Energy storage battery anomaly analysis method and system based on retrieval enhancement generation

Abstract

The invention relates to an energy storage battery anomaly analysis method and system based on retrieval enhancement generation. The method comprises the following steps: acquiring real-time operation data of an energy storage battery, extracting a statistical feature vector, performing sliding window segmentation according to a division condition, and determining an abnormal window; constructing a knowledge database; taking abnormal window data at the abnormal window as input, searching corresponding energy storage battery abnormal data and contextual information from the knowledge database, and combining the energy storage battery abnormal data and the contextual information with the statistical feature vector of the abnormal window to obtain combined information; and inputting the combined information into a generation model, and generating a battery abnormity report corresponding to the energy storage battery through the generation model. Through the sliding window detection technology, the exceptions can be quickly screened, and the real-time performance and accuracy of exception analysis are greatly improved in cooperation with the reasoning ability of knowledge base retrieval and model generation.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage battery status detection, and in particular to an energy storage battery abnormality analysis method and system based on retrieval enhancement generation. Background Art

[0002] With the rapid development of energy storage technology, energy storage batteries, as core components of new energy systems, play a vital role in scenarios such as power peak regulation, distributed energy access and microgrid operation. The safety and stability of energy storage batteries directly affect the reliability of the entire energy storage system. However, due to the complex working conditions in long-term operation and the performance degradation of the battery itself, energy storage batteries frequently have abnormal problems, such as over-discharge, over-charge, over-temperature, capacity attenuation, etc. If these abnormalities are not detected and handled in a timely and effective manner, they may lead to reduced system efficiency and even cause safety accidents.

[0003] At present, the methods of abnormal analysis of energy storage batteries are mainly divided into two categories: rule-based anomaly detection and predictive analysis based on machine learning. Among them, rule-based anomaly detection often relies on artificially set rules and thresholds (such as upper and lower limits of voltage, current, and temperature) to monitor the operating status of the battery. When a certain indicator exceeds the predetermined range, it is judged to be abnormal. This method is simple to implement and has low computational overhead, and has been widely used in early energy storage systems. However, due to the singleness and static nature of the rules, this method is difficult to cope with complex operating conditions. For example, the operating characteristics of energy storage batteries are affected by multiple factors such as ambient temperature, charge and discharge rate, and historical aging degree. Static rules cannot adapt to these dynamic changes and are prone to false positives or omissions. Predictive analysis based on machine learning uses historical operating data to train models (such as support vector machines, random forests, or deep learning models) to predict the battery status and determine whether there is an abnormality. This type of method solves the limitations of rule methods to a certain extent and can adapt to more diverse operating scenarios.

[0004] However, with the rapid development of the new energy industry, the types of anomalies in the operation of energy storage systems are constantly increasing, and the traditional methods of analyzing anomalies in energy storage batteries lack flexibility and scalability. For example, for new abnormal modes of energy storage batteries (such as anomalies caused by the unique characteristics of new material batteries), traditional rules or static models are difficult to update and adapt in a timely manner. Therefore, the traditional methods of analyzing anomalies in energy storage batteries are not flexible enough, and there are problems such as insufficient dynamic adaptability of anomaly detection, insufficient explanation of anomalies, and insufficient practicality. Summary of the invention

[0005] Based on this, in order to solve the above technical problems, a method and system for analyzing energy storage battery anomalies based on retrieval enhanced generation are provided, which can realize efficient analysis and explanatory output of energy storage battery anomalies and ensure the accuracy of detection.

[0006] A method for analyzing abnormality of an energy storage battery based on retrieval enhancement generation, the method comprising:

[0007] Acquire real-time operation data of the energy storage battery, extract statistical feature vectors in the real-time operation data, and perform sliding window segmentation according to the statistical feature vectors according to the segmentation conditions to determine the abnormal window;

[0008] Collecting historical abnormal data and experience knowledge corresponding to the energy storage battery, and performing standardization and vectorization on the historical abnormal data and experience knowledge to build a knowledge database;

[0009] Taking the abnormal window data at the abnormal window as input, searching the corresponding energy storage battery abnormal data and context information from the knowledge database, and combining the energy storage battery abnormal data and context information with the statistical feature vector of the abnormal window to obtain combined data;

[0010] The combined data is input into a generation model, and a battery abnormality report corresponding to the energy storage battery is generated through the generation model.

[0011] In one embodiment, performing sliding window segmentation according to the statistical feature vector according to the segmentation condition to determine the abnormal window includes:

[0012] Determining the acquisition time of the statistical feature vector, and using the acquisition time as a division condition;

[0013] The statistical feature vector is segmented into sliding windows according to the segmentation condition, and adjacent sliding windows have an overlap, so as to obtain individual windows;

[0014] Respectively acquiring sliding window statistical features corresponding to each of the windows, and comparing the sliding window statistical features with a dynamic threshold to obtain a comparison result;

[0015] Determine an abnormal window according to the comparison result;

[0016] The dynamic threshold is dynamically updated according to the data in the knowledge database.

[0017] In one embodiment, the historical abnormal data and experience knowledge are standardized and vectorized to construct a knowledge database, including:

[0018] Perform data cleaning on the historical abnormal data and experience knowledge, and perform sentence segmentation on the text data behind the information to obtain cleaned data;

[0019] Normalizing the numerical data in the cleaned data, and preprocessing the text data in the cleaned data to obtain standardized data;

[0020] The standardized data is organized according to a fixed format to obtain target data, and the target data is vectorized to construct a knowledge database.

[0021] In one embodiment, vectorization is performed on the target data to construct a knowledge database, including:

[0022] Determine the text data and the numerical data in the target data, and select the embedding methods corresponding to the text data and the numerical data respectively;

[0023] Use the embedding method to vectorize the text data and the numerical data to obtain generated vector data;

[0024] The generated vector data is verified, and the verified vector data is stored in a vector database to obtain a knowledge database.

