Sodium-ion battery management methods and battery management systems
By filtering data from the sodium-ion battery management system and performing hierarchical retrieval of electrochemical scenario constraints, the problem of inaccurate anomaly diagnosis in sodium-ion batteries has been solved, achieving accurate anomaly identification and safe battery management.
Patent Information
- Application Number
- CN202611590355.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-14
- Publication Date
- 2026-11-10
AI Technical Summary
Existing battery management systems struggle to accurately identify abnormal states in sodium-ion batteries, leading to false alarms, missed alarms, or inaccurate attribution, which affects the accuracy and safety of battery management.
By acquiring battery operating status data, matching and filtering are performed based on normal and abnormal operating feature tables to extract abnormal characterization features. Then, a hierarchical search is performed in conjunction with electrochemical scenario constraints to generate diagnostic results and execute battery control actions.
It enables precise diagnosis and control of abnormal states in sodium-ion batteries, improves the accuracy of anomaly identification and the safety of handling, and avoids misjudgment and redundant calculations.
Smart Images

Figure CN123073229A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of battery management technology, and in particular to a sodium-ion battery management method and battery management system. Background Technology
[0002] In the field of battery management, battery management systems typically use fixed thresholds or single rules to determine abnormal battery states. However, due to the complex electrochemical characteristics of batteries (especially sodium-ion batteries), such as flat discharge curves, wide operating temperature ranges, and high self-discharge rates, the same abnormal signal may correspond to completely different failure mechanisms at different states of charge, temperature ranges, or aging stages. Existing solutions struggle to accurately extract diagnostically valuable abnormal features from massive amounts of high-frequency operational data and cannot combine multi-dimensional electrochemical scenarios for differentiated attribution. This leads to problems such as false alarms, missed alarms, or inaccurate attribution in anomaly diagnosis, thereby affecting the accuracy and safety of battery management. Summary of the Invention
[0003] The purpose of this invention is to provide a sodium-ion battery management method and a battery management system, which aims to solve the technical problem of inaccurate attribution of existing battery anomaly diagnosis.
[0004] In a first aspect, this application provides a sodium-ion battery management method, comprising the following steps: S10: Obtain battery operating status data; S20: Based on the normal operation feature table and the abnormal operation feature table, the operation status data is matched and filtered to identify abnormal operation segments; S30: Extract the abnormal characterization features of the abnormal operation segment and determine the electrochemical scenario constraints of the battery; S40: Based on the abnormal characterization features and the electrochemical scenario constraints, perform a hierarchical retrieval in the historical knowledge base to obtain the target diagnostic knowledge; S50: Determine the generation strategy based on the matching confidence of the target diagnostic knowledge, and generate a diagnostic result based on the generation strategy and the target diagnostic knowledge to execute the corresponding battery management action.
[0005] In one embodiment, step S20 includes the following steps: S21: Divide the running status data into multiple data segments and extract the segment features of each data segment; S22: Based on the fragment features, match them with the abnormal operation feature table, and determine the successfully matched data fragments as known abnormal fragments; S23: Based on the fragment features of the unmatched data fragments, match them with the normal operation feature table, determine the successfully matched data fragments as normal background fragments and filter them, and determine the unmatched data fragments as unknown abnormal fragments; S24: The known abnormal segment and the unknown abnormal segment are identified as the abnormal running segment.
[0006] In one embodiment, step S21 includes the following steps: S211: Divide the operating status data into the multiple data segments according to at least one of the preset time window, charging / discharging stage switching point, and signal characteristic mutation point; S212: Perform feature extraction on each of the data segments to obtain the segment features; wherein, the segment features include at least one time-series waveform feature or electrochemical feature among voltage change slope, polarization internal resistance, and temperature change rate, or at least one statistical feature among mean, variance, and peak value.
[0007] In one embodiment, the electrochemical scenario constraints include at least one of the following: state of charge range, operating temperature range, charge / discharge rate, cyclic aging stage, and resting time; step S40 includes the following steps: S41: Using the abnormal characterization features as the retrieval entry point, recall at least one candidate diagnostic knowledge from the historical knowledge base; S42: Match the electrochemical scenario constraints with the at least one candidate diagnostic knowledge one by one, and determine the candidate diagnostic knowledge with the highest comprehensive matching degree as the target diagnostic knowledge.
[0008] In one embodiment, the historical knowledge base includes a vector database and a relational database, wherein the elements in the vector database are vector representations of the elements in the relational database in the feature space; Step S41 includes: performing similarity matching between the feature vector corresponding to the anomaly representation feature and the vector database to obtain at least one candidate knowledge vector; Step S42 includes: S421: Based on the constraint vector corresponding to the electrochemical scenario constraint conditions, perform precise localization on the at least one candidate knowledge vector to determine the target knowledge vector; S422: Based on the target knowledge vector, retrieve the complete target diagnostic knowledge from the relational database.
[0009] In one embodiment, step S50 includes: S51: Determine the matching confidence level based on the similarity between the target diagnostic knowledge and the current abnormal scene features; wherein, the current abnormal scene features include at least one of the problem description, the abnormal running segment, the abnormal characterization features, and the electrochemical scene constraints; S52: In response to the matching confidence meeting the first preset condition, the first generation strategy is used to drive the generative model to output a deterministic diagnostic result, the deterministic diagnostic result including a clear cause of the anomaly and a solution; S53: In response to the matching confidence not meeting the first preset condition, the second generation strategy is used to drive the generative model to output a reference diagnostic result, which includes a reference anomaly cause and a reference processing solution.
