Battery risk assessment method and system based on long-term abnormal evolution

Abstract

This invention discloses a battery risk assessment method and system based on long-term time-series anomaly evolution. The method includes: collecting voltage data of each battery cell from a battery management system; calculating the voltage variation coefficient at each stage based on the voltage data and converting it into a consistency score; identifying abnormal battery cells in the battery pack using the DBSCAN algorithm for cells with consistency scores below a set threshold; analyzing the anomaly detection results within a set number of days, combining spatial proximity characteristics, and adjusting and calculating the risk score of each battery cell using common-mode penalty and thermal propagation gain operators; constructing a multi-dimensional risk matrix based on the risk scores of all battery cells, and calculating the overall risk score of the battery pack. By implementing this method, it is possible to combine long-term operational data for continuous tracking and analysis of abnormal cells, and establish a mechanism for quantifying individual cell risk and assessing the overall risk of the battery pack, thereby improving the accuracy and reliability of battery system fault early warning.

Description

Technical Field

[0001] This invention relates to battery risk assessment methods, and more specifically to battery risk assessment methods and systems based on long-term time-series anomalous evolution. Background Technology

[0002] A battery system is composed of multiple battery cells connected in series and parallel. Due to differences in manufacturing processes, operating environments, and aging rates, these battery cells are prone to problems such as inconsistent voltage, asynchronous capacity decay, and differences in thermal characteristics during long-term operation. With the continuous expansion of new energy vehicles and energy storage systems, accurate identification and safety warnings of battery cell anomalies have become an important research direction in battery safety management. Effectively managing and preventing these problems is key to ensuring the efficient and safe operation of battery systems.

[0003] Current technologies primarily rely on operational data such as voltage and temperature of individual battery cells, using methods like consistency analysis, threshold judgment, statistical analysis, or clustering algorithms to screen and locate abnormal battery cells. For example, by analyzing the voltage deviation, voltage dispersion, or temperature distribution characteristics of each cell within the battery pack, abnormal cells deviating from normal operating conditions can be identified, and fault warnings can be issued. However, most existing methods focus on anomaly detection within a single instance or short time window, only identifying anomalies at the current moment. They lack effective modeling capabilities for the persistence, accumulation, and evolutionary trends of abnormal behavior over long-term operation. Furthermore, these methods typically only provide a binary judgment of "whether it is abnormal," making it difficult to further quantify the risk differences between different battery cells and failing to meet the needs for more accurate risk assessment.

[0004] In addition to the aforementioned issues, existing solutions also have significant shortcomings in terms of anomaly propagation characteristics and comprehensive system-level risk assessment. When multiple cells within a battery pack exhibit abnormal fluctuations over a long period, existing methods struggle to accurately assess the impact of these abnormal cells on the overall safety of the battery pack, thus failing to achieve a dynamic quantitative assessment of the overall risk level of the battery pack.

[0005] Therefore, it is necessary to design a new method that can combine long-term operational data to continuously track and analyze abnormal individual cells, and establish a mechanism for quantifying individual cell risk and assessing the overall risk of the battery pack, in order to improve the accuracy and reliability of battery system fault early warning. This not only solves the problem that existing technologies can only identify and locate abnormal battery cells, but also effectively quantifies the degree of fault risk of individual cells, achieving overall risk assessment of the battery pack. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a battery risk assessment method and system based on long-term time-series abnormal evolution.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a battery risk assessment method based on long-term time-series anomalous evolution, comprising: Voltage data of each battery cell is collected from the battery management system during four stages: charging, charging rest, discharging, and discharging rest. Based on the voltage data, the voltage variation coefficient for each stage is calculated and converted into a consistency score; For battery packs with consistency scores below a set threshold, the voltage data during the charging and resting phase is normalized, and the DBSCAN algorithm is used to identify abnormal battery cells to obtain anomaly detection results. The anomaly detection results within a set number of days are analyzed, and combined with spatial proximity characteristics, the risk score of each battery cell is adjusted and calculated using common-mode penalty and thermal spread gain operators. Based on the risk scores of all individual battery cells, a multi-dimensional risk matrix is ​​constructed, including critical risk values, diffusion risk values, and concentration risk values, and the overall risk score of the battery pack is calculated.

[0008] This invention also provides a battery risk assessment system based on long-term time-series anomalous evolution, comprising: The acquisition unit is used to collect voltage data of each battery cell from the battery management system during the four stages of charging, charging rest, discharging, and discharging rest. The consistency score calculation unit is used to calculate the voltage variation coefficient of each stage based on the voltage data and convert it into a consistency score; The anomaly detection unit is used to normalize the voltage data of battery packs with consistency scores below a set threshold during the charging and resting phase, and to use the DBSCAN algorithm to identify abnormal battery cells in order to obtain anomaly detection results. The risk score calculation unit is used to analyze the anomaly detection results within a set number of days, combine spatial proximity characteristics, and adjust and calculate the risk score of each battery cell using common mode penalty and thermal spread gain operators. The overall risk scoring unit is used to construct a multi-dimensional risk matrix, including key risk values, diffusion risk values, and concentration risk values, based on the risk scores of all individual battery cells, and to calculate the overall risk score of the battery pack.

[0009] The present invention also provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described above.

[0010] The advantages of this invention compared to existing technologies are as follows: This invention collects voltage data of each cell at different operating stages from the battery management system, calculates the voltage variation coefficient for each stage and converts it into a consistency score, normalizes the voltage data during the charging and resting stages of the low-scoring battery pack, and uses the DBSCAN algorithm to identify abnormal cells. Then, combining the anomaly detection results within a set number of days with spatial proximity characteristics, it adjusts and calculates the risk score of each cell using common-mode penalty and thermal propagation gain operators. Finally, it constructs a multi-dimensional risk matrix including critical risk values, diffusion risk values, and concentration risk values ​​to calculate the overall risk score of the battery pack. This method not only achieves continuous tracking and analysis of abnormal cells over a long period but also solves the problem of existing technologies only being able to qualitatively identify and locate anomalies. It achieves effective quantification of fault risk levels and overall risk assessment of the battery pack, thereby significantly improving the accuracy and reliability of battery system fault early warning.

[0011] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A schematic flowchart of a battery risk assessment method based on long-term time-series anomalous evolution provided in an embodiment of the present invention; Figure 2 A schematic diagram of the original time type provided in an embodiment of the present invention; Figure 3 A schematic diagram illustrating the addition of a time type as provided in an embodiment of the present invention; Figure 4 A schematic diagram of consistency scoring provided for an embodiment of the present invention; Figure 5 This is a schematic diagram of the voltage curve during the charging resting stage provided in an embodiment of the present invention; Figure 6 A schematic diagram of the risk score of a single battery cell provided in an embodiment of the present invention; Figure 7 A schematic block diagram of a battery risk assessment system based on long-term time-series anomalous evolution provided in an embodiment of the present invention; Figure 8 A schematic block diagram of a computer device provided for an embodiment of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0016] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0017] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0018] Please see Figure 1 , Figure 1 This is a flowchart illustrating a battery risk assessment method based on long-term anomaly evolution provided in an embodiment of the present invention. This method is applied in a server. It collects voltage data of each cell at different operating stages from the battery management system, calculates the voltage variation coefficient and converts it into a consistency score. The DBSCAN algorithm is used to identify potentially abnormal battery cells during the charging and resting phase. The risk score of each battery cell is adjusted by combining spatial proximity features, common-mode penalty, and thermal propagation gain operators. Subsequently, a multi-dimensional risk matrix is ​​constructed based on the risk scores of all cells. A dynamic boundary adjustment mechanism is used to integrate the internal evolution comprehensive risk and anomaly coverage factor, ultimately calculating the overall risk score of the battery pack. This method not only achieves accurate identification and location of abnormal battery cells but also quantifies their fault risk level, providing an overall risk assessment of the battery pack, thereby significantly improving the accuracy and reliability of battery system fault early warning.

[0019] Figure 1 This is a flowchart illustrating the battery risk assessment method based on long-term time-series anomalous evolution provided in an embodiment of the present invention. Figure 1 As shown, the method includes the following steps S110 to S150.

[0020] S110. Collect voltage data of each battery cell from the battery management system during the four stages of charging, charging rest, discharging, and discharging rest.

[0021] In this embodiment, the voltage data of each battery cell at different operating stages needs to be obtained from the battery management system (BMS). These stages include: charging stage, charging resting stage, discharging stage, and discharging resting stage. Each stage has its specific meaning and function, as detailed below: Charging phase: In this phase, the battery receives current from an external power source to charge. Voltage changes during this process can reflect some fundamental characteristics of the battery, such as capacity differences and self-discharge rate.

[0022] Charging resting phase: After charging is completed, the battery enters a resting state for a period of time. Without external current flowing in or out, the internal voltage of the battery gradually stabilizes. Data from this phase is crucial for detecting voltage consistency between individual battery cells, as the voltage at this stage better reflects the battery's true health condition.

[0023] Discharge phase: The process by which the battery provides energy to the load. By observing the voltage changes during discharge, we can understand the battery's performance and remaining capacity.

[0024] Discharge resting phase: Similar to the charging resting phase, but after discharge. This phase can help identify battery cells that behave abnormally during discharge, as any inconsistencies may be more apparent under no-load conditions.

[0025] Time Division: Based on the battery's operation log, the battery pack's timeline is divided into the four stages mentioned above. The duration of each stage may vary depending on the application.

[0026] Voltage data acquisition: At each stage, the voltage of all individual cells within the battery pack is measured and recorded. To ensure the accuracy and representativeness of the data, high-frequency voltage sampling is typically required.

[0027] Data preprocessing: Since the data obtained directly from the BMS may contain noise or other interference factors, it is necessary to preprocess the raw data before analysis, such as filtering and smoothing, to remove unnecessary noise and improve the accuracy of subsequent analysis.

[0028] The main purpose of this step is to provide foundational data support for subsequent analysis and evaluation. By accurately collecting and preliminarily processing the voltage data from these four stages, the voltage differences between individual cells within the battery pack can be effectively identified, laying a solid foundation for further calculations of voltage variation coefficients, consistency scores, and the identification of potentially abnormal battery cells. Furthermore, this meticulous data collection method enables a deeper understanding of the battery's operating state and its long-term evolution trends, thereby improving the battery system's fault prediction capabilities and safety.

[0029] S120. Based on the voltage data, calculate the voltage variation coefficient for each stage and convert it into a consistency score.

[0030] In this embodiment, the voltage variation coefficient refers to the quantification of the voltage difference between individual cells within the battery pack by analyzing the maximum and minimum voltage values ​​of each cell over a specific time period. The consistency score transforms this degree of difference into an easily understandable rating to evaluate the consistency performance within the battery pack.

[0031] By using consistency scores, battery packs with poor consistency performance from a large number of battery pack clusters were identified as requiring close monitoring. The standard was a consistency score below 90 points during the charging and resting phase. The rationale for using the charging and resting phase score for screening is as follows: During the charging and discharging operation, the active intervention of the BMS balancing strategy can smooth out the original voltage differences between individual cells; however, during the discharging and resting phase, due to the inherent electrochemical characteristics of lithium batteries, such as inconsistent voltage rebound at the end of polarization and amplified voltage sensitivity in the low-charge region, voltage data is prone to non-fault fluctuations.

[0032] In contrast, batteries in the charging resting phase are in a high state of charge (SOC) and free from current interference. Voltage data from this phase most accurately reflects the intrinsic differences in individual cells (capacity, internal resistance, etc.). Therefore, cluster analysis using only voltage data from the charging resting phase can effectively eliminate spurious outliers caused by electrochemical background noise, thus more accurately identifying potentially abnormal battery cells.

[0033] In one embodiment, step S120 described above may include steps S121 to S125.

[0034] S121. Divide the battery pack's timeline into four stages: charging, charging rest, discharging, and discharging rest.

[0035] Based on the operation logs recorded by the Battery Management System (BMS), the time intervals of the battery pack under different operating modes are identified and segmented: charging, charging rest, discharging, and discharging rest. This step ensures that subsequent analysis can be performed on each specific stage, thereby more accurately reflecting the battery's performance under different operating conditions.

[0036] S122. For each stage, calculate the maximum and minimum voltage values ​​of all cells in the battery pack at each moment.

[0037] Within a defined time period, the voltage value of each individual cell in the battery pack is recorded at each sampling point.

[0038] For each stage, find the maximum voltage value U of all individual cells within that stage. max-j and minimum voltage value U min-j , where j represents the j-th sampling time.

[0039] S123. Calculate the root mean square error using the maximum and minimum voltage values.

[0040] The root mean square error (RMSE) measures the magnitude of the deviation between a set of values ​​and their mean. Here, it is used to represent the difference between the highest voltage curve and the lowest voltage curve.

[0041] S124. The voltage variation coefficient is obtained by dividing the root mean square error by the average voltage value.

[0042] In this embodiment, the average voltage value can be obtained by averaging the voltage values ​​at all sampling points.

[0043] The coefficient of variation (CV) is defined as the ratio of the root mean square error to the average voltage value.

[0044] S125. Based on the coefficient of variation, a segmented scoring model is adopted to map voltage differences into consistency score intervals.

[0045] Using a pre-defined segmented scoring model, based on the calculated voltage variation coefficient... Convert it into a consistency score. For example, different threshold ranges can be set to correspond to different score segments: when The score is 90-100; when When, the score is 70-90; when When, the score is 40-70; when At that time, the score was below 40. Here... It is a preset baseline value, such as 2.5%, which can be adjusted according to the actual situation.

[0046] The above steps effectively quantify the voltage consistency level among individual cells within the battery pack and convert it into an intuitive consistency score, facilitating subsequent anomaly detection and risk assessment. This process not only improves the understanding of battery health but also provides strong support for maintaining and optimizing the battery system's operation.

[0047] In this embodiment, please refer to Figure 2 and Figure 3 When analyzing the battery pack, two basic operational phases were first identified and defined: charging and discharging. Based on these two basic operational modes, two derivative types were further introduced: charging rest and discharging rest. Specifically, when an operational phase ends and enters a resting state, the classification of this resting period depends on the type of operation in the preceding phase. If the preceding phase was charging, this resting interval is marked as charging rest; conversely, if the preceding phase was discharging, this resting period is classified as discharging rest. The duration of these resting intervals is determined by calculating the time difference between adjacent operational phases and then categorized according to the above rules. In this way, the timeline of each battery cluster is reconstructed into a unified sequence containing four types: charging, charging rest, discharging, and discharging rest. This reorganization of the timeline provides solid time reference data for the subsequent calculation of key indicators such as voltage consistency, thereby enabling a more accurate assessment of the consistency performance and health status of individual cells within the battery pack. This method not only improves the accuracy of data processing but also lays the foundation for in-depth analysis of battery performance.

[0048] exist Figure 2 The graph shows the voltage time sequence trend of all cells in a certain battery pack, starting from midnight on December 7th and ending at midnight on December 9th. The original time type only has two types: charging and discharging. The green block indicates that charging is in progress during this period, and the voltage of all cells shows an upward trend. The blue block indicates that discharging is in progress during this period, and the voltage of all cells shows a downward trend.

[0049] exist Figure 3 In addition to the two basic stages of charging and discharging, charging rest and discharging rest types derived from these two are introduced. The rest interval is calculated by the time difference between adjacent operating segments and is classified according to the type of the preceding segment: if the preceding stage is charging, the rest interval is marked as charging rest; if the preceding stage is discharging, it is classified as discharging rest.

[0050] Ultimately, there are four time types: charging (green block), charging rest (red block), discharging (blue block), and discharging rest (yellow block). As can be seen, during the rest phase, the voltage of all cells remains stable, with the voltage value remaining almost unchanged.

[0051] The voltage variation coefficient for each stage is calculated using the following formula: ; ; In the formula, Indicates the number of each battery cell in the battery pack The maximum voltage value at any given moment, and all of its values ​​constitute the highest voltage curve; Indicates the number of each battery cell in the battery pack The minimum voltage value at any given time, and all of its values ​​form the minimum voltage curve; This represents the root mean square error between the highest voltage curve data and the lowest voltage curve data. This represents the coefficient of variation of the battery pack voltage.

[0052] Finally, based on this coefficient, the algorithm adopts a piecewise scoring model, mapping voltage differences of different magnitudes to a scoring range of 0–100 using linear interpolation, making the scoring interpretable. This yields the voltage consistency scores for each battery pack at four time stages, as shown in Table 1.

[0053] Table 1. Voltage Consistency Scores

[0054] threshold The threshold is set at 2.5%, which roughly corresponds to a range of 95 millivolts (mV). This setting is used to evaluate the consistency score of the battery pack, and its calculation method is existing technology. By adopting this threshold, the voltage differences between individual cells within the battery pack can be effectively quantified, thereby evaluating the consistency status of the entire battery pack. This method provides a scientific and practical standard that helps ensure the accuracy and reliability of evaluation results when using charging data collected in the cloud for lithium battery pack performance analysis.

[0055] Please see Figure 4 The low consistency score during the discharge resting period reflects poor voltage consistency during this period.

[0056] The small plot on the right of the image shows the calculated voltage consistency scores for each of the four stages. A higher score indicates better voltage consistency in that stage. It can be seen that the voltage consistency scores are high during the charging, charging rest, and discharging stages, indicating good voltage consistency, with the voltage trends of all cells almost overlapping into a single line. However, during the discharging rest stage, the voltage consistency score is only 30.989, a low score indicating poor voltage consistency. As seen from the yellow blocks (i.e., the discharging rest stage), the voltage trends of all cells are relatively dispersed. Different colored lines represent the voltage trends of different cells. During the discharging rest stage, the highest voltage cell in the battery pack is approximately 3.1V, and the lowest voltage cell is approximately 3.03V, with a voltage difference of about 0.07V. Compared to the other three stages, there is almost no voltage difference, indicating poor consistency.

[0057] S130. For battery packs with consistency scores below a set threshold, normalize the voltage data during the charging and resting phase, and use the DBSCAN algorithm to identify abnormal battery cells to obtain anomaly detection results.

[0058] In this embodiment, the anomaly detection result refers to the judgment output after performing density clustering analysis on the voltage time sequence characteristics of each battery cell in the battery pack using the DBSCAN clustering algorithm during the charging resting stage. The battery cells marked as noise points (cluster label is -1) and whose voltage change trends deviate significantly from the mainstream pattern are identified as potential abnormal cells, and their identification information, battery pack to which they belong, and the time of anomaly occurrence are recorded.

[0059] The voltage curves of all cells in battery packs with a charging-resting consistency score below 90 during the charging-resting phase are displayed. Due to the high frequency of voltage data acquisition and the large number of cells, clustering is performed using data from the past two days to reduce memory pressure during data reading. The calculation is performed daily, with results for abnormal cells displayed daily. The output shows whether each cell in the battery pack is identified as abnormal or normal by the DBSCAN clustering algorithm.

[0060] Step S130 aims to identify potential abnormal battery cells that differ significantly from most battery cells from the voltage time-series data during the charging resting phase using a density-based unsupervised clustering algorithm, thereby achieving accurate location of abnormal cells inside the battery pack.

[0061] In one embodiment, step S130 described above may include steps S131 to S133.

[0062] S131. Select battery packs whose consistency score is lower than a set threshold during the charging and resting phase as the analysis object, and obtain the voltage curve data of all battery cells of the analysis object during the charging and resting phase.

[0063] Specifically, during the operation of an energy storage system, the battery management system typically activates voltage balancing strategies only in the high SOC range. Once the individual cell voltage drops to a certain value, the balancing function usually ceases. Therefore, voltage consistency at the end of discharge and in the low SOC phase is easily affected by SOC nonlinearity, voltage plateau characteristics, and operating condition disturbances, limiting its ability to characterize true abnormal states. In contrast, the charging and resting charging phases are typically in a higher SOC range. At this time, cell voltage is more sensitive to capacity differences, self-discharge differences, and polarization characteristics. Simultaneously, the operating current gradually decreases, weakening the influence of dynamic polarization and ohmic voltage drop, allowing the voltage to more accurately reflect the internal state differences of the cells. Therefore, if a battery cluster still exhibits poor voltage consistency during the resting charging phase, it often indicates abnormal voltage recovery behavior or potential performance degradation in some cells, making its anomaly indication more significant. Based on these characteristics, this step first selects battery packs with low voltage consistency scores (specifically, scores below 90) during the resting charging phase as the analysis object, and selects the voltage time series of this phase as the clustering input data. This screening mechanism effectively focuses on battery packs with poor consistency and potential anomaly risks, avoiding the waste of computing resources caused by indiscriminate analysis of all battery packs, while ensuring that the data targeted by cluster analysis has stronger anomaly indication significance.

[0064] S132. Perform global normalization and transpose processing on the voltage curve data to obtain the processed voltage sequence.

[0065] After obtaining the voltage curve data of all battery cells during the charging and resting phase of the analysis object, the data needs to be preprocessed to eliminate the influence of different dimensions and numerical ranges on the clustering results and to construct an input data structure suitable for the DBSCAN clustering algorithm. Specifically, firstly, the voltage data of the battery pack during the charging and resting phase is obtained, and a time-series-based data matrix is ​​constructed, where each column corresponds to a battery cell, each row corresponds to a sampling time, and each element represents the voltage characteristic value of the corresponding battery cell at the corresponding time. Subsequently, the data matrix is ​​globally normalized, and the global maximum and minimum values ​​of all cells and all time points are statistically analyzed. Based on the maximum and minimum values, all data are mapped to a unified numerical range, thereby ensuring that the time-series characteristics of each battery cell have a consistent scale standard. After normalization, the data matrix is ​​transposed so that each battery cell corresponds to an independent sample, and each sample consists of the normalized characteristic values ​​of the battery cell at multiple consecutive time points. At this point, each battery cell is represented as a high-dimensional time-series feature vector, with the vector dimension equal to the number of effective time points, thus forming a sample set suitable for cluster analysis. This modeling approach uses "cell battery" as the sample dimension and "time series" as the feature dimension. It can directly utilize the overall morphological information of voltage changes over time for cluster analysis, rather than relying on statistical features at a single moment, thus capturing the dynamic behavior characteristics of cell battery more comprehensively.

[0066] S133. The DBSCAN algorithm is applied to cluster the processed voltage sequence to identify potential abnormal battery cells whose changing trends do not meet the requirements, so as to obtain the abnormality detection results.

[0067] After data preprocessing, DBSCAN clustering analysis is performed on the constructed cell time-series feature set. DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a density-based unsupervised clustering algorithm. Its core idea is to discover clusters of arbitrary shapes through the density connections between samples and to mark samples that do not belong to any cluster as noise points. In this embodiment, most battery cells will be grouped into several density clusters due to their similar voltage change trends. Battery cells whose voltage time-series features are significantly different from the mainstream pattern and are difficult to be absorbed by clustering (labeled -1) will be marked as abnormal candidate cells with risk potential.

[0068] In one embodiment, step S133 described above may include steps S1331 to S1335.

[0069] S1331, Set the neighborhood radius and minimum number of samples.

[0070] Before applying the DBSCAN algorithm, two key parameters need to be pre-set: the neighborhood radius parameter is set to 2.5%, which roughly corresponds to a range of 95 millivolts (mV). This setting is used to evaluate the consistency score of the battery pack, and its calculation method is existing technology. By adopting this threshold, the voltage difference between individual cells within the battery pack can be effectively quantified, thereby evaluating the consistency status of the entire battery pack. This method provides a scientific and practical standard, helping to ensure the accuracy and reliability of evaluation results when using charging data collected in the cloud for lithium battery pack performance analysis.

[0071] Please see Figure 4 The low consistency score during the discharge resting period reflects poor voltage consistency during this period.

[0072] S130. For battery packs with consistency scores below a set threshold, normalize the voltage data during the charging and resting phase, and use the DBSCAN algorithm to identify abnormal battery cells to obtain anomaly detection results.

[0073] In this embodiment, the anomaly detection result refers to the judgment output after performing density clustering analysis on the voltage time sequence characteristics of each battery cell in the battery pack using the DBSCAN clustering algorithm during the charging resting stage. The battery cells marked as noise points (cluster label is -1) and whose voltage change trends deviate significantly from the mainstream pattern are identified as potential abnormal cells, and their identification information, battery pack to which they belong, and the time of anomaly occurrence are recorded.

[0074] The voltage curves of all cells in battery packs with a charging-resting consistency score below 90 during the charging-resting phase are displayed. Due to the high frequency of voltage data acquisition and the large number of cells, clustering is performed using data from the past two days to reduce memory pressure during data reading. The calculation is performed daily, with results for abnormal cells displayed daily. The output shows whether each cell in the battery pack is identified as abnormal or normal by the DBSCAN clustering algorithm.

[0075] Step S130 aims to identify potential abnormal battery cells that differ significantly from most battery cells from the voltage time-series data during the charging resting phase using a density-based unsupervised clustering algorithm, thereby achieving accurate location of abnormal cells inside the battery pack.

[0076] In one embodiment, step S130 described above may include steps S131 to S133.

[0077] S131. Select battery packs whose consistency score is lower than a set threshold during the charging and resting phase as the analysis object, and obtain the voltage curve data of all battery cells of the analysis object during the charging and resting phase.

[0078] Specifically, during the operation of an energy storage system, the battery management system typically activates voltage balancing strategies only in the high SOC range. Once the individual cell voltage drops to a certain value, the balancing function usually ceases. Therefore, voltage consistency at the end of discharge and in the low SOC phase is easily affected by SOC nonlinearity, voltage plateau characteristics, and operating condition disturbances, limiting its ability to characterize true abnormal states. In contrast, the charging and resting charging phases are typically in a higher SOC range. At this time, cell voltage is more sensitive to capacity differences, self-discharge differences, and polarization characteristics. Simultaneously, the operating current gradually decreases, weakening the influence of dynamic polarization and ohmic voltage drop, allowing the voltage to more accurately reflect the internal state differences of the cells. Therefore, if a battery cluster still exhibits poor voltage consistency during the resting charging phase, it often indicates abnormal voltage recovery behavior or potential performance degradation in some cells, making its anomaly indication more significant. Based on these characteristics, this step first selects battery packs with low voltage consistency scores (specifically, scores below 90) during the resting charging phase as the analysis object, and selects the voltage time series of this phase as the clustering input data. This screening mechanism effectively focuses on battery packs with poor consistency and potential anomaly risks, avoiding the waste of computing resources caused by indiscriminate analysis of all battery packs, while ensuring that the data targeted by cluster analysis has stronger anomaly indication significance.

[0079] S132. Perform global normalization and transpose processing on the voltage curve data to obtain the processed voltage sequence.

[0080] After obtaining the voltage curve data of all battery cells during the charging and resting phase of the analysis object, the data needs to be preprocessed to eliminate the influence of different dimensions and numerical ranges on the clustering results and to construct an input data structure suitable for the DBSCAN clustering algorithm. Specifically, firstly, the voltage data of the battery pack during the charging and resting phase is obtained, and a time-series-based data matrix is ​​constructed, where each column corresponds to a battery cell, each row corresponds to a sampling time, and each element represents the voltage characteristic value of the corresponding battery cell at the corresponding time. Subsequently, the data matrix is ​​globally normalized, and the global maximum and minimum values ​​of all cells and all time points are statistically analyzed. Based on the maximum and minimum values, all data are mapped to a unified numerical range, thereby ensuring that the time-series characteristics of each battery cell have a consistent scale standard. After normalization, the data matrix is ​​transposed so that each battery cell corresponds to an independent sample, and each sample consists of the normalized characteristic values ​​of the battery cell at multiple consecutive time points. At this point, each battery cell is represented as a high-dimensional time-series feature vector, with the vector dimension equal to the number of effective time points, thus forming a sample set suitable for cluster analysis. This modeling approach uses "cell battery" as the sample dimension and "time series" as the feature dimension. It can directly utilize the overall morphological information of voltage changes over time for cluster analysis, rather than relying on statistical features at a single moment, thus capturing the dynamic behavior characteristics of cell battery more comprehensively.

[0081] S133. The DBSCAN algorithm is applied to cluster the processed voltage sequence to identify potential abnormal battery cells whose changing trends do not meet the requirements, so as to obtain the abnormality detection results.

[0082] After data preprocessing, DBSCAN clustering analysis is performed on the constructed cell time-series feature set. DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a density-based unsupervised clustering algorithm. Its core idea is to discover clusters of arbitrary shapes through the density connections between samples and to mark samples that do not belong to any cluster as noise points. In this embodiment, most battery cells will be grouped into several density clusters due to their similar voltage change trends. Battery cells whose voltage time-series features are significantly different from the mainstream pattern and are difficult to be absorbed by clustering (labeled -1) will be marked as abnormal candidate cells with risk potential.

[0083] In one embodiment, step S133 described above may include steps S1331 to S1335.

[0084] S1331, Set the neighborhood radius and minimum number of samples.

[0085] Before applying the DBSCAN algorithm, two key parameters need to be pre-defined: the neighborhood radius parameter ε and the minimum sample size parameter MinPts. The neighborhood radius parameter ε defines the density connectivity between samples; that is, for a given sample point, other sample points within a spherical region centered on it and with a radius of ε are considered its neighborhood samples. The minimum sample size parameter MinPts is used as a threshold to determine whether a sample point is a core sample; that is, when the number of samples contained in the neighborhood of a sample point reaches or exceeds MinPts, the sample point is identified as a core sample. The setting of these two parameters has a significant impact on the clustering results and needs to be reasonably selected based on the actual data distribution characteristics and clustering requirements. In this embodiment, by setting appropriate neighborhood radius and minimum sample size, abnormal candidate cells whose voltage timing characteristics differ significantly from most other cells can be effectively identified, while avoiding misclassifying normal voltage fluctuations as abnormal.

[0086] S1332. Traverse each battery cell sample, calculate the number of neighboring samples in its neighborhood according to the set neighborhood radius and minimum sample number. If the number of samples in the neighborhood reaches or exceeds the minimum sample number, mark the battery cell as a core sample.

[0087] After setting the neighborhood radius and minimum sample size parameters, the number of neighboring samples within the neighborhood range is calculated for each cell sample. Specifically, for each battery cell sample, with the sample as the center and the set neighborhood radius as the radius, all other battery cell samples falling within that neighborhood range are searched in the feature space, and their numbers are counted. When the number of samples within the neighborhood reaches the preset minimum sample size threshold, the battery cell sample is determined as a core sample. Core samples are the foundation for cluster formation in the DBSCAN algorithm. Only core samples can serve as the starting point for cluster expansion, gradually attracting other samples to join the cluster through density reachability relationships. Through this step, representative core samples located in high-density areas can be selected from all battery cell samples, laying the foundation for subsequent cluster expansion.

[0088] S1333. For each core sample, a cluster is gradually formed by finding other samples that satisfy the density reachability relationship with it.

[0089] After identifying the core samples, the algorithm further searches for other samples that satisfy the density reachability relation with the core samples, and gradually expands these samples to form clusters. Density reachability is one of the core concepts of the DBSCAN algorithm. Specifically, if there exists a sample sequence p1, p2, ..., p... n Where p1 is the core sample, and for any two adjacent samples p i and p i+1 p i+1Located at p i Within the neighborhood of p, the sample is called p. n The density is reachable from the core sample p1. By continuously performing the density expansion process, starting from each core sample, all samples that are density reachable are recursively searched and grouped into the same cluster. During the expansion process, if a non-core sample is located in the neighborhood of a core sample, it is assigned to the cluster to which the core sample belongs, but it cannot be used as the starting point for expansion itself; if a core sample is located in the neighborhood of another core sample, the clusters corresponding to these two core samples are merged into one cluster. By continuously performing the above density expansion process, multiple density clustering results are finally formed, and the battery cell samples in each cluster have similar trends in voltage time-series characteristics.

[0090] S1334. Assign a corresponding cluster label to each battery cell. Battery cells belonging to a certain density cluster are assigned a corresponding cluster number. Battery cells that cannot establish an effective density connection with any core sample and do not meet the conditions for forming an independent cluster are marked as noise points, forming potential abnormal battery cells.

[0091] After clustering, each battery cell is assigned a corresponding cluster label. Specifically, battery cells belonging to a certain density cluster are assigned a corresponding cluster number. Battery cells within the same cluster have similar voltage timing characteristics, indicating that their internal states are relatively consistent. Battery cells that cannot establish an effective density connection with any core sample and do not meet the conditions for forming an independent cluster are marked as noise points. The cluster label corresponding to these noise points is recorded as a preset anomaly label (usually -1). These noise points have significantly different voltage timing characteristics from most battery cells, and their voltage change patterns do not conform to the mainstream pattern. Therefore, they are considered as anomaly candidate cells with a risk tendency. Through this labeling mechanism, all battery cells in the battery pack can be clearly divided into two categories: normal clusters and anomalous noise points, providing a clear basis for subsequent anomaly determination.

[0092] S1335. When the cluster label corresponding to a battery cell is a noise point, the battery cell is identified as a potential abnormal battery cell, and the corresponding information is recorded to obtain the abnormality detection result.

[0093] Finally, the cell status is determined based on the clustering labels. When the clustering label for a battery cell is a noise point (label -1), the cell's temporal variation pattern is considered significantly different from most other battery cells, and it is identified as a potentially abnormal battery cell. Corresponding information is recorded, including but not limited to the cell's identification information, its battery pack information, the time of the anomaly, and the clustering label, to form a complete anomaly detection result. These cells marked as potentially abnormal exhibit voltage curves that show different trends from most normal cells during the charging and resting phase, potentially reflecting internal capacity decay, abnormal self-discharge, changes in polarization characteristics, or other potential performance degradation issues, requiring close attention and further analysis. When a battery cell belongs to a normal cluster, it is considered to have a similar trend to other cells of the same type and is not output as an anomaly. Through these steps, battery cell anomaly localization based on the DBSCAN clustering algorithm is achieved, providing fundamental data support for subsequent long-term anomaly tracking analysis and risk quantification assessment.

[0094] Therefore, during the operation of energy storage systems, the high SOC range is more sensitive to the remaining energy and is usually an important window for triggering voltage equalization. At the end of the discharge phase and in the low SOC phase, due to the multiple influences of SOC nonlinearity, voltage plateau characteristics, and operating condition disturbances, the voltage consistency index in these phases has relatively limited ability to characterize real abnormal states.

[0095] In contrast, during the charging and resting phases, batteries are typically in a higher SOC range. During this phase, cell voltage is more sensitive to differences in capacity, self-discharge, and polarization characteristics. Simultaneously, as the operating current gradually decreases, the effects of dynamic polarization and ohmic voltage drop weaken accordingly, allowing the voltage to more accurately reflect the internal state differences within the cells. Therefore, if a battery cluster still exhibits poor voltage consistency during the resting phase, it often indicates that some cells may have abnormal voltage recovery behavior or potential performance degradation, making its anomaly indication more significant. Based on these characteristics, this invention first selects battery packs with consistency scores below a set threshold (e.g., 90 points) during the resting phase as objects for further analysis, and selects the voltage time series data from this phase as input to the clustering algorithm.

[0096] DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a density-based unsupervised clustering algorithm. In this invention, the normalized voltage sequence of all individual cells within each battery pack during the charging and resting phase is treated as a high-dimensional feature vector. By treating the "individual cell" as the sample dimension and the "time series" as the feature dimension, an input data structure for DBSCAN clustering is constructed. This method can directly utilize the overall morphological information of voltage changes over time, rather than relying on statistical features at a single moment.

[0097] Using the DBSCAN clustering algorithm, most battery cells are grouped into several density clusters due to their similar voltage change trends. Battery cells with voltage timing characteristics that are significantly different from the mainstream pattern (marked as noise points, i.e., labeled -1) are considered as anomalous candidate cells with risk potential.

[0098] The specific implementation steps are as follows: Data Acquisition and Processing: First, voltage data of the battery pack during the charging and resting phases are collected, and a time-series-based data matrix is ​​constructed. In this matrix, each column represents a single battery cell, each row represents a sampling time, and each element represents the voltage value of the corresponding battery cell at a specific time. To eliminate the influence of different units and numerical ranges on the clustering results, the data matrix needs to be globally normalized. Specifically, the global maximum and minimum values ​​for all battery cells and time points are calculated, and all data are mapped to a unified numerical range based on these values, ensuring that the time-series characteristics of each battery cell have a consistent scaling standard.

[0099] Data Preparation: After normalization, the data matrix is ​​transposed so that each battery cell becomes an independent sample. Each sample consists of the normalized eigenvalues ​​of that battery cell at multiple consecutive time points. In this way, each battery cell is represented as a high-dimensional time-series feature vector, the dimension of which is equal to the number of valid time points, forming a sample set suitable for cluster analysis.

[0100] Cluster Analysis: Next, the DBSCAN algorithm is used to perform cluster analysis on the prepared set of battery cell time-series features. First, a neighborhood radius parameter and a minimum sample number parameter are set to define the density connectivity between samples. Then, for each battery cell sample, the number of neighboring samples within its neighborhood is calculated. When the number of samples within the neighborhood reaches the preset minimum sample number threshold, the battery cell sample is identified as a core sample. Based on the core samples, other samples that satisfy the density reachability relationship are further searched, and clusters are gradually formed. By continuously repeating this process, multiple density clustering results are ultimately generated.

[0101] Clustering Label Assignment: After clustering is completed, a corresponding clustering label is assigned to each battery cell. Battery cells belonging to a certain density cluster are assigned a corresponding cluster number; battery cells that cannot establish an effective density connection with any core sample and do not meet the conditions for forming an independent cluster are marked as noise points (preset anomaly labels).

[0102] Cell Status Determination: Finally, the status of individual battery cells is determined based on cluster labels. If a battery cell's cluster label is a noise point (label -1), its temporal variation pattern is considered significantly different from that of most battery cells, and it is identified as a potential abnormal battery cell, with relevant information recorded. If a battery cell belongs to a normal cluster, it is considered to have a similar variation trend to other battery cells of the same type and is not output as an abnormal object.

[0103] Please see Figure 5 The three voltage curves highlighted in color represent three battery cells identified as noise points by DBSCAN, i.e., potentially abnormal battery cells. These voltage curves only show data during the charging resting phase and do not include data from the other three phases. The gray lines represent normal battery cells, that is, those battery cells successfully clustered by the DBSCAN algorithm. Figure 5 As can be clearly seen, the voltage curves of these three potentially abnormal battery cells deviate from the voltage curves of most battery cells to some extent. Specifically, the figure shows the voltage trends of all cells in a battery pack during the "charging and resting" phase. The three colored voltage curves (cells 17, 42, and 48) represent the voltage curves of three cells identified as noise points, i.e., potentially abnormal cells, by DBSCAN. The multiple gray voltage curves represent normal cells, i.e., cells identified by DBSCAN as clusterable. As can be seen from the example figure, the voltage curves of the three potentially abnormal cells do indeed deviate from the voltage curves of most cells to some extent, and therefore were identified by the algorithm as potentially abnormal cells.

[0104] S140. Analyze the anomaly detection results within the set number of days, combine spatial proximity characteristics, and adjust and calculate the risk score of each battery cell using common mode penalty and thermal spread gain operator.

[0105] In this embodiment, the risk score is a multi-dimensional quantitative evaluation index that comprehensively reflects the severity, continuous deterioration trend, and timeliness characteristics of abnormal behavior of a battery cell within a set observation period. This index adaptively corrects the original number of abnormal days by introducing a common-mode penalty and thermal propagation gain mechanism under spatial topological constraints, and integrates the composite characteristics of three dimensions: abnormality ratio, continuous deterioration sequence, and timeliness decay. The final comprehensive risk evolution score is obtained by weighting the score using an adaptive empirical correction entropy weight method. The higher the score, the greater the possibility that the battery cell has potential performance degradation or thermal runaway risk, and it needs to be prioritized for maintenance or replacement.

[0106] Input the daily abnormal cell detection results for the battery pack over 90 days. If the battery pack has 50 cells, there will be a total of 50*90 data entries. The data content is 0 or 1, indicating whether the cell was identified as abnormal on that day. The data also includes the distribution of cells within the battery pack, used to calculate the physical distance between cells.

[0107] In addition, the common-mode penalty operator is designed for confidence self-correction and adaptive suppression of common-mode noise in the preceding clustering algorithm. If a battery pack detects a large number of abnormal cells on a certain day, it is determined that the clustering algorithm on that day is at risk of generalization failure. The score weight of each individual cell on that day is adaptively reduced through the exponential penalty term. Alternatively, if a large number of abnormal cells are discretely distributed in physical space (with low spatial proximity), it is further confirmed that they belong to global system-level common-mode noise interference.

[0108] The thermal propagation gain operator can dynamically amplify the potential collaborative risks caused by the propagation of thermal runaway when there are a large number of spatially adjacent abnormal cells. Based on the two operators, the effective abnormal days for each battery cell are obtained after adjustment, and then the effective abnormality ratio is calculated.

[0109] Step S140 aims to extract collaborative features of the spatial distribution of abnormal cells in a specific battery pack within a set observation period (e.g., 90 days). By constructing spatial coupling features under physical topological constraints, common-mode penalty operators and thermal propagation gain operators are introduced to perform confidence self-correction and physical risk amplification on the previous DBSCAN clustering results. Then, the number of spatially collaborative effective abnormal days for each battery cell is calculated, and multi-dimensional composite time-series features are extracted based on this. Finally, the weights of each dimension are solved by the adaptive empirical correction entropy weight method, and the comprehensive risk score of each battery cell is calculated, realizing the upgrade from single-point anomaly identification to long-term risk quantification.

[0110] In one embodiment, step S140 described above may include steps S141 to S144.

[0111] S141. Convert the position information of individual battery cells inside the battery pack into grid points in the physical coordinate system.

[0112] Specifically, to establish the spatial correlation between abnormal battery cells, it is first necessary to construct a spatial mapping function. This maps the battery distribution within the battery pack to grid points in a three-dimensional or two-dimensional physical coordinate system. Let the cell arrangement within the battery pack be R×C, where R is the number of rows and C is the number of columns, and each module contains m×n cells. For any cell with logical number id, its corresponding two-dimensional physical coordinates... satisfy The mapping function f depends on the actual wiring topology, such as different cell series-parallel connection methods like S-shaped or Z-shaped wiring. This spatial mapping transforms cells that originally only had logical numbers into grid points with clear physical location information, laying the foundation for subsequent calculations of the physical topological distance and spatial proximity between abnormal cells. The core significance of this step lies in combining electrical connections with physical spatial layout, enabling subsequent risk assessments to fully consider the physical propagation path of heat spread and the thermal field distribution characteristics within the battery pack.

[0113] S142. For the abnormal detection results within a set number of days, calculate the spatial distance and proximity between abnormal battery cells.

[0114] After completing the physical coordinate mapping of the battery cells, the spatial distance and spatial proximity between abnormal battery cells are calculated based on the anomaly detection results within a set number of days (e.g., 90 days). Specifically, at time t, the set of battery cells that experienced anomalies is defined as... , where k is the number of battery cells identified as abnormal on that day. For any two abnormal battery cells in the set... and The physical topological distance between them Calculated using Euclidean distance: Based on the aforementioned physical topological distance, the spatial proximity of the abnormal cell set at time t is further calculated. The calculation formula is as follows: Where C(k,2) represents the number of combinations of selecting 2 cells from k abnormal cells. This spatial proximity index uses a reciprocal weighting method, so that abnormal cells that are physically closer contribute more to the proximity, thus effectively characterizing the degree of spatial clustering of abnormal cells. When abnormal cells are densely clustered in physical space, the spatial proximity is high, indicating a potential risk of heat spread; conversely, if abnormal cells are discretely distributed, the spatial proximity is low, which may indicate common-mode noise interference or systematic measurement errors.

[0115] S143. Based on the spatial distance and proximity, the effective number of abnormal days for each abnormal battery cell is adjusted using the common-mode penalty operator and the thermal spread gain operator.

[0116] Obtaining the spatial proximity of daily abnormal battery cells and the number of abnormal battery cells Then, for a single target cell n, first extract the set of all times when anomalies occurred within a set time window T. Subsequently, spatial proximity was considered. and the number of abnormal battery cells The spatial cooperative effective score of the battery cell is calculated using the common-mode penalty operator and the thermal spread gain operator. .

[0117] Common mode penalty operator The aim is to establish a confidence self-calibration and common-mode noise adaptive suppression mechanism for the preceding DBSCAN clustering algorithm. The calculation formula is as follows: ,in This represents the common-mode penalty sensitivity coefficient. The core logic of this operator is: if the number of abnormal cells detected in a battery pack on a certain day... There are many abnormal cells, and these abnormal cells are discretely distributed in physical space (i.e., spatial proximity). If the score is low, the clustering algorithm is considered to be at risk of generalization failure that day, possibly due to global system-level common-mode noise interference. In this case, the score weight of each individual node is adaptively reduced by an exponential penalty term. Specifically, when... Larger and When smaller, Common-mode penalty operator approaching 1 If the value is small, a penalty reduction is applied to the effective number of abnormal days for all abnormal cells on that day; conversely, if the number of abnormal cells is small and they are spatially clustered, then... When the value approaches 1, the punitive effect is weak.

[0118] Heat spread gain operator This is used to dynamically amplify the potential collaborative risks caused by the propagation of thermal runaway when the number of abnormal cells in the vicinity of space exceeds the trigger threshold. The calculation formula is as follows: ;in This is the thermal spread gain coefficient. As the trigger threshold, This is an indicator function that takes a value of 1 when a certain condition is met, and 0 otherwise. The initial design intent of this operator is to indicate the number of abnormal battery cells detected on a given day. Exceeding the trigger threshold Furthermore, these abnormal battery cells are spatially adjacent to each other. A high gain coefficient indicates a possible localized thermal anomaly or early signs of thermal runaway propagation within the battery pack. In this case, the gain coefficient can be used to assess the situation. Spatial proximity The product term dynamically amplifies the effective number of abnormal days for the abnormal cell on that day, thereby improving the sensitivity of identifying potential early thermal spread hazards in the battery pack.

[0119] Based on the two operators mentioned above, the number of spatially coordinated effective abnormal days The calculation formula is: This indicator, by weighting and correcting the original number of abnormal days daily, not only suppresses false alarms caused by common-mode noise but also highlights the real threat posed by thermal spread risk, making the effective number of abnormal days for each battery cell more accurately reflect its actual risk level.

[0120] S144. Based on the adjusted effective number of abnormal days and multi-dimensional time series characteristics, calculate the risk score for each battery cell.

[0121] The multidimensional time series features include anomaly ratio indicators, continuous deterioration time series indicators, and timeliness decay indicators.

[0122] After calculating the number of effective abnormal days in spatial coordination, the cell-level multidimensional composite time-series features are further extracted, and the final risk score of each battery cell is calculated based on the adaptive empirical correction entropy weight method.

[0123] Specifically, for all battery cells participating in the evaluation within the time window T (assuming there are N battery cells in total), a multi-dimensional time-series feature matrix is ​​constructed. For the nth cell, its eigenvector is... The calculation method is as follows: Abnormal ratio index This indicator reflects the effective frequency of anomalies occurring in a single battery cell within the observation period. The calculation formula is as follows: ,in The number of effective anomaly days for spatial coordination is calculated in step S143, and T is the total number of days in the set observation period. The higher the value of this index, the higher the frequency of the cell being identified as an anomaly over a long period of time, and the greater the potential risk.

[0124] Continuous deterioration time series indicators This indicator reflects the continuity and deterioration trend of abnormal states in a single battery cell. First, it calculates the maximum number of consecutive days that the cell experiences continuous abnormalities within a window T. Then the calculation formula is: This indicator can detect whether there is a continuous deterioration trend in the battery cell. The longer the consecutive abnormal days, the more likely the battery cell's performance degradation is in an accelerated phase, and the higher the risk level.

[0125] Time-related decay index This indicator reflects the proximity of the most recent anomaly in a battery cell to the current assessment date. A time-dependent decay mechanism is introduced to highlight the importance of recent anomalies. Let the current assessment end date be... The last time this battery cell experienced an anomaly was on [date missing]. Time interval in days decay time constant ,but This indicator uses an exponential decay function, which gives higher weight to anomalies that are closer to the current time point, while the contribution of anomalies that are older gradually decreases, thