Energy storage device fault analysis method and device, computer device and storage medium

Abstract

The application provides a fault analysis method and device of an energy storage device, a computer device and a storage medium. The fault analysis method comprises the following steps: obtaining basic data and running time sequence data of the energy storage device, and extracting time domain features and frequency domain features of the running time sequence data; according to the basic data, the running time sequence data, the time domain features and the frequency domain features, searching a historical fault database to determine a plurality of first historical fault cases, wherein the historical fault database comprises historical fault cases split according to fault time ranges; determining a plurality of second historical fault cases according to a multi-dimensional feature matching degree between each first historical fault case and the energy storage device; generating prompt information according to the basic data, the running time sequence data and the plurality of second historical fault cases, inputting the prompt information into a fault analysis model, and outputting a fault analysis result of the energy storage device. In this way, the interference information in the prompt information is reduced, the accuracy and comprehensiveness of the prompt information are improved, and the accuracy of fault diagnosis is improved.

Description

Technical Field

[0001] This invention relates to the field of energy storage fault analysis technology, and more specifically, to a fault analysis method and apparatus for energy storage devices, a computer device, and a storage medium. Background Technology

[0002] Since large language models do not possess specific operational knowledge of energy storage devices, nor do they understand the physical meaning of the operational sequence data of energy storage devices, when large language models are used for fault diagnosis of energy storage devices, directly inputting the original operational sequence data of energy storage devices into the large language model in text form will result in extremely poor fault diagnosis results.

[0003] Based on this, in related technologies, when large language models are used for fault diagnosis of energy storage equipment, the following approach is typically adopted: knowledge documents such as maintenance manuals and fault case texts are sliced ​​into text segments of fixed length or paragraphs; the sliced ​​text blocks are converted into vectors and stored in a vector database; when a user asks a question, the question is vectorized, and similar text blocks are retrieved from the vector database; the retrieved text blocks are used as context to incorporate prompt information to guide the large language model to answer.

[0004] However, the above-mentioned fault diagnosis schemes have shortcomings such as text slicing, single retrieval mode, insufficient ability of vector pairs to represent temporal features, and simple recall result processing, which result in low accuracy of the obtained prompt information and reduce the accuracy of fault diagnosis by large language models. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art or related art.

[0006] Therefore, the first aspect of the present invention is to provide a fault analysis method for energy storage devices.

[0007] A second aspect of the present invention is to provide a fault analysis device for energy storage equipment.

[0008] A third aspect of the invention is to provide a computer device.

[0009] A fourth aspect of the present invention is to provide a readable storage medium.

[0010] In view of this, according to a first aspect of the present invention, a fault analysis method for energy storage devices is proposed, comprising: acquiring basic data and runtime sequence data of the energy storage device, and extracting time-domain and frequency-domain features of the runtime sequence data, wherein the basic data includes at least one or a combination of the following: device number, device model, hardware version, device runtime, fault time range, and fault type label of the energy storage device; retrieving a historical fault database based on the basic data, runtime sequence data, time-domain features, and frequency-domain features to determine multiple first historical fault cases, wherein the historical fault database includes historical fault cases segmented according to fault time range; sorting the multiple first historical fault cases according to the matching degree between each first historical fault case and the multidimensional features of the energy storage device to determine multiple second historical fault cases among the multiple first historical fault cases; generating prompt information based on the basic data, runtime sequence data, and multiple second historical fault cases, inputting the prompt information into a fault analysis model, and outputting the fault analysis results of the energy storage device.

[0011] The technical solution of the fault analysis method for energy storage devices provided by this invention can be executed by a computer device, a fault analysis device, or something else depending on actual usage requirements; no specific limitation is made here. To more clearly describe the fault analysis method for energy storage devices provided by this invention, the following description uses a fault analysis device as the executing entity.

[0012] Specifically, in the fault analysis method for energy storage devices provided by this invention, when diagnosing faults in energy storage devices, the fault analysis device acquires basic data of the energy storage device. This basic data includes at least one or a combination of the following: device number, device model, hardware version, device runtime, fault time range, and fault type label. Furthermore, the fault analysis device acquires runtime sequence data of the energy storage device and extracts its time-domain and frequency-domain features. Based on this, the fault analysis device retrieves a pre-constructed historical fault database according to the basic data, runtime sequence data, time-domain features, and frequency-domain features. The historical fault database includes historical fault cases segmented according to fault time ranges. By retrieving the historical fault database, the fault analysis device identifies multiple first historical fault cases within the database. Based on this, the fault analysis device sorts multiple first historical fault cases according to the matching degree of each first historical fault case with the multidimensional features of the energy storage device, and selects multiple second historical fault cases from the multiple first historical fault cases. Then, based on the basic data, runtime sequence data and multiple second historical fault cases, it generates prompt information for the fault analysis model and inputs the prompt information into the fault analysis model, so that the fault analysis model outputs the fault analysis results of the energy storage device.

[0013] The fault analysis method for energy storage devices provided by this invention, on the one hand, improves the accuracy of the recalled historical fault data (i.e., the first historical fault case) by retrieving historical fault data from a historical fault database based on the basic data, operational sequence data, and their time-domain and frequency-domain characteristics of the energy storage device. The historical fault database includes historical fault cases segmented according to fault time ranges, thus improving the completeness of the recalled historical fault data. On the other hand, it refines and filters the recalled historical fault data based on multi-dimensional feature matching, improving the adaptability of the final adopted historical fault data (i.e., the second historical fault case) to the current operating condition of the energy storage device. Thus, when performing fault diagnosis based on the fault analysis model, it reduces interfering information in the generated prompts, improves the accuracy and comprehensiveness of the prompts, and ultimately enhances the accuracy of fault diagnosis.

[0014] In some technical solutions, optionally, the historical fault database includes a historical fault vector library and a historical fault text library. Before retrieving the historical fault database, the fault analysis method further includes: acquiring multiple historical fault case data of different energy storage devices, each historical fault case data including at least one or a combination of the following: device number, device model, hardware version, fault type label, fault evolution information, fault diagnosis information, and fault handling information; generating corresponding historical fault description text based on each historical fault case data to obtain a historical fault text library; splitting each historical fault case data into early-stage fault data, mid-stage fault data, and late-stage fault data according to the fault time range; generating corresponding historical fault vectors based on the early-stage fault data, mid-stage fault data, and late-stage fault data to obtain a historical fault vector library.

[0015] In this technical solution, the aforementioned historical fault database may specifically include a historical fault vector database and a historical fault text database.

[0016] Based on this, before retrieving the historical fault database, the fault analysis device also acquires multiple historical fault case data from different energy storage devices. Each historical fault case data includes at least one or a combination of the following: device number, device model, hardware version, fault type tag, fault evolution information, fault diagnosis information, and fault handling information. Based on this, the fault analysis device generates corresponding historical fault description text for each historical fault case data, resulting in the aforementioned historical fault text library. Furthermore, the fault analysis device splits each historical fault case data into early-stage fault data, mid-stage fault data, and late-stage fault data according to the fault time range. Then, based on the early-stage, mid-stage, and late-stage fault data, it generates corresponding historical fault vectors for each, resulting in the aforementioned historical fault vector library. This structured splitting of historical fault case data based on the fault time range improves the completeness of the data splitting results, facilitating the comprehensiveness of subsequent historical fault data retrieval. Simultaneously, it retains the complete temporal characteristic description of the historical fault case data, making it easier to combine semantic information to further improve the accuracy of the retrieved historical fault data.