[0025] In one embodiment, taking the abnormal window data at the abnormal window as input, searching the knowledge database for the corresponding energy storage battery abnormal data and context information includes:

[0026] Encoding the abnormal window data using an embedding model, and converting the abnormal window data into a vector representation;

[0027] The vector representation is input into the knowledge database, and the energy storage battery abnormal data and context information most relevant to the abnormal window data are retrieved using similarity.

[0028] In one embodiment, the abnormal data of the energy storage battery is combined with the statistical feature vector of the abnormal window to obtain the combined data, including:

[0029] Determine an abnormal performance description corresponding to the statistical feature vector of the abnormal window;

[0030] Integrate the energy storage battery abnormal data into structured information;

[0031] A task generation instruction is obtained, and according to the task generation instruction, information is extracted from the abnormal performance description and the structured information to form prompt information, and combined data is obtained.

[0032] In one embodiment, the method further comprises:

[0033] Input the prompt information into the model to be trained, perform reasoning questions and answers through the model to be trained, complete model training and obtain a generated model;

[0034] The generation model is adjusted according to different types of the prompt information, and the generation model is adaptively expanded according to the search results in the knowledge database to obtain a final generation model.

[0035] In one embodiment, the combined data is input into a generation model, and a battery abnormality report corresponding to the energy storage battery is generated by the generation model, including:

[0036] Inputting the combined data into a generation model, and determining battery abnormality description information based on the energy storage battery abnormality data and context information through the generation model;

[0037] Determine summary information corresponding to the energy storage battery abnormality data according to the battery abnormality description information;

[0038] Analyze the cause of the battery abnormality according to the battery abnormality description information to obtain an analysis result;

[0039] Determine corresponding abnormality resolution suggestions based on the analysis results;

[0040] The battery anomaly description information, summary information, analysis results, and anomaly resolution suggestions are used to generate a battery anomaly report.

[0041] In one embodiment, the method further comprises:

[0042] Outputting and pushing the battery abnormality report;

[0043] The battery abnormality report is visualized and displayed on a display interface.

[0044] A system for analyzing abnormalities of energy storage batteries by retrieval enhancement generation, the system comprising:

[0045] A sliding window statistical feature detection module is used to obtain real-time operation data of the energy storage battery, extract statistical feature vectors in the real-time operation data, and perform sliding window segmentation according to the statistical feature vectors according to the segmentation conditions to determine the abnormal window;

[0046] A knowledge base construction module, used to collect historical abnormal data and experience knowledge corresponding to the energy storage battery, and perform standardization and vectorization on the historical abnormal data and experience knowledge to construct a knowledge database;

[0047] A retrieval enhancement generation analysis module is used to take the abnormal window data at the abnormal window as input, search for the corresponding energy storage battery abnormal data and context information from the knowledge database, and combine the energy storage battery abnormal data with the statistical feature vector of the abnormal window to obtain the combined data;

[0048] The report generation and output module is used to input the combined data into the generation model, and generate a battery abnormality report corresponding to the energy storage battery through the generation model.

[0049] The above-mentioned energy storage battery abnormality analysis method and system based on retrieval-enhanced generation can realize rapid screening of abnormalities through sliding window detection technology, and cooperate with the reasoning ability of knowledge base retrieval and generation model to greatly improve the real-time and accuracy of abnormality analysis; retrieval-enhanced generation technology combines abnormal context data to generate detailed battery abnormality reports, which are highly interpretable and meet actual needs; the knowledge base can be dynamically updated to ensure rapid response to new working conditions and new abnormal modes, which is suitable for the complex and changeable operating environment of energy storage batteries. Combining sliding window detection technology, knowledge base retrieval and generation model, and retrieval-enhanced generation technology, efficient analysis and explanatory output of energy storage battery abnormalities can be achieved to ensure the accuracy of detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 An application environment diagram of an energy storage battery abnormality analysis method based on retrieval enhancement generation in one embodiment;

[0051] Figure 2 A schematic diagram of a process of analyzing an abnormality of an energy storage battery based on retrieval enhancement generation in one embodiment;

[0052] Figure 3 A schematic block diagram of a sliding window statistical feature detection process in one embodiment;

[0053] Figure 4 It is a structural block diagram of an energy storage battery abnormality analysis system based on retrieval enhancement generation in one embodiment;

[0054] Figure 5 It is a structural block diagram of an energy storage battery abnormality analysis system based on retrieval enhancement generation in another embodiment;

[0055] Figure 6 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0057] The energy storage battery abnormality analysis method based on retrieval enhancement generation provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Figure 1 As shown, the application environment includes a computer device 110. The computer device 110 can obtain the real-time operation data of the energy storage battery, extract the statistical feature vector in the real-time operation data, and perform sliding window segmentation according to the statistical feature vector according to the division condition to determine the abnormal window; the computer device 110 can collect the historical abnormal data and experience knowledge corresponding to the energy storage battery, and perform standardization and vectorization on the historical abnormal data and experience knowledge to build a knowledge database; the computer device 110 can use the abnormal window data at the abnormal window as input, search for the corresponding energy storage battery abnormal data and context information from the knowledge database, and combine the energy storage battery abnormal data and context information with the statistical feature vector of the abnormal window to obtain the combined data; the computer device 110 can input the combined data into the generation model, and generate a battery abnormality report corresponding to the energy storage battery through the generation model. Among them, the computer device 110 can be, but not limited to, various personal computers, laptops, smart phones, robots, unmanned aerial vehicles and other devices.

[0058] In one embodiment, Figure 2 As shown, a method for analyzing abnormality of an energy storage battery based on retrieval enhancement generation is provided, comprising the following steps:

[0059] Step 202, acquiring real-time operation data of the energy storage battery, extracting statistical feature vectors in the real-time operation data, and performing sliding window segmentation according to the statistical feature vectors and the segmentation conditions to determine the abnormal window.