[0010] In one embodiment, step S52 includes: constructing a first prompt word based on the first instruction, the target diagnostic knowledge, and the current abnormal scene features, and inputting the first prompt word into the generative model to instruct the generative model to output the deterministic diagnostic result in a deterministic tone; Step S53 includes: constructing a second prompt word based on the second instruction, the target diagnostic knowledge, and the current abnormal scene features; inputting the second prompt word into the generative model to instruct the generative model to output the reference diagnostic result in a speculative tone and generate a prompt for supplementary information.
[0011] In one embodiment, the method further includes: S61: In response to the diagnostic result meeting the preset reference conditions, generate candidate diagnostic knowledge based on the diagnostic result and supplementary investigation information; S62: In response to the candidate diagnostic knowledge passing the review, the candidate diagnostic knowledge is written into the historical knowledge base, and the retrieval index of the historical knowledge base is updated.
[0012] Secondly, this application provides a battery management system for use in sodium-ion batteries, comprising: The data acquisition module is used to acquire battery operating status data; The matching and filtering module is used to match and filter the running status data based on the normal operation feature table and the abnormal operation feature table, and to filter out abnormal operation segments. The feature and constraint determination module is used to extract the abnormal characterization features of the abnormal operating segment and determine the electrochemical scenario constraints of the battery. The retrieval and positioning module is used to perform hierarchical retrieval in the historical knowledge base based on the abnormal characterization features and the electrochemical scenario constraints to obtain target diagnostic knowledge. The diagnosis generation module is used to determine a generation strategy based on the matching confidence of the target diagnostic knowledge, and to generate a diagnostic result based on the generation strategy and the target diagnostic knowledge. The control execution module is used to perform corresponding battery control actions based on the diagnostic results.
[0013] In one embodiment, the battery management system further includes: The cloud communication module is used to upload the operating status data and the diagnostic results to the cloud server, and to receive firmware online upgrade instructions and visual configuration data issued by the cloud server.
[0014] The beneficial effects of the sodium-ion battery management method and battery management system provided by this invention are as follows: By matching and filtering operating status data based on normal operation feature tables and abnormal operation feature tables, abnormal operating segments can be accurately extracted from massive time-series data, effectively filtering out normal background noise and improving the utilization efficiency of abnormal features; By extracting abnormal characterization features and combining them with electrochemical scenario constraints for hierarchical retrieval, it is possible to overcome the fixed threshold misjudgment caused by characteristics such as flat battery discharge curves, wide temperature range, and high self-discharge, and achieve differentiated and accurate attribution of the same abnormal characterization under different electrochemical scenarios; By using a matching confidence determination generation strategy based on target diagnostic knowledge, deterministic diagnostic results can be output with high confidence to achieve rapid handling. Attached Figure Description
[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 This is a flowchart illustrating the sodium-ion battery management method provided in an embodiment of the present invention. Figure 2 This is a flowchart illustrating step S20 of the sodium-ion battery management method provided in an embodiment of the present invention. Figure 3 This is a flowchart illustrating step S50 of the sodium-ion battery management method provided in an embodiment of the present invention. Figure 4 This is a schematic diagram of the battery management system provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0017] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.
[0018] Throughout this specification, references to "an embodiment" or "an embodiment" mean that a particular feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment of this application. Therefore, the phrases "in one embodiment" or "in some embodiments" appear in various places throughout the specification, and not all refer to the same embodiment. Furthermore, in one or more embodiments, particular features, structures, or characteristics may be combined in any suitable manner.
[0019] Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature.
[0020] Example 1 Combination Figure 1 The sodium-ion battery management method provided in this application achieves accurate diagnosis and control of abnormal battery states by integrating time-series data feature analysis with electrochemical scenario constraints. Specifically, the method includes the following steps: S10: Obtain battery operating status data.
[0021] In this embodiment, operational status data refers to time-series signals of physical quantities reflecting the real-time electrochemical behavior of the battery, rather than discrete text records or event codes. Specifically, operational status data includes at least continuous sampling sequences of the voltage, current, and temperature of individual battery cells over time. For example, the battery management system can collect these physical quantities using high-precision sensors at sampling periods of 10ms to 1s, forming a multi-dimensional time-series data stream. It should be understood that although this embodiment uses voltage, current, and temperature as examples, in other embodiments, operational status data may further include derived parameters or auxiliary sensing data that characterize the internal state of the battery, such as estimated internal resistance, expansion force, and electrolyte pressure, as long as these data can be used for subsequent anomaly analysis.
[0022] S20: Match and filter the operating status data based on the normal operation feature table and the abnormal operation feature table to identify abnormal operation segments.
[0023] The normal operation feature table and the abnormal operation feature table are template libraries storing typical time-series waveform patterns or statistical distribution characteristics of the battery under different states. Specifically, the matching and filtering process involves calculating the waveform similarity or measuring the statistical feature distance between the currently acquired operating state data and the templates in these two feature tables. For example, a dynamic time warping algorithm can be used to calculate the similarity between the current voltage curve and the "micro-short-circuit voltage drop" template in the abnormal operation feature table, or to calculate the Euclidean distance between the current temperature change rate and the "steady-state heat dissipation" template in the normal operation feature table. When the matching degree of a certain data segment with the abnormal operation feature table exceeds a preset threshold, it is marked as abnormal; when it highly matches the normal operation feature table, it is considered normal background and filtered out. The abnormal operation segment determined in this way refers to the continuous data interval extracted on the time axis that is judged to deviate from the normal benchmark or conform to a known abnormal pattern. This filtering mechanism based on physical feature similarity can effectively overcome the defect of the fixed threshold method that is prone to false alarms under the nonlinear characteristics of the battery, while avoiding the computational redundancy and noise interference caused by directly sending the entire raw data segment into the subsequent analysis stage.