[0017] In some technical solutions, optionally, based on basic data, runtime sequence data, time-domain features, and frequency-domain features, a historical fault database is retrieved to determine multiple first historical fault cases. This includes: retrieving a historical fault vector database based on basic data, time-domain features, and frequency-domain features to obtain a first retrieval result, which includes multiple third historical fault cases; performing semantic retrieval on a historical fault text database based on runtime sequence data to obtain a second retrieval result, which includes multiple fourth historical fault cases; and deduplicating the multiple third and fourth historical fault cases to obtain multiple first historical fault cases.

[0018] In this technical solution, during the process of retrieving historical fault databases and identifying multiple first historical fault cases, specifically, the fault analysis device searches the historical fault vector database based on basic data, time-domain features, and frequency-domain features to obtain a first search result, which includes multiple third historical fault cases. Furthermore, the fault analysis device performs semantic retrieval on the historical fault text database based on runtime sequence data to obtain a second search result, which includes multiple fourth historical fault cases. Based on this, the fault analysis device further deduplicates the multiple third and fourth historical fault cases to obtain multiple first historical fault cases. Thus, by using both vector retrieval and semantic retrieval, historical fault data is retrieved through a dual-path retrieval system, improving the accuracy of the retrieved historical fault data.

[0019] In some technical solutions, optionally, a first search result is obtained by retrieving a historical fault vector library based on basic data, time-domain features, and frequency-domain features. This includes: generating a time-series feature vector of the energy storage device based on the time-domain and frequency-domain features; the time-series data includes the parameter values ​​of the energy storage device's operating parameters changing over time; the operating parameters include at least one or a combination of the following: cell voltage, cell current, cell temperature, cell charge, and inverter efficiency of the energy storage device; and the time-domain features include at least one or a combination of the following: mean, variance, standard deviation, skewness, kurtosis, peak value, valley value, rate of change, and cumulative value of the time-series data; filtering the historical fault vector library based on the basic data to obtain a candidate vector library; and retrieving the candidate vector library based on the vector similarity between the time-series feature vector and each historical fault vector in the candidate vector library to obtain the first search result.

[0020] In this technical solution, during the process of retrieving the historical fault vector database to obtain the first retrieval result, specifically, the runtime sequence data includes the parameter values ​​of the energy storage device's operating parameters changing over time. The operating parameters include at least one or a combination of the following: cell voltage, cell current, cell temperature, cell charge, and inverter efficiency. The time-domain features include at least one or a combination of the following: mean, variance, standard deviation, skewness, kurtosis, peak value, valley value, rate of change, and cumulative value of the runtime sequence data. The fault analysis device generates a time-series feature vector of the energy storage device based on the time-domain and frequency-domain features, and filters the historical fault vector database based on the basic data to obtain a candidate vector database. On this basis, the fault analysis device further calculates the vector similarity between the time-series feature vector and each historical fault vector in the candidate vector database, and retrieves the candidate vector database based on the calculated vector similarity to obtain the first retrieval result. In this way, the historical fault vector library is first filtered through basic data, and then the candidate vector library is searched based on vector similarity. This reduces the search space for vector retrieval and can filter fault vectors with large differences, thereby improving the speed and accuracy of subsequent vector retrieval.

[0021] In some technical solutions, optionally, semantic retrieval is performed on a historical fault text library based on runtime sequence data to obtain a second retrieval result, including: generating fault phenomenon description information of the energy storage device based on runtime sequence data, wherein the fault phenomenon description information is used to describe the evolution trend of the operating parameters of the energy storage device; and performing semantic retrieval on the historical fault text library based on the semantic relevance between the fault phenomenon description information and each historical fault description text in the historical fault text library to obtain a second retrieval result.

[0022] In this technical solution, during the process of semantically searching the historical fault text library to obtain the second search result, specifically, the fault analysis device generates fault phenomenon description information of the energy storage device based on runtime sequence data. This fault phenomenon description information describes the evolution trend of the energy storage device's operating parameters. Based on this, the fault analysis device further analyzes the semantic correlation between the fault phenomenon description information and each historical fault description text in the historical fault text library, and performs a semantic search on the historical fault text library based on the analyzed semantic correlation to obtain the second search result. Thus, by leveraging the semantic correlation between the fault phenomenon description information of the energy storage device and the temporal feature descriptions of historical fault case data, the device assists in recalling historical fault data, further improving the accuracy of the recalled historical fault data.

[0023] In some technical solutions, optionally, multiple first historical fault cases are sorted based on the multi-dimensional feature matching degree between each first historical fault case and the energy storage device to determine multiple second historical fault cases among the multiple first historical fault cases. This includes: calculating the vector similarity, device model matching degree, time window matching degree, and fault type matching degree between each first historical fault case and the energy storage device based on basic data, runtime sequence data, and each first historical fault case; performing weighted calculations on the vector similarity, device model matching degree, time window matching degree, and fault type matching degree corresponding to each first historical fault case to obtain a comprehensive score for each first historical fault case; sorting the multiple first historical fault cases in descending order of comprehensive score, and determining the top N first historical fault cases as second historical fault cases, where N is a positive integer.

[0024] In this technical solution, during the process of selecting multiple second historical failure cases from multiple first historical failure cases, specifically, the failure analysis device calculates the vector similarity, equipment model matching degree, time window matching degree, and failure type matching degree between each first historical failure case and the energy storage device based on basic data, runtime sequence data, and each first historical failure case. Based on this, the failure analysis device performs a weighted calculation of the vector similarity, equipment model matching degree, time window matching degree, and failure type matching degree corresponding to each first historical failure case to obtain a comprehensive score for each first historical failure case. Based on this, the failure analysis device then sorts the multiple first historical failure cases according to their comprehensive scores from highest to lowest, and determines the top N first historical failure cases as second historical failure cases. Here, N is a positive integer. In this way, based on the multi-dimensional feature matching degree such as vector similarity, equipment model matching degree, time window matching degree and fault type matching degree between each first historical fault case and energy storage device, the recalled historical fault data is finely sorted and filtered, which improves the adaptability of the final historical fault data to the current operating condition of the energy storage device, reduces the interference information in the generated prompt information, and improves the accuracy of the prompt information.

[0025] In some technical solutions, optionally, a prompt message is generated based on basic data, runtime sequence data, and multiple second historical fault cases. This includes: obtaining a preset template for the prompt message, the preset template including device description information, information to be analyzed, and reference information; filling the device description information with basic data, filling the information to be analyzed with runtime sequence data, and filling the reference information with multiple second historical fault cases to obtain the prompt message.

[0026] In this technical solution, during the generation of prompt information, specifically, the fault analysis device acquires a preset template for the prompt information. This preset template includes equipment description information, information to be analyzed, and reference information. Based on this, the fault analysis device fills the equipment description information with basic data, the information to be analyzed with runtime sequence data, and the reference information with multiple second historical fault cases, thus obtaining the prompt information for the fault analysis model. In this way, by injecting the finely filtered historical fault data into the prompt information of the fault analysis model according to the preset template structure, it ensures that the fault analysis model obtains factual anchors before reasoning, thereby improving the accuracy of fault diagnosis by the fault analysis model.

[0027] According to a second aspect of the present invention, a fault analysis device for an energy storage device is proposed, comprising: an acquisition unit, configured to acquire basic data and runtime sequence data of the energy storage device, and extract time-domain and frequency-domain features of the runtime sequence data, wherein the basic data includes at least one or a combination of the following: device number, device model, hardware version, device runtime, fault time range, and fault type label of the energy storage device; a processing unit, configured to retrieve a historical fault database based on the basic data, runtime sequence data, time-domain features, and frequency-domain features, and determine multiple first historical fault cases, wherein the historical fault database includes historical fault cases segmented according to fault time range; the processing unit is further configured to sort the multiple first historical fault cases based on the matching degree between each first historical fault case and the multidimensional features of the energy storage device, and determine multiple second historical fault cases among the multiple first historical fault cases; the processing unit is further configured to generate prompt information based on the basic data, runtime sequence data, and multiple second historical fault cases, and input the prompt information into a fault analysis model, and output the fault analysis result of the energy storage device.

