A cell risk assessment generation system, method, and readable storage medium

Abstract

The application relates to a battery cell risk assessment generation system, a method and a readable storage medium, wherein the battery cell risk assessment generation system comprises the following steps: extracting multi-dimensional risk features of each single battery cell in a battery cluster based on historical time-series operation data of the single battery cell by a risk calculation module, and determining a corresponding risk value based on the multi-dimensional risk features; performing vector embedding on the original risk data of each single battery cell by a vector embedding module to obtain a semantic vector of each single battery cell, and storing the semantic vector in a vector database; and based on a risk query request, searching in the vector database by an evaluation generation module to obtain a search result, and generating a natural language report of the risk query request based on the search result, thereby improving the accuracy of battery cell risk assessment, combining risk data with a knowledge base, and realizing automatic overvoltage risk identification and risk explanation of single battery cells.

Description

Technical Field

[0001] This application relates to the field of battery technology, and in particular to a cell risk assessment generation system, method, and readable storage medium. Background Technology

[0002] With the rapid development of electric vehicles, the safety requirements for them are also gradually increasing. Because battery packs involve complex electrochemical reactions and are susceptible to external environmental interference, they pose significant safety risks. Therefore, real-time monitoring of battery pack safety risks is crucial to ensuring the safe operation of electric vehicles.

[0003] Current battery cell risk assessment schemes mostly use threshold judgment or simple statistical methods, such as detecting whether the voltage exceeds the voltage limit. These methods cannot reflect the risk intensity of the battery cell, have low accuracy, and lack the ability to explain the causes or provide decision-making suggestions. Summary of the Invention

[0004] Therefore, it is necessary to provide a battery cell risk assessment generation system, method, and readable storage medium to address the aforementioned technical problems.

[0005] In a first aspect, embodiments of this application provide a battery cell risk assessment generation system, the system comprising:

[0006] The risk calculation module is used to extract multi-dimensional risk features of each individual cell in the battery cluster based on the historical time-series operation data of each individual cell, and to determine the corresponding risk value based on the multi-dimensional risk features; wherein, the multi-dimensional risk features and the risk value constitute the original risk data;

[0007] The vector embedding module is used to embed the original risk data of each individual battery cell into a vector to obtain the semantic vector of each individual battery cell, and store it in the vector database.

[0008] An assessment generation module is used to perform a search in the vector database based on a risk query request, obtain search results, and generate a natural language report of the risk query request based on the search results.

[0009] In one embodiment, the historical time-series operation data includes at least one of voltage, current, and temperature, and the risk calculation module is further used to input the multidimensional risk characteristics of each individual cell into the risk prediction model to obtain the risk value corresponding to each individual cell.

[0010] In one embodiment, the multidimensional risk features include at least severity, kurtosis, and exposure rate; the risk calculation module is further used to perform weighted fusion of the severity, kurtosis, and exposure rate of each individual cell to obtain the risk value corresponding to each individual cell.

[0011] In one embodiment, the risk calculation module is further configured to determine the severity based on the peak values ​​of voltage and / or current and / or temperature; determine the spuriousness based on the voltage and / or current and / or temperature values ​​at preset points; and determine the exposure rate based on the proportion of voltage and / or current and / or temperature in overvoltage and / or overcurrent and / or overtemperature regions.

[0012] In one embodiment, the evaluation generation module includes a retrieval module and a question-and-answer generation module;

[0013] The retrieval module is used to perform a search in the vector database based on the risk query request and obtain the search results.

[0014] The question-and-answer generation module is used to generate a natural language report of the risk query request based on the search results.

[0015] In one embodiment, the vector embedding module is further configured to construct an index for the semantic vectors stored in the vector database, and the retrieval module is specifically configured to: convert the risk query request into a risk query vector, and calculate the cosine similarity between the risk query vector and each of the semantic vectors in the vector database based on the index, to obtain the top K semantic vectors, so as to obtain the retrieval results.

