A Battery Material Screening and Performance Prediction Method Based on a Dual-Driven Deep Learning Model

Abstract

This invention provides a method for battery material screening and performance prediction based on a dual-drive deep learning model, belonging to the interdisciplinary fields of battery material genome engineering, artificial intelligence, and materials science. The method includes the following steps: 1) Preprocessing the collected data, including data cleaning and normalization; 2) Extracting physicochemical descriptors of the materials and using a Filter-Wrapper-Embedded cascaded feature screening framework to identify key feature subsets; 3) Constructing a dual-drive deep learning model combining graph neural networks and key physicochemical descriptors to predict the specific capacity, average voltage, and volume change of the materials; 4) Designing a multi-objective collaborative evaluation system that balances high energy density and mechanical stability for comprehensive scoring; 5) Finally, selecting materials based on the scores and performing DFT verification. This invention effectively solves the problems of lost characterization information and difficulty in balancing high capacity and stability in traditional screening methods, achieving efficient and customized screening of battery materials.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary fields of battery material genome engineering, artificial intelligence and materials science, and relates to a method for screening and predicting the performance of battery materials based on a dual-drive deep learning model. Background Technology

[0002] With the rapid development of new energy vehicles and large-scale energy storage technologies, the development of novel battery electrode materials that combine high energy density and long cycle life has become an urgent industry need. Traditional materials research and development mainly relies on trial-and-error experiments or density functional theory (DFT) calculations. This approach is not only time-consuming and costly, but also difficult to efficiently traverse the vast chemical space to find the optimal solution. In recent years, data-driven machine learning methods have become a powerful tool for accelerating materials discovery because they can quickly extract structure-property relationships from massive amounts of data.

[0003] However, existing machine learning-based battery material screening methods still face many challenges in practical applications. First, the feature representation methods are relatively simplistic. Existing models often extract features solely based on the chemical composition of the material (ignoring crystal topological information) or only utilize graph neural networks to extract structural features (lacking clear physicochemical interpretability). This makes it difficult for the models to comprehensively capture the structure-property relationships of the materials, limiting prediction accuracy and generalization ability. Second, the data quality in publicly available material databases varies greatly, especially volume change rate data, which often contains extreme outliers caused by conversion reactions or calculation errors. These constitute serious noise interference for the screening of electrode materials, which are mainly based on intercalation reactions. Without targeted processing, this will greatly mislead model training. Finally, existing screening strategies often suffer from one-sided evaluation metrics. Most focus only on screening materials with high specific capacity, ignoring the inherent contradiction of "high capacity accompanied by large volume strain" commonly found in lithium-rich or high-nickel materials. This leads to the selected high-capacity materials experiencing severe volume expansion during actual charge and discharge, causing particle breakage and electrical contact failure, failing to meet the practical application requirements for long cycle life. Therefore, there is an urgent need for a battery material screening method that can deeply integrate structural and compositional information, effectively eliminate data noise, and take into account both high energy density and mechanical stability. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a battery material screening and performance prediction method based on a dual-drive deep learning model, so as to solve the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for battery material screening and performance prediction based on a dual-drive deep learning model, the method specifically includes the following steps: S1. Data collection and preprocessing: Crystal structure data and electrochemical performance data of battery electrode materials are collected from the material database. The collected data are preprocessed, including data cleaning, outlier removal and data normalization. S2. Cascaded Feature Screening: Extract explicit physicochemical descriptors of materials, construct a cascaded feature screening framework based on filter-wrapper-embedded, reduce and optimize the extracted physicochemical descriptors, eliminate redundant features, and lock in the key feature subset that affects electrochemical performance. S3. Dual-drive model construction and prediction: Construct a dual-drive deep learning model, which includes a first drive branch and a second drive branch; the first drive branch uses a graph neural network CGCNN to extract the topological features of the material's crystal structure, and the second drive branch uses the Catboost algorithm to process the key feature subset selected in step S2; the graph neural network and the Catboost algorithm are used together to predict the specific capacity, average voltage and volume change rate of the candidate material. S4. Multi-objective collaborative evaluation: Design a multi-objective collaborative evaluation system that balances high energy density and mechanical stability, and use a weighted scoring function to calculate the comprehensive performance score of each candidate material; S5. Based on the comprehensive performance score obtained in step S4, the candidate materials are sorted and screened, and the preferred materials are selected for density functional theory (DFT) verification to complete the battery material screening.