[0024] S30: Extract abnormal characterization features from abnormal operating segments and determine the electrochemical scenario constraints of the battery.
[0025] Anomaly characterization features are mathematical abstractions of the intrinsic morphology of abnormal operating segments, used to characterize the external signal manifestations of faults. For example, anomaly characterization features can include time-series waveform features such as voltage change slope, polarization resistance increment, and temperature rise rate, as well as statistical features such as the mean, variance, and peak factor of the segment. Meanwhile, given the characteristics of sodium-ion batteries, such as flat discharge curves and wide temperature range operation, the same anomaly characterization often corresponds to drastically different failure mechanisms under different electrochemical states. Therefore, electrochemical scenario constraints are a key dimension for distinguishing failure types, specifically including at least one of the following: state of charge range, operating temperature range, charge / discharge rate, cycle aging stage, and resting time. For example, the same anomaly characterization of "abnormal voltage rise during initial charging" may point to sodium deposition risk if it occurs in a low-temperature, long-resting scenario; while if it occurs in a high-temperature, high-SOC scenario, it may point to a phase transition in the cathode material structure. By decoupling and parallelizing the acquisition of anomaly characterization features with electrochemical scenario constraints, the attribution confusion problem caused by neglecting scenario differences in traditional methods is solved.
[0026] S40: Based on the abnormal characterization features and electrochemical scenario constraints, perform hierarchical retrieval in the historical knowledge base to obtain the target diagnostic knowledge.
[0027] Specifically, firstly, using abnormal characterization features as the retrieval entry point, the system quickly recalls a set of candidate diagnostic knowledge sets similar to the current signal morphology from the historical knowledge base. Secondly, using electrochemical scenario constraints as filters, the system performs one-by-one matching and precise positioning of candidate diagnostic knowledge, selecting the target diagnostic knowledge that best matches the current physical state of the battery, ensuring the accuracy of the results. For example, the system first recalls all fault cases involving voltage drops based on the feature of "voltage drop," and then uses the constraint set of "high SOC, 25℃, 0.5C discharge" to lock in the specific knowledge item "excessive concentration polarization under high-rate discharge," while excluding other cases such as "low SOC over-discharge" or "loose connectors" that, although manifesting as voltage drops, do not match the scenario. This hierarchical retrieval mechanism effectively balances retrieval efficiency and accuracy, ensuring that the final target diagnostic knowledge is both a match in signal morphology and self-consistent in physical cause.
[0028] S50: Determine the generation strategy based on the matching confidence of the target diagnostic knowledge, and generate diagnostic results based on the generation strategy and the target diagnostic knowledge to execute the corresponding battery management actions.
[0029] The matching confidence level reflects the degree of agreement between the target diagnostic knowledge and the current actual abnormal scenario. It is typically derived by weighting the similarity of the abnormal characterization features and the matching degree of the electrochemical scenario constraints. When the matching confidence level is high, it indicates that historical experience has a strong explanatory power for the current problem. The system adopts a first generation strategy, driving the generative model to output a deterministic diagnostic result containing a clear cause of the abnormality and a solution, and accordingly executes direct control actions such as power reduction or circuit disconnection. Conversely, when the matching confidence level is low, it indicates that the current situation may have variations or be a new type of fault. The system adopts a second generation strategy, outputting a reference diagnostic result containing speculative causes and suggested troubleshooting directions, and executes more conservative protective measures or prompts for manual intervention. For example, at high confidence, the diagnostic result might directly indicate "A single-cell micro-short circuit has been detected; please immediately shut down for maintenance"; while at low confidence, the result might be stated as "Suspected insulation performance degradation; it is recommended to supplement leakage current data." This dynamic adjustment strategy ensures efficient closed-loop handling of known faults while avoiding the safety risks of blindly executing aggressive operations under uncertain conditions, thus achieving a balance between diagnostic intelligence and control safety.
[0030] In summary, based on steps S10 to S50, this method constructs a complete closed loop for battery anomaly diagnosis and control. This approach completely departs from the text processing paradigm of general log analysis, anchoring data processing logic deeply to the physical characteristics and electrochemical evolution of battery time-series signals. Through the synergistic effect of dual-template filtering and noise reduction, scene-constrained hierarchical retrieval, and confidence-based graded decision-making, it effectively overcomes the diagnostic challenges posed by battery nonlinear characteristics, significantly improving the accuracy of anomaly identification and the safety of handling.
[0031] In one embodiment, combined Figure 2 Step S20 includes the following steps: S21: Divide the running status data into multiple data segments and extract the segment features of each data segment.
[0032] S22: Match the fragment features with the abnormal operation feature table, and identify the successfully matched data fragments as known abnormal fragments.
[0033] S23: Match the fragment features of the unmatched data fragments with the normal operation feature table, identify the successfully matched data fragments as normal background fragments and filter them, and identify the unmatched data fragments as unknown abnormal fragments.
[0034] S24: Identify known and unknown abnormal segments as abnormal runtime segments.
[0035] Based on this, this method prioritizes matching the current data segment with the abnormal operation feature table. If the similarity exceeds a preset abnormal threshold, it is directly marked as a known abnormal segment, indicating that the segment conforms to the confirmed fault modes in the historical database, such as typical micro-short-circuit voltage drops or thermal runaway temperature rise curves. For data segments that do not match the abnormal template, they are further matched with the normal operation feature table. If they highly match the templates such as steady-state charging, constant power discharging, or static self-discharging in the normal operation feature table, they are judged as normal background segments and filtered out. This step is crucial because the battery is in normal operation for most of its life cycle. If all the massive amounts of data generated are used in subsequent diagnostic analysis, it will not only cause a huge waste of computing power but also introduce a large amount of redundant information to interfere with the identification of weak abnormal signals. Only those data segments that neither conform to known abnormal modes nor belong to the normal background category are classified as unknown abnormal segments. These segments often indicate the emergence of new failure mechanisms or early variations of known faults. Including them in the abnormal operation segment set can effectively improve the system's ability to perceive new problems and avoid missed detections.
