets the viscosity threshold of the battery model;

[0045] Read the SMILES string corresponding to the molecular structure using ASE (Atomic Simulation Environment);

[0046] Perform dehydrogenation processing on the SMILES string to obtain a dehydrogenated SMILES string;

[0047] Use ASE to identify the Lewis base sites in the dehydrogenated SMILES string, and splice the Lewis base sites with standard Lewis acids to generate the SMILES representation of the Lewis base target molecule.

[0048] Further, construct the DN value prediction model based on a graph convolutional neural network, where the data processing method of the DN value prediction model includes:

[0049] Convert the SMILES representation of the Lewis base target molecule into a graph structure representation, and extract node features and edge features. The node features include atomic number, chirality, degree, formal charge, number of hydrogen atoms, number of radical electrons, orbital hybridization, aromaticity, and ring determination. The edge features include bond type, bond stereoconfiguration, and conjugation;

[0050] Use the input layer to receive the node features and the edge features and output them to the graph convolutional layer;

[0051] Use the graph convolutional layer to weighted aggregate the neighbor nodes of the node features and update the node features;

[0052] Use the output layer to map the output of the graph convolutional layer to a single DN value and output the DN value of the electrolyte formulation.

[0053] Further, the training process of the second large language model includes:

[0054] Collect electrolyte formulations that meet the viscosity threshold of the battery model and the system properties of the electrolyte formulations;

[0055] Construct an electrolyte according to the electrolyte formulation and simulate the electrochemical reaction kinetic trajectory of the electrolyte and the battery material;

[0056] Use the electrochemical reaction kinetic trajectory of the electrolyte and the battery material, the system properties of the electrolyte formulation, and the electrolyte formulation as a training set to train the second large language model for prediction and output an optimized electrolyte formulation.

[0057] Compared with the prior art, the beneficial effects achieved by the present invention:

[0058] 1. Through a data acquisition module, a formulation prediction module, a trajectory simulation module, a formulation property calculation module, and a formulation output module, the present invention realizes the full-chain intelligent prediction from the input of property parameters of the target electrolyte to the output of the optimized electrolyte formulation. The present invention significantly improves the R & D efficiency and formulation quality of the electrolyte formulation, reduces the experimental cost and time cost, accelerates the R & D process of battery materials, solves the problems of difficult deployment, difficult use, and inaccurate prediction results faced by the prior art in the R & D of electrolyte materials, realizes the rapid screening and optimization of electrolyte materials, and reduces the R & D cost and improves the R & D efficiency.

[0059] 2. Through a pre-trained first large language model, combined with chemical rule constraints and reinforcement learning strategies, the present invention can efficiently generate reasonable molecular structures and molecular ratios as electrolyte formulations according to the elemental composition categories and additive types of electrolytes. At the same time, using the electrolyte formulation-viscosity correlation database as a supervision signal, the cross-entropy loss function is used to calculate the sequence generation loss and property prediction loss for training, ensuring that the predicted electrolyte formulation not only has a reasonable structure but also has a viscosity value close to the actual value, thus greatly improving the accuracy and efficiency of electrolyte formulation prediction.

[0060] 3. Through molecular dynamics simulation, an initial simulation box is constructed according to the predicted electrolyte formulation, and after steps such as energy minimization and pre-equilibration, the actual viscosity value of the electrolyte is finally calculated. In addition, the present invention can also pre-construct a battery model according to the electrolyte formulation that meets the viscosity threshold of the battery model, simulate the electrochemical reaction kinetic trajectory of the electrolyte and battery materials, and calculate key performance indicators such as ion migration energy barrier and interface impedance parameters, providing reliable data support for the further optimization of the electrolyte formulation.

[0061] 4. Based on the training process of the second large language model, the present invention can collect electrolyte formulations that meet the viscosity threshold of the battery model and the properties of the electrolyte formulation system, predict through the second large language model, and output the optimized electrolyte formulation, which not only realizes the intelligent optimization of the electrolyte formulation but also greatly improves the R & D efficiency and reduces the R & D cost. At the same time, through the DN value prediction model constructed by the graph convolutional neural network, the system properties such as the DN value of the electrolyte formulation can be accurately predicted, providing a more comprehensive evaluation index for the screening and optimization of the electrolyte formulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 is a schematic diagram of the module composition of a lithium battery electrolyte formulation prediction platform based on a large language model provided by an embodiment of the present invention;

[0063] Figure 2 is a schematic diagram of the working process of a lithium battery electrolyte formulation prediction platform based on a large language model provided by an embodiment of the present invention. Detailed implementation manners

[0064] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. Without conflict, the technical features in the embodiments of the present invention and the embodiments can be combined with each other.

