Performance prediction method, system and equipment for porous carbon skeleton in silicon-carbon negative electrode and medium

Abstract

The invention relates to the technical field of porous structure modeling analysis, in particular to a method, system, equipment and medium for predicting the performance of a porous carbon skeleton in a silicon-carbon negative electrode, and the method comprises the following steps: S1, establishing a bicontinuous nano porous structure model of the porous carbon skeleton in the silicon-carbon negative electrode based on a Gaussian random field; s2, performing model training by taking the structural information of the porous carbon material as a feature vector of machine learning training and taking the mechanical property of the porous carbon material as a target value; and S3, carrying out gradient lifting decision tree model training to obtain model performance prediction. According to the method, the bicontinuous nano-porous structure model is established by using the Gaussian random field, the randomness of the porous structure is considered, the continuity and consistency of the model are ensured through the periodic boundary condition, and a reliable basis is provided for subsequent performance prediction. Meanwhile, the gradient boosting decision tree model is utilized to predict the mechanical property of the porous carbon material, so that the prediction accuracy is improved, the prediction time is shortened, and the working efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of porous structure modeling and analysis, and specifically to a method, system, equipment and medium for predicting the performance of a porous carbon skeleton in a silicon-carbon negative electrode. Background Art

[0002] Lithium-ion battery is a rechargeable battery with the advantages of high discharge voltage platform, low self-discharge, high energy density, long life, lightweight, environmental protection, low cost, etc. It is widely used in mobile electronic devices, electric vehicles, smart grids, energy storage systems, etc. However, traditional lithium-ion batteries have reached the limit of energy density, and irreversible capacity loss still restricts the further improvement of lithium-ion battery performance. With the popularization of new energy vehicles, people's demand for batteries is growing, which puts forward new requirements for the development of batteries.

[0003] The performance of lithium-ion batteries depends mainly on the properties of their anode materials, which affect key indicators such as energy density, power density and cycle stability. Traditional anode materials such as graphite have good electrochemical stability and coulombic efficiency, but have a low theoretical capacity of 372 mAhg -1 , which is not enough to meet modern high energy density storage needs. The theoretical capacity of silicon is 4200mAhg -1 , is a promising alternative to graphite. However, its large volume change (up to 300%) during the charge and discharge cycle can cause structural damage and reduce battery performance. These problems make the application and promotion of silicon anodes face major challenges.

[0004] At present, there have been studies on the composite of silicon and nanoporous amorphous carbon materials. The porous carbon skeleton provides good conductivity and mechanical properties while buffering the volume expansion of silicon and stabilizing the electrode structure. However, the relationship between the microstructure of the porous carbon skeleton, its core structure and mechanical properties is still unclear. This gap limits the optimization of material properties and hinders its wider application in high-performance lithium-ion batteries. Summary of the invention

[0005] In view of the problem that the porous amorphous carbon skeleton model in the silicon-carbon negative electrode in the prior art is difficult to model and has low simulation accuracy, the present invention provides a method, system, equipment and medium for predicting the performance of the porous carbon skeleton in the silicon-carbon negative electrode.

[0006] The present invention is achieved through the following technical solutions: A method for predicting the performance of a porous carbon skeleton in a silicon-carbon negative electrode comprises the following steps: S1, a bicontinuous nanoporous structure model of porous carbon skeleton in silicon-carbon anode is established based on Gaussian random field; S2, using the structural information of the porous carbon material as the feature vector for machine learning training and the mechanical properties of the porous carbon material as the target value for model training to obtain the target model; S3, training a gradient boosting decision tree model for the target model obtained in S2 to obtain a model performance prediction; S4, analyze the characteristics of the obtained prediction data to determine the importance of different structural parameters of the model.

[0007] Preferably, in S1, the process of establishing the bicontinuous nanoporous structure model of the porous carbon skeleton in the silicon-carbon negative electrode includes: S11, Gaussian random field generated by superposition of sinusoidal waves with fixed wavelength and amplitude but random direction and phase; S12, a bicontinuous porous structure with periodic boundary conditions is constructed in a cuboid space in a Gaussian random field.