[0036] In one embodiment, step S21 includes the following steps: S211: Divide the operating status data into multiple data segments according to at least one of the following: a preset time window, a charging / discharging stage switching point, and a signal characteristic mutation point.
[0037] The preset time window can be divided into segments every 30 seconds or 1 minute. The charge / discharge stage switching point refers to the moment when the current direction changes, the rate jumps, or the operating condition transitions from constant current to constant voltage. These moments are nodes where the internal state of the battery undergoes significant changes, and the segments divided by these boundaries can completely encompass an independent electrochemical response process. The signal characteristic abrupt change point refers to the moment when the first or second derivative of parameters such as voltage and temperature exceeds a preset threshold, usually corresponding to the triggering of internal side reactions or drastic fluctuations in external load.
[0038] S212: Perform feature extraction on each data segment to obtain segment features; wherein, the segment features include at least one time-series waveform feature or electrochemical feature among voltage change slope, polarization internal resistance, and temperature change rate, and / or include at least one statistical feature among mean, variance, and peak value.
[0039] Among them, the slope of voltage change ( (Ion insertion / extraction) is an important indicator characterizing the phase transition process and ion insertion / extraction kinetics of electrode materials. In sodium-ion batteries, specific (Ion insertion / extraction) parameters are crucial. Anomalies often precede absolute voltage over-limit occurrences and can serve as early warning signals for sodium deposition or positive electrode structure collapse. Polarization resistance reflects the charge transfer and diffusion impedance within the battery; an increase in its value usually indicates thickening of the SEI (Solid Electrolyte Interphase) film, electrolyte drying, or loosening of connectors, making it a core parameter for assessing battery aging and health. The rate of temperature change directly relates to the balance between the battery's heat generation rate and heat dissipation conditions; abnormal temperature rise rates are often precursors to the initiation of a thermal runaway chain reaction. In terms of statistical characteristics, variance effectively quantifies signal fluctuations, distinguishing normal sensor noise from random voltage jitter caused by micro-short circuits; the peak factor helps capture transient overshoot or drop events.
[0040] Furthermore, by combining time-series waveform features with statistical features, the system can comprehensively characterize the attributes of data segments from both morphological and distributional dimensions, providing a solid data foundation for subsequent accurate matching with the feature table.
[0041] In some embodiments, due to the significant nonlinear and time-varying characteristics of sodium-ion batteries, their discharge curves are relatively flat, leading to the possibility that the same voltage reading or abnormal waveform may correspond to drastically different physical causes under different electrochemical states. For example, the same abnormal characteristic of "abnormal voltage rise in the initial stage of charging" may indicate sodium deposition at the negative electrode if it occurs at low temperatures (e.g., -10°C to 0°C) and after a long period of rest; however, if it occurs at high temperatures (e.g., above 45°C) and with a high number of cycles (e.g., >2000 cycles), it is more likely to indicate a phase transition in the cathode material structure or an increase in internal resistance due to electrolyte drying. Therefore, the abnormal characteristic alone is often insufficient to definitively determine the root cause of the failure; a joint judgment must be made based on the electrochemical constraints that characterize the current thermodynamic and kinetic state of the battery.
[0042] Specifically, the constraints of the electrochemical scenario include at least one of the following: state of charge range, operating temperature range, charge / discharge rate, cyclic aging stage, and resting time.
[0043] Among them, the state of charge range is used to define the thermodynamic equilibrium point of the battery and solve the problem of voltage ambiguity in the flat region; the operating temperature range is used to distinguish failure modes dominated by reaction kinetics (such as low-temperature sodium precipitation and high-temperature side reactions); the charge / discharge rate reflects the current polarization and heat generation rate; the cycle aging stage characterizes the cumulative effect of internal battery material degradation and interfacial film growth; and the resting time is related to the self-discharge recovery and concentration gradient relaxation process.
[0044] In one embodiment, step S40 includes the following steps: S41: Using anomaly representation features as the retrieval entry point, at least one candidate diagnostic knowledge is recalled from the historical knowledge base. Specifically, the extracted anomaly representation features (such as voltage change slope sequences, polarization internal resistance values, etc.) are mapped into high-dimensional feature vectors through a pre-trained encoding model, and a similarity search is performed in the vector database. This step leverages the distance metric advantage of the vector space model in a continuous feature space, enabling rapid recall of all candidate knowledge vectors that are similar to the current anomaly in signal morphology, regardless of whether their historical labels are completely identical. This ensures both the breadth and recall of the retrieval, avoiding omissions due to missing historical records or non-standard naming.
[0045] In one specific embodiment, the historical knowledge base includes a vector database and a relational database, where each element in the vector database is a vector representation of an element in the relational database in the feature space. Step S41 specifically includes: performing similarity matching between the feature vector corresponding to the anomaly representation features and the vector database to obtain at least one candidate knowledge vector. The candidate knowledge vector is a vector representation of candidate diagnostic knowledge in the feature space.