[0065] Embodiment 1

[0066] As Figure 1 shown, this embodiment introduces a prediction platform for the lithium battery electrolyte formula based on a large language model, including:

[0067] A data acquisition module, configured to acquire the property parameters of the target lithium battery electrolyte, including the viscosity range, element composition category, and additive type.

[0068] In the present invention, the data acquisition module is responsible for collecting the property parameters of the target lithium battery electrolyte, including the viscosity range, element composition category such as the content of chemical elements, and additive type such as thickeners and conductive agents.

[0069] A formula prediction module, configured to predict based on the property parameters of the target lithium battery electrolyte using a trained first large language model, and output the electrolyte formula and the theoretical viscosity value of the electrolyte formula. The electrolyte formula includes the molecular structure and molecular ratio; when the difference between the theoretical viscosity value of the electrolyte formula and the actual viscosity value of the electrolyte formula is within a preset threshold range, the electrolyte formula is stored in the electrolyte database.

[0070] In the present invention, the formula prediction module, according to the property parameters of the target lithium battery electrolyte, uses the trained first large language model through its powerful data processing and pattern recognition capabilities to predict the electrolyte formula including the molecular structure and molecular ratio and the theoretical viscosity value of the electrolyte formula. The output electrolyte formula may be a set or multiple sets.

[0071] When the difference between the theoretical viscosity value of the electrolyte formula and the actual viscosity value of the electrolyte formula is within a preset threshold range, it indicates that the electrolyte formula has good prediction accuracy. Therefore, it is stored in the electrolyte database to accumulate reliable electrolyte formula data.

[0072] A trajectory simulation module, configured to pre-construct a battery model according to the electrolyte formula in the electrolyte database that meets the viscosity threshold of the battery model, and simulate the electrochemical reaction kinetic trajectory between the electrolyte and the battery materials.

[0073] In the present invention, the trajectory simulation module pre-constructs a battery model using an electrolyte formulation in the electrolyte database that meets the viscosity threshold of the battery model, and simulates the electrochemical reaction kinetics trajectory between the electrolyte and the battery materials, capable of evaluating the actual performance of the electrolyte constructed according to the electrolyte formulation in the battery, including its charge-discharge performance, cycle stability, etc.

[0074] A formulation property calculation module for calculating the system properties of the electrolyte formulation that meets the viscosity threshold of the battery model.

[0075] For an electrolyte formulation that meets the viscosity threshold of the battery model, calculate key properties such as the DN value, viscosity, diffusion coefficient, etc. of the electrolyte constructed according to the electrolyte formulation, which are important indicators for evaluating electrolyte performance.

[0076] A formulation output module for predicting based on the trained second large language model according to the electrochemical reaction kinetics trajectory between the electrolyte and the battery materials and the system properties of the electrolyte formulation that meets the viscosity threshold of the battery model, and outputting an optimized electrolyte formulation.

[0077] In the present invention, the formulation output module uses the electrochemical reaction kinetics trajectory provided by the trajectory simulation module and the system properties provided by the formulation property calculation module to perform comprehensive analysis based on the trained second large language model, predict and output an optimized electrolyte formulation, provide the final electrolyte formulation optimization result, and provide a practical guidance scheme for battery manufacturers and R & D personnel.

[0078] Example 2

[0079] Based on the same inventive concept as Example 1, as Figure 2 shown, this example introduces the working process of a lithium battery electrolyte formulation prediction platform based on a large language model, including:

[0080] Step 1: Obtain the property parameters of the target lithium battery electrolyte, including the viscosity range, element composition category, and additive type.

[0081] Step 2: Based on the property parameters of the target lithium battery electrolyte, perform prediction using the trained first large language model, and output the electrolyte formulation and the theoretical viscosity value of the electrolyte formulation. The electrolyte formulation includes the molecular structure and molecular ratio; when the difference between the theoretical viscosity value of the electrolyte formulation and the actual viscosity value of the electrolyte formulation is within a preset threshold range, the electrolyte formulation is stored in the electrolyte database.