[0008] Preferably, in S12, the expression of the Gaussian random field is:

[0009] in, is the position vector, represents the wave number in the truncated series, is the normalization factor, and Respectively represent The direction and phase of each wave; is a set of uniformly distributed Random phase within range. Must meet , is a constant; represents the surface of a solid, The region is defined as the solid phase, and The area represents the hole; Used to express the porosity of bicontinuous porous structures. The porosity is the number of deleted atoms divided by the original number of atoms.

[0010] Preferably, in S12, Must meet: , a, b, c are the edge vectors of the model, and m, n, p are integers.

[0011] Preferably, in S2, during training, the loss function is the root mean square error.

[0012] Preferably, in S4, the importance of different structural parameters of the model is calibrated using the average impurity reduction method and the Shapley value method.

[0013] Preferably, model evaluation and reliability check are also included, and the evaluation indicators include correlation coefficient R, determination coefficient R2, mean absolute error MAE, mean square error MSE and root mean square error RMSE, and the Pearson correlation coefficient PCC is used to measure the linear correlation between structural parameters and mechanical properties.

[0014] A system for predicting the performance of a porous carbon skeleton in a silicon-carbon negative electrode comprises a model building module, a model training module, a model prediction module and a model post-processing module, wherein the model building module is used to establish a bicontinuous nanoporous structure model of a porous carbon skeleton in a silicon-carbon negative electrode based on a Gaussian random field; the model training module is used to perform model training using the structural information of a porous carbon material as a feature vector for machine learning training and the mechanical properties of the porous carbon material as a target value to obtain a target model; the model prediction module is used to perform gradient boosting decision tree model training on the obtained target model to obtain a model performance prediction; and the model post-processing module is used to perform feature analysis on the obtained prediction data to determine the importance of different structural parameters of the model.

[0015] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and wherein the processor implements the steps of the modeling method when executing the computer program.

[0016] A storage medium stores a computer program, which implements the steps of any one of the modeling methods when executed by a processor.

[0017] Compared with the prior art, the present invention has the following beneficial effects: The present invention discloses a method for predicting the performance of a porous carbon skeleton in a silicon-carbon negative electrode. The method uses a Gaussian random field to establish a bicontinuous nanoporous structure model, which not only takes into account the randomness of the porous structure, but also ensures the continuity and consistency of the model through periodic boundary conditions, providing a reliable basis for subsequent performance predictions. At the same time, the gradient boosting decision tree model is used to predict the mechanical properties of porous carbon materials, which not only improves the accuracy of the prediction, but also shortens the prediction time and improves work efficiency. The importance of different structural parameters of the model is calibrated by means of the average impurity reduction method, the Shapley value method, etc., in order to deeply understand the relationship between the structure and performance of porous carbon materials, and provide guidance for the optimal design and preparation of materials. It is not only suitable for the performance prediction of porous carbon skeletons in silicon-carbon negative electrodes, but can also be extended to the performance prediction of other porous materials. Its general modeling and prediction methods provide a powerful tool for the research and application of porous materials.

[0018] Furthermore, the performance of the model was comprehensively evaluated through multiple indicators such as correlation coefficient R, determination coefficient R2, mean absolute error MAE, mean square error MSE and root mean square error RMSE. At the same time, the Pearson correlation coefficient PCC was used to measure the linear correlation between structural parameters and mechanical properties to ensure the reliability and accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of the method for predicting the mechanical properties of the model in the present invention; Figure 2 It is a schematic diagram of the flow chart of the modeling method of the present invention; Figure 3 is a cross-sectional view of a typical model with different pore sizes constructed in the present invention; Figure 4 This is a comparison chart of the importance of model features calibrated by the average impurity reduction method used in the machine learning model of the present invention; Figure 5 This is a comparison chart of the importance of model features calibrated using the Shapley value method for the machine learning model of the present invention; Figure 6 is an evaluation diagram of the degree of fit and reliability of the machine learning model of the present invention; Figure 7 It is a linear correlation diagram between the structural parameters of the machine learning model of the present invention and the mechanical properties. DETAILED DESCRIPTION

[0020] The present invention is further described in detail below in conjunction with specific embodiments, which are intended to explain the present invention rather than to limit it.