[0028] The fault analysis device for energy storage equipment provided by this invention includes an acquisition unit and a processing unit. When diagnosing faults in the energy storage equipment, the acquisition unit acquires basic data of the energy storage equipment, wherein the basic data includes at least one or a combination of the following: equipment number, equipment model, hardware version, equipment runtime, fault time range, and fault type label of the energy storage equipment. Furthermore, the acquisition unit acquires runtime sequence data of the energy storage equipment and extracts the time-domain and frequency-domain features of the runtime sequence data. Based on this, the processing unit retrieves a pre-constructed historical fault database according to the basic data, runtime sequence data, time-domain features, and frequency-domain features. The historical fault database includes historical fault cases segmented according to fault time ranges. By retrieving the historical fault database, the processing unit identifies multiple first historical fault cases in the historical fault database. Based on this, the processing unit sorts the multiple first historical fault cases according to the matching degree between each first historical fault case and the multidimensional features of the energy storage device, and selects multiple second historical fault cases from the multiple first historical fault cases. Then, based on the basic data, runtime sequence data and multiple second historical fault cases, it generates prompt information for the fault analysis model and inputs the prompt information into the fault analysis model, so that the fault analysis model outputs the fault analysis results of the energy storage device.

[0029] The fault analysis device for energy storage devices provided by this invention, on the one hand, improves the accuracy of the recalled historical fault data (i.e., the first historical fault case) by retrieving historical fault data from a historical fault database based on the basic data, operational sequence data, and their time-domain and frequency-domain characteristics of the energy storage device. The historical fault database includes historical fault cases segmented according to fault time ranges, thus improving the completeness of the recalled historical fault data. On the other hand, it refines and filters the recalled historical fault data based on multi-dimensional feature matching, improving the adaptability of the final adopted historical fault data (i.e., the second historical fault case) to the current operating condition of the energy storage device. Thus, when performing fault diagnosis based on the fault analysis model, it reduces interfering information in the generated prompts, improves the accuracy and comprehensiveness of the prompts, and ultimately enhances the accuracy of fault diagnosis.

[0030] According to a third aspect of the present invention, a computer device is provided, comprising a processor and a memory, wherein the memory stores a program or instructions executable on the processor, and the program or instructions, when executed by the processor, implement the steps of the fault analysis method for energy storage devices as described in any of the above-described technical solutions. Therefore, the computer device proposed in the third aspect of the present invention possesses all the beneficial effects of the fault analysis method for energy storage devices in any of the technical solutions of the first aspect described above, which will not be elaborated further here.

[0031] According to a fourth aspect of the present invention, a readable storage medium is provided, on which a program or instructions are stored, which, when executed by a processor, implement the fault analysis method for energy storage devices as described in any of the above-described technical solutions. Therefore, the readable storage medium proposed in the fourth aspect of the present invention possesses all the beneficial effects of the fault analysis method for energy storage devices in any of the technical solutions of the first aspect, and will not be elaborated further here.

[0032] Additional aspects and advantages of the invention will become apparent in the following description or may be learned by practice of the invention. Attached Figure Description

[0033] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0034] Figure 1 One of the flowcharts of the fault analysis method for energy storage devices according to an embodiment of the present invention is shown;

[0035] Figure 2 A second schematic flowchart of the fault analysis method for energy storage devices according to an embodiment of the present invention is shown;

[0036] Figure 3 The third flowchart illustrates the fault analysis method for energy storage devices according to an embodiment of the present invention.

[0037] Figure 4 The fourth flowchart illustrates the fault analysis method for energy storage devices according to an embodiment of the present invention.

[0038] Figure 5 One of the schematic diagrams of the fault analysis method for energy storage devices according to an embodiment of the present invention is shown;

[0039] Figure 6 The diagram below shows a second schematic of the fault analysis method for energy storage devices according to an embodiment of the present invention.

[0040] Figure 7 The third schematic diagram illustrates the principle of the fault analysis method for energy storage devices according to an embodiment of the present invention;

[0041] Figure 8 The fourth schematic diagram illustrates the fault analysis method for energy storage devices according to an embodiment of the present invention.

[0042] Figure 9 A structural block diagram of a fault analysis device for an energy storage device according to an embodiment of the present invention is shown;

[0043] Figure 10 A structural block diagram of a computer device according to an embodiment of the present invention is shown. Detailed Implementation

[0044] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0045] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0046] The following is combined Figures 1 to 10 The present application provides a detailed description of the fault analysis method and apparatus, computer device and storage medium for energy storage devices through specific embodiments and application scenarios.

[0047] In one embodiment of the present invention, such as Figure 1 As shown, the fault analysis method for energy storage devices may specifically include the following steps 102 to 108:

[0048] Step 102: Obtain the basic data and runtime sequence data of the energy storage device, and extract the time domain and frequency domain features of the runtime sequence data;

[0049] Step 104: Based on the basic data, runtime sequence data, time domain characteristics and frequency domain characteristics, retrieve the historical fault database to identify multiple first historical fault cases;

[0050] Step 106: Based on the matching degree of each first historical fault case with the multidimensional features of the energy storage device, sort the multiple first historical fault cases to determine multiple second historical fault cases among the multiple first historical fault cases;

[0051] Step 108: Generate prompt information based on basic data, runtime sequence data and multiple second historical fault cases, input the prompt information into the fault analysis model, and output the fault analysis results of the energy storage device;

[0052] The basic data includes at least one or a combination of the following: the device number, device model, hardware version, device runtime, fault time range, and fault type label of the energy storage device; the historical fault database includes historical fault cases broken down according to fault time range.

[0053] The technical solution of the fault analysis method for energy storage devices provided by this invention can be executed by a computer device, a fault analysis device, or something else depending on actual usage requirements; no specific limitation is made here. To more clearly describe the fault analysis method for energy storage devices provided by this invention, the following description uses a fault analysis device as the executing entity.

[0054] Specifically, in the fault analysis method for energy storage devices provided by this invention, when diagnosing faults in energy storage devices, the fault analysis device acquires basic data of the energy storage device. This basic data includes at least one or a combination of the following: device number, device model, hardware version, device runtime, fault time range, and fault type label. Furthermore, the fault analysis device acquires runtime sequence data of the energy storage device and extracts its time-domain and frequency-domain features. Based on this, the fault analysis device retrieves a pre-constructed historical fault database according to the basic data, runtime sequence data, time-domain features, and frequency-domain features. The historical fault database includes historical fault cases segmented according to fault time ranges. By retrieving the historical fault database, the fault analysis device identifies multiple first historical fault cases within the database. Based on this, the fault analysis device sorts multiple first historical fault cases according to the matching degree of each first historical fault case with the multidimensional features of the energy storage device, and selects multiple second historical fault cases from the multiple first historical fault cases. Then, based on the basic data, runtime sequence data and multiple second historical fault cases, it generates prompt information for the fault analysis model and inputs the prompt information into the fault analysis model, so that the fault analysis model outputs the fault analysis results of the energy storage device.

[0055] The fault analysis method for energy storage devices provided by this invention, on the one hand, improves the accuracy of the recalled historical fault data (i.e., the first historical fault case) by retrieving historical fault data from a historical fault database based on the basic data, operational sequence data, and their time-domain and frequency-domain characteristics of the energy storage device. The historical fault database includes historical fault cases segmented according to fault time ranges, thus improving the completeness of the recalled historical fault data. On the other hand, it refines and filters the recalled historical fault data based on multi-dimensional feature matching, improving the adaptability of the final adopted historical fault data (i.e., the second historical fault case) to the current operating condition of the energy storage device. Thus, when performing fault diagnosis based on the fault analysis model, it reduces interfering information in the generated prompts, improves the accuracy and comprehensiveness of the prompts, and ultimately enhances the accuracy of fault diagnosis.