[0016] In one embodiment, the question-answering generation module is specifically used to: combine the first K semantic vectors and the corresponding original risk data to form a context, input the context into a large language model, and use the large language model to generate a natural language report of the risk query request, wherein the natural language report includes a natural language answer and a corresponding solution.

[0017] In one embodiment, both the large language model and the vector database are deployed on an edge device.

[0018] Secondly, embodiments of this application also provide a cell risk assessment generation method, applied to the cell risk assessment generation system described in the first aspect above, the method comprising:

[0019] Based on the historical time-series operation data of each individual cell in the battery cluster, multidimensional risk features of each individual cell are extracted, and the corresponding risk value is determined based on the multidimensional risk features; wherein, the multidimensional risk features and the risk value constitute the original risk data;

[0020] The original risk data of each individual battery cell is vector-embedded to obtain the semantic vector of each individual battery cell, and stored in the vector database;

[0021] Based on the risk query request, a search is performed in the vector database to obtain the search results, and a natural language report of the risk query request is generated based on the search results.

[0022] Thirdly, embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the second aspect above.

[0023] The aforementioned cell risk assessment generation system, method, and readable storage medium utilize a risk calculation module to extract multi-dimensional risk features from each individual cell in a battery cluster based on historical time-series operational data, and determine the corresponding risk value based on these features. A vector embedding module is used to embed the original risk data of each individual cell into a semantic vector, which is then stored in a vector database. An assessment generation module is used to retrieve data from the vector database based on a risk query request, obtain the retrieval results, and generate a natural language report of the risk query request based on the retrieval results. This improves the accuracy of cell risk assessment and combines risk data with a knowledge base to achieve automated overvoltage risk identification and interpretation for individual cells.

[0024] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0025] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0026] Figure 1 This is a structural block diagram of a battery cell risk assessment generation system in one embodiment;

[0027] Figure 2 This is a structural block diagram of a battery cell risk assessment generation system in another embodiment;

[0028] Figure 3 This is a flowchart illustrating the cell risk assessment generation method in one embodiment. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0030] Obviously, the accompanying drawings described below are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0031] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0032] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0033] This application provides a battery cell risk assessment generation system. Please refer to [link / reference]. Figure 1 The diagram shows the structure of the cell risk assessment generation system, which includes a risk calculation module 100, a vector embedding module 200, and an assessment generation module 300.

[0034] The risk calculation module 100 is used to extract multi-dimensional risk features of each individual cell in the battery cluster based on historical time-series operational data, and to determine the corresponding risk value based on the multi-dimensional risk features. The historical time-series operational data includes data acquired by each individual cell during multiple complete charging stages, including at least one of voltage, current, and temperature, and may be a combination of one or more of these. Then, the historical time-series operational data of each individual cell undergoes data preprocessing and feature extraction. Data preprocessing includes filtering invalid numbers and abnormal voltage / current / temperature values ​​to ensure that the data is sampled from the charging stage, standardizing the sampling time field to a uniform format, and performing feature extraction on the cleaned raw data to obtain the multi-dimensional risk features of each individual cell. Based on these multi-dimensional risk features, the corresponding risk value is determined. The multi-dimensional risk features and the risk value of each individual cell constitute the original risk data, which can be output as a CSV file.

[0035] The vector embedding module 200 is used to embed the original risk data of each individual battery cell into a semantic vector, which is then stored in a vector database. A local vector database is constructed using Milvus-Lite. The original risk data file (e.g., .csv) serves as the input data source for the vector embedding module 200. A local embedding model (e.g., bge-small-zh-v1.5) is used to semantically vectorize each record, converting text data into a semantic vector representation. The generated semantic vectors and corresponding text data are stored in the local vector database for subsequent similarity retrieval.

[0036] The local vector database is continuously updated, with semantic vectors and text data corresponding to newly added individual battery cells being updated continuously.

[0037] Furthermore, the vector embedding module 200 is also used to build an index for the semantic vectors stored in the vector database. This index improves retrieval efficiency, storing the correspondence between semantic vectors and the original text, thus enabling index-based semantic similarity retrieval. For example, when a user inputs "find the highest-risk battery cell," the system will quickly retrieve the set of vectors most relevant to the high-risk semantics based on the index.

[0038] The assessment generation module 300 is used to perform a search in the vector database based on the risk query request, obtain the search results, and generate a natural language report of the risk query request based on the search results.

[0039] After receiving a risk query request, the assessment generation module 300 queries the vector database for semantic vectors similar to the request based on the retrieval enhancement generation logic, obtains matching retrieval results, and then uses a large language model to generate a natural language report that meets the requirements based on the retrieval results, providing an intuitive and accurate reference and corresponding solutions for battery cell risk assessment.

[0040] The cell risk assessment generation system of this application embodiment uses a risk calculation module 100 to extract multi-dimensional risk features of each individual cell in a battery cluster based on historical time-series operational data, and determines the corresponding risk value based on the multi-dimensional risk features; a vector embedding module 200 embeds the original risk data of each individual cell into a vector to obtain a semantic vector of each individual cell, and stores it in a vector database; an assessment generation module 300 searches the vector database based on a risk query request to obtain search results, and generates a natural language report of the risk query request based on the search results, thereby improving the accuracy of cell risk assessment. By combining risk data with a knowledge base, it realizes automated overvoltage risk identification and risk interpretation of individual cells.

[0041] In one embodiment, the historical time-series operation data includes at least one of voltage, current, and temperature, and the risk calculation module 100 is further used to input the multidimensional risk characteristics of each individual cell into the risk prediction model to obtain the risk value corresponding to each individual cell.

[0042] The historical time-series operational data includes time-series voltage, current, and temperature data acquired by a single battery cell during a full charge phase. Specifically, the voltage, current, and temperature values ​​of a single battery cell can be measured with high precision using a data acquisition chip. In some embodiments, the historical time-series operational data can be the time-series voltage data acquired by a single battery cell during a full charge phase. The risk prediction model is trained using a training set, which consists of multidimensional risk characteristics of multiple single battery cells and their corresponding true risk values.

[0043] In one embodiment, the multidimensional risk features include at least severity, peak intensity, and exposure rate; the risk calculation module 100 is further used to perform weighted fusion of the severity, peak intensity, and exposure rate of each individual battery cell to obtain the risk value corresponding to each individual battery cell.

[0044] Severity indicates the degree to which voltage / current / temperature peaks approach the safety limit; kurtosis indicates the intensity of voltage / current / temperature fluctuations at the high quantile; and exposure rate indicates the proportion of voltage / current / temperature that remains in the high-voltage zone for an extended period. Therefore, severity, kurtosis, and exposure rate can be used to calculate the risk value corresponding to a single battery cell. In some embodiments, the multidimensional risk characteristics may also include other risk characteristics.

[0045] The severity is determined based on the peak values ​​of voltage and / or current and / or temperature; the spury is determined based on the voltage and / or current and / or temperature values ​​at a preset point; and the exposure rate is determined based on the proportion of voltage and / or current and / or temperature in the overvoltage and / or overcurrent and / or overtemperature regions.

[0046] In one example embodiment, for instance when the historical time-series running data includes voltage data, the extracted multidimensional risk features are shown in Table 1.

[0047] Table 1

[0048]

[0049] The maximum voltage, representing the highest value among the time-series voltage data acquired in each complete charging stage, indicates severity. The 99.2% voltage point represents the voltage value at the 99.2% mark among the time-series voltage data acquired in each complete charging stage, indicating spike. Assuming a complete charging voltage sequence has 1000 data points, P992v refers to the voltage value corresponding to the 992nd data point. Of course, the 99.2% point can also be set to any value between 99.0% and 99.5%, corresponding to the point near the end of the charging process. Generally, a voltage exceeding 3.65V in a single battery cell during the charging stage indicates overvoltage. The exposure rate in this embodiment can be determined by the proportion of time-series voltage data exceeding 3.64V in each complete charging stage.

[0050] Furthermore, to eliminate the dimensional differences between different battery cells, the extracted feature data is normalized using the following formula:

[0051]

[0052] Here, clip01() represents the clipping function, ensuring that the output is between [0, 1].

[0053] For example, the risk value of each individual battery cell is calculated using the following formula:

[0054] Risk value = 0.5 × severity + 0.3 × exposure rate + 0.2 × kurtosis;

[0055] The weighting combination was determined through multiple rounds of data experiments and reflects the comprehensive risk of the battery cell across three dimensions. In the calculation results, the risk value ranges from [0,1], with a higher value indicating a higher risk.

[0056] After risk calculation, the system can automatically output a result file that sorts all cells from high to low risk value. The result data file includes the cell's cluster number, sub-cluster number, cell ID, maximum voltage (max_v), voltage 99.2% point (p992_v), severity, exposure rate, peak intensity, risk value (risk), and last high voltage time (last_high_ts). The cell's cluster number and sub-cluster number are used to locate the battery pack in which the cell is located, and the cell ID is used to locate the cell's position in the battery pack.

[0057] The system exports the results file (e.g., .csv) as the input data source for the vector embedding module 200 for semantic analysis. This includes five steps: data loading, vector embedding, vector indexing, retrieval enhancement, and LLM generation. Specifically, the system loads the original risk data using the PandasCSVReader component of the large language model and parses the file using the pandas engine in Python software. This step automatically converts the structured CSV file to a large language model file object and serves as the input for the entire semantic analysis. Vector embedding generates semantic vectors from the original risk data using an embedding model (bge-small-zh-v1.5). Vector indexing involves building a local vector database and creating an index for the semantic vectors. Retrieval enhancement includes retrieving relevant information from the local vector database based on similarity search. LLM generation involves calling the DeepSeek-R1 model on the RKLLM Runtime to generate natural language explanations and suggestions.

[0058] In one embodiment, such as Figure 2 As shown, the assessment generation module 300 includes a retrieval module 310 and a question-and-answer generation module 320; the retrieval module 310 is used to perform a retrieval in the vector database based on the risk query request to obtain retrieval results; the question-and-answer generation module 320 is used to generate a natural language report of the risk query request based on the retrieval results.

[0059] In some embodiments, this application employs Retrieval Augmentation Generation (RAG) technology at the edge to achieve retrieval and question answer generation. Retrieval Augmentation Generation (RAG) is a framework that combines a text retrieval module with a text generation module, aiming to improve the quality and accuracy of Large Language Model (LLM) responses.

[0060] The logic of generating a natural language report for a risk query request using the Retrieval Enhanced Generation (RAG) method in this application is as follows: Each record in the risk result file is semantically vectorized using a local embedding model to generate a corresponding semantic vector. An index is built for the semantic vectors, and the semantic vectors are stored in a local vector database. When a risk query request is received, the risk query request is converted into a risk query vector. Based on the index, the cosine similarity between the risk query vector and each semantic vector in the vector database is calculated. According to the cosine similarity results between the risk query vector and each semantic vector in the vector database, the top K semantic vectors are retained to obtain the retrieval results. Then, the top K semantic vectors and the corresponding original risk data are combined to form a context. This context is input into a large language model (e.g., using the RKLLM Runtime provided by the Rockchip official SDK, and calling librkllmrt.so through Pythonctypes to implement model loading, inference, and streaming output). The large language model infers and answers based on the context, generating a natural language report for the risk query request. The natural language report includes a natural language answer and a corresponding solution. The natural language report is returned to the host computer in JSON format, and the host computer displays the output results.

[0061] In one example embodiment, a user poses a natural language question, a risk query request that asks, "Which battery cells have a risk value exceeding 0.8? And provide relevant risk solutions." The assessment generation module 300 then performs the following steps:

[0062] Step 1: Convert the risk query request into a query vector.

[0063] Step 2: Retrieve the top-K semantic vectors with the highest similarity to the query vector in the vector database (Milvus-Lite) to obtain the search results.

[0064] Step 3: Combine the search results with the original risk data to form a context;

[0065] Step 4: Package the context into prompts and input them into the Lightweight Large Language Model (LLM). By adding relevant risk fields to the context, the large language model can use real data to generate interpretive results.

[0066] This embodiment uses the RKLLM Runtime to load the DeepSeek-R1-1.5B-w8a8 model. The model performs generative inference and outputs a natural language report. Output example: Cell number 96 in sub-cluster 1 has the highest risk, with an overvoltage of 3.67V, which lasted for 8 minutes. It is recommended to check the contact resistance at the sampling end or perform equalization correction.

[0067] In one embodiment, both the large language model and the vector database are deployed on edge devices. Compared to related methods that rely on cloud computing, the battery cell risk assessment generation system of this application is implemented on edge computing devices. This allows for both quantitative risk calculation and real-time generation of interpretive diagnostics on the edge device, ensuring the efficient operation of the risk assessment algorithm and the large language model knowledge enhancement system. Through model quantization, algorithm fusion, and local caching, reasoning and retrieval tasks are completed under limited computing power, constructing a low-latency, high-reliability intelligent risk assessment generation system while avoiding internal data leakage.

[0068] This application also provides a method for generating battery cell risk assessments, applicable to the battery cell risk assessment generation system described in any of the above embodiments, such as... Figure 3 As shown, the method includes the following steps:

[0069] Step S201: Based on the historical time-series operation data of each individual cell in the battery cluster, extract the multidimensional risk features of each individual cell, and determine the corresponding risk value based on the multidimensional risk features; wherein, the multidimensional risk features and the risk value constitute the original risk data.

[0070] Step S202: The original risk data of each individual battery cell is vector-embedded to obtain the semantic vector of each individual battery cell, and stored in the vector database.

[0071] Step S203: Based on the risk query request, perform a search in the vector database to obtain the search results, and generate a natural language report of the risk query request based on the search results.

[0072] In one embodiment, the historical time-series operation data includes at least one of voltage, current, and temperature, and the step of determining the corresponding risk value based on the multidimensional risk characteristics includes: inputting the multidimensional risk characteristics of each individual cell into the risk prediction model to obtain the risk value corresponding to each individual cell.

[0073] In one embodiment, the multidimensional risk features include at least severity, kurtosis, and exposure rate; determining the corresponding risk value based on the multidimensional risk features includes: weighting and fusing the severity, kurtosis, and exposure rate of each individual battery cell to obtain the risk value corresponding to each individual battery cell.

[0074] In one embodiment, the severity is determined based on the peak values ​​of voltage and / or current and / or temperature; the spury is determined based on the voltage and / or current and / or temperature values ​​at preset points; and the exposure rate is determined based on the proportion of voltage and / or current and / or temperature in overvoltage and / or overcurrent and / or overtemperature regions.

[0075] In one embodiment, the method further includes constructing an index for semantic vectors stored in the vector database. The step of searching the vector database based on a risk query request to obtain search results includes: converting the risk query request into a risk query vector, and calculating the cosine similarity between the risk query vector and each of the semantic vectors in the vector database based on the index to obtain the top K semantic vectors to obtain search results.

[0076] In one embodiment, generating a natural language report for the risk query request based on the search results includes: combining the first K semantic vectors and the corresponding original risk data to form a context, inputting the context into a large language model, and using the large language model to generate a natural language report for the risk query request, wherein the natural language report includes a natural language answer and a corresponding solution.

[0077] In one example embodiment, the cell risk assessment generation method includes the following steps:

[0078] Step S301: Based on the historical time-series voltage data of each individual cell, extract the severity, peak intensity, and exposure rate of each individual cell, perform a weighted summation of the severity, peak intensity, and exposure rate to obtain the corresponding risk value, and output the multidimensional risk features and the risk value as a risk result file.

[0079] Step S302: Load the risk results file using PandasCSVReader.