[0021] The present invention discloses a method for predicting the performance of porous carbon skeleton in silicon-carbon negative electrode, referring to Figure 1 , including the following steps: S1, based on Gaussian random field, a bicontinuous nanoporous structure model of porous carbon skeleton in silicon-carbon negative electrode is established; specifically: S11, the Gaussian random field is generated by the superposition of sine waves with fixed wavelength and amplitude but random direction and phase. The expression of the Gaussian random field is:

[0022] in, is the position vector, represents the wave number in the truncated series, is the normalization factor, and Respectively represent The direction and phase of each wave; is a set of uniformly distributed Random phase within range. Must meet , is a constant; represents the surface of a solid, The region is defined as the solid phase, and The area represents the hole; Used to express the porosity of bicontinuous porous structures. The porosity is the number of deleted atoms divided by the original number of atoms.

[0023] S12, a bicontinuous porous structure with periodic boundary conditions is constructed in a cuboid space in a Gaussian random field.

[0024] Must meet: , a, b, c are the edge vectors of the model, and m, n, p are integers.

[0025] In order to meet the above conditions, we have:

[0026] The above formula means:

[0027] , , is an integer. Therefore, Must meet:

[0028] in,

[0029] a, b, c are the lengths of vectors a, b, c respectively, , , They are , , Angle.

[0030] Reference Figure 2 , the wave vector is determined as follows: In the first step, the wave vectors of these waves are evenly distributed on a grid of size The solid angle of is the entire sphere. The cylindrical coordinates of the grid points are ,in is the golden angle (~137.508º), n is the total number of grid points, here we take .

[0031] The second step is to convert the above cylindrical coordinates into Cartesian coordinates and multiply them by a desired Value, take , thus we can get the accurate wave vector In order to show periodicity in a parallelepiped, we need to make ,in .

[0032] In the third step, in order to force the final structure to be a periodic structure, the formula is used to find .

[0033] Depend on Get the nearest integer , and then calculate according to the following formula :

[0034] Secondly, by Get the nearest integer , and then calculate according to the following formula :

[0035] Depend on Get the nearest integer , and then calculate according to the following formula :

[0036] In order to ensure that the porous structure has a clear ligament size distribution, only exist Between . S2, processing the available structural information of amorphous porous carbon materials into format data that can be input into the machine learning model through data preprocessing, using the structural information of the porous carbon materials as the feature vector of machine learning training, and using the mechanical properties of the porous carbon materials as the target value for model training to obtain the target model, and the loss function is the root mean square error; S3, performs gradient boosting decision tree model training on the target model obtained in S2 to obtain model performance prediction.

[0037] The GBDT gradient boosting algorithm is as follows: Input: Training data , a differentiable loss function , number of iterations ; Output: Regression model corresponding to training data

[0038] Algorithm steps: Step 1: Initialization is a constant:

[0039] Step 2: Order , follow the 4 steps below to find : (a) Calculate the “pseudo residual”:

[0040] (b) Fitting the data using a CART regression tree , get the The leaf node area of ​​a tree ,in .

[0041] (c) Yes calculate:

[0042] (d) Update for:

[0043] Step 3: The regression model corresponding to the training data is .

[0044] S4, analyze the characteristics of the obtained prediction data to determine the importance of different structural parameters of the model.

[0045] In S4, the average impurity reduction method and the Shapley value method were used to calibrate the importance of different structural parameters of the model.