[0056] The aforementioned energy storage devices can specifically be portable energy storage devices, home photovoltaic devices, etc., without specific restrictions.

[0057] Optionally, runtime timing data includes parameter values ​​of the energy storage device's operating parameters as they change over time.

[0058] The operating parameters include at least one or a combination of the following: cell voltage, cell current, cell temperature, cell charge, and inverter efficiency of the energy storage device.

[0059] In other words, the operating sequence data of energy storage devices includes, but is not limited to: the battery cell's charge time-series data, voltage time-series data, current time-series data, temperature time-series data, inverter conversion efficiency time-series data, etc., without specific limitations.

[0060] Optionally, the time-domain characteristics of runtime series data include, but are not limited to, the mean, variance, standard deviation, skewness, kurtosis, peak value, valley value, rate of change, and cumulative value of runtime series data, etc., without specific limitations.

[0061] Based on this, the time-domain characteristics of runtime sequence data can characterize the waveform shape of runtime sequence data of energy storage devices, such as rise acceleration and fluctuation frequency, and enable the measurement of the similarity of time-series waveforms, thereby effectively measuring the morphological similarity of different time-series waveforms.

[0062] Optionally, the frequency domain characteristics of the runtime sequence data can specifically be the FFT (Fast Fourier Transform) characteristics of the runtime sequence data in the time domain, including the main frequency components and spectral energy distribution based on FFT, without specific limitations.

[0063] Optionally, the aforementioned fault analysis model can be a large language model, or it can be replaced with other AI tools, without any specific restrictions.

[0064] Optionally, the hardware version of the energy storage device may specifically be the firmware version of the energy storage device.

[0065] Furthermore, in practical applications, after outputting the fault analysis results of the energy storage device, and assuming the results are correct, the fault analysis device can automatically generate new historical fault cases and add them to the historical fault database based on the current basic data, runtime sequence data, and fault analysis results of the energy storage device. This achieves automatic evolution of the knowledge base, thereby continuously improving retrieval coverage and accuracy.

[0066] In some embodiments of the present invention, optionally, the historical fault database includes a historical fault vector database and a historical fault text database. Before the step of retrieving the historical fault database, the above-mentioned fault analysis method may further include the following steps 110 to 116:

[0067] Step 110: Obtain historical fault case data for different energy storage devices;

[0068] Step 112: Generate corresponding historical fault description text based on each historical fault case data to obtain a historical fault text library;

[0069] Step 114: Based on the time range of the failure, split the data of each historical failure case into early failure data, mid-term failure data, and late failure data;

[0070] Step 116: Generate corresponding historical fault vectors based on the early-stage fault data, mid-stage fault data, and late-stage fault data to obtain a historical fault vector library;

[0071] Each historical fault case data includes at least one or a combination of the following: device number, device model, hardware version, fault type label, fault evolution information, fault diagnosis information, and fault handling information.

[0072] In this embodiment, the aforementioned historical fault database may specifically include a historical fault vector database and a historical fault text database.

[0073] Based on this, before retrieving the historical fault database, the fault analysis device also acquires multiple historical fault case data from different energy storage devices. Each historical fault case data includes at least one or a combination of the following: device number, device model, hardware version, fault type tag, fault evolution information, fault diagnosis information, and fault handling information. Based on this, the fault analysis device generates corresponding historical fault description text for each historical fault case data, resulting in the aforementioned historical fault text library. Furthermore, the fault analysis device splits each historical fault case data into early-stage fault data, mid-stage fault data, and late-stage fault data according to the fault time range. Then, based on the early-stage, mid-stage, and late-stage fault data, it generates corresponding historical fault vectors for each, resulting in the aforementioned historical fault vector library. This structured splitting of historical fault case data based on the fault time range improves the completeness of the data splitting results, facilitating the comprehensiveness of subsequent historical fault data retrieval. Simultaneously, it retains the complete temporal characteristic description of the historical fault case data, making it easier to combine semantic information to further improve the accuracy of the retrieved historical fault data.

[0074] Among them, the early-stage fault data is the data during the normal attenuation stage of the energy storage device, the mid-stage fault data is the data during the abnormal evolution stage of the energy storage device, and the late-stage fault data is the data during the fault confirmation and handling stage of the energy storage device.

[0075] Optionally, for each historical fault case data, the three fault vectors corresponding to the early, mid, and late fault data can form a fault evolution trajectory vector group for the current historical fault case data.

[0076] Optionally, by splitting historical failure case data into time-based segments, during subsequent searches of the historical failure database, historical failure case data within the corresponding time window can be prioritized based on the actual failure time range, thereby improving search efficiency. For example, when an energy storage device only exhibits early symptoms, the earlier failure data in the historical failure case data should be matched first, rather than the entire historical failure case data.

[0077] Optionally, in the process of generating corresponding historical fault vectors based on early-stage fault data, mid-stage fault data, and late-stage fault data, the temporal and frequency domain features of the early-stage, mid-stage, and late-stage fault data are extracted respectively. The extracted temporal and frequency domain features are then standardized. Subsequently, through a configurable time-series embedding model, the standardized temporal and frequency domain features are mapped into a fixed-dimensional dense vector to obtain the corresponding fault vector.

[0078] Among them, the fault vector can simultaneously capture the statistical shape of the timing waveform, the periodic characteristics in the frequency domain, and the correlation pattern with other operating parameters.

[0079] In practical applications, the historical fault vector library can be replaced by Faiss (an open-source algorithm library that provides a professional tool for efficient similarity search and clustering of massive high-dimensional vectors), Elasticsearch (an open-source distributed search and analysis engine) vector plugin, Qdrant (an open-source vector database), etc., without specific restrictions.

[0080] Optionally, the time series embedding model may specifically be TS2Vec (a general representation learning framework for time series), Time2Vec (a lightweight, general-purpose time coding plugin), PatchTST (Patch Time Series Transformer), and TimesNet (a general-purpose time series analysis model for tasks), etc., without specific limitations.

[0081] For example, such as Figure 6 As shown, the process of splitting and vectorizing a historical failure case can be as follows: obtain the original text of the failure case, split the original text of the failure case into three stages according to the time window: the early stage of failure, the middle stage of failure, and the late stage of failure, vectorize the original text fragments of the three stages respectively, number them and store them in the vector library.

[0082] In some embodiments of the present invention, optionally, such as Figure 2 As shown, step 104 above may specifically include steps 104a to 104c as follows:

[0083] Step 104a: Based on the basic data, time-domain features and frequency-domain features, the historical fault vector database is searched to obtain the first search result;

[0084] Step 104b: Based on the runtime sequence data, perform semantic retrieval on the historical fault text database to obtain the second retrieval result;

[0085] Step 104c: Deduplicate multiple third-historical fault cases and multiple fourth-historical fault cases to obtain multiple first-historical fault cases;

[0086] The first search result includes multiple third-generation historical fault cases, and the second search result includes multiple fourth-generation historical fault cases.

[0087] In this embodiment, during the process of retrieving the historical fault database and identifying multiple first historical fault cases, specifically, the fault analysis device searches the historical fault vector database based on basic data, time-domain features, and frequency-domain features to obtain a first search result, which includes multiple third historical fault cases. Furthermore, the fault analysis device performs semantic retrieval on the historical fault text database based on runtime sequence data to obtain a second search result, which includes multiple fourth historical fault cases. Based on this, the fault analysis device further deduplicates the multiple third and fourth historical fault cases to obtain multiple first historical fault cases. Thus, by using both vector retrieval and semantic retrieval, a dual-path recall of historical fault data is achieved, improving the accuracy of the recalled historical fault data.