[0060] Computer equipment can obtain the real-time operation data of the energy storage battery in real time, and perform sliding window segmentation on the real-time operation data of the energy storage battery, extract statistical features such as average value, change rate, jump amplitude, etc., and combine dynamic thresholds to determine abnormal windows. That is, sliding window statistical feature detection is used to perform preliminary processing and screening of the real-time operation data of the energy storage battery, quickly locate potential abnormal windows by dynamically extracting key features, and quickly screen potential abnormal fragments to provide efficient input for subsequent analysis.

[0061] Specifically, in one embodiment, a method for analyzing abnormalities of energy storage batteries based on retrieval enhancement generation is provided, which may also include a sliding window statistical feature detection process, and the specific process includes: determining the acquisition time of the statistical feature vector, and using the acquisition time as a division condition; performing sliding window segmentation on the statistical feature vector according to the division condition, and having an overlap between adjacent sliding windows, to obtain individual windows; respectively obtaining the sliding window statistical features corresponding to each window, and comparing the sliding window statistical features with a dynamic threshold to obtain a comparison result; determining the abnormal window according to the comparison result; wherein the dynamic threshold is dynamically updated according to the data in the knowledge database.

[0062] The process of sliding window statistical feature detection is as follows Figure 3 As shown, it mainly includes data input, data processing and analysis, abnormal output, and adaptive update. Among them, the input data is the real-time operation data of the energy storage battery; data processing and analysis can include sliding window segmentation, statistical feature extraction, feature type division, and dynamic threshold judgment. Among them, the feature type can include the mean / maximum value of the concentration feature, the standard deviation / rate of change of the volatility feature, and the trend analysis of the time series feature; the abnormal output can be divided into abnormal window marking and feature value recording; the adaptive update mainly includes parameter and threshold update.

[0063] Specifically, in this embodiment, sliding window segmentation is the basis of statistical feature extraction, which is to segment the real-time collected energy storage battery operation data (such as voltage, current, SOC status, etc.) according to fixed time windows. Each sliding window covers a fixed time range, and the window size is set according to the operating characteristics of the energy storage battery (such as 1 minute or 5 minutes) to capture the local characteristics of abnormal changes. A certain degree of overlap can be set between sliding windows to avoid missing rapid anomalies. Among them, the real-time data stream input is the sliding window that directly reads the operating data of the current time period from the data stream. In this embodiment, the window coverage strategy adopted is mainly to use a sliding window of fixed length, while supporting flexible adjustment of the window size to adapt to different operating scenarios.

[0064] Next, the computer device can perform statistical feature extraction, specifically calculating a set of statistical features for each data segment in the sliding window to describe the overall trend and fluctuation of the data. Common statistical features include but are not limited to: centrality features: such as average, maximum, minimum, etc., used to reflect the overall level of the data; volatility features: such as standard deviation, rate of change, jump amplitude, etc., used to identify drastic changes; time series features: such as the trend of changes in the values ​​in the sliding window, used to determine whether there are abnormal fluctuations. These features can quickly extract key information of the operating data and identify potential anomalies through subsequent steps.

[0065] Computer equipment can compare the sliding window statistical features with the dynamic threshold to determine whether the current sliding window is abnormal. Among them, the dynamic threshold is generated by historical operation data analysis or domain expert knowledge, and is adjusted in real time according to different operating conditions. Compared with the traditional fixed threshold method, the dynamic threshold is more adaptable to changes in the operating status of the energy storage battery. The computer equipment can compare the sliding window statistical feature value with the corresponding threshold, and mark it as an abnormal window if it exceeds the set range. Through the comprehensive judgment of multiple features, the risk of false positives or false negatives of a single feature is reduced.

[0066] For sliding windows that are judged to be abnormal, the computer device can mark them as "potential abnormal windows" and output the time range of the window and its statistical characteristics to provide input for subsequent processing. The abnormal window description includes: recording the time range of the abnormal window and related characteristic values. The computer device can retain the original data fragments of the abnormal window for subsequent detailed analysis.

[0067] In this embodiment, the configured sliding window parameters may include: the sliding window length, sliding step size and overlap rate can all be flexibly configured according to the operating requirements of the energy storage battery to ensure that key abnormal characteristics are captured; and as the system runs, the threshold of the sliding window statistical feature can be dynamically updated through incremental learning of historical data, further improving the adaptability and accuracy of detection.

[0068] Step 204, collect historical abnormal data and experience knowledge corresponding to the energy storage battery, and perform standardization and vectorization on the historical abnormal data and experience knowledge to build a knowledge database.

[0069] Knowledge base construction belongs to the data preprocessing part, which can be used to store and manage the domain knowledge and historical data required for energy storage battery anomaly analysis, build a knowledge database containing historical anomaly cases of energy storage batteries, domain technical documents and related experience, and achieve efficient retrieval through vectorized representation. It supports dynamic updates to adapt to new working conditions and new abnormal modes, and can achieve efficient retrieval and real-time updates to provide data support for anomaly analysis.

[0070] In one embodiment, a method for analyzing abnormalities of energy storage batteries based on retrieval enhancement generation is provided, which may also include a process for constructing a knowledge database, and the specific process includes: performing data cleaning processing on historical abnormal data and empirical knowledge, and performing sentence operations on the text data after the information to obtain cleaned data; normalizing the numerical data in the cleaned data, and performing preprocessing operations on the text data in the cleaned data to obtain standardized data; organizing the standardized data in a fixed format to obtain target data, and vectorizing the target data to construct a knowledge database.