[0046] Example S1, select the density as 1.0 g·cm -3 , simple cubic lattices with 8000, 10000 and 12000 atoms were used as the initial configuration, and the liquid-quenching method was used to obtain three different sizes of amorphous carbon. The cells were expanded N×N×N in the three directions of xyz (1<=N<=9). In order to ensure the convenience and feasibility of subsequent calculations, certain restrictions were imposed on the expansion: on the one hand, the number of atoms must be within 50w atoms, and on the other hand, it must be ensured that the expanded cells are still in blocks rather than sheets or strips (the ratio of the three directions of xyz must not be greater than 3 or less than 1 / 3), and then the initial configuration was obtained; S2, using the above Gaussian random field method to dig holes, Lammps provides Python with model parameters (a, b, c and , , ) and the atomic coordinates ( ), calculate the corresponding value of each atom through Python programming , and The value is returned in Lammps, and deleted in Lammps of atoms; S3, adjust the size of parameters H and ξ to obtain different models. (i.e. H) Theoretically, the pore distribution and size cannot be controlled in a more intuitive way. Therefore, a large number of porous carbon structures are constructed by traversal. The huge configuration space will cover a wider range of distribution and pore size. The porosity is controlled, especially =0, the porosity is 50%. In the present invention, porosity refers to the number of atoms deleted divided by the original number of atoms, that is, when the porosity is 40%, the number of atoms is equal to the density of 2.0 g·cm -3 The amorphous carbon structure is reduced by 40%, and the density of the amorphous porous carbon structure is 1.2 g cm -3 .

[0047] S4, use Zeo++ software to perform model pore size analysis, and ensure that all characterization parameters are consistent (chan_radius=1.7 Å, probe_radius=1.7 Å, num_samples=10000), Figure 3 Shown is a cross section of a typical resulting aperture configuration.

[0048] The weak classifier of the gradient boosting decision tree is composed of multiple decision trees. Each node of the decision tree is a condition about a certain feature, in order to divide the data set into two according to different response variables. By continuously dividing the data set according to the features, the purity of the data can be continuously improved (or information gain), so that classification or regression judgments can be made based on the divided pure data. And each additional division represents the influence of this feature on the model and data purity. Therefore, when training a decision tree, it is possible to calculate how much each feature reduces the impurity of the tree.

[0049] For a gradient boosting decision tree, we can calculate how much impurity is reduced on average for each feature, and use the average reduced impurity as the value of feature importance. Impurity can be used to determine nodes (optimal conditions). For classification problems, Gini impurity or information gain is usually used, and for regression problems, variance or least squares fitting is usually used. The ranking results based on impurity are very clear. The scores of features after the highest-scoring features drop sharply.

[0050] The ability to extract feature importance directly from the training process is an important reason for choosing the gradient boosting decision tree as the machine learning algorithm of the present invention. Figure 4 As shown in Figure 3, this evaluation method shows that the mean pore size significantly affects the predictive ability of the model.

[0051] The SHAP (SHapley Additive exPlanations) value method is a method for explaining the prediction results of the model, which can be used to evaluate the contribution of each feature to the prediction results. This method is based on the concept of Shapley value and is a feature importance evaluation method based on game theory. The SHAP value method can generate a SHAP value for each feature, which indicates the degree of influence of the feature on the prediction result of a certain sample. Features with positive SHAP values ​​will have a positive impact on the prediction, while features with negative values ​​will have a negative impact. The amplitude is a measure of the strength of the effect. SHAP values ​​are model-independent, which means that they can be used to explain any machine learning model, including: linear models, tree models, neural networks, and so on. The advantage of the SHAP value method is that it can take into account the interaction between features and can be interpreted for different types of models. The SHAP value actually considers the marginal contribution rate of each feature and averages all possible results. The marginal contribution rate can be expressed as:

[0052] SHAP values ​​have several useful properties that make them effective for interpreting models while also making the evaluation method more fair: Additivity: SHAP values ​​are additive, meaning that the contribution of each feature to the final prediction can be calculated independently and then summed. This property allows for efficient computation of SHAP values, even for high-dimensional datasets.

[0053] Local Accuracy: SHAP values ​​add up to the difference between the expected model output for a given input and the actual output. This means that SHAP values ​​can provide an accurate local explanation for the model predictions for a given input.

[0054] Robustness: For features where the predictions are missing or irrelevant, the SHAP value is zero. This makes the SHAP value robust to missing data and ensures that irrelevant features do not distort the interpretation.

[0055] Consistency: SHAP values ​​do not change when the model changes, unless the contribution of the features changes. This means that SHAP values ​​can provide a consistent explanation of the model's behavior even if the model architecture or parameters change.