[0088] In some embodiments of the present invention, optionally, such as Figure 3 As shown, step 104a above may specifically include steps 104a1 to 104a3 as follows:

[0089] Step 104a1: Generate the time-series feature vector of the energy storage device based on the time-domain and frequency-domain features;

[0090] Step 104a2: Filter the historical fault vector library based on the basic data to obtain a candidate vector library;

[0091] Step 104a3: Based on the vector similarity between the time-series feature vector and each historical fault vector in the candidate vector library, the candidate vector library is searched to obtain the first search result;

[0092] The runtime sequence data includes the parameter values ​​of the energy storage device's operating parameters as a function of time. The operating parameters include at least one or a combination of the following: cell voltage, cell current, cell temperature, cell charge, and inverter efficiency of the energy storage device. The time-domain characteristics include at least one or a combination of the following: mean, variance, standard deviation, skewness, kurtosis, peak value, valley value, rate of change, and cumulative value of the runtime sequence data.

[0093] In this embodiment, during the process of retrieving the historical fault vector library to obtain the first retrieval result, specifically, the runtime sequence data includes parameter values ​​of the energy storage device's operating parameters that change over time. The operating parameters include at least one or a combination of the following: cell voltage, cell current, cell temperature, cell charge, and inverter efficiency. The time-domain features include at least one or a combination of the following: mean, variance, standard deviation, skewness, kurtosis, peak value, valley value, rate of change, and cumulative value of the runtime sequence data. The fault analysis device generates a time-series feature vector of the energy storage device based on the time-domain and frequency-domain features, and filters the historical fault vector library based on the basic data to obtain a candidate vector library. On this basis, the fault analysis device further calculates the vector similarity between the time-series feature vector and each historical fault vector in the candidate vector library, and retrieves the candidate vector library based on the calculated vector similarity to obtain the first retrieval result. In this way, the historical fault vector library is first filtered through basic data, and then the candidate vector library is searched based on vector similarity. This reduces the search space for vector retrieval and can filter fault vectors with large differences, thereby improving the speed and accuracy of subsequent vector retrieval.

[0094] Among them, the time-series feature vector is a dense vector that can characterize the time-series waveform shape of the running time-series data of the energy storage device.

[0095] Specifically, in the fault analysis method for energy storage devices provided by this invention, the process of retrieving the historical fault vector database includes a scalar filtering stage and a vector retrieval stage. In the scalar filtering stage, the search space is narrowed down from tens of billions to millions. In the vector retrieval stage, similarity retrieval is performed on the fault vectors.

[0096] In some embodiments of the present invention, optionally, such as Figure 3 As shown, step 104b above may specifically include the following steps 104b1 and 104b2:

[0097] Step 104b1: Generate fault description information of the energy storage device based on runtime sequence data;

[0098] Step 104b2: Based on the semantic relevance between the fault phenomenon description information and the historical fault description texts in the historical fault text library, perform semantic retrieval on the historical fault text library to obtain the second retrieval result;

[0099] Among them, the fault phenomenon description information is used to describe the evolution trend of the operating parameters of the energy storage device.

[0100] In this embodiment, during the semantic retrieval of the historical fault text library to obtain the second retrieval result, specifically, the fault analysis device generates fault phenomenon description information of the energy storage device based on runtime sequence data. This fault phenomenon description information describes the evolution trend of the energy storage device's operating parameters. Based on this, the fault analysis device further analyzes the semantic correlation between the fault phenomenon description information and each historical fault description text in the historical fault text library, and performs a semantic retrieval of the historical fault text library based on the analyzed semantic correlation to obtain the second retrieval result. Thus, by leveraging the semantic correlation between the fault phenomenon description information of the energy storage device and the temporal feature descriptions of historical fault case data, the device assists in recalling historical fault data, further improving the accuracy of the recalled historical fault data.

[0101] In practical applications, when performing semantic retrieval on a historical fault text database, retrieval can be performed directly based on the text, or the text can be converted into semantic vectors before retrieval; no specific restrictions are imposed here.

[0102] In some embodiments of the present invention, optionally, such as Figure 4 As shown, step 106 above may specifically include steps 106a to 106c as follows:

[0103] Step 106a: Based on the basic data, runtime sequence data, and each first historical fault case, calculate the vector similarity, equipment model matching degree, time window matching degree, and fault type matching degree between each first historical fault case and the energy storage device.

[0104] Step 106b: Perform a weighted calculation on the vector similarity, equipment model matching degree, time window matching degree and fault type matching degree corresponding to each first historical fault case to obtain the comprehensive score of each first historical fault case;

[0105] Step 106c: Sort the multiple first historical failure cases in descending order of comprehensive score, and determine the first N first historical failure cases as second historical failure cases;

[0106] Where N is a positive integer.

[0107] In this embodiment, during the process of selecting multiple second historical failure cases from multiple first historical failure cases, specifically, the fault analysis device calculates the vector similarity, equipment model matching degree, time window matching degree, and failure type matching degree between each first historical failure case and the energy storage device based on basic data, runtime sequence data, and each first historical failure case. Based on this, the fault analysis device performs a weighted calculation of the vector similarity, equipment model matching degree, time window matching degree, and failure type matching degree corresponding to each first historical failure case to obtain a comprehensive score for each first historical failure case. Based on this, the fault analysis device then sorts the multiple first historical failure cases according to their comprehensive scores from highest to lowest, and determines the top N first historical failure cases as second historical failure cases. Here, N is a positive integer. In this way, based on the multi-dimensional feature matching degree such as vector similarity, equipment model matching degree, time window matching degree and fault type matching degree between each first historical fault case and energy storage device, the recalled historical fault data is finely sorted and filtered, which improves the adaptability of the final historical fault data to the current operating condition of the energy storage device, reduces the interference information in the generated prompt information, and improves the accuracy of the prompt information.

[0108] In the weighted calculation of vector similarity, equipment model matching degree, time window matching degree, and fault type matching degree for each first historical fault case, the weight of vector similarity is greater than that of equipment model matching degree, which is greater than that of fault type matching degree, which is greater than that of time window matching degree. For example, the weight of vector similarity is 40%, the weight of equipment model matching degree is 25%, the weight of fault type matching degree is 20%, and the weight of time window matching degree is 15%.

[0109] In practical applications, those skilled in the art can set the specific weights for vector similarity, device model matching degree, time window matching degree, and fault type matching degree according to the actual situation, and no specific restrictions are imposed here.

[0110] For example, such as Figure 7 As shown, the complete retrieval chain based on multi-source hybrid retrieval and re-ranking can include four stages: retrieval input, retrieval recall, candidate fusion, and fine-ranking output. Retrieval input serves as the routing and recall entry point; retrieval recall is used to retrieve multiple candidates; candidate fusion is used to merge and deduplicate multiple candidates; and fine-ranking output is the re-ranking and output, realizing the retrieval process of "scalar filtering → vector retrieval → semantic retrieval → merging and deduplication → re-ranking → output results".

[0111] In some embodiments of the present invention, optionally, the step of generating prompt information based on basic data, runtime sequence data, and multiple second historical fault cases may specifically include steps 118 and 120 below:

[0112] Step 118: Obtain a preset template for the prompt information. The preset template includes device description information, information to be analyzed, and reference information.

[0113] Step 120: Fill in the basic data into the device description information, fill in the runtime sequence data into the analysis information, fill in multiple second historical fault cases into the reference information, and obtain the prompt information.