[0071] Among them, the data sources for constructing the knowledge database are diverse, covering different types of information required for abnormal analysis of energy storage batteries, mainly including: historical abnormal cases, field technical documents, real-time new data, etc. Among them, historical abnormal cases come from abnormal data recorded during the operation of the energy storage system, and the main contents are: abnormal characteristic values ​​(such as voltage, SOC changes), abnormal cause analysis, corresponding solutions, etc. The field technical documents come from battery operation manuals, technical standards, research papers and experience summaries of field experts, and the main contents are: professional knowledge about the operating characteristics of energy storage batteries, fault handling procedures and battery performance. The real-time new data comes from the abnormal window and its characteristic values ​​output by the sliding window statistical feature detection module, and the main contents are: potential abnormal fragments detected in real time and related descriptions. These data are sorted and labeled according to categories to ensure clear and structured content, providing high-quality input for subsequent retrieval and generation.

[0072] The key to building a knowledge database is to standardize and vectorize the raw data to support efficient retrieval. The specific processing flow may include: data cleaning, data standardization, and data structuring. Among them, data cleaning includes: removing redundant or invalid data, such as incomplete abnormal records or domain documents unrelated to energy storage batteries; processing text data into sentences to ensure that each sentence independently expresses complete semantics. Data standardization includes: normalizing numerical data (such as voltage, current, SOC) to facilitate unified modeling with other features; performing preprocessing operations such as word segmentation, stop word removal, and spelling correction on text data. Data structuring includes: organizing data in a fixed format, such as: {"type": "abnormal case", "feature": {"voltage": 2.5, "SOC":}, "cause": "overdischarge", "solution": "stop discharging and recharge to the normal range"}, ensuring that all data have clear field names for subsequent vectorization processing.

[0073] Next, in one embodiment, a method for analyzing abnormalities of energy storage batteries based on retrieval enhanced generation may also include a process of vectorizing data, and the specific process includes: determining text data and numerical data in the target data, and selecting embedding methods corresponding to the text data and numerical data respectively; using the embedding method to vectorize the text data and numerical data respectively to obtain generated vector data; verifying the generated vector data, and storing the verified vector data in a vector database to obtain a knowledge database.

[0074] After obtaining the target data, it needs to be converted into a vector representation that can be used for efficient retrieval. Specifically, in this embodiment, for text data, the computer device can use a sentence vector model (text2vec-paragraphs-chinese) fine-tuned based on domain knowledge as an embedding model; for numerical features (such as voltage changes, SOC), the computer device can use a specific feature embedding method or directly merge it with text embedding.

[0075] Next, the computer device can input text data (such as exception descriptions and solutions) into the embedding model to generate a vector representation of fixed dimensions; for numerical data, the computer device can first normalize it and then generate a composite vector by simple embedding or combining it with the text context. Then verify whether the generated vector accurately expresses the semantics or characteristics of the original data; and evaluate the embedding quality through similarity calculation (such as cosine similarity).

[0076] The knowledge database constructed in this embodiment can be a Milvus vector library. Among them, Milvus is a high-performance vector database that supports efficient storage and retrieval of massive vectors. In this embodiment, Milvus is the core component of knowledge base storage and retrieval. The specific implementation process may include: initializing the Milvus database, data insertion, index construction, and dynamic update. Among them, initializing the Milvus database includes: creating a collection (Collection) for storing embedded vectors, specifying the dimension and index type of the vector; for example, creating an abnormal case library collection, setting the dimension to 768 (the default dimension of the output vector of the embedding model), and the index type to HNSW (supporting efficient similarity retrieval). Data insertion includes: associating the embedded vector of each piece of knowledge data with its metadata (such as text content, numerical features), and inserting it into Milvus. Index construction includes: using the index construction function provided by Milvus to create an efficient retrieval index for the vector collection; the commonly used index types are HNSW or IVFPQ, which can be selected according to the data scale and retrieval scenario. Dynamic updates include: when new abnormal fragments or new knowledge are detected, they are vectorized and inserted into Milvus; ensuring that the new data can participate in retrieval in real time and maintaining the dynamic update capability of the knowledge base.

[0077] In this embodiment, the knowledge database also has query, maintenance, and expansion functions. Among them, the query function includes: according to the embedded vector of the input abnormal window, the knowledge entry with the highest similarity is retrieved through Milvus; the retrieval results contain the text content and context information of the relevant cases. The maintenance function includes: regularly cleaning up outdated or redundant data to maintain the efficiency and reliability of the knowledge base. The expansion function includes: supporting vectorization processing of new data types, such as images or time series vectors when combining multimodal data.

[0078] Step 206, taking the abnormal window data at the abnormal window as input, searching the knowledge database for the corresponding abnormal data and context information of the energy storage battery, and combining the abnormal data and context information of the energy storage battery with the statistical feature vector of the abnormal window to obtain combined data.

[0079] The computer device can use the statistical feature vector of the abnormal window as input to match relevant historical cases and technical documents from the knowledge database; then, it can combine the retrieval results and the abnormal context to generate a detailed abnormal analysis report through a generation model, including abnormal feature description, cause speculation and processing suggestions.

[0080] By deeply analyzing the abnormal windows marked by sliding window statistical feature detection, combining the characteristic values ​​of the abnormal windows with relevant cases and domain knowledge retrieved from the knowledge database, an abnormal cause analysis and corresponding solutions are generated. The organic combination of retrieval and generation models not only achieves efficient information matching, but also provides highly interpretable output.

[0081] Specifically, the process of retrieval enhancement to generate RGA can include feature vectorization and retrieval, context construction of retrieval results, generative model reasoning and report generation, wherein feature vectorization and retrieval include: the abnormal window marked by the sliding window statistical feature detection module is processed by feature vectorization and input into the knowledge base for retrieval; the retrieval returns the most relevant historical cases and domain knowledge for the current abnormal window. The context construction of the retrieval results includes: combining the retrieval results with the statistical features of the abnormal window to generate a structured prompt; the prompt is the key to generative model reasoning, providing abnormal context and task instructions. Generative model reasoning and report generation include: using a large language model (QWen1.5-32B) to generate an abnormal analysis report based on the prompt content; the report content includes abnormal feature description, cause speculation and detailed solution suggestions.