[0046] In battery fault diagnosis, abnormal characteristics often manifest as continuous time-series waveforms or complex statistical distributions. Traditional keyword-based or fixed-threshold-based retrieval methods struggle to capture waveform morphological similarities, easily leading to missed detections due to differences in description or minor deformations. While pure vector retrieval performs excellently in semantic and morphological matching, it often suffers from accuracy drift when handling precise numerical constraints (such as specific SOC percentages, precise temperature values, or cycle counts), making it difficult to strictly meet the hard boundary conditions of electrochemical scenarios. Therefore, this embodiment utilizes a vector database for unstructured feature semantic retrieval and a relational database for structured constraint filtering and complete knowledge storage, with both working together to achieve optimal retrieval results.
[0047] S42: Match the electrochemical scenario constraints with at least one candidate diagnostic knowledge one by one, and determine the candidate diagnostic knowledge with the highest comprehensive matching degree as the target diagnostic knowledge.
[0048] Based on this, step S42 uses the constraint vector corresponding to the electrochemical scenario constraint conditions to perform secondary screening and reordering of the aforementioned recalled candidate knowledge vectors.
[0049] In one specific embodiment, step S42 includes: S421: Based on the constraint vectors corresponding to the constraints of the electrochemical scenario, perform precise localization on at least one candidate knowledge vector to determine the target knowledge vector. Specifically, based on the constraint vectors corresponding to the constraints of the electrochemical scenario, perform precise localization on at least one candidate knowledge vector to determine the target knowledge vector. Here, the constraint vectors can be vector representations of discrete or continuous values such as SOC, temperature, and multiplier, or they can be precise field matching performed directly in a relational database using a structured query language.
[0050] S422: Retrieve complete target diagnostic knowledge from the relational database based on the target knowledge vector.
[0051] Step S42 first uses constraint vectors to perform nearest-neighbor rearrangement in the vector space, eliminating candidates that, although similar in waveform, clearly do not match the scenario (e.g., a low-temperature fault template matching a high-temperature scenario). Then, based on the unique identifier carried by the rearranged target knowledge vector, it retrieves complete target diagnostic knowledge from the relational database. This target diagnostic knowledge not only includes the final fault cause and handling solution, but also complete evidence chain information, such as detailed environmental parameters at the time of occurrence, associated original waveform segments, and confidence weights annotated by experts.
[0052] Compared to single-dimensional retrieval methods, step S40 retains the ability of vector retrieval to generalize and identify complex time-series signal patterns, while introducing electrochemical physical constraints as a rigid filter to ensure the self-consistency of the diagnostic results in terms of physical mechanisms.
[0053] In some embodiments, combined with Figure 3 Step S50 includes: S51: Determine the matching confidence level based on the similarity between the target diagnostic knowledge and the current abnormal scene features; wherein, the current abnormal scene features include at least one of the following: problem description, abnormal operation segment, abnormal characterization features, and electrochemical scene constraints.
[0054] Specifically, the system weighted and fused the waveform similarity of anomaly characteristics with the matching degree of electrochemical scene constraints. For example, if the voltage drop waveform in the target diagnostic knowledge has a very small dynamic time warp distance from the current anomaly segment, and the constraints such as SOC range and temperature range fall completely within the effective range of historical records, the matching confidence level approaches 1. Conversely, if only the waveforms are similar but the temperature conditions deviate significantly, or if the constraints match but the waveform morphology has significant distortion, the confidence level will decrease accordingly. This multi-dimensional confidence assessment mechanism effectively avoids the risk of blindly applying historical conclusions based solely on signal morphology similarity, ensuring the physical reliability of subsequent diagnostic generation.
[0055] S52: In response to the matching confidence level meeting the first preset condition, the first generation strategy is used to drive the generative model to output a deterministic diagnostic result. The deterministic diagnostic result includes a clear cause of the anomaly and a solution. The first preset condition is set to a high confidence threshold (e.g., 0.85 or 0.9). When this condition is met, it indicates that the current anomaly scenario has been fully covered by the historical knowledge base and the evidence is conclusive.
[0056] In practice, a first prompt word is constructed based on the first instruction, target diagnostic knowledge, and current abnormal scenario characteristics. This first prompt word is input into the generative model to instruct it to output a deterministic diagnostic result with a deterministic tone. The structure of the first prompt word typically includes three key elements: first, a clear role and task instruction, such as "As a battery safety expert, please provide a final diagnosis based on the following conclusive evidence"; second, a complete evidence chain injection, which embeds the historical causes, solutions, and currently matched abnormal fragments and constraints from the target diagnostic knowledge as immutable factual background into the prompt word; and third, strict output format constraints, requiring the model to directly output standardized fault codes, a definite failure mechanism description, and executable control parameters (such as "limit the discharge current to 0.3C"), and strictly prohibiting the use of vague terms such as "possibly" or "suspected." This affirmative generation strategy can minimize the model's room for maneuver, ensuring that the output diagnostic result has a high degree of determinism and operability, facilitating the battery management system to directly execute corresponding protection actions and achieve rapid closed-loop processing of known faults.
[0057] S53: In response to the matching confidence level not meeting the first preset condition, the second generation strategy is used to drive the generative model to output a reference diagnostic result. The reference diagnostic result includes a reference anomaly cause and a reference handling solution. When the confidence level is lower than the threshold, it means that the current situation may belong to a new type of fault, a compound fault, or a scenario with poor data quality. Forcing a definitive conclusion is very likely to cause erroneous operations.