[0114] In this embodiment, during the generation of prompt information, specifically, the fault analysis device obtains a preset template for the prompt information. This preset template includes equipment description information, information to be analyzed, and reference information. Based on this, the fault analysis device fills the equipment description information with basic data, the information to be analyzed with runtime sequence data, and the reference information with multiple second historical fault cases, thus obtaining the prompt information for the fault analysis model. In this way, by injecting the finely filtered historical fault data into the prompt information of the fault analysis model according to the preset template structure, it ensures that the fault analysis model obtains factual anchors before reasoning, improving the accuracy of fault diagnosis performed by the fault analysis model.

[0115] For example, the second historical fault case includes Reference Case 1 and Reference Case 2, and the prompt information may be:

[0116] [Character Setting] You are an expert in diagnosing energy storage equipment faults.

[0117] [Device Background] Device Model: {model}, Firmware Version: {firmware}, Runtime: {hours}h.

[0118] [Abnormal Data] The currently detected abnormal features: {feature_vector_description}.

[0119] [Reference Case 1]:

[0120] Time: {date}, Fault Type: {type}, Similarity: {score};

[0121] Fault evolution process: {pre_stage} → {mid_stage} → {post_stage};

[0122] Final diagnosis conclusion: {conclusion};

[0123] Solution: {solution}

[0124] [Reference Case 2]: ……

[0125] [Analysis Requirements] Based on the above reference case, please deduce the possible causes of the current equipment malfunction step by step, and provide suggestions for handling the situation.

[0126] [Negative Constraint] If the reference case differs significantly from the current situation, please clearly point out the uncertainties and do not force a connection.

[0127] In summary, such as Figure 5 As shown, the end-to-end link of the fault analysis method for energy storage devices provided by this invention can be as follows: device data source → time series feature extraction → vectorization → data storage → hybrid retrieval → reordering → hint injection → model inference.

[0128] Based on this, the system architecture of the fault analysis method for energy storage devices provided by the present invention may specifically include a time-aware document splitting module, a time-series feature engine, a multi-source hybrid retrieval module, a reordering module, and a dynamic prompt injection module.

[0129] The time-aware document segmentation module does not use fixed-length text slicing. Instead, it uses a structured segmentation strategy based on time windows and semantic similarity to segment historical failure cases of different energy storage devices into three time stages: early stage, middle stage, and late stage of failure, while retaining complete time-series feature descriptions within each time window.

[0130] Optionally, the time series feature engine supports the vectorization of time-domain features (such as mean, variance, peak value, and kurtosis) and frequency-domain features, and can plug and play dedicated time series embedding models such as Time2Vec, TS2Vec, and PatchTST to generate dense vectors that can characterize the shape of time series waveforms.

[0131] Optionally, the multi-source hybrid retrieval module supports hybrid queries of scalar filtering and vector retrieval. Scalar filtering quickly narrows the search scope, and vector similarity matching is performed within the limited scope. The multi-source hybrid retrieval module also supports semantic retrieval to achieve dual-path retrieval of historical fault data.

[0132] Optionally, the reordering module is used to fine-rank the historical fault data of dual-path recall. By comprehensively considering multi-dimensional features such as vector similarity, temporal proximity, device model matching degree, and fault type consistency, it outputs the optimal context set after sorting.

[0133] Optionally, the dynamic prompt injection module is used to inject prompt information into the reordered context according to a preset template structure, ensuring that the fault analysis model obtains fact anchors before reasoning.

[0134] Among them, the temporal embedding model, vector library and reordering module can all be designed as replaceable abstract interfaces, supporting adaptation to different model selections and storage engines, and realizing the flexible pluggable characteristics of the system architecture.

[0135] Based on this, the fault analysis method for energy storage devices provided by this invention can achieve the following beneficial effects: Through a multi-level filtering mechanism of time-aware document splitting + scalar filtering + vector retrieval + reordering, the retrieval recall accuracy is greatly improved, and the retrieval recall rate can be increased to 95% or above, providing a high-quality contextual basis for subsequent model inference; the context injected into the model consists of real historical cases of the same model, similar operating conditions, and matching time windows, so the model inference is based on evidence, which significantly reduces the model illusion rate, which can be reduced to below 3%; the fault diagnosis accuracy is greatly improved. Compared with the method of directly calling the model for time-series data analysis, the fault analysis accuracy of this invention can be improved by more than 40%, and compared with the scheme that only uses general text RAG (Retrieval-Augmented Generation), the fault analysis accuracy of this invention can be improved by more than 20%.

[0136] For example, an enhanced RAG retrieval scenario in the early diagnosis of thermal runaway in portable energy storage batteries: During use, cloud monitoring detects that the temperature of cell 3 reported by its BMS (Battery Management System) rises continuously from 35°C to 52°C within 30 minutes, and the temperature difference with other cells widens to 4°C, but the 60°C threshold alarm has not yet been triggered. The fault diagnosis process includes:

[0137] Step 1: Time-series feature extraction and vectorization. Obtain the reported data from the energy storage device over the past 2 hours, i.e., the runtime sequence data. Calculate 15-dimensional time-domain features such as the rate of temperature change, the speed of temperature difference expansion, and the amplitude of voltage fluctuation within the window in real time. Also calculate the frequency-domain features of the runtime sequence data. Vectorize the obtained time-domain and frequency-domain features to generate a time-series feature vector.

[0138] Step 2: Hybrid Search. Scalar Filtering: Set the filtering conditions device_model=="HB-2000" AND fault_type=="thermal" in the historical fault vector library to filter candidate vector libraries with device model HB-2000 and fault type thermal; Vector Search: Based on time-series feature vectors, perform similarity search on fault vectors in the candidate vector library and output the top 5 fault cases with the highest similarity; Semantic Search Supplement: Using "cell temperature difference continues to widen, temperature rises abnormally" as the text, search the historical fault text library based on semantic relevance and output the top 3 fault cases with the highest semantic relevance.

[0139] Step 3: Re-sorting. A total of 8 fault cases were recalled from both approaches. After deduplication, 6 fault cases were obtained. Based on the vector similarity, equipment model matching degree, time window matching degree, and fault type matching degree between each fault case and the energy storage device, a comprehensive score was calculated for each fault case. The top 3 fault cases with the highest comprehensive scores were output, as shown in Table 1 below:

[0140] Table 1

[0141]

[0142] Step 4: Hint Injection and Model Inference. For example... Figure 8 As shown, the three cases in Table 1 are injected into a preset standardized template, and the three cases are assembled to obtain the prompt information of the fault analysis model. Based on this, the fault analysis model infers that "it is highly likely that cell No. 3 has a micro-short circuit, and the increase in temperature difference and accelerated temperature rise are typical early signs" based on the high similarity between case 1 and case 2, and points out that the reference weight of case 3 needs to be reduced due to the mismatch of the time window.

[0143] Step 5: Diagnostic Output and Knowledge Accumulation. The diagnostic conclusion is output as follows: "Cell No. 3 is suspected of having a micro-short circuit; thermal runaway risk level: medium-high. It is recommended to limit the module's charging and discharging current to 0.5C (charge / discharge rate) and arrange for return to the factory for testing." After manual confirmation, the complete abnormal evolution trajectory is vectorized and stored in the candidate vector library as a new historical fault case for the same model of equipment.

[0144] In practical applications, during the retrieval process, knowledge transfer across device models is supported. That is, when a new device model lacks historical cases, the similarity of time-series feature vectors of different device models under similar working conditions can be calculated, and cross-model retrieval can refer to cases to achieve zero-sample diagnosis.