[0082] In some embodiments, the training method of the first large language model includes:

[0083] Step 2.1: Pretrain the first large language model to generate a molecular structure and molecular ratio as an electrolyte formulation based on the elemental composition category and additive type of the given electrolyte.

[0084] In this embodiment, pretraining the first large language model to generate a molecular structure and molecular ratio as an electrolyte formulation based on the elemental composition category and additive type of the given electrolyte includes:

[0085] Step 2.1.1: Obtain the SMILES string corresponding to the electrolyte and the molecular structure and molecular ratio corresponding to the SMILES string from a professional database in the electrolyte field.

[0086] Step 2.1.2: Annotate the SMILES string corresponding to the electrolyte according to the molecular structure and molecular ratio corresponding to the SMILES string to construct a Prompt dataset.

[0087] Step 2.1.3: Use the Prompt dataset to pretrain the first large language model to generate a molecular structure and molecular ratio as an electrolyte formulation based on the elemental composition category and additive type of the given electrolyte.

[0088] Step 2.2: Introduce chemical rule constraints to perform domain adaptation training on the pretrained first large language model.

[0089] Step 2.3: Establish an electrolyte formulation-viscosity correlation database as a supervision signal to guide the training process according to the electrolyte formulation and the viscosity range of the electrolyte.

[0090] During the training process, an enhanced learning strategy cross-entropy loss function is adopted to calculate the sequence generation loss and property prediction loss, and viscosity value reward feedback is performed on the electrolyte formulation according to the supervision signal to guide the first large language model to consider the viscosity range of the electrolyte when generating the molecular structure and molecular ratio, and output the electrolyte formulation and the theoretical viscosity value of the electrolyte formulation.

[0091] In this embodiment, the sequence generation loss is used to measure the rationality of generating the molecular structure and molecular ratio.

[0092] In this embodiment, the property prediction loss is used to measure the difference between the theoretical viscosity value of the electrolyte formulation and the actual viscosity value of the electrolyte formulation.

[0093] In some embodiments, calculating the actual viscosity value of the electrolyte formulation through molecular dynamics simulation includes:

[0094] Construct an initial simulation box according to the electrolyte formulation and the theoretical viscosity value of the electrolyte formulation, and add solvents, cations, and anions according to the molecular structure and molecular ratio of the electrolyte formulation and the theoretical viscosity value of the electrolyte formulation.

[0095] Optimize the molecular structure and interactions in the initial simulation box using the energy minimization method;

[0096] Set the NVT simulation parameters to pre-equilibrate the initial simulation box and generate an initial conformation;

[0097] Judge whether the initial simulation box reaches a stable state according to the energy, temperature, pressure and density of the initial simulation box;

[0098] If the initial simulation box does not reach a stable state, adjust the NVT simulation parameters or extend the equilibration time until the initial simulation box is stable, record and output the conformation in the stable state as the electrolyte formula obtained from the simulation;

[0099] Use the liquid shear viscosity formula to calculate the actual viscosity value of the electrolyte formula according to the acceleration amplitude of the conformation in the stable state, the fitting velocity, the density of the initial simulation box, and the length in the Z-axis direction.

[0100] In this embodiment, the liquid shear viscosity formula is expressed as:

[0101] ;

[0102] In the formula, represents the actual viscosity value of the electrolyte formula, represents the acceleration amplitude of the conformation in the stable state, represents the density of the initial simulation box, represents the length in the Z-axis direction of the initial simulation box, represents the fitting velocity of the conformation in the stable state.

[0103] Step 3: Pre-construct a battery model according to the electrolyte formula in the electrolyte database that meets the viscosity threshold of the battery model, and simulate the electrochemical reaction kinetic trajectory of the electrolyte and the battery materials.

[0104] In electrolyte research, theoretical calculation methods such as density functional theory (DFT) and molecular dynamics (MD) simulations play important roles. Density functional theory (DFT) calculations can reveal the interactions between electrolyte components from the quantum chemistry level. For example, by calculating the binding energy between the solvent and lithium ions, the differences in the solvation ability of different solvents for lithium ions can be explained. Molecular dynamics (MD) simulations focus on studying the dynamic behavior of electrolytes from the atomic and molecular levels, can calculate macroscopic physicochemical properties such as the ionic diffusion coefficient and viscosity of electrolytes, and reveal the relationships between these properties and the microscopic structure.

