Lithium battery safety performance prediction and attribution system based on big data analysis and GPT model
The lithium battery safety performance prediction and attribution system, which utilizes big data analytics and the GPT model, solves the problems of data silos and subjective attribution in lithium battery safety assessment. It achieves high-precision safety performance prediction and accurate attribution analysis, thereby improving system reliability and user experience.
Patent Information
- Application Number
- CN202510937421.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-08
- Publication Date
- 2025-10-28
AI Technical Summary
Existing lithium battery safety performance assessments suffer from problems such as data silos, strong subjective attribution, insufficient prediction accuracy, and delayed response, resulting in a high assessment error rate.
A lithium battery safety performance prediction and attribution system based on big data analysis and GPT model is adopted, including data acquisition, data preprocessing, safety performance prediction, attribution analysis and human-computer interaction modules. Through multi-model fusion and rich feature extraction, combined with GPT text analysis and expert knowledge fusion, the system can achieve accurate assessment of lithium battery safety performance.
It improves the accuracy of lithium battery safety performance prediction and attribution results, enhances system reliability and user experience, provides a visual interface and risk warning function, and achieves more efficient and accurate safety performance assessment.
Smart Images
Figure CN120850239A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of lithium battery technology, specifically relating to a lithium battery safety performance prediction and attribution system based on big data analysis and GPT model. Background Technology
[0002] Lithium-ion batteries are widely used in new energy vehicles and portable electronic devices. Their safety performance assessment mainly relies on the limited testing methods available on the market or on human experience.
[0003] The testing methods can be broadly categorized into two types: non-destructive testing (NDT) and destructive testing. NDT techniques include:
[0004] X-ray imaging utilizes X-rays to penetrate the interior of a battery and generates two-dimensional (DR) or three-dimensional (CT) images by receiving attenuated signals through a detector; neutron scattering imaging, which is highly sensitive to lithium, is used to observe dynamic processes such as lithium-ion migration and electrolyte decomposition during charging and discharging in real time; ultrasonic detection utilizes the phenomenon that high-frequency ultrasound waves generate reflected waves when they encounter defects (such as bubbles or delamination) inside the battery; and machine vision inspection, etc.
[0005] Destructive testing involves abuse tests such as overcharging, over-discharging, heating, squeezing, needle puncture, and short circuits, which are not very applicable.
[0006] For a long time, large-scale lithium battery safety performance assessments have been plagued by problems such as data silos, strong subjectivity in attribution, insufficient prediction accuracy, and delayed response. Specifically, the lithium battery production data, usage data, and fault data that need to be tested are stored in a scattered manner, lacking cross-domain correlation analysis capabilities. Furthermore, fault analysis based on human experience lacks a quantitative chain of evidence, and the attribution results are easily affected by human factors, ultimately leading to a high trend in assessment error rate. Summary of the Invention
[0007] To address the problems mentioned in the background art, one aspect of the present invention provides a lithium battery safety performance prediction and attribution system based on big data analysis and GPT model, including a data acquisition module, a data preprocessing module, a safety performance prediction module, an attribution analysis module, and a human-computer interaction module.
[0008] The data acquisition module is used to acquire lithium battery production data from the MES system, usage data from the T-BOX terminal, and fault description text from the fault database.
[0009] The data preprocessing module includes a data cleaning unit, a standardization unit, and a feature extraction unit;
[0010] The data cleaning unit is used to perform outlier removal and missing value imputation.
[0011] The standardization unit is used to normalize the production data, usage data, and fault description text.
[0012] The feature extraction unit is used to extract temporal features, statistical features, and image features;
[0013] The safety performance prediction module consists of multiple learning models and model fusion units, and is used to output prediction results of thermal runaway risk probability and bulging probability.
[0014] The attribution analysis module is used to establish a mapping relationship between the risk probability output by the prediction module and production data and usage data through multi-source data association, including a GPT text analysis unit, an association analysis unit and an expert knowledge fusion unit.
[0015] The GPT text analysis unit parses the fault description text based on the fine-tuned GPT model and generates a preliminary attribution report.