[0082] In one embodiment, a method for analyzing energy storage battery anomalies based on retrieval-enhanced generation may also include a process of feature vectorization and retrieval, the specific process including: encoding the abnormal window data using an embedding model, converting the abnormal window data into a vector representation; inputting the vector representation into a knowledge database, and using similarity to retrieve the energy storage battery abnormal data and context information most relevant to the abnormal window data.

[0083] In this embodiment, the computer device can convert the abnormal window data (such as average value, change rate, jump amplitude, etc.) output by the sliding window statistical feature detection into a vector representation to support knowledge base retrieval. Then, the abnormal feature can be encoded using an embedding model to generate a vector of fixed length. If the feature is numerical (such as voltage change), a composite vector can be generated by direct embedding or combining with text context.

[0084] Among them, the vector of the abnormal window is input into the Milvus vector database, and similarity retrieval can be used to find the historical cases and domain documents most relevant to the current abnormality. The retrieved results can include: characteristic values, causes and solutions of historical abnormal cases, and content fragments in domain documents related to the current abnormality (such as specific instructions in technical manuals). For example, input the characteristic values ​​of the abnormal window, including: voltage average value 2.5V, SOC change rate 30%, jump amplitude 1.2; the retrieval results obtained through the knowledge database can be: Case A: The voltage is too low may be caused by over-discharge, and the solution is to stop discharging and charging; Document fragment: Over-discharge may cause irreversible damage, it is recommended to charge to a safe range in time.

[0085] After the retrieval is completed, it is necessary to combine the retrieved content with the characteristic value of the abnormal window to generate the input prompt of the generation model. Specifically, in this embodiment, the context construction includes the following steps: Abnormal description: briefly describe the statistical characteristics of the abnormal window, and clarify the main manifestations of the abnormal data. For example, the abnormal window characteristics of the current energy storage battery operation data are as follows: voltage average value: 2.5V, jump amplitude: 1.2, SOC change rate: 30%; Retrieval result integration: integrate the historical cases returned by the retrieval and the relevant field knowledge fragments into structured information as supplementary content of the prompt, such as the retrieved related cases and knowledge: Case A: Low voltage may be caused by over-discharge. The solution is to stop discharging and charge the document fragment: Over-discharge may cause irreversible damage to the battery. It is recommended to charge it to a safe range in time; Generate task instructions: clarify to the generation model Task objectives, indicating the content that needs to be analyzed and the format of the output. For example, please combine the above abnormal features and search content to analyze the cause of this abnormality and propose specific solutions; Complete Prompt example: Combine the above content to form a complete Prompt: The abnormal window characteristics of the current energy storage battery operation data are as follows: voltage average: 2.5V, jump amplitude: 1.2, SOC change rate: 30%, related cases and knowledge retrieved: Case A: Low voltage may be caused by over-discharge. The solution is to stop discharging and charge. Document fragment: Over-discharge may cause irreversible damage to the battery. It is recommended to charge it to a safe range in time. Please combine the above abnormal features and search content to analyze the cause of this abnormality and propose specific solutions.

[0086] In one embodiment, a method for analyzing anomalies of energy storage batteries based on retrieval enhanced generation may also include a data combination process, the specific process including: determining an abnormal performance description corresponding to a statistical feature vector of an abnormal window; integrating abnormal data of the energy storage battery into structured information; obtaining a task generation instruction, and according to the task generation instruction, extracting information from the abnormal performance description and the structured information to form prompt information, and obtaining combined data.

[0087] In one embodiment, the process of generating model reasoning and report generation may also be included, and the specific process includes: inputting prompt information into the model to be trained, performing reasoning and question answering through the model to be trained, completing model training to obtain a generated model; adjusting the generated model according to different types of prompt information, and adaptively expanding the generated model according to the search results in the knowledge database to obtain the final generated model.

[0088] Among them, the generation model can use large language models such as QWen, combined with energy storage battery field data for reasoning answers. Specifically, after entering the complete prompt, the model performs reasoning answers. In this embodiment, the generated result example may include several parts: abnormal description, cause analysis, and solution suggestions. Among them, the abnormal description includes: the energy storage battery operating voltage is detected to be too low, the SOC change rate is 30%, and over-discharge may occur; the cause analysis includes: according to historical cases, the average voltage is lower than 2.7V and the SOC drops rapidly, which is usually caused by over-discharge. In addition, the retrieved domain knowledge shows that over-discharge may cause irreversible damage to the battery; the solution suggestions include: A. Immediately stop discharging and recharge to the normal range; B. Check whether there is abnormal contact in the discharge circuit; configure the battery management system (BMS) to avoid the voltage being lower than 2.7V.

[0089] In this embodiment, the prompt can be flexibly adjusted. Specifically, a variety of prompt templates can be designed according to different abnormality types (such as over-discharge, temperature abnormality, battery aging, etc.). For example: over-discharge abnormality: focus on voltage, SOC and time series characteristics; temperature abnormality: combine ambient temperature and internal temperature change analysis.

[0090] Among them, when the content of the knowledge database is updated, new cases and domain knowledge can be directly included in the analysis, and the knowledge database and generation model can continuously optimize the generation quality based on user feedback.

[0091] In this embodiment, the sliding window statistical features are closely combined with the domain knowledge base to achieve efficient and intelligent abnormality analysis, providing strong technical support for the operation and maintenance of energy storage batteries.

[0092] Step 208, input the combined data into the generation model, and generate a battery abnormality report corresponding to the energy storage battery through the generation model.

[0093] Computer equipment can generate corresponding battery abnormality reports, specifically, structured and readable abnormality analysis reports, convert the results of retrieval enhanced generation (RAG) analysis into a structured report format, provide clear and practical abnormality analysis conclusions and processing suggestions, and provide real-time and highly interpretable abnormality detection and processing solutions, which are convenient for energy storage battery operation and maintenance personnel to quickly understand and take action, and meet the engineering practice needs of energy storage batteries.