[0145] For example, cross-model knowledge transfer for efficiency degradation of a residential photovoltaic inverter: When a residential photovoltaic inverter of model HB-INV-3000 first experiences a slow decline in efficiency, since there are no historical failure cases for the current model, calculations show that the similarity of the time-series feature vectors of HB-INV-3000 and the existing model HB-INV-2000 under the same operating conditions reaches 0.89. At this point, a "DC bus capacitor aging" case of HB-INV-2000 is retrieved across models as a reference, successfully achieving zero-sample diagnosis.

[0146] In one embodiment of the present invention, a fault analysis device for energy storage equipment is also proposed. For example... Figure 9 As shown, Figure 9A structural block diagram of a fault analysis device 200 for an energy storage device according to an embodiment of the present invention is shown. Specifically, the fault analysis device 200 for the energy storage device may include an acquisition unit 202 and a processing unit 204.

[0147] The acquisition unit 202 is used to acquire the basic data and runtime sequence data of the energy storage device, and extract the time domain characteristics and frequency domain characteristics of the runtime sequence data. The basic data includes at least one of the following or a combination thereof: the device number, device model, hardware version, device runtime, fault time range and fault type label of the energy storage device;

[0148] Processing unit 204 is used to retrieve historical fault database based on basic data, runtime sequence data, time domain features and frequency domain features, and determine multiple first historical fault cases. The historical fault database includes historical fault cases split according to fault time range.

[0149] The processing unit 204 is also used to sort multiple first historical fault cases according to the matching degree of each first historical fault case with the multidimensional features of the energy storage device, and determine multiple second historical fault cases among the multiple first historical fault cases.

[0150] The processing unit 204 is also used to generate prompt information based on basic data, runtime sequence data and multiple second historical fault cases, input the prompt information into the fault analysis model, and output the fault analysis results of the energy storage device.

[0151] The fault analysis device 200 for energy storage devices provided in this embodiment of the invention includes an acquisition unit 202 and a processing unit 204. When diagnosing faults in the energy storage device, the acquisition unit 202 acquires basic data of the energy storage device, wherein the basic data includes at least one or a combination of the following: device number, device model, hardware version, device runtime, fault time range, and fault type label. Furthermore, the acquisition unit 202 acquires runtime sequence data of the energy storage device and extracts the time-domain and frequency-domain features of the runtime sequence data. Based on this, the processing unit 204 retrieves a pre-constructed historical fault database according to the basic data, runtime sequence data, time-domain features, and frequency-domain features. The historical fault database includes historical fault cases segmented according to fault time ranges. By retrieving the historical fault database, the processing unit 204 identifies multiple first historical fault cases in the historical fault database. Based on this, the processing unit 204 sorts the multiple first historical fault cases according to the matching degree of each first historical fault case with the multidimensional features of the energy storage device, and selects multiple second historical fault cases from the multiple first historical fault cases. Then, based on the basic data, runtime sequence data and multiple second historical fault cases, it generates prompt information for the fault analysis model and inputs the prompt information into the fault analysis model, so that the fault analysis model outputs the fault analysis results of the energy storage device.

[0152] The fault analysis device 200 for energy storage devices provided by this invention, on the one hand, retrieves historical fault databases based on the basic data, operational sequence data, and their time-domain and frequency-domain characteristics of the energy storage device, improving the accuracy of the recalled historical fault data, i.e., the first historical fault case. The historical fault database includes historical fault cases segmented according to fault time ranges, enhancing the completeness of the recalled historical fault data. On the other hand, based on multi-dimensional feature matching, the recalled historical fault data is refined and filtered, improving the adaptability of the final adopted historical fault data, i.e., the second historical fault case, to the current operating conditions of the energy storage device. Thus, when performing fault diagnosis based on the fault analysis model, interference information in the generated prompts is reduced, improving the accuracy and comprehensiveness of the prompts, and ultimately enhancing the accuracy of fault diagnosis.

[0153] In some embodiments of the present invention, optionally, the historical fault database includes a historical fault vector database and a historical fault text database. Before retrieving the historical fault database, the processing unit 204 is further configured to: acquire multiple historical fault case data of different energy storage devices, each historical fault case data including at least one or a combination of the following: device number, device model, hardware version, fault type label, fault evolution information, fault diagnosis information, and fault handling information; generate corresponding historical fault description text based on each historical fault case data to obtain a historical fault text database; divide each historical fault case data into early-stage fault data, mid-stage fault data, and late-stage fault data according to the fault time range; and generate corresponding historical fault vectors based on the early-stage fault data, mid-stage fault data, and late-stage fault data to obtain a historical fault vector database.

[0154] In some embodiments of the present invention, optionally, the processing unit 204 is specifically configured to: search the historical fault vector library based on basic data, time-domain features and frequency-domain features to obtain a first search result, the first search result including multiple third historical fault cases; perform semantic search on the historical fault text library based on runtime sequence data to obtain a second search result, the second search result including multiple fourth historical fault cases; and perform deduplication processing on the multiple third historical fault cases and the multiple fourth historical fault cases to obtain multiple first historical fault cases.

[0155] In some embodiments of the present invention, optionally, the processing unit 204 is specifically configured to: generate a time-series feature vector of the energy storage device based on time-domain features and frequency-domain features, wherein the time-series data includes parameter values ​​of the operating parameters of the energy storage device that change over time, and the operating parameters include at least one or a combination of the following: cell voltage, cell current, cell temperature, cell charge, and inverter efficiency of the energy storage device; and the time-domain features include at least one or a combination of the following: mean, variance, standard deviation, skewness, kurtosis, peak value, valley value, rate of change, and cumulative value of the time-series data; filter the historical fault vector library based on the basic data to obtain a candidate vector library; and search the candidate vector library based on the vector similarity between the time-series feature vector and each historical fault vector in the candidate vector library to obtain a first search result.

[0156] In some embodiments of the present invention, optionally, the processing unit 204 is specifically configured to: generate fault phenomenon description information of the energy storage device based on runtime sequence data, wherein the fault phenomenon description information is used to describe the evolution trend of the operating parameters of the energy storage device; and perform semantic retrieval on the historical fault text library based on the semantic relevance between the fault phenomenon description information and each historical fault description text in the historical fault text library to obtain a second retrieval result.

[0157] In some embodiments of the present invention, optionally, the processing unit 204 is specifically configured to: calculate the vector similarity, equipment model matching degree, time window matching degree, and fault type matching degree between each first historical fault case and the energy storage device based on the basic data, runtime sequence data, and each first historical fault case; perform weighted calculation on the vector similarity, equipment model matching degree, time window matching degree, and fault type matching degree corresponding to each first historical fault case to obtain a comprehensive score for each first historical fault case; sort the multiple first historical fault cases in descending order of comprehensive score, and determine the first N first historical fault cases as second historical fault cases, where N is a positive integer.

[0158] In some embodiments of the present invention, optionally, the processing unit 204 is specifically used to: obtain a preset template for prompt information, the preset template including device description information, information to be analyzed and reference information; fill basic data into the device description information, fill runtime sequence data into the information to be analyzed, fill multiple second historical fault cases into the reference information, and obtain prompt information.

[0159] In one embodiment of the invention, a computer device is also provided. For example... Figure 10 As shown, Figure 10 A structural block diagram of a computer device 300 provided in an embodiment of the present invention is shown. The computer device 300 includes:

[0160] Memory 302, on which programs or instructions are stored;

[0161] The processor 304 executes the above-described program or instructions to implement the steps of the fault analysis method for the energy storage device as described in any of the above embodiments.

[0162] The computer device 300 provided in this embodiment includes a memory 302 and a processor 304. When the program or instructions in the memory 302 are executed by the processor 304, they implement the steps of the fault analysis method for energy storage devices as described in any of the above embodiments. Therefore, the computer device 300 has all the beneficial effects of the fault analysis method for energy storage devices as described in any of the above embodiments, which will not be repeated here.