[0056] Therefore, the SHAP value can be expressed as: .

[0057] like Figure 5 As shown in Figure 3, this evaluation method also shows that the mean pore size significantly affects the predictive ability of the model.

[0058] Based on the constructed model and the existing mechanical properties data of molecular dynamics tests, the structural parameters of amorphous porous carbon materials are used as features for model training to train the machine learning model to predict performance. In terms of feature selection, the main features include the total number of atoms (Num Atoms), the average pore size (Mean Size), the pore size variance (Variance), the number of internal atoms (Internal Atoms), the number of surface atoms (Surface Atoms), the internal atomic ratio (Internal Ratio), the surface atomic ratio (Surface Ratio), the length (Length), width (Width), height (Height), volume (Volumes), etc.

[0059] The structural data and mechanical properties data of 100 amorphous porous carbon materials with a porosity of 50% were randomly divided into 10 parts, 8 of which were used as training sets for model training and participated in the model training throughout the process; the remaining 2 were used as test sets for model training and did not participate in the model training at all, but were used as the main reference data for model evaluation. At the same time, the distribution of the training set and the test set was guaranteed to be basically consistent.

[0060] In the training set, the model was trained using shear modulus as an example, and a GBDT model that can accurately predict mechanical properties was obtained. Figure 6 The model fit and reliability were evaluated.

[0061] Evaluation indicators include correlation coefficient R, determination coefficient R 2 , mean absolute error MAE, mean square error MSE and root mean square error RMSE. The calculation expressions of each indicator are as follows:

[0062]

[0063]

[0064] in, Indicates response, represents the predicted value, represents the mean of the response values, and N represents the size of the data set. If the values ​​of RMSE and MAE are close to zero, R or R 2 The closer it is to 1, the higher the degree of model fit, and the better the reliability and predictive ability of the model.

[0065] The training model of the present invention performed R=0.968, R2=0.931 on the test set, and the mean square error was also kept below 1 GPa, indicating that the model has good generalization ability.

[0066] The Pearson correlation coefficient PCC measures the linear correlation between structural parameters and mechanical properties. The PCC value ranges from -1 to 1. When the Pearson correlation coefficient is greater than 0.9 or less than -0.9, it means that the correlation between the characteristic structures is too high and one of them should be eliminated. When applied to samples, PCC is usually represented by rxy, which can be called the sample Pearson correlation coefficient. Given data consisting of n pairs (x1, y1), ..., (xn, yn), rxy is defined as:

[0067] in n is the sample size, , So i For each sample point of the index, is the mean of variable x, and similarly for y.

[0068] The results are as follows Figure 7 Obviously, the Pearson correlation coefficient in the model rarely exceeds the range, which shows that the features are independent and can well express their impact on the model.

[0069] The modeling method of the present invention is simple and efficient, easy to understand and master, and can change the characteristics of the model structure by simply modifying the program parameters, so as to achieve a wide range of adjustment of the pore parameters of the porous structure and realize strong controllability; Experiments have shown that the three-dimensional model of the porous carbon skeleton in the silicon-carbon negative electrode constructed by the present invention is closer to the actual object, which improves the accuracy of subsequent analysis to a certain extent and can accurately describe the mechanical properties of the actual object.

[0070] When the present invention conducts mechanical property testing on a three-dimensional model, a model is built based on existing structure and performance data through a machine learning algorithm, which can quickly and accurately predict material properties, effectively saving manpower, time and cost.

[0071] A system for predicting the performance of a porous carbon skeleton in a silicon-carbon negative electrode comprises a model building module, a model training module, a model prediction module and a model post-processing module, wherein the model building module is used to establish a bicontinuous nanoporous structure model of a porous carbon skeleton in a silicon-carbon negative electrode based on a Gaussian random field; the model training module is used to perform model training using the structural information of a porous carbon material as a feature vector for machine learning training and the mechanical properties of the porous carbon material as a target value to obtain a target model; the model prediction module is used to perform gradient boosting decision tree model training on the obtained target model to obtain a model performance prediction; and the model post-processing module is used to perform feature analysis on the obtained prediction data to determine the importance of different structural parameters of the model.