[0105] Step 3.1: Pre-construct a battery model according to the electrolyte formula in the electrolyte database that meets the viscosity threshold of the battery model, including:

[0106] Step 3.1.1: Construct an electrolyte according to the electrolyte formula that meets the viscosity threshold of the battery model.

[0107] Step 3.1.2: Define the initial parameters of the battery model, the lattice structure of the electrode material, the electrolyte infiltration layer, and the multi-scale model of the SEI film.

[0108] In this embodiment, the initial parameters include the molecular structure and molecular ratio of the electrolyte, the type of the electrode, and the size of the initial simulation box.

[0109] Step 3.1.3: For the known molecular structures in the electrolyte, obtain the stable conformations of the known molecular structures from the professional database of electrolyte molecules.

[0110] Step 3.1.4: For the unknown molecular structures in the electrolyte, calculate the stable conformations of the unknown molecular structures through density functional theory software.

[0111] Step 3.1.5: Calculate the filling amounts of the individual molecular structures in the initial simulation box according to the stable conformations of the known molecular structures and the stable conformations of the unknown molecular structures.

[0112] Step 3.1.6: Distribute them in the battery model space according to the molecular structure and ratio according to the filling amounts of the individual molecular structures in the initial simulation box.

[0113] Step 3.1.7: Add an electrode layer in the reserved electrode space to obtain a pre-constructed battery model.

[0114] Step 3.2: Simulate the electrochemical reaction kinetic trajectory of the electrolyte and the battery material, including:

[0115] Step 3.2.1: Map the molecular distribution of the electrolyte in the space of the pre-constructed battery model.

[0116] Step 3.2.2: Simulate the charge transfer process at the interface between the electrolyte and the electrode through reactive molecular dynamics, calculate the ion migration energy barrier and the interface impedance parameters, and output the voltage-time curve and the interface structure evolution trajectory as the electrochemical reaction kinetic trajectory of the electrolyte and the battery material.

[0117] Step 4: Calculate the system properties of the electrolyte formula that meets the viscosity threshold of the battery model, including the DN value, viscosity, and diffusion coefficient.

[0118] In the field of electrochemistry, researchers usually rely on the Gutmann donor number (DN) to explain the Lewis basicity of chemical substances as electrolytes and additives, in order to estimate the degree of their solvation or the degree of interaction with Li+ as a Lewis acid. Therefore, the Gutmann donor number (DN) is often cited as an important reference index for the research of electrolytes in advanced lithium-oxygen and lithium-sulfur batteries. At the same time, in the latest research, it is found that the Gutmann donor number (DN) is an effective index for screening the main solvents and diluents of locally high-concentration electrolytes, where the DN value of the diluent ≤ 10 and the DN value of the main solvent ≥ 10.

[0119] In some embodiments, calculating the system properties of the electrolyte formulation that meets the battery model viscosity threshold includes:

[0120] Convert the electrolyte formulation that meets the battery model viscosity threshold into the SMILES representation of the Lewis base target molecule, and predict based on the trained DN value prediction model to obtain the DN value of the electrolyte formulation.

[0121] Step 4.1: Convert the electrolyte formulation that meets the battery model viscosity threshold into the SMILES representation of the Lewis base target molecule, including:

[0122] Step 4.1.1: Extract the molecular structure in the electrolyte formulation that meets the battery model viscosity threshold;

[0123] Step 4.1.2: Use ASE to read the SMILES string corresponding to the molecular structure;

[0124] Step 4.1.3: Perform dehydrogenation processing on the SMILES string to obtain the dehydrogenated SMILES string;

[0125] Step 4.1.4: Use ASE to identify the Lewis base sites in the dehydrogenated SMILES string, and splice the Lewis base sites with the standard Lewis acid to generate the SMILES representation of the Lewis base target molecule.

[0126] Step 4.2: Construct the DN value prediction model based on the graph convolutional neural network.