[0006] Furthermore, in step S1, the collected data is preprocessed to remove extreme values ​​of the volume change data. Specifically, considering that the volume change rate of embedded electrode materials is usually low, and the extremely high volume change data in the database are mostly noise caused by conversion reactions or calculation errors, a statistical threshold method is used to remove the largest 2% of the volume change data, retaining the remaining 98% of the data as an effective training set to improve the model's prediction accuracy of the volume stability of embedded electrode materials.

[0007] Furthermore, in step S2, the specific execution flow of the cascaded feature filtering framework is as follows: S21. Filtering stage: Calculate the Pearson correlation coefficient between all physicochemical descriptors and the target performance, sort them according to correlation, and retain the top 100 features with the highest correlation. S22. Wrapper stage: The recursive feature elimination (RFE) method is used to iteratively eliminate the features retained in the filtering stage, eliminating the 50% of features with the lowest contribution and retaining the remaining features. S23. Embedded stage: Based on the feature importance score of the tree model, the remaining features in the packaging stage are finally optimized, and the optimal feature set is retained as the key feature subset.

[0008] Furthermore, in step S3, the specific construction and prediction methods of the dual-drive model are as follows: S31, Structural Feature Extraction Branch: CGCNN is used to read the atomic coordinates and lattice constants of the material, multi-layer graph convolution operations are used to capture the local chemical environment between atoms, and a fixed-length crystal structure feature vector is output through a pooling layer; S32, Physical Feature Mapping Branch: Input the key physicochemical feature subset locked in step S2 into the fully connected layer and map it into a high-dimensional physicochemical feature vector; S33, Joint Prediction Module: A feature fusion layer is set up to concatenate the crystal structure feature vector with the physicochemical feature vector to form a dual-drive hybrid feature vector; S34. Input the hybrid feature vector into a multilayer perceptron (MLP) and output the final prediction result through nonlinear transformation.

[0009] Furthermore, in step S4, the multi-objective collaborative evaluation system adopts the following weighted scoring function: calculating the comprehensive performance distance of the material. and overall score :

[0010]

[0011] in, , , These are the normalized predicted specific capacity, predicted average voltage, and volume stability index, respectively. , , These are the corresponding weighting coefficients.

[0012] The present invention also provides a battery material screening and performance prediction system based on a dual-drive deep learning model, which employs the method described above.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described above.

[0014] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0015] The beneficial effects of this invention are as follows: 1. Targeted noise reduction: An innovative strategy is proposed to remove the top 2% extreme values ​​of volume change data, which effectively eliminates the interference of non-embedded reaction data and significantly improves the model's robustness in predicting volume changes of embedded electrode materials.

[0016] 2. Precise Feature Mining: By constructing a three-level cascaded screening framework of Filter-Wrapper-Embedded, the most physically meaningful key features are accurately identified from massive descriptors, which not only significantly reduces the model dimensionality but also reveals the contribution mechanism of specific elements or properties to electrochemical performance.

[0017] 3. Dual-drive high-precision prediction: The dual-drive joint prediction architecture of "CGCNN structural features + physicochemical features" is adopted. Through feature splicing and joint MLP deep fusion, the problem of information loss in a single model is solved, and the advantages of micro-crystal structure and macro-physical properties are complemented.

[0018] 4. Multi-objective balance: The established multi-objective collaborative evaluation system forcibly introduces volume stability as a screening constraint, which overcomes the limitation of traditional screening that only focuses on high capacity. It can effectively screen out excellent materials that have both high energy density and low volume strain, providing an efficient and reliable screening path for the development of long-life battery materials.

[0019] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 An overall architecture diagram of a preferred embodiment of the present invention is provided; Figure 2 This is a comparison chart showing the implementation effects of the cascaded feature filtering framework in this invention; Figure 3 The diagram shows the training process and final prediction results of the dual-drive model of this invention. Figure 4 This is a comparison chart of the actual and predicted values ​​of the dual-drive model of this invention on the full test sample sequence; Figure 5 This is a histogram showing the distribution of predicted comprehensive scores for candidate materials under the multi-objective collaborative evaluation system of this invention. Detailed Implementation

[0021] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0022] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0023] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0024] This invention relates to a method for battery material screening and performance prediction based on a dual-drive deep learning model, which utilizes machine learning techniques for feature mining and performance prediction.