[0058] In practice, a second prompt word is constructed based on the second instruction, target diagnostic knowledge, and current abnormal scenario characteristics. This second prompt word is input into the generative model to instruct it to output a speculative diagnostic result and generate supplementary information prompts. Unlike the first prompt word, the second prompt word explicitly requires the model to perform "hypothetical reasoning" rather than "factual statements" at the instruction level. For example, it instructs the model to "analyze the most likely potential causes based on limited information and list the data items that need further verification." At the content level, the second prompt word not only injects the target diagnostic knowledge with the highest matching degree as a reference anchor, but also explicitly adds risk warning paragraphs to inform the model of the incompleteness of the current evidence. The resulting speculative diagnostic result typically includes two parts: first, a speculative conclusion with probabilistic properties, such as "Combining the characteristics of low temperature and long static placement, it is speculated that the risk of sodium precipitation at the negative electrode is high, but the possibility of sensor drift cannot be ruled out"; second, a structured list of supplementary information, such as "It is recommended to supplement the collection of open-circuit voltage curves during static placement" or "Please check the temperature sensor calibration record." This suggestive generation strategy not only fully utilizes the reasoning capabilities of generative models to provide troubleshooting directions for operations and maintenance personnel, but also prevents AI illusions from misleading decisions through clear boundary constraints, thus ensuring the safety baseline of the battery system under uncertain conditions.
[0059] It should be understood that the specific threshold of the first preset condition, the template structure of the prompt words, and the specific type of the generative model (such as a large language model, a time-series prediction model, or a multimodal model) can all be adjusted according to the actual safety level and computing resources of the battery system. For example, in high-safety-requirement scenarios such as energy storage power stations, the threshold of the first preset condition can be increased to allow more edge cases to enter the reference diagnostic process to ensure absolute safety; while in scenarios with high real-time requirements such as two-wheeled vehicles, the token length of the first prompt word can be appropriately optimized to improve response speed. Regardless of the specific configuration adopted, the essence of this embodiment, which drives a differentiated generation strategy through confidence level, lies in establishing an intelligent decision-making defense line adapted to battery safety risk management. This enables the system to efficiently handle known problems and prudently deal with unknown risks when facing massive and complex data, achieving a dynamic balance between intelligent diagnosis and safe management.
[0060] In some embodiments, the method further includes: S61: In response to the diagnostic results meeting the preset reference conditions, generate candidate diagnostic knowledge based on the diagnostic results and supplementary investigation information.
[0061] In practical implementation, the preset reference conditions can include various scenarios: for example, when the diagnostic result generated in step S53 is a reference diagnostic result and supplementary data that can confirm the cause of the fault is subsequently obtained; or when maintenance personnel correct the deterministic diagnostic result output by the system, and the correction content differs significantly from the original diagnosis; or when the same abnormal characteristic is marked as "unresolved" or "false alarm" multiple times in a short period of time. These scenarios all indicate that the existing historical knowledge base has coverage blind spots or cognitive biases in the current scenario, and has the value of being precipitated into new knowledge. On this basis, the system will automatically or semi-automatically integrate the original reference diagnostic results with the subsequently obtained supplementary investigation information to form structured candidate diagnostic knowledge. Among them, the sources of supplementary investigation information are diverse, including verification data collected by the battery management system during subsequent operation (such as the open circuit voltage recovery curve after resting, the capacity retention rate of the next charge), physical inspection reports after offline maintenance and disassembly (such as photos of sodium precipitation on the electrode, records of separator damage), and statistical patterns derived from cloud big data analysis. By combining vague speculative conclusions with solid empirical data, candidate diagnostic knowledge has completed the qualitative leap from "suspected" to "certain".
[0062] S62: In response to the approval of candidate diagnostic knowledge, write the candidate diagnostic knowledge into the historical knowledge base and update the retrieval index of the historical knowledge base.
[0063] In practice, the first step is to verify the accuracy of the anomaly characteristics. Reviewers or automated verification programs must confirm whether the waveform features or statistical indicators extracted from the candidate knowledge truly reflect the physical nature of the fault, eliminating false features caused by sensor noise, sampling anomalies, or data processing algorithm defects, and preventing accidental signal interference from being solidified as fault criteria. The second step is to verify the completeness of the electrochemical scenario constraints, focusing on whether the candidate knowledge fully records all key boundary conditions at the time of the fault. For example, whether specific low-temperature ranges, high-rate pulses, or long static periods are omitted, avoiding the generalization and misuse of new knowledge in subsequent applications due to missing constraints. Finally, the effectiveness of the anomaly causes and solutions must be verified. This requires combining actual maintenance verification records or simulation test results to confirm that the recorded measures actually eliminated the fault or mitigated the risk, rather than merely reflecting a temporal sequence. Only after all three dimensions of verification pass will the candidate diagnostic knowledge be approved for the next stage of database entry.
[0064] After approval, the system performs a dual-database synchronous update operation to complete the knowledge loop. Specifically, on the one hand, candidate diagnostic knowledge is written as a new structured record into the relational database of the historical knowledge base, assigned a unique knowledge identifier, and an association index is established between it and metadata such as anomaly type, applicable model, and battery system to ensure the traceability and manageability of the knowledge. On the other hand, a pre-trained semantic encoding model is used to map the anomaly representation features and constraints in the candidate diagnostic knowledge into a high-dimensional vector, which is then inserted or updated into the vector database of the historical knowledge base, and the corresponding approximate nearest neighbor search index is reconstructed or incrementally updated. This synchronous update mechanism ensures that the newly added knowledge is not only logically stored but also becomes "retrievalable" in the feature space. For example, when a similar voltage anomaly waveform reappears in the future, vector retrieval can immediately recall this newly added knowledge without waiting for periodic full model retraining. Through the aforementioned positive cycle of "diagnosis-feedback-review-learning", this embodiment enables the battery management system to have lifelong learning capabilities while ensuring safety. As the deployment scale expands and the operating time accumulates, it can continuously improve the diagnostic coverage and accuracy of new failure modes and complex compound faults.