[0163] Specifically, the memory 302 and the processor 304 can be connected via a bus or other means. The processor 304 may include one or more processing units, and the processor 304 may be a chip such as a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or a field-programmable gate array (FPGA).

[0164] In one embodiment of the present invention, a readable storage medium is also provided. The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the fault analysis method for the energy storage device as described in any of the above embodiments.

[0165] The readable storage medium provided in this embodiment of the invention, when its stored program or instructions are executed by a processor, can implement the steps of the fault analysis method for energy storage devices as described in any of the above embodiments. Therefore, the readable storage medium possesses all the beneficial effects of the fault analysis method for energy storage devices in any of the above embodiments, which will not be elaborated further here.

[0166] Specifically, the aforementioned readable storage media can include any medium capable of storing or transmitting information. Examples of readable storage media include electronic circuits, semiconductor memory devices, read-only memory (ROM), random access memory (RAM), compact disc read-only memory (CD-ROM), flash memory, erasable ROM (EROM), magnetic tape, floppy disk, optical disk, hard disk, fiber optic media, radio frequency (RF) links, optical data storage devices, etc. Code segments can be downloaded via computer networks such as the Internet and intranets.

[0167] In the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance, unless otherwise expressly specified and limited. The terms "connection," "installation," and "fixing," etc., should be interpreted broadly. For example, "connection" can mean a fixed connection, a detachable connection, or an integral connection; it can mean a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0168] In the description of this specification, the terms "one embodiment," "some embodiments," "specific embodiment," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0169] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0170] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A fault analysis method for energy storage equipment, characterized in that, include: Acquire basic data and runtime sequence data of the energy storage device, and extract the time domain features and frequency domain features of the runtime sequence data. The basic data includes at least one or a combination of the following: device number, device model, hardware version, device runtime, fault time range, and fault type label of the energy storage device. Based on the basic data, the runtime sequence data, the time domain features and the frequency domain features, the historical fault database is retrieved to determine multiple first historical fault cases. The historical fault database includes historical fault cases that are split according to the fault time range. Based on the matching degree of each first historical fault case with the multidimensional features of the energy storage device, the multiple first historical fault cases are sorted to determine multiple second historical fault cases among the multiple first historical fault cases. Based on the basic data, the runtime sequence data, and multiple second historical fault cases, a prompt message is generated, and the prompt message is input into the fault analysis model to output the fault analysis result of the energy storage device.

2. The fault analysis method for energy storage equipment according to claim 1, characterized in that, The historical fault database includes a historical fault vector database and a historical fault text database. Before retrieving the historical fault database, the fault analysis method further includes: Acquire multiple historical fault case data of different energy storage devices, each of the historical fault case data includes at least one or a combination of the following: device number, device model, hardware version, fault type label, fault evolution information, fault diagnosis information and fault handling information; Based on each historical fault case data, a corresponding historical fault description text is generated to obtain the historical fault text library; Based on the time range of the failure, each historical failure case data is divided into early failure data, mid-term failure data, and late failure data; Based on the early-stage fault data, the mid-stage fault data, and the late-stage fault data, corresponding historical fault vectors are generated to obtain the historical fault vector library.

3. The fault analysis method for energy storage equipment according to claim 2, characterized in that, The step involves retrieving historical fault databases based on the basic data, runtime sequence data, time-domain features, and frequency-domain features to identify multiple first historical fault cases, including: Based on the basic data, the time-domain features, and the frequency-domain features, the historical fault vector library is searched to obtain a first search result, which includes multiple third historical fault cases. Based on the runtime sequence data, a semantic search is performed on the historical fault text library to obtain a second search result, which includes multiple fourth historical fault cases. The multiple third historical fault cases and multiple fourth historical fault cases are deduplicated to obtain multiple first historical fault cases.

4. The fault analysis method for energy storage equipment according to claim 3, characterized in that, The step of retrieving the historical fault vector database based on the basic data, the time-domain features, and the frequency-domain features to obtain a first retrieval result includes: Based on the time-domain features and the frequency-domain features, a time-series feature vector of the energy storage device is generated. The runtime sequence data includes parameter values ​​of the operating parameters of the energy storage device as a function of time. The operating parameters include at least one or a combination of the following: cell voltage, cell current, cell temperature, cell charge, and inverter efficiency of the energy storage device. The time-domain features include at least one or a combination of the following: mean, variance, standard deviation, skewness, kurtosis, peak value, valley value, rate of change, and cumulative value of the runtime sequence data. Based on the aforementioned basic data, the historical fault vector library is filtered to obtain a candidate vector library; Based on the vector similarity between the time-series feature vector and each historical fault vector in the candidate vector library, the candidate vector library is searched to obtain the first search result.

5. The fault analysis method for energy storage equipment according to claim 3, characterized in that, The step of performing semantic retrieval on the historical fault text database based on the runtime sequence data to obtain a second retrieval result includes: Based on the runtime sequence data, fault phenomenon description information of the energy storage device is generated, and the fault phenomenon description information is used to describe the evolution trend of the operating parameters of the energy storage device. Based on the semantic relevance between the fault phenomenon description information and the historical fault description texts in the historical fault text library, a semantic search is performed on the historical fault text library to obtain the second search result.

6. The fault analysis method for energy storage equipment according to claim 1, characterized in that, The step of sorting multiple first historical fault cases based on the multidimensional feature matching degree between each first historical fault case and the energy storage device, and determining multiple second historical fault cases among the multiple first historical fault cases, includes: Based on the basic data, the runtime sequence data, and each first historical fault case, calculate the vector similarity, equipment model matching degree, time window matching degree, and fault type matching degree between each first historical fault case and the energy storage device; The vector similarity, device model matching degree, time window matching degree and fault type matching degree corresponding to each first historical fault case are weighted and calculated to obtain the comprehensive score of each first historical fault case; According to the comprehensive score from high to low, the multiple first historical fault cases are sorted, and the first N first historical fault cases are determined as the second historical fault cases, where N is a positive integer.

7. The fault analysis method for energy storage equipment according to claim 1, characterized in that, The step of generating prompt information based on the basic data, the runtime sequence data, and multiple second historical fault cases includes: A preset template for obtaining the prompt information is obtained, the preset template including device description information, information to be analyzed, and reference information; The basic data is filled into the device description information, the runtime sequence data is filled into the information to be analyzed, and multiple second historical fault cases are filled into the reference information to obtain the prompt information.

8. A fault analysis device for energy storage equipment, characterized in that, include: The acquisition unit is used to acquire basic data and runtime sequence data of the energy storage device, and extract the time domain features and frequency domain features of the runtime sequence data. The basic data includes at least one of the following or a combination thereof: the device number, device model, hardware version, device runtime, fault time range, and fault type label of the energy storage device. The processing unit is configured to retrieve historical fault databases based on the basic data, the runtime sequence data, the time-domain features, and the frequency-domain features, and determine multiple first historical fault cases. The historical fault database includes historical fault cases split according to fault time ranges. The processing unit is further configured to sort multiple first historical fault cases according to the multidimensional feature matching degree between each first historical fault case and the energy storage device, and determine multiple second historical fault cases among the multiple first historical fault cases. The processing unit is further configured to generate prompt information based on the basic data, the runtime sequence data, and multiple second historical fault cases, input the prompt information into the fault analysis model, and output the fault analysis results of the energy storage device.

9. A computer device, characterized in that, It includes a processor and a memory, the memory storing programs or instructions that can run on the processor, the programs or instructions being executed by the processor to implement the steps of the fault analysis method for the energy storage device as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the fault analysis method for the energy storage device as described in any one of claims 1 to 7.

Images