[0127] In this embodiment, the data processing method of the DN value prediction model includes:

[0128] Step 4.2.1: Convert the SMILES representation of the target molecule of the Lewis base into a graph structure representation, and extract node features and edge features. The node features include atomic number, chirality, degree, formal charge, number of hydrogen atoms, number of radical electrons, orbital hybridization, aromaticity, and ring determination. The edge features include bond type, bond stereoconfiguration, and conjugation;

[0129] Step 4.2.2: Use the input layer to receive the node features and the edge features and output them to the graph convolutional layer;

[0130] Step 4.2.3: Use the graph convolutional layer to weighted-aggregate the neighboring nodes of the node features and update the node features;

[0131] Step 4.2.4: Use the output layer to map the output of the graph convolutional layer to a single DN value and output the DN value of the electrolyte formulation.

[0132] Step 5: Based on the electrochemical reaction kinetic trajectory of the electrolyte and the battery material and the system properties of the electrolyte formulation that meet the viscosity threshold of the battery model, perform prediction based on the trained second large language model and output the optimized electrolyte formulation.

[0133] In this embodiment, the training process of the second large language model includes:

[0134] Step 5.1: Collect electrolyte formulations that meet the viscosity threshold of the battery model and the system properties of the electrolyte formulations;

[0135] Step 5.2: Construct an electrolyte according to the electrolyte formulation and simulate the electrochemical reaction kinetic trajectory of the electrolyte and the battery material;

[0136] Step 5.3: Use the electrochemical reaction kinetic trajectory of the electrolyte and the battery material, the system properties of the electrolyte formulation, and the electrolyte formulation as a training set to train the second large language model for prediction and output the optimized electrolyte formulation.

[0137] Example 3

[0138] Based on the same inventive concept as other embodiments, this embodiment introduces a computer-readable storage medium, on which computer instructions are stored, and when the computer instructions are executed by a processor, the steps of the method in the above-mentioned Embodiment 1 or 2 are implemented.

[0139] Example 4

[0140] Based on the same inventive concept as other embodiments, this embodiment introduces a computer program product, including computer instructions, and when the computer instructions are executed by a processor, the steps of the method in the above-mentioned Embodiment 1 or 2 are implemented.

[0141] In summary, through the data acquisition module, formula prediction module, trajectory simulation module, formula property calculation module and formula output module, the present invention realizes the full-chain intelligent prediction from the input of the property parameters of the target electrolyte to the output of the optimized electrolyte formula. The present invention significantly improves the R & D efficiency and formula quality of the electrolyte formula, reduces the experimental cost and time cost, accelerates the R & D process of battery materials, solves the problems of difficult deployment, difficult use and inaccurate prediction results faced by the prior art in the R & D of electrolyte materials, realizes the rapid screening and optimization of electrolyte materials, and reduces the R & D cost and improves the R & D efficiency.

[0142] Through the pre-trained first large language model, combined with chemical rule constraints and reinforcement learning strategies, the present invention can efficiently generate reasonable molecular structures and molecular ratios as electrolyte formulas according to the element composition categories and additive types of electrolytes. At the same time, using the electrolyte formula-viscosity correlation database as a supervision signal, the cross-entropy loss function is used to calculate the sequence generation loss and property prediction loss for training, ensuring that the predicted electrolyte formula not only has a reasonable structure, but also the viscosity value is close to the actual value, thus greatly improving the accuracy and efficiency of electrolyte formula prediction.

[0143] Through molecular dynamics simulation, an initial simulation box is constructed according to the predicted electrolyte formula, and after steps such as energy minimization and pre-equilibration, the actual viscosity value of the electrolyte is finally calculated. In addition, the present invention can also pre-construct a battery model according to the electrolyte formula that meets the viscosity threshold of the battery model, simulate the electrochemical reaction kinetic trajectory of the electrolyte and battery materials, and calculate key performance indicators such as ion migration energy barrier and interface impedance parameters, providing reliable data support for the further optimization of the electrolyte formula.

[0144] Based on the training process of the second large language model, the present invention can collect electrolyte formulas that meet the viscosity threshold of the battery model and the properties of the electrolyte formula system, and through prediction by the second large language model, output the optimized electrolyte formula, which not only realizes the intelligent optimization of the electrolyte formula, but also greatly improves the R & D efficiency and reduces the R & D cost. At the same time, through the DN value prediction model constructed by the graph convolutional neural network, the system properties such as the DN value of the electrolyte formula can be accurately predicted, providing a more comprehensive evaluation index for the screening and optimization of the electrolyte formula.