[0072] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and wherein the processor implements the steps of the modeling method when executing the computer program.

[0073] A storage medium stores a computer program, which implements the steps of any one of the modeling methods when executed by a processor.

[0074] The above description is only a preferred embodiment of the present invention and is not intended to impose any limitation on the technical solution of the present invention. Those skilled in the art should understand that, without departing from the spirit and principles of the present invention, the technical solution can also be subjected to several simple modifications and substitutions, and these modifications and substitutions are also within the scope of protection covered by the claims.

Claims

1. A method for predicting the performance of porous carbon skeleton in silicon-carbon negative electrode, characterized in that: The following steps are involved: S1, a bicontinuous nanoporous structure model of porous carbon skeleton in silicon-carbon anode is established based on Gaussian random field; S2, using the structural information of the porous carbon material as the feature vector for machine learning training and the mechanical properties of the porous carbon material as the target value for model training to obtain the target model; S3, training a gradient boosting decision tree model for the target model obtained in S2 to obtain a model performance prediction; S4, perform feature analysis on the obtained prediction data to determine the importance of different structural parameters of the model.

2. The method for predicting the performance of porous carbon skeleton in silicon-carbon negative electrode according to claim 1, characterized in that: In S1, the process of establishing the bicontinuous nanoporous structure model of the porous carbon skeleton in the silicon-carbon anode includes: S11, Gaussian random field generated by superposition of sinusoidal waves with fixed wavelength and amplitude but random direction and phase; S12, a bicontinuous porous structure with periodic boundary conditions is constructed in a cuboid space in a Gaussian random field.

3. The method for predicting the performance of porous carbon skeleton in silicon-carbon negative electrode according to claim 2, characterized in that: In S12, the expression of Gaussian random field is: in, is the position vector, represents the wave number in the truncated series, is the normalization factor, and Respectively represent The direction and phase of each wave; is a set of uniformly distributed Random phase within range; Must meet , is a constant; represents the surface of a solid, The region is defined as the solid phase, and The area represents the hole; Used to express the porosity of bicontinuous porous structures. The porosity is the number of deleted atoms divided by the original number of atoms.

4. The method for predicting the performance of porous carbon skeleton in silicon-carbon negative electrode according to claim 2, characterized in that: In S12, Must meet: ， a, b, c are the edge vectors of the model, and m, n, p are integers.

5. The method for predicting the performance of porous carbon skeleton in silicon-carbon negative electrode according to claim 1, characterized in that: In S2, during training, the loss function is the root mean square error.

6. The method for predicting the performance of porous carbon skeleton in silicon-carbon negative electrode according to claim 1, characterized in that: In S4, the average impurity reduction method and the Shapley value method were used to calibrate the importance of different structural parameters of the model.

7. The method for predicting the performance of porous carbon skeleton in silicon-carbon negative electrode according to claim 1, characterized in that: It also includes model evaluation and reliability check. The evaluation indicators include correlation coefficient R, determination coefficient R2, mean absolute error MAE, mean square error MSE and root mean square error RMSE. The Pearson correlation coefficient PCC is used to measure the linear correlation between structural parameters and mechanical properties.

8. A system for predicting the performance of porous carbon skeletons in silicon-carbon negative electrodes, characterized in that: The method comprises a model building module, a model training module, a model prediction module and a model post-processing module. The model building module is used to establish a dual-continuous nanoporous structure model of a porous carbon skeleton in a silicon-carbon negative electrode based on a Gaussian random field; the model training module is used to train a model using the structural information of the porous carbon material as a feature vector for machine learning training and the mechanical properties of the porous carbon material as a target value to obtain a target model; the model prediction module is used to train a gradient boosting decision tree model for the obtained target model to obtain a model performance prediction; the model post-processing module is used to analyze the characteristics of the obtained prediction data to determine the importance of different structural parameters of the model.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the modeling method according to any one of claims 1 to 7 are implemented.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the modeling method according to any one of claims 1 to 7 are implemented.

Images