[0025] In this embodiment, the battery material screening and performance prediction method based on a dual-drive deep learning model provided by the present invention includes the following steps: 1. Data Collection and Specific Preprocessing. Crystal structure data (CIF files) and electrochemical performance data (specific capacity, average voltage, volume change) of battery electrode materials were collected from materials databases such as Materials Project. Preprocessing of the collected data was necessary, including data cleaning, outlier removal, and data normalization. Specifically, quantile-based extreme value removal was performed on the volume change data, eliminating the top 2% of values ​​to remove noise from non-intercalation reactions or calculation errors, ensuring data quality and consistency. 2. Cascaded Feature Screening. Explicit physicochemical descriptors of the material (such as atomic radius, electronegativity, etc.) are extracted. A three-level cascaded feature screening framework based on Filter-Wrapper-Embedded is constructed, sequentially using Pearson correlation screening, recursive feature elimination (RFE) simplification, and tree model importance scoring for confirmation, to identify the optimal subset of key features affecting electrochemical performance from the original descriptors. 3. Dual-Driven Model Construction and Prediction. A dual-driven deep learning model based on feature fusion is constructed. The first drive uses a graph neural network (CGCNN) to extract topological features from the crystal structure, while the second drive uses a neural network to map and process the selected key physicochemical features. The two sets of features are concatenated and fused, and then input into a joint multilayer perceptron (MLP) to jointly predict the specific capacity, average voltage, and volume change rate of candidate materials. 4. Multi-objective collaborative evaluation and verification. A multi-objective collaborative evaluation system balancing high energy density and mechanical stability is designed, and a weighted scoring function is used to calculate the comprehensive performance score of each candidate material. Finally, materials are selected based on the scores, and DFT verification is performed.

[0026] Figure 1 An overall architecture diagram of a preferred embodiment of the present invention is provided, such as... Figure 1 The diagram illustrates the processing flow of this invention. First, data collection and preprocessing are performed, extracting crystal structure data and physicochemical descriptors from the database. Second, cascaded feature selection is conducted, extracting the optimal feature set through three stages: filtering, wrapping, and embedding. Third, a dual-drive model is trained and predicted, inputting the crystal structure and the optimal feature set into a graph neural network (CGCNN) and a traditional machine learning (ML) model, respectively. The two branches are then fused and fed into a multilayer perceptron (MLP) to jointly predict the material's specific capacity, average voltage, and volume change. Following this is a multi-objective collaborative evaluation, scoring the material's overall performance. Finally, the optimal material is selected through DFT verification.

[0027] Figure 2 This figure shows a comparison of the implementation effects of the cascaded feature selection framework in this invention. Taking the prediction of average voltage as an example, the figure illustrates the difference in the distribution of the model's coefficient of determination when using "all features" versus using the selected "top 100 features." It can be seen that after feature selection, the median R² score of the model prediction is improved, and the interquartile range of the box plot is narrower, indicating a more concentrated overall distribution and fewer outliers. This proves that the feature selection framework of this invention not only effectively eliminates redundant noise but also significantly improves the accuracy and stability of model prediction.

[0028] Figure 3This figure illustrates the training process and final prediction results of the dual-drive model of this invention. The figure consists of three sub-figures: the left figure shows the training loss curve, illustrating the error reduction of the CGCNN branch, CatBoost branch, and fused MLP during the iteration process, all of which converge rapidly and tend to stabilize; the middle figure shows the ensemble weights learned by the model, intuitively reflecting the relative contributions of structural features and physicochemical features in joint prediction; the right figure is a scatter plot of the final regression prediction results on the test set (taking voltage prediction as an example), with the horizontal axis representing the true value and the vertical axis representing the predicted value. Most data points are closely distributed near the diagonal, and the coefficient of determination (R²) reaches 0.758, indicating that the dual-drive model combining microstructure and macroscopic properties has good accuracy in predicting battery material performance.

[0029] Figure 4 This figure shows a comparison between the actual and predicted values ​​of the dual-drive model of this invention on the full test sample sequence. As shown in the figure, the horizontal axis represents the index of the test sample, and the vertical axis represents the number of targets. The blue curve represents the actual calculation result, and the orange curve represents the model's prediction result. It can be seen that within a wide range covering thousands of samples, the orange curve representing the predicted value and the blue curve representing the actual value highly overlap. Especially when facing extreme samples with very few values ​​exhibiting huge fluctuations, the model can still closely track and make accurate predictions. This further intuitively verifies that the dual-drive model constructed in this invention has excellent generalization ability and extremely high prediction stability when processing large-scale battery material data.