[0016] The correlation analysis unit uses the Apriori algorithm to mine correlation rules between production-usage-failure data;
[0017] The expert knowledge fusion unit is used to correct the attribution results through the engine;
[0018] The human-computer interaction module provides a visual interface that supports querying prediction results, interactive correction of attribution reports, and push of safety recommendations.
[0019] Preferably, the data acquisition module synchronizes data with the MES system, T-BOX terminal and fault database in real time or at regular intervals via API interface.
[0020] Preferably, the timing features in the feature extraction unit are extracted from the charge-discharge curve data, and the timing features include: capacity change rate, internal resistance change rate, and voltage fluctuation rate.
[0021] Preferably, the image features in the feature extraction unit are extracted from the lithium battery X-ray inspection image, and the image features include texture features or defect area features, specifically visual features such as cracks, bulges, and leakage.
[0022] Preferably, the model fusion unit adopts a stacking strategy, using the output of the machine learning model as the input features of the deep learning model, and finally outputs the fused prediction result through the logistic regression layer.
[0023] Preferably, the GPT text analysis unit implements the attribution report through the following steps:
[0024] 1) Input the fault description text into the fine-tuned GPT model and output a list of candidate fault causes;
[0025] 2) Based on the association rules output by the association analysis unit, filter candidate reasons that are strongly correlated with production / usage data;
[0026] 3) Generate the final attribution report after verification by the expert knowledge base.
[0027] Preferably, the expert knowledge fusion unit has a built-in editable rule base that supports dynamic adjustment of attribution logic based on input from domain experts.
[0028] Preferably, the human-computer interaction module also provides a risk warning function, which triggers multi-level alarm notifications when the prediction result exceeds a preset threshold.
[0029] On the other hand, a method for predicting the safety performance of lithium batteries based on the attribution system is provided, characterized by the following steps:
[0030] S1 collects production data, usage data, and fault description text;
[0031] S2 performs data cleaning, standardization, and time series / image feature extraction;
[0032] S3 predicts safety performance indicators using a hybrid model;
[0033] S4 uses the GPT model to generate a preliminary attribution report, and after correction by association analysis, outputs the final attribution results.
[0034] S5 pushes the prediction results and attribution reports to the user interface and records user feedback for model iteration.
[0035] Compared with the prior art, the beneficial effects of the present invention are:
[0036] This invention integrates production, usage, and fault data, breaking down data silos and improving data quality. Simultaneously, it significantly enhances the accuracy of safety performance prediction by utilizing multi-model fusion and rich feature extraction. Its attribution analysis module combines GPT text analysis, data association rule mining, and expert knowledge fusion to improve the accuracy and objectivity of attribution results. Furthermore, the system's human-computer interaction module provides a visual interface, risk warning functions, and a user feedback mechanism, not only improving user experience and system usability but also further enhancing the reliability of prediction and attribution through model iteration and optimization. In short, this system effectively solves the problems of data dispersion, strong subjectivity in attribution, and insufficient prediction accuracy in traditional lithium battery safety assessments, providing a more efficient, accurate, and intelligent technical solution for lithium battery safety performance assessment. Attached Figure Description
[0037] Figure 1 This is a system framework diagram of the prediction and attribution system of the present invention;
[0038] Figure 2 This is a flowchart of the system-based process in this invention.
[0039] Label in the diagram: Data acquisition module 1;
[0040] Data preprocessing module 2, data cleaning unit 201, standardization unit 202, feature extraction unit 203;
[0041] Safety performance prediction module 3;
[0042] Attribution analysis module 4, GPT text analysis unit 401, association analysis unit 402, expert knowledge fusion unit 403;
[0043] Human-computer interaction module 5. Detailed Implementation
[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0045] Example 1
[0046] like Figures 1 to 2 The lithium battery safety performance prediction and attribution system based on big data analysis and GPT model shown includes a data acquisition module 1, a data preprocessing module 2, a safety performance prediction module 3, an attribution analysis module 4, and a human-computer interaction module 5.