[0145] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0146] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to generate a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0147] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0148] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0149] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the spirit and scope protected by the present invention and the claims. These all fall within the protection scope of the present invention.

Claims

1. A prediction platform for lithium battery electrolyte formulations based on large language models, characterized in that, Including: A data acquisition module for acquiring the property parameters of the target lithium battery electrolyte; A formula prediction module for predicting based on the property parameters of the target lithium battery electrolyte using a trained first large language model, and outputting the electrolyte formula and the theoretical viscosity value of the electrolyte formula; when the difference between the theoretical viscosity value of the electrolyte formula and the actual viscosity value of the electrolyte formula is within a preset threshold range, storing the electrolyte formula in the electrolyte database; A trajectory simulation module for pre-constructing a battery model according to the electrolyte formula in the electrolyte database that meets the battery model viscosity threshold, and simulating the electrochemical reaction kinetic trajectory of the electrolyte and the battery materials; A formula property calculation module for calculating the system properties of the electrolyte formula that meets the battery model viscosity threshold; A formula output module for predicting based on the electrochemical reaction kinetic trajectory of the electrolyte and the battery materials and the system properties of the electrolyte formula that meets the battery model viscosity threshold using a trained second large language model, and outputting the optimized electrolyte formula; Pre-constructing a battery model according to the electrolyte formula in the electrolyte database that meets the battery model viscosity threshold includes: Constructing an electrolyte according to the electrolyte formula that meets the battery model viscosity threshold; Defining the initial parameters of the battery model, the lattice structure of the electrode material, the electrolyte infiltration layer, and the multi-scale model of the SEI film, where the initial parameters include the molecular structure and molecular ratio of the electrolyte, the type of the electrode, and the size of the initial simulation box; For the known molecular structures in the electrolyte, obtaining the stable conformations of the known molecular structures from the electrolyte molecule professional database; For the unknown molecular structures in the electrolyte, calculating the stable conformations of the unknown molecular structures through density functional theory software; Calculating the filling amounts of each molecular structure in the initial simulation box according to the stable conformations of the known molecular structures and the stable conformations of the unknown molecular structures; Distributing according to the filling amounts of each molecular structure in the initial simulation box in the battery model space according to the molecular structure and ratio; Adding an electrode layer in the reserved electrode space to obtain the pre-constructed battery model.

2. The prediction platform for lithium battery electrolyte formulations based on large language models according to claim 1, wherein The training method of the first large language model includes: Pre-training the first large language model to generate a molecular structure and molecular ratio as the electrolyte formula according to the element composition category and additive type of the given electrolyte; Introducing chemical rule constraints to perform domain adaptation training on the pre-trained first large language model; Establishing an electrolyte formula-viscosity correlation database according to the electrolyte formula and the viscosity range of the electrolyte as a supervision signal to guide the training process; During the training process, adopting a reinforcement learning strategy cross-entropy loss function to calculate the sequence generation loss and the property prediction loss, and performing viscosity value reward feedback on the electrolyte formula according to the supervision signal to guide the first large language model to consider the viscosity range of the electrolyte when generating the molecular structure and molecular ratio, and outputting the electrolyte formula and the theoretical viscosity value of the electrolyte formula; The sequence generation loss is used to measure the rationality of generating the molecular structure and molecular ratio; The property prediction loss is used to measure the difference between the theoretical viscosity value of the electrolyte formula and the actual viscosity value of the electrolyte formula.

3. The prediction platform for lithium battery electrolyte formulations based on large language models according to claim 2, characterized in that Pre-train the first large language model to generate a molecular structure and molecular ratio as an electrolyte formula based on the elemental composition category and additive type of the given electrolyte, including: Obtain the SMILES string corresponding to the electrolyte and the molecular structure and molecular ratio corresponding to the SMILES string from a professional database in the field of electrolytes; Annotate the SMILES string corresponding to the electrolyte according to the molecular structure and molecular ratio corresponding to the SMILES string to construct a Prompt dataset; Use the Prompt dataset to pre-train the first large language model to generate a molecular structure and molecular ratio as an electrolyte formula based on the elemental composition category and additive type of the given electrolyte.