[0030] Figure 5 This is a histogram showing the distribution of predicted comprehensive scores for candidate materials under the multi-objective collaborative evaluation system of this invention. As shown in the figure, the horizontal axis represents the predicted comprehensive score, and the vertical axis represents the frequency of materials in each score interval. It can be seen that the comprehensive scores of the massive number of candidate materials exhibit an approximately normal distribution, with most materials scoring between 0.4 and 0.6, while the number of materials in the high-score region is sparse. This indicates that the scoring system designed in this invention has excellent discriminative power and can effectively and quickly identify a very small number of high-quality electrode materials with both high energy density and low volume change potential from a vast material library.

[0031] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0032] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0033] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for battery material screening and performance prediction based on a dual-drive deep learning model, characterized in that, The method specifically includes the following steps: S1. Data collection and preprocessing: Crystal structure data and electrochemical performance data of battery electrode materials are collected from the material database. The collected data are preprocessed, including data cleaning, outlier removal and data normalization. S2. Cascaded Feature Screening: Extract explicit physicochemical descriptors of materials, construct a cascaded feature screening framework based on filtering-packaging-embedding, reduce the dimensionality and optimize the extracted physicochemical descriptors, eliminate redundant features, and lock in the key feature subset that affects electrochemical performance. S3. Dual-Driven Model Construction and Prediction: A dual-drive deep learning model is constructed, which includes a first drive branch and a second drive branch. The first drive branch uses a graph neural network (CGCNN) to extract the topological features of the material's crystal structure, and the second drive branch uses the Catboost algorithm to process the key feature subset selected in step S2. The graph neural network and the Catboost algorithm are used together to predict the specific capacity, average voltage, and volume change rate of the candidate material. S4. Multi-objective collaborative evaluation: Design a multi-objective collaborative evaluation system that balances high energy density and mechanical stability, and use a weighted scoring function to calculate the comprehensive performance score of each candidate material; S5. Based on the comprehensive performance score obtained in step S4, the candidate materials are sorted and screened, and the preferred materials are selected for density functional theory verification to complete the battery material screening.

2. The battery material screening and performance prediction method based on a dual-drive deep learning model according to claim 1, characterized in that, In step S1, the collected data is preprocessed to remove extreme values ​​from the volume change data. Specifically, a statistical threshold method is used to remove the 2% of data with the largest values ​​in the volume change data, and the remaining 98% of the data is retained as an effective training set to improve the model's prediction accuracy for the volume stability of embedded electrode materials.

3. The battery material screening and performance prediction method based on a dual-drive deep learning model according to claim 2, characterized in that, In step S2, the specific execution flow of the cascaded feature filtering framework is as follows: S21. Filtering stage: Calculate the Pearson correlation coefficient between all physicochemical descriptors and target performance, sort them according to correlation, and retain the top 100 features with the highest correlation. S22. Packaging stage: The recursive feature elimination method is used to iteratively remove the features retained in the filtering stage, eliminating the 50% of features with the lowest contribution and retaining the remaining features. S23. Embedding stage: Based on the feature importance score of the tree model, the remaining features in the packaging stage are finally optimized, and the optimal feature set is retained as the key feature subset.

4. The battery material screening and performance prediction method based on a dual-drive deep learning model according to claim 3, characterized in that, In step S3, the specific construction and prediction methods of the dual-drive model are as follows: S31, Structural Feature Extraction Branch: CGCNN is used to read the atomic coordinates and lattice constants of the material, multi-layer graph convolution operations are used to capture the local chemical environment between atoms, and a fixed-length crystal structure feature vector is output through a pooling layer; S32, Physical Feature Mapping Branch: Input the key physicochemical feature subset locked in step S2 into the fully connected layer and map it into a high-dimensional physicochemical feature vector; S33, Joint Prediction Module: A feature fusion layer is set up to concatenate the crystal structure feature vector with the physicochemical feature vector to form a dual-drive hybrid feature vector; S34. Input the hybrid feature vector into a multilayer perceptron and output the final prediction result through nonlinear transformation.

5. The battery material screening and performance prediction method based on a dual-drive deep learning model according to claim 4, characterized in that, In step S4, the multi-objective collaborative evaluation system uses the following weighted scoring function: calculating the comprehensive performance distance of the material. and overall score : in, , , These are the normalized predicted specific capacity, predicted average voltage, and volume stability index, respectively. , , These are the corresponding weighting coefficients.

6. A battery material screening and performance prediction system based on a dual-drive deep learning model, characterized in that, The system employs the method as described in any one of claims 1 to 5.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 5.

Images