[0080] Step S303: Use the local embedding model (bge-small-zh-v1.5) to semantically vectorize each record in the risk results file to generate the corresponding semantic vector.

[0081] Step S304: Construct a local vector database (Milvus-Lite), store the semantic vectors of each individual battery cell and the corresponding original risk data in the vector database, and build an index.

[0082] Step S305: Receive risk query request.

[0083] Step S306: Perform a retrieval based on the index, calculate the similarity between the risk query request and each semantic vector in the vector database, and obtain the retrieval results.

[0084] Step S307: Call the DeepSeek-R1 model on the RKLLM Runtime, combine the search results with the original risk data to form a context input into the DeepSeek-R1 model, and generate natural language explanations and suggestions.

[0085] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in any of the above embodiments of the battery cell risk assessment generation method.

[0086] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0087] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0088] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A battery cell risk assessment generation system, characterized in that, The system includes: The risk calculation module is used to extract multi-dimensional risk features of each individual cell in the battery cluster based on the historical time-series operation data of each individual cell, and to determine the corresponding risk value based on the multi-dimensional risk features; wherein, the multi-dimensional risk features and the risk value constitute the original risk data; The vector embedding module is used to embed the original risk data of each individual battery cell into a vector to obtain the semantic vector of each individual battery cell, and store it in the vector database. An assessment generation module is used to perform a search in the vector database based on a risk query request, obtain search results, and generate a natural language report of the risk query request based on the search results.

2. The system according to claim 1, characterized in that, The historical time-series operation data includes at least one of voltage, current, and temperature. The risk calculation module is also used to input the multidimensional risk characteristics of each individual cell into the risk prediction model to obtain the risk value corresponding to each individual cell.

3. The system according to claim 2, characterized in that, The multidimensional risk characteristics include at least severity, peak intensity, and exposure rate; the risk calculation module is also used to perform weighted fusion of the severity, peak intensity, and exposure rate of each individual cell to obtain the risk value corresponding to each individual cell.

4. The system according to claim 3, characterized in that, The risk calculation module is also used to determine the severity based on the peak values ​​of voltage and / or current and / or temperature; to determine the peak intensity based on the voltage and / or current and / or temperature values ​​at preset points; and to determine the exposure rate based on the proportion of voltage and / or current and / or temperature in the overvoltage zone and / or overcurrent zone and / or overtemperature zone.

5. The system according to claim 1, characterized in that, The evaluation generation module includes a retrieval module and a question-and-answer generation module; The retrieval module is used to perform a search in the vector database based on the risk query request and obtain the search results. The question-and-answer generation module is used to generate a natural language report of the risk query request based on the search results.

6. The system according to claim 5, characterized in that, The vector embedding module is also used to build an index for the semantic vectors stored in the vector database. The retrieval module is specifically used to: convert the risk query request into a risk query vector, and calculate the cosine similarity between the risk query vector and each of the semantic vectors in the vector database based on the index, so as to obtain the top K semantic vectors and thus obtain the retrieval results.

7. The system according to claim 6, characterized in that, The question-and-answer generation module is specifically used to: combine the first K semantic vectors and the corresponding original risk data to form a context, input the context into a large language model, and use the large language model to generate a natural language report of the risk query request, the natural language report including a natural language answer and a corresponding solution.

8. The system according to claim 7, characterized in that, Both the large language model and the vector database are deployed on edge devices.

9. A method for generating battery cell risk assessments, applied to the battery cell risk assessment generation system as described in any one of claims 1 to 8, characterized in that, The method includes: Based on the historical time-series operation data of each individual cell in the battery cluster, multidimensional risk features of each individual cell are extracted, and the corresponding risk value is determined based on the multidimensional risk features; wherein, the multidimensional risk features and the risk value constitute the original risk data; The original risk data of each individual battery cell is vector-embedded to obtain the semantic vector of each individual battery cell, and stored in the vector database; Based on the risk query request, a search is performed in the vector database to obtain the search results, and a natural language report of the risk query request is generated based on the search results.

10. 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 claim 9.

Images