4. The prediction platform for lithium battery electrolyte formulations based on large language models according to claim 1, wherein, Calculate the actual viscosity value of the electrolyte formula through molecular dynamics simulation, including: Construct an initial simulation box according to the electrolyte formula and the theoretical viscosity value of the electrolyte formula, and add solvents, cations, and anions according to the molecular structure and molecular ratio of the electrolyte formula and the theoretical viscosity value of the electrolyte formula; Optimize the molecular structure and interactions in the initial simulation box using the energy minimization method; Set NVT simulation parameters to pre-equilibrate the initial simulation box to generate an initial conformation; Judge whether the initial simulation box reaches a stable state according to the energy, temperature, pressure, and density of the initial simulation box; If the initial simulation box does not reach a stable state, adjust the NVT simulation parameters or extend the equilibration time until the initial simulation box is stable, record and output the conformation in the stable state as the simulated electrolyte formula; Use the liquid shear viscosity formula to calculate the actual viscosity value of the electrolyte formula according to the acceleration amplitude and fitting velocity of the conformation in the stable state, the density of the initial simulation box, and the length in the Z-axis direction.

5. The large language model-based lithium battery electrolyte formula prediction platform according to claim 1, simulating the electrochemical reaction kinetic trajectory of the electrolyte and battery materials, including: Map the molecular distribution of the electrolyte in the space of the pre-constructed battery model; Simulate the charge transfer process between the electrolyte and the electrode interface through reactive molecular dynamics, calculate the ion migration energy barrier and interface impedance parameters, and output the voltage-time curve and interface structure evolution trajectory as the electrochemical reaction kinetic trajectory of the electrolyte and battery materials.

6. The prediction platform for lithium battery electrolyte formulations based on large language models according to claim 5, wherein, Calculate the system properties of the electrolyte formula that meets the viscosity threshold of the battery model, including: Convert the electrolyte formula that meets the viscosity threshold of the battery model into the SMILES representation of the Lewis base target molecule, and predict based on the trained DN value prediction model to obtain the DN value of the electrolyte formula.

7. The prediction platform for lithium battery electrolyte formulations based on large language models according to claim 6, wherein, Convert the electrolyte formula that meets the viscosity threshold of the battery model into the SMILES representation of the Lewis base target molecule, including: Extract the molecular structure in the electrolyte formula that meets the viscosity threshold of the battery model; Use ASE to read the SMILES string corresponding to the molecular structure; Perform dehydrogenation processing on the SMILES string to obtain the dehydrogenated SMILES string; Use ASE recognition to identify the Lewis base sites in the dehydrogenated SMILES string, and splice the Lewis base sites with standard Lewis acids to generate the SMILES representation of the Lewis base target molecule.

8. The prediction platform for lithium battery electrolyte formulations based on large language models according to claim 6, wherein, Construct the DN value prediction model based on the graph convolutional neural network, where the data processing method of the DN value prediction model includes: Convert the SMILES representation of the Lewis base target molecule into a graph structure representation, and extract node features and edge features. The node features include atomic number, chirality, degree, formal charge, number of hydrogen atoms, number of radical electrons, orbital hybridization, aromaticity, and ring determination. The edge features include bond type, bond stereoconfiguration, and conjugation; Use the input layer to receive the node features and the edge features and output them to the graph convolutional layer; Use the graph convolutional layer to weighted aggregate the neighbor nodes of the node features and update the node features; Use the output layer to map the output of the graph convolutional layer to a single DN value and output the DN value of the electrolyte formulation.

9. The prediction platform for lithium battery electrolyte formulations based on large language models according to claim 1, wherein The training process of the second large language model includes: Collect electrolyte formulations that meet the viscosity threshold of the battery model and the system properties of the electrolyte formulations; Construct an electrolyte according to the electrolyte formulation and simulate the electrochemical reaction kinetic trajectory of the electrolyte and the battery materials; Use the electrochemical reaction kinetic trajectory of the electrolyte and the battery materials, the system properties of the electrolyte formulation, and the electrolyte formulation as a training set to train the second large language model for prediction and output an optimized electrolyte formulation.

Images