[0094] In one embodiment, a method for analyzing anomalies in an energy storage battery based on retrieval-enhanced generation is provided, which may also include a process for generating a battery anomaly report, the specific process including: inputting the combined data into a generation model, determining battery anomaly description information based on the energy storage battery anomaly data and context information through the generation model; determining summary information corresponding to the energy storage battery anomaly data according to the battery anomaly description information; analyzing the cause of the battery anomaly according to the battery anomaly description information to obtain an analysis result; determining corresponding anomaly resolution suggestions based on the analysis result; and generating a battery anomaly report using the battery anomaly description information, summary information, analysis results, and anomaly resolution suggestions.

[0095] Among them, the abnormal analysis report can be presented in a structured manner, and the key content parts may include: abnormal description, cause analysis, solution suggestions, and search result summary. Among them, the abnormal description includes: describing the specific characteristics of the current abnormal window and its background information; including the time range of the abnormality, abnormal characteristic values ​​(such as voltage, SOC, jump amplitude, etc.) and abnormality category. For example, the abnormal description includes: time range: 2024-11-2110:00 to 10:05; voltage average value: 2.5V (lower than the normal range), SOC change rate: 30% (beyond normal fluctuations); abnormal category: over-discharge risk.

[0096] The cause analysis includes: combining the retrieved historical cases and domain knowledge to make detailed inferences on the possible causes of the anomaly; including direct causes (such as over-discharge) and potential related factors (such as poor line contact, battery aging); for example, the cause analysis: the voltage was detected to be lower than 2.7V, and the SOC dropped rapidly, which may be caused by over-discharge; the search results show that similar features have appeared in Case A, and the reason is long-term high-rate discharge; at the same time, the technical documentation mentions that poor contact of the discharge circuit may further aggravate the voltage drop.

[0097] The solution suggestions include: providing actionable solutions, with specific operation steps, so that engineers can directly apply them; including short-term solutions (such as stopping discharge and recharging) and long-term optimization strategies (such as adjusting the parameters of the battery management system); giving examples, the solution suggestions include: immediately stop discharging and recharge to a safe range (the recommended voltage is greater than 3.0V); check whether there is poor contact in the discharge circuit to ensure a stable connection; adjust the battery management system (BMS) and set the minimum discharge voltage threshold to 2.7V.

[0098] The search result summary includes: a summary of relevant cases and documents retrieved from the knowledge base is listed to ensure traceability of the report content; for example, the search result summary: Case A: The voltage is lower than 2.7V, and the SOC drops rapidly. The reason is: over-discharge; Solution: Charge to the normal range in time; Document fragment: Over-discharge may cause irreversible damage to the battery. It is recommended to stop discharging and charge immediately.

[0099] In this embodiment, in order to improve the practicality and applicability of the report, the abnormal analysis report can support multiple output forms. In this embodiment, the output form may include a structured text report and a data file, wherein the structured text report is output in a clear text structure, which is suitable for direct viewing and recording. For example, the abnormal analysis report: time: 2024-11-2110:00 to 10:05; abnormal description: the voltage is lower than the normal range, and the SOC drops rapidly; cause analysis: over-discharge may cause this abnormality; solution suggestion: stop discharging, charge immediately to a safe range; check the discharge circuit. The data file is output in a standardized data format (such as JSON or CSV) to facilitate system integration and automated management; for example, JSON output: {"time":"2024-11-2110:00to10:05","description":"The voltage is lower than the normal range, and the SOC drops rapidly.","analysis":"Over-discharge may cause this abnormality.","suggestions":["Stop discharging, charge immediately to a safe range.","Check the discharge circuit."]}.

[0100] In one embodiment, a method for analyzing energy storage battery anomalies based on retrieval enhancement generation may also include a process of report pushing and displaying, and the specific process includes: outputting and pushing the battery anomaly report; visualizing the battery anomaly report and displaying it on a display interface.

[0101] Computer equipment can send abnormal analysis results to operation and maintenance personnel in real time through API or message push (such as email, SMS), improving response speed. For example, the abnormal analysis notification content may include: time: 2024-11-21 10:00 to 10:05; abnormality: voltage is too low, possible over-discharge; suggestion: stop discharging and check the line.

[0102] In this embodiment, a visual report in the form of a chart or dashboard is also provided to intuitively present abnormal characteristics and analysis results, wherein the chart content includes an abnormal value trend chart, a characteristic distribution chart, the execution status of the processing suggestion, etc.

[0103] In one embodiment, the technical implementation process of report generation may also be included, which may specifically include several parts including generation model output processing, report format template, and multi-channel output support. Among them, generation model output processing includes: the results generated by LLM are post-processed and converted into a structured format to ensure that the report content is clear and easy to output. For example, the generated text paragraphs are decomposed into three fields: "Description", "Cause Analysis", and "Suggestions", and formatted. The report format template includes: based on the predefined template, the generated content is filled into the report framework; the template design supports flexible expansion, such as dynamically adjusting the template structure for different exception types. Multi-channel output support includes: according to user needs or system settings, support for real-time switching of multiple output channels (such as text, files, notifications).

[0104] In this embodiment, it is possible to provide operation and maintenance personnel with clear and practical abnormality analysis results in various forms, effectively improve the efficiency of abnormality handling, reduce potential risks, and fully support the safe and stable operation of energy storage batteries. Through the sliding window statistical feature detection technology, the real-time operation data of the energy storage battery is dynamically monitored to quickly identify potential abnormalities. Compared with the traditional static rule method, this technology can extract local features in real time according to data changes, improve the adaptability and sensitivity of detection; apply the retrieval enhancement generation technology to the abnormal analysis of energy storage batteries, combined with the dynamically constructed domain knowledge base, it can retrieve key information from similar historical cases, and generate highly explanatory abnormal cause analysis and specific solutions; through real-time retrieval and automatic knowledge base update mechanism, it can quickly adapt to new abnormal patterns and improve flexibility and scalability.