[0047] Data acquisition module 1 is used to acquire lithium battery production data from the MES system, usage data from the T-BOX terminal, and fault description text from the fault database;
[0048] Data preprocessing module 2 includes a data cleaning unit 201, a standardization unit 202, and a feature extraction unit 203;
[0049] Data cleaning unit 201 is used to perform outlier removal and missing value imputation;
[0050] Standardization unit 202 is used to normalize production data, usage data, and fault description text;
[0051] Feature extraction unit 203 is used to extract temporal features, statistical features and image features;
[0052] Safety performance prediction module 3 consists of multiple learning models and model fusion units, used to output prediction results of thermal runaway risk probability and bulging probability;
[0053] Attribution analysis module 4 is used to establish a mapping relationship between the risk probability output by the prediction module and production data and usage data through multi-source data association, including GPT text analysis unit 401, association analysis unit 402 and expert knowledge fusion unit 403.
[0054] GPT text analysis unit 401 parses the fault description text based on the fine-tuned GPT model and generates a preliminary attribution report;
[0055] Association analysis unit 402 uses the Apriori algorithm to mine association rules between production, usage, and failure data;
[0056] Expert knowledge fusion unit 403 is used to correct attribution results through the engine;
[0057] The human-computer interaction module 5 provides a visual interface that supports querying prediction results, interactive correction of attribution reports, and push of safety recommendations.
[0058] In simple terms, data acquisition module 1 provides the raw data foundation for the system; data preprocessing module 2 cleans, standardizes, and extracts features from the data, providing high-quality input for subsequent modules; safety performance prediction module 3 performs risk prediction based on feature data; attribution analysis module 4 conducts in-depth analysis of the prediction results to identify the causes; and human-computer interaction module 5 presents the prediction and attribution results to the user and supports user interaction and feedback. All modules of the entire system work closely together to realize the prediction and attribution functions of lithium battery safety performance.
[0059] Example 2
[0060] like Figures 1 to 2 The system shown is a lithium battery safety performance prediction and attribution system based on big data analysis and the GPT model. It includes a data acquisition module 1, a data preprocessing module 2, a safety performance prediction module 3, an attribution analysis module 4, and a human-computer interaction module 5. The data acquisition module 1 provides the system with raw data; the data preprocessing module 2 cleans, standardizes, and extracts features from the data, providing high-quality input for subsequent modules; the safety performance prediction module 3 performs risk prediction based on feature data; the attribution analysis module 4 conducts in-depth analysis of the prediction results to identify the causes; and the human-computer interaction module 5 presents the prediction and attribution results to the user and supports user interaction and feedback. All modules of the system work closely together to achieve the prediction and attribution functions of lithium battery safety performance.
[0061] Specifically, the data acquisition module 1 is used to acquire lithium battery production data from the MES system, usage data from the T-BOX terminal, and fault description text from the fault database; it transmits the acquired production data, usage data, and fault description text to the data preprocessing module 2; and the data acquisition module 1 performs real-time or timed data synchronization with the MES system, T-BOX terminal, and fault database through the API interface. During data transmission and synchronization, the module will automatically detect the data update status according to preset rules and protocols, and efficiently transmit the newly added or changed data to the data preprocessing module 2.
[0062] During the data acquisition process, the production data provided by the MES system includes, but is not limited to, battery production process parameters, raw material batch information, production equipment operating status, production environment data (such as temperature and humidity), and quality inspection data during the production process, providing basic information for subsequent safety performance analysis.
[0063] The usage data collected by the T-BOX terminal covers the operating status of lithium batteries in actual applications, including but not limited to the battery's charging and discharging current, voltage, temperature, SOC (state of charge) change curve, environmental parameters (such as vehicle speed, road conditions, etc.), and user usage habits. This data is used to help the system understand the battery's performance and potential risks under actual working conditions.
[0064] The fault description text in the fault database is used to record various fault phenomena, the time and place of the fault, the fault handling measures, and the experience summary of relevant technical personnel that occur during the production, testing and use of lithium batteries. The above data is used to provide a reference for the system's attribution analysis module 4, so as to realize the rapid location of the fault cause.
[0065] After completing the collection and synchronization of the above data, the data acquisition module 1 performs preliminary formatting on the collected production data, usage data, and fault description text to ensure that the data format is uniform and meets the input requirements of the data preprocessing module 2. After the data is transmitted to the data preprocessing module 2, it enters the next processing step.
[0066] Example 3
[0067] like Figures 1 to 2The system shown is a lithium battery safety performance prediction and attribution system based on big data analysis and the GPT model. It includes a data acquisition module 1, a data preprocessing module 2, a safety performance prediction module 3, an attribution analysis module 4, and a human-computer interaction module 5. The data acquisition module 1 provides the system with raw data; the data preprocessing module 2 cleans, standardizes, and extracts features from the data, providing high-quality input for subsequent modules; the safety performance prediction module 3 performs risk prediction based on feature data; the attribution analysis module 4 conducts in-depth analysis of the prediction results to identify the causes; and the human-computer interaction module 5 presents the prediction and attribution results to the user and supports user interaction and feedback. All modules of the system work closely together to achieve the prediction and attribution functions of lithium battery safety performance.
[0068] The data preprocessing module 2 includes a data cleaning unit 201, a standardization unit 202, and a feature extraction unit 203. The data cleaning unit 201 performs outlier removal and missing value imputation on the collected data to remove noise and incomplete parts, and then passes the cleaned data to the standardization unit 202. The standardization unit 202 receives the data and normalizes the production data, usage data, and fault description text. The feature extraction unit 203 extracts time-series features, statistical features, and image features from the standardized data, and then transmits the extracted feature data to the safety performance prediction module 3 and the attribution analysis module 4 to provide important basis for subsequent prediction and attribution analysis.
[0069] Furthermore, the extraction of time-series features can help the system capture the dynamic changes of the battery during operation. The time-series feature extraction in the feature extraction unit 203 is based on key data in the charging and discharging process of the lithium battery. The key data is usually presented in the form of charging and discharging curves, which record the changes in parameters such as voltage, current, and capacity of the battery at different time points, specifically including: capacity change rate, internal resistance change rate, and voltage fluctuation rate.
[0070] Capacity change rate, calculated by analyzing the capacity changes of a battery over multiple charge-discharge cycles, primarily reflects the aging degree and health status of the battery. A higher capacity change rate indicates a greater risk of battery performance degradation. Internal resistance change rate, monitored by observing the rate of change of internal resistance over time, helps detect potential abnormalities within the battery. A higher internal resistance change rate suggests a greater likelihood of performance degradation. Voltage fluctuation rate reflects the stability of the battery voltage during charge and discharge. Greater voltage fluctuations indicate potential uneven chemical reactions or localized short circuits within the battery, ultimately increasing the safety risk.
[0071] Furthermore, the image features in the feature extraction unit 203 are extracted from the lithium battery X-ray inspection image to reflect the internal condition of the battery. First, the internal material structure and surface condition of the battery will exhibit different texture features. By comparing and analyzing the texture in the X-ray image, it can be determined whether there are problems such as uneven material distribution, electrode coating peeling, and separator deformation inside the battery. Second, visual inspection is performed on the defective areas inside the battery, specifically including visual features such as cracks, bulges, and leakage. Among these visual features, cracks may lead to leakage of chemical substances or short circuits inside the battery, bulges may be caused by gas accumulation inside the battery, which will increase the internal pressure of the battery after long-term use, leading to the rupture of the battery casing, and leakage will cause chemical corrosion and short circuits. By extracting these defective area features, the system can quickly locate potential problems inside the battery, providing an intuitive basis for attribution analysis and safety performance prediction.
[0072] Example 4
[0073] like Figures 1 to 2 The system shown is a lithium battery safety performance prediction and attribution system based on big data analysis and the GPT model. It includes a data acquisition module 1, a data preprocessing module 2, a safety performance prediction module 3, an attribution analysis module 4, and a human-computer interaction module 5. The data acquisition module 1 provides the system with raw data; the data preprocessing module 2 cleans, standardizes, and extracts features from the data, providing high-quality input for subsequent modules; the safety performance prediction module 3 performs risk prediction based on feature data; the attribution analysis module 4 conducts in-depth analysis of the prediction results to identify the causes; and the human-computer interaction module 5 presents the prediction and attribution results to the user and supports user interaction and feedback. All modules of the system work closely together to achieve the prediction and attribution functions of lithium battery safety performance.
[0074] Among them, the safety performance prediction module 3 is used to receive the feature data output by the feature extraction unit 203, analyze it through multiple learning models, and integrate the prediction results of each model by the model fusion unit, and finally output the prediction results of thermal runaway risk probability and bulging probability; then transmit the prediction results to the human-computer interaction module 5, and transmit the data related to the prediction results to the attribution analysis module 4.
[0075] The model fusion unit employs a stacking strategy, first constructing multiple learning models of different types, including but not limited to traditional machine learning models and deep learning models, to analyze and learn from the acquired feature data from different perspectives. Secondly, the feature data output by the feature extraction unit 203 is input into each basic model. The machine learning model performs preliminary analysis and prediction based on this feature data, generating a set of intermediate prediction results, including the model's estimation of the target variables, namely the probability of thermal runaway risk and the probability of bulging. These intermediate prediction results are used as new features input into the deep learning model. Through further learning and analysis, the deep learning model uncovers deeper feature relationships and patterns, and transmits the output results to the logistic regression layer. The logistic regression layer weights and fuses the prediction results of each model, ultimately generating a comprehensive prediction result.
[0076] In a specific technical solution, the machine learning model mentioned above can be any of the following models, or a combination thereof: Random Forest, Support Vector Machine (SVM), or Gradient Boosting Tree; the deep learning model mentioned above can be a combination of: Convolutional Neural Network (CNN), Long Short-Term Memory Network (LSTM), or Transformer architecture.
[0077] Example 5
[0078] like Figures 1 to 2 The system shown is a lithium battery safety performance prediction and attribution system based on big data analysis and the GPT model. It includes a data acquisition module 1, a data preprocessing module 2, a safety performance prediction module 3, an attribution analysis module 4, and a human-computer interaction module 5. The data acquisition module 1 provides the system with raw data; the data preprocessing module 2 cleans, standardizes, and extracts features from the data, providing high-quality input for subsequent modules; the safety performance prediction module 3 performs risk prediction based on feature data; the attribution analysis module 4 conducts in-depth analysis of the prediction results to identify the causes; and the human-computer interaction module 5 presents the prediction and attribution results to the user and supports user interaction and feedback. All modules of the system work closely together to achieve the prediction and attribution functions of lithium battery safety performance.
[0079] The attribution analysis module 4 is used to establish a mapping relationship between the risk probability output by the prediction module and the production and usage data through multi-source data association. Specifically, it includes a GPT text analysis unit 401, an association analysis unit 402, and an expert knowledge fusion unit 403. The GPT text analysis unit 401 parses the fault description text based on the fine-tuned GPT model, generates a preliminary attribution report, and passes the preliminary attribution report to the association analysis unit 402. The association analysis unit 402 uses the Apriori algorithm to mine the association rules between the production data, usage data, and fault data, further analyzes them in conjunction with the preliminary attribution report, and passes the mined association rules and analysis results to the expert knowledge fusion unit 403. The expert knowledge fusion unit 403 corrects the attribution results through the engine, optimizes and improves the attribution report by combining expert experience and knowledge, and transmits the final attribution analysis results to the human-computer interaction module 5.
[0080] The GPT text analysis unit 401 implements the attribution report through the following steps:
[0081] 1) Input the fault description text into the fine-tuned GPT model and output a list of candidate fault causes;
[0082] 2) Based on the association rules output by the association analysis unit 402, filter out candidate reasons that are strongly correlated with production / usage data;
[0083] 3) Generate the final attribution report after verification by the expert knowledge base.
[0084] After training, the GPT model can understand the technical terms and contextual information related to lithium battery failures, thereby outputting a list containing multiple candidate failure causes. These candidate failure causes cover various problems that may occur in the production and use stages, including raw material defects, production process deviations, and improper use environments. Subsequently, the above candidate causes are matched with association rules to filter out candidate causes that are strongly correlated with production and use data. When the association rule indicates that a specific production process parameter is highly correlated with the occurrence of failure, the entries related to that process parameter in the candidate causes will be retained first.
[0085] Subsequently, the selected candidate causes are submitted to an expert knowledge base for verification. This knowledge base is a database integrating the experience and knowledge of domain experts, including professional knowledge and experience rules related to lithium battery production, use, and fault handling. The system can further verify and correct candidate causes by comparing them with the expert knowledge base, ensuring the accuracy and reliability of the attribution. Furthermore, the expert knowledge fusion unit 403 also has a built-in editable rule base. The rules in the rule base can be dynamically adjusted based on the input of domain experts. When experts discover that a new production process defect may lead to a specific fault, they can add this discovery to the rule base. The system will automatically apply this new rule in subsequent attribution analysis, maintaining a high degree of accuracy and adaptability.
[0086] Furthermore, the association analysis unit 402 specifically employs the Apriori algorithm to mine association rules between production data, usage data, and fault data. The Apriori algorithm is a relatively classic algorithm, and its calculation process is as follows: first, the production data, usage data, and fault data are integrated to form a unified dataset; then, the Apriori algorithm is used to mine frequent itemsets in the integrated dataset; by searching frequent itemsets layer by layer, statistically significant association rules are generated; finally, the mined association rules are combined with the preliminary attribution report generated by the GPT text analysis unit 401 for analysis.
[0087] Example 6
[0088] like Figures 1 to 2 The system shown is a lithium battery safety performance prediction and attribution system based on big data analysis and the GPT model. It includes a data acquisition module 1, a data preprocessing module 2, a safety performance prediction module 3, an attribution analysis module 4, and a human-computer interaction module 5. The data acquisition module 1 provides the system with raw data; the data preprocessing module 2 cleans, standardizes, and extracts features from the data, providing high-quality input for subsequent modules; the safety performance prediction module 3 performs risk prediction based on feature data; the attribution analysis module 4 conducts in-depth analysis of the prediction results to identify the causes; and the human-computer interaction module 5 presents the prediction and attribution results to the user and supports user interaction and feedback. All modules of the system work closely together to achieve the prediction and attribution functions of lithium battery safety performance.
[0089] Among them, the human-computer interaction module 5 provides a visual interface to display the prediction results of the security performance prediction module 3 and the attribution report of the attribution analysis module 4. It supports users to query prediction results, interactively correct attribution reports, and push security suggestions. Based on user operations and feedback, the interactive experience and functions of the system will be further optimized.
[0090] Specifically, the visual interface provided by the human-computer interaction module 5 clearly displays the prediction results of the safety performance prediction module 3 and the attribution report of the attribution analysis module 4 to the user. Users can easily query specific prediction results and understand the battery's safety performance status through this interface. They can also interactively correct the attribution report. If users have questions or need supplementary information regarding certain analysis results in the report, they can provide feedback directly on the interface. The system will dynamically adjust the attribution logic and report content based on the user's input to ensure that the analysis results are more consistent with the actual situation. Furthermore, the safety suggestion push function of the human-computer interaction module 5 can provide users with targeted preventative measures and improvement suggestions based on the prediction results and attribution analysis, helping users take proactive measures to reduce safety risks.
[0091] Furthermore, the human-computer interaction module 5 also provides a risk warning function. When the prediction result exceeds a preset threshold, it triggers multi-level alarm notifications. Specifically, for low-risk situations, the system may display a yellow warning message on the interface to remind the user to pay attention to the operating status of the relevant battery; for high-risk situations, the system will send a red emergency alarm and push the alarm information to designated managers or technicians through various means such as SMS, email, or instant messaging tools to ensure that they can receive the notification in a timely manner and take appropriate emergency measures. Ultimately, this improves the system's response speed and provides comprehensive protection for the safe operation of lithium batteries.
[0092] Example 7
[0093] like Figure 2 The lithium battery safety performance prediction method based on the above system, as shown, includes the following steps:
[0094] S1 collects production data, usage data, and fault description text;
[0095] S2 performs data cleaning, standardization, and time series / image feature extraction;
[0096] S3 predicts safety performance indicators using a hybrid model;
[0097] S4 uses the GPT model to generate a preliminary attribution report, and after correction by association analysis, outputs the final attribution results.
[0098] S5 pushes the prediction results and attribution reports to the user interface and records user feedback for model iteration.
[0099] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0100] The above description is only used to illustrate the technical solution of the present invention and is not intended to limit it. Any other modifications or equivalent substitutions made by those skilled in the art to the technical solution of the present invention, as long as they do not depart from the spirit and scope of the technical solution of the present invention, should be covered within the scope of the claims of the present invention.
Claims
1. A lithium battery safety performance prediction and attribution system based on big data analysis and GPT model, characterized in that, It includes a data acquisition module, a data preprocessing module, a security performance prediction module, an attribution analysis module, and a human-computer interaction module; The data acquisition module is used to acquire lithium battery production data from the MES system, usage data from the T-BOX terminal, and fault description text from the fault database. The data preprocessing module includes a data cleaning unit, a standardization unit, and a feature extraction unit; The data cleaning unit is used to perform outlier removal and missing value imputation. The standardization unit is used to normalize the production data, usage data, and fault description text. The feature extraction unit is used to extract temporal features, statistical features, and image features; The safety performance prediction module consists of multiple learning models and model fusion units, and is used to output prediction results of thermal runaway risk probability and bulging probability. The attribution analysis module is used to establish a mapping relationship between the risk probability output by the prediction module and production data and usage data through multi-source data association, including a GPT text analysis unit, an association analysis unit and an expert knowledge fusion unit. The GPT text analysis unit parses the fault description text based on the fine-tuned GPT model and generates a preliminary attribution report. The correlation analysis unit uses the Apriori algorithm to mine correlation rules between production-usage-failure data; The expert knowledge fusion unit is used to correct the attribution results through the engine; The human-computer interaction module provides a visual interface that supports querying prediction results, interactive correction of attribution reports, and push of safety recommendations.
2. The lithium battery safety performance prediction and attribution system based on big data analysis and GPT model according to claim 1, characterized in that, The data acquisition module synchronizes data with the MES system, T-BOX terminal, and fault database in real time or at regular intervals via API interface.
3. The lithium battery safety performance prediction and attribution system based on big data analysis and GPT model according to claim 1, characterized in that, The time-series features in the feature extraction unit are extracted from the charge-discharge curve data, and the time-series features include: capacity change rate, internal resistance change rate, and voltage fluctuation rate.
4. The lithium battery safety performance prediction and attribution system based on big data analysis and GPT model according to claim 3, characterized in that, The image features in the feature extraction unit are extracted from the lithium battery X-ray inspection image. The image features include texture features or defect area features, specifically visual features such as cracks, bulges, and leakage.
5. The lithium battery safety performance prediction and attribution system based on big data analysis and GPT model according to claim 1, characterized in that, The model fusion unit adopts a stacking strategy, using the output of the machine learning model as the input features of the deep learning model, and finally outputs the fused prediction result through the logistic regression layer.
6. The lithium battery safety performance prediction and attribution system based on big data analysis and GPT model according to claim 1, characterized in that, The GPT text analysis unit generates an attribution report through the following steps: 1) Input the fault description text into the fine-tuned GPT model and output a list of candidate fault causes; 2) Based on the association rules output by the association analysis unit, filter candidate reasons that are strongly correlated with production / usage data; 3) Generate the final attribution report after verification by the expert knowledge base.
7. The lithium battery safety performance prediction and attribution system based on big data analysis and GPT model according to claim 6, characterized in that, The expert knowledge fusion unit has a built-in editable rule base that supports dynamic adjustment of attribution logic based on input from domain experts.
8. The lithium battery safety performance prediction and attribution system based on big data analysis and GPT model according to claim 1, characterized in that, The human-computer interaction module also provides a risk warning function, which triggers multi-level alarm notifications when the prediction result exceeds a preset threshold.
9. A method for predicting the safety performance of lithium batteries based on the system described in claims 1-8, characterized in that, Includes the following steps: S1 collects production data, usage data, and fault description text; S2 performs data cleaning, standardization, and time series / image feature extraction; S3 predicts safety performance indicators using a hybrid model; S4 uses the GPT model to generate a preliminary attribution report, and after correction by association analysis, outputs the final attribution results. S5 pushes the prediction results and attribution reports to the user interface and records user feedback for model iteration.