[0065] Steps S61 and S62 aim to address the dynamic adaptability problem of battery failure modes evolving with material system iterations and changes in operating conditions. At the same time, by introducing a controlled human review process, the risk of erroneous knowledge accumulation and model drift that may be brought about by purely automated machine learning is fundamentally avoided, ensuring the safety and reliability of the diagnostic system in long-term operation.
[0066] Example 2 Combination Figure 4This application provides a battery management system for use in sodium-ion batteries, comprising: The data acquisition module 21 is used to acquire the battery's operating status data.
[0067] The matching and filtering module 22 is used to match and filter the running status data based on the normal operation feature table and the abnormal operation feature table, and to filter out abnormal operation segments.
[0068] The feature and constraint determination module 23 is used to extract abnormal characterization features of abnormal operating segments and determine the electrochemical scenario constraints of the battery.
[0069] The retrieval and positioning module 24 is used to perform hierarchical retrieval in the historical knowledge base based on abnormal characterization features and electrochemical scenario constraints to obtain target diagnostic knowledge.
[0070] The diagnosis generation module 25 is used to determine the generation strategy based on the matching confidence of the target diagnostic knowledge, and generate diagnostic results based on the generation strategy and the target diagnostic knowledge.
[0071] The control execution module 26 is used to execute corresponding battery control actions based on the diagnostic results.
[0072] In one embodiment, the battery management system further includes a cloud communication module, which is used to upload operating status data and diagnostic results to a cloud server, and to receive firmware online upgrade instructions and visual configuration data issued by the cloud server.
[0073] Specifically, the cloud communication module supports MQTT, HTTPS, or proprietary encryption protocols, enabling secure bidirectional data transmission between the device and the cloud. In the uplink direction, in addition to regular monitoring data, this module synchronizes locally generated diagnostic results, abnormal segment snapshots, and confidence information to the cloud, providing empirical samples for big data analysis and knowledge base updates. In the downlink direction, the cloud communication module not only supports traditional BMS firmware code OTA upgrades but also incremental updates to local normal operation characteristic tables, abnormal operation characteristic tables, and historical knowledge bases. This means that as the cloud continuously learns from massive amounts of operational data and identifies new failure modes, the diagnostic capabilities of terminal devices can be continuously improved through dual-channel OTA of "data + knowledge" without frequent hardware replacements or factory re-flashing of complete firmware. Furthermore, the visualized configuration data allows maintenance personnel to customize the thresholds of local diagnostic parameters, adjust alarm sensitivity, or configure the display style of diagnostic results through the cloud interface, thereby achieving flexible adaptation for different application scenarios.
[0074] In terms of visualization, the cloud server or local human-machine interface provides multi-dimensional status presentation based on the received diagnostic results and confidence information. For example, the interface can trace back the original waveform of the abnormal operation segment in the form of a timeline and overlay the abnormal characterization feature markers identified by the system; at the same time, it can intuitively display the matching confidence of the current diagnostic results in the form of color coding or dashboards, helping maintenance personnel to quickly judge the credibility of the diagnostic conclusions. For reference diagnostic results, the interface can also link to display a list of supplementary information and suggested troubleshooting steps, forming a human-machine collaborative fault handling guide. This transparent information display mechanism effectively enhances users' trust in the intelligent diagnostic system and provides sufficient basis for manual review and decision-making.
[0075] Through the modular architecture and cloud collaboration mechanism described above, the battery management system provided in this embodiment not only fully implements the various technical means in the aforementioned method embodiments, but also endows the system with high flexibility, scalability and continuous evolution capability through the design of hardware and software decoupling and edge-cloud linkage, thereby better adapting to the actual needs of rapid iteration of battery technology and diversified application scenarios.
[0076] Example 3 Combination Figure 5 An electronic device 1 includes a memory 11, a processor 10, and a computer program 12 stored in the memory 11 and executable on the processor 10. When the processor 10 executes the computer program 12, it implements any of the sodium-ion battery management methods in Embodiment 1.
[0077] For example, the computer program 12 may be divided into one or more units, which are stored in the memory 11 and executed by the processor 10 to complete the present application. The one or more units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program 12 in the electronic device 1.
[0078] The electronic device 1 can be any electronic device with control functions. The electronic device 1 may include, but is not limited to, a processor 10 and a memory 11. Those skilled in the art will understand that... Figure 5 This is merely an example of electronic device 1 and does not constitute a limitation on electronic device 1. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 1 may also include input / output devices, network access devices, buses, etc.
[0079] The processor 10 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.
[0080] The memory 11 can be an internal storage unit of the electronic device 1, such as a hard disk or RAM. The memory 11 can also be an external storage device of the electronic device 1, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory 11 can include both internal and external storage units of the electronic device 1. The memory 11 is used to store the computer program and other programs and data required by the terminal device. The memory 11 can also be used to temporarily store data that has been output or will be output.
[0081] Example 4 This application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the above-described sodium-ion battery management methods. Specific implementations of each of the above operations can be found in the preceding embodiments and will not be repeated here.
[0082] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.
[0083] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.
[0084] Since the instructions stored in the computer-readable storage medium can execute the steps of any sodium-ion battery management method provided in the embodiments of this application, the beneficial effects that any sodium-ion battery management method provided in the embodiments of this application can achieve can be realized, as detailed in the preceding embodiments, and will not be repeated here.
[0085] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0086] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0087] In the embodiments provided in this application, it should be understood that the disclosed apparatus / terminal devices and methods can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.
[0088] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0089] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0090] If an integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. A computer-readable medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in a computer-readable medium can be appropriately added to or subtracted according to the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.
[0091] Example 5 This application provides a computer program that, when executed by a processor, implements any of the above-described sodium-ion battery management methods.