[0105] It should be understood that, although the various steps in the above-mentioned flow chart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the above-mentioned flow chart may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0106] In one embodiment, Figure 4 As shown, a system for analyzing abnormality of energy storage batteries based on retrieval enhancement generation is provided, including: a sliding window statistical feature detection module 410, a knowledge base construction module 420, a retrieval enhancement generation analysis module 430 and a report generation and output module 440, wherein:

[0107] The sliding window statistical feature detection module 410 is used to obtain the real-time operation data of the energy storage battery, extract the statistical feature vector in the real-time operation data, and perform sliding window segmentation according to the statistical feature vector according to the segmentation condition to determine the abnormal window;

[0108] The knowledge base construction module 420 is used to collect historical abnormal data and experience knowledge corresponding to the energy storage battery, and perform standardization and vectorization on the historical abnormal data and experience knowledge to construct a knowledge database;

[0109] The retrieval enhancement generation analysis module 430 is used to take the abnormal window data at the abnormal window as input, search for the corresponding energy storage battery abnormal data and context information from the knowledge database, and combine the energy storage battery abnormal data with the statistical feature vector of the abnormal window to obtain the combined data;

[0110] The report generation and output module 440 is used to input the combined data into the generation model, and generate a battery abnormality report corresponding to the energy storage battery through the generation model.

[0111] In one embodiment, the sliding window statistical feature detection module 410 is also used to determine the acquisition time of the statistical feature vector, and use the acquisition time as a division condition; the statistical feature vector is divided into sliding windows according to the division condition, and adjacent sliding windows have an overlap, and each window is obtained by division; the sliding window statistical features corresponding to each window are obtained respectively, and the sliding window statistical features are compared with a dynamic threshold to obtain a comparison result; the abnormal window is determined according to the comparison result; wherein the dynamic threshold is dynamically updated according to the data in the knowledge database.

[0112] In one embodiment, the knowledge base construction module 420 is also used to perform data cleaning on historical abnormal data and empirical knowledge, and perform sentence segmentation operations on the text data after the information to obtain cleaned data; normalize the numerical data in the cleaned data, and perform preprocessing operations on the text data in the cleaned data to obtain standardized data; organize the standardized data according to a fixed format to obtain target data, and vectorize the target data to construct a knowledge database.

[0113] In one embodiment, the knowledge base construction module 420 is also used to determine the text data and numerical data in the target data, and select the embedding methods corresponding to the text data and numerical data respectively; use the embedding methods to vectorize the text data and numerical data respectively to obtain generated vector data; verify the generated vector data, and store the verified vector data in the vector database to obtain a knowledge database.

[0114] In one embodiment, the retrieval enhancement generation analysis module 430 is also used to encode the abnormal window data using an embedding model, converting the abnormal window data into a vector representation; inputting the vector representation into a knowledge database, and using similarity to retrieve the energy storage battery abnormal data and context information that are most relevant to the abnormal window data.

[0115] In one embodiment, the retrieval enhancement generation analysis module 430 is also used to determine the abnormal performance description corresponding to the statistical feature vector of the abnormal window; integrate the abnormal data of the energy storage battery into structured information; obtain the task generation instruction, and according to the task generation instruction, extract information from the abnormal performance description and the structured information to form prompt information, and obtain the combined data.

[0116] In one embodiment, the report generation and output module 440 is also used to input prompt information into the model to be trained, perform reasoning questions and answers through the model to be trained, complete model training and obtain a generated model; adjust the generated model according to different types of prompt information, and adaptively expand the generated model according to the search results in the knowledge database to obtain the final generated model.

[0117] In one embodiment, the report generation and output module 440 is also used to input the combined data into the generation model, and determine the battery abnormality description information based on the energy storage battery abnormality data and context information through the generation model; determine the summary information corresponding to the energy storage battery abnormality data according to the battery abnormality description information; analyze the cause of the battery abnormality according to the battery abnormality description information to obtain the analysis result; determine the corresponding abnormality solution suggestion based on the analysis result; and generate a battery abnormality report with the battery abnormality description information, summary information, analysis result, and abnormality solution suggestion.

[0118] In one embodiment, the report generation and output module 440 is also used to output and push the battery abnormality report; visualize the battery abnormality report and display it on the display interface.

[0119] like Figure 5 As shown, in one embodiment, a system for analyzing abnormalities of energy storage batteries based on retrieval enhancement generation is provided, including: a sliding window statistical feature detection module, a knowledge base construction module, a retrieval enhancement generation analysis module, and a report generation and output module. Figure 5 As shown, the real-time operation data of the energy storage battery is input into the data input layer. After the sliding window statistical feature detection module performs sliding window data segmentation, statistical feature extraction, and dynamic threshold judgment, the abnormal window is transmitted to the retrieval enhancement generation analysis module to perform abnormal feature vector extraction, related case retrieval, context fusion, and generation model analysis; wherein, the knowledge base construction module contains historical abnormal cases, field technical documents, and related experience libraries, and these data are vectorized to realize related case retrieval; the report generation and output module generates abnormal feature descriptions, cause speculations, and processing suggestions, and generates structured reports, thereby feeding back and updating the data in the knowledge database.

[0120] In this embodiment, the various modules cooperate with each other to quickly detect potential anomalies through sliding window statistical features, and combine the retrieval and generation capabilities of the knowledge base to achieve efficient analysis and explanatory output of energy storage battery anomalies, which not only ensures the accuracy of detection, but also improves adaptability and scalability.

[0121] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for analyzing abnormalities of an energy storage battery based on retrieval enhancement generation is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0122] Those skilled in the art will understand that Figure 6The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0123] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of the energy storage battery abnormality analysis method based on retrieval enhanced generation are implemented.

[0124] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the energy storage battery abnormality analysis method based on retrieval enhancement generation are implemented.

[0125] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0126] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, 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.

[0127] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for analyzing abnormality of energy storage batteries based on retrieval enhancement generation, characterized in that: The method comprises: Acquire real-time operation data of the energy storage battery, extract statistical feature vectors in the real-time operation data, and perform sliding window segmentation according to the statistical feature vectors according to the segmentation conditions to determine the abnormal window; Collecting historical abnormal data and experience knowledge corresponding to the energy storage battery, and performing standardization and vectorization on the historical abnormal data and experience knowledge to build a knowledge database; Taking the abnormal window data at the abnormal window as input, searching the corresponding energy storage battery abnormal data and context information from the knowledge database, and combining the energy storage battery abnormal data and context information with the statistical feature vector of the abnormal window to obtain combined data; The combined data is input into a generation model, and a battery abnormality report corresponding to the energy storage battery is generated through the generation model.

2. The energy storage battery abnormality analysis method based on retrieval enhancement generation according to claim 1 is characterized in that: Performing sliding window segmentation according to the statistical feature vector and the segmentation condition to determine the abnormal window includes: Determining the acquisition time of the statistical feature vector, and using the acquisition time as a division condition; The statistical feature vector is segmented into sliding windows according to the segmentation condition, and adjacent sliding windows have an overlap, so as to obtain individual windows; Respectively acquiring sliding window statistical features corresponding to each of the windows, and comparing the sliding window statistical features with a dynamic threshold to obtain a comparison result; Determine an abnormal window according to the comparison result; The dynamic threshold is dynamically updated according to the data in the knowledge database.

3. The energy storage battery abnormality analysis method based on retrieval enhancement generation according to claim 1 is characterized in that: The historical abnormal data and experience knowledge are standardized and vectorized to construct a knowledge database, including: Perform data cleaning on the historical abnormal data and experience knowledge, and perform sentence segmentation on the text data behind the information to obtain cleaned data; Normalizing the numerical data in the cleaned data, and preprocessing the text data in the cleaned data to obtain standardized data; The standardized data is organized according to a fixed format to obtain target data, and the target data is vectorized to construct a knowledge database.

4. The energy storage battery abnormality analysis method based on retrieval enhanced generation according to claim 3 is characterized in that: The target data is vectorized to construct a knowledge database, including: Determine the text data and the numerical data in the target data, and select embedding methods corresponding to the text data and the numerical data respectively; Use the embedding method to vectorize the text data and the numerical data to obtain generated vector data; The generated vector data is verified, and the verified vector data is stored in a vector database to obtain a knowledge database.

5. The energy storage battery abnormality analysis method based on retrieval enhancement generation according to claim 1 is characterized in that: Taking the abnormal window data at the abnormal window as input, searching the corresponding energy storage battery abnormal data and context information from the knowledge database, including: Encoding the abnormal window data using an embedding model, and converting the abnormal window data into a vector representation; The vector representation is input into the knowledge database, and the energy storage battery abnormal data and context information most relevant to the abnormal window data are retrieved using similarity.

6. The energy storage battery abnormality analysis method based on retrieval enhancement generation according to claim 1 is characterized in that: The abnormal data of the energy storage battery is combined with the statistical feature vector of the abnormal window to obtain combined data, including: Determine an abnormal performance description corresponding to the statistical feature vector of the abnormal window; Integrate the energy storage battery abnormal data into structured information; A task generation instruction is obtained, and according to the task generation instruction, information is extracted from the abnormal performance description and the structured information to form prompt information, and combined data is obtained.

7. The energy storage battery abnormality analysis method based on retrieval enhancement generation according to claim 6 is characterized in that: The method further comprises: Input the prompt information into the model to be trained, perform reasoning questions and answers through the model to be trained, complete model training and obtain a generated model; The generation model is adjusted according to different types of the prompt information, and the generation model is adaptively expanded according to the search results in the knowledge database to obtain a final generation model.

8. The energy storage battery abnormality analysis method based on retrieval enhancement generation according to claim 1 is characterized in that: The combined data is input into a generation model, and a battery abnormality report corresponding to the energy storage battery is generated by the generation model, including: Inputting the combined data into a generation model, and determining battery abnormality description information based on the energy storage battery abnormality data and context information through the generation model; Determine summary information corresponding to the energy storage battery abnormality data according to the battery abnormality description information; Analyze the cause of the battery abnormality according to the battery abnormality description information to obtain an analysis result; Determine corresponding abnormality resolution suggestions based on the analysis results; The battery anomaly description information, summary information, analysis results, and anomaly resolution suggestions are used to generate a battery anomaly report.

9. The energy storage battery abnormality analysis method based on retrieval enhancement generation according to claim 1 is characterized in that: The method further comprises: Outputting and pushing the battery abnormality report; The battery abnormality report is visualized and displayed on a display interface.

10. A retrieval enhanced generation energy storage battery abnormality analysis system, characterized in that: The system comprises: A sliding window statistical feature detection module is used to obtain real-time operation data of the energy storage battery, extract statistical feature vectors in the real-time operation data, and perform sliding window segmentation according to the statistical feature vectors according to the segmentation conditions to determine the abnormal window; A knowledge base construction module, used to collect historical abnormal data and experience knowledge corresponding to the energy storage battery, and perform standardization and vectorization on the historical abnormal data and experience knowledge to construct a knowledge database; A retrieval enhancement generation analysis module is used to take the abnormal window data at the abnormal window as input, search for the corresponding energy storage battery abnormal data and context information from the knowledge database, and combine the energy storage battery abnormal data with the statistical feature vector of the abnormal window to obtain the combined data; The report generation and output module is used to input the combined data into the generation model, and generate a battery abnormality report corresponding to the energy storage battery through the generation model.

Images