[0092] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for managing sodium-ion batteries, characterized in that: Includes the following steps: S10: Obtain battery operating status data; S20: Based on the normal operation feature table and the abnormal operation feature table, the operation status data is matched and filtered to identify abnormal operation segments; S30: Extract the abnormal characterization features of the abnormal operation segment and determine the electrochemical scenario constraints of the battery; S40: Based on the abnormal characterization features and the electrochemical scenario constraints, perform a hierarchical retrieval in the historical knowledge base to obtain the target diagnostic knowledge; S50: Determine the generation strategy based on the matching confidence of the target diagnostic knowledge, and generate a diagnostic result based on the generation strategy and the target diagnostic knowledge to execute the corresponding battery management action.
2. The sodium-ion battery management method according to claim 1, characterized in that: Step S20 includes the following steps: S21: Divide the running status data into multiple data segments and extract the segment features of each data segment; S22: Based on the fragment features, match them with the abnormal operation feature table, and determine the successfully matched data fragments as known abnormal fragments; S23: Based on the fragment features of the unmatched data fragments, match them with the normal operation feature table, determine the successfully matched data fragments as normal background fragments and filter them, and determine the unmatched data fragments as unknown abnormal fragments; S24: The known abnormal segment and the unknown abnormal segment are identified as the abnormal running segment.
3. The sodium-ion battery management method according to claim 2, characterized in that: Step S21 includes the following steps: S211: Divide the operating status data into the multiple data segments according to at least one of the preset time window, charging / discharging stage switching point, and signal characteristic mutation point; S212: Perform feature extraction on each of the data segments to obtain the segment features; wherein, the segment features include at least one time-series waveform feature or electrochemical feature among voltage change slope, polarization internal resistance, and temperature change rate, or at least one statistical feature among mean, variance, and peak value.
4. The sodium-ion battery management method according to claim 1, characterized in that: The electrochemical scenario constraints include at least one of the following: state of charge range, operating temperature range, charge / discharge rate, cyclic aging stage, and resting time; step S40 includes the following steps: S41: Using the abnormal characterization features as the retrieval entry point, recall at least one candidate diagnostic knowledge from the historical knowledge base; S42: Match the electrochemical scenario constraints with the at least one candidate diagnostic knowledge one by one, and determine the candidate diagnostic knowledge with the highest comprehensive matching degree as the target diagnostic knowledge.
5. The sodium-ion battery management method according to claim 4, characterized in that: The historical knowledge base includes a vector database and a relational database, wherein the elements in the vector database are vector representations of the elements in the relational database in the feature space; Step S41 includes: performing similarity matching between the feature vector corresponding to the anomaly representation feature and the vector database to obtain at least one candidate knowledge vector; Step S42 includes: S421: Based on the constraint vector corresponding to the electrochemical scenario constraint conditions, perform precise localization on the at least one candidate knowledge vector to determine the target knowledge vector; S422: Based on the target knowledge vector, retrieve the complete target diagnostic knowledge from the relational database.
6. The sodium-ion battery management method according to claim 1, characterized in that: Step S50 includes: S51: Determine the matching confidence level based on the similarity between the target diagnostic knowledge and the current abnormal scene features; wherein, the current abnormal scene features include at least one of the problem description, the abnormal running segment, the abnormal characterization features, and the electrochemical scene constraints; S52: In response to the matching confidence meeting the first preset condition, the first generation strategy is used to drive the generative model to output a deterministic diagnostic result, the deterministic diagnostic result including a clear cause of the anomaly and a solution; S53: In response to the matching confidence not meeting the first preset condition, the second generation strategy is used to drive the generative model to output a reference diagnostic result, which includes a reference anomaly cause and a reference processing solution.
7. The sodium-ion battery management method according to claim 6, characterized in that: Step S52 includes: constructing a first prompt word based on the first instruction, the target diagnostic knowledge, and the current abnormal scene features; inputting the first prompt word into the generative model to instruct the generative model to output the deterministic diagnostic result in a deterministic tone; Step S53 includes: constructing a second prompt word based on the second instruction, the target diagnostic knowledge, and the current abnormal scene features; inputting the second prompt word into the generative model to instruct the generative model to output the reference diagnostic result in a speculative tone and generate a prompt for supplementary information.
8. The sodium-ion battery management method according to claim 1, characterized in that: The method further includes: S61: In response to the diagnostic result meeting the preset reference conditions, generate candidate diagnostic knowledge based on the diagnostic result and supplementary investigation information; S62: In response to the candidate diagnostic knowledge passing the review, the candidate diagnostic knowledge is written into the historical knowledge base, and the retrieval index of the historical knowledge base is updated.
9. A battery management system applied to sodium-ion batteries, characterized in that: include: The data acquisition module is used to acquire battery operating status data; The matching and filtering module is used to match and filter the running status data based on the normal operation feature table and the abnormal operation feature table, and to filter out abnormal operation segments. The feature and constraint determination module is used to extract the abnormal characterization features of the abnormal operation segment and determine the electrochemical scenario constraints of the battery. The retrieval and positioning module is used to perform hierarchical retrieval in the historical knowledge base based on the abnormal characterization features and the electrochemical scenario constraints to obtain target diagnostic knowledge. A diagnosis generation module is used to determine a generation strategy based on the matching confidence of the target diagnostic knowledge, and to generate a diagnostic result based on the generation strategy and the target diagnostic knowledge. The control execution module is used to perform corresponding battery control actions based on the diagnostic results.
10. The battery management system according to claim 9, characterized in that: The battery management system also includes: The cloud communication module is used to upload the operating status data and the diagnostic results to the cloud server, and to receive firmware online upgrade instructions and visual configuration data issued by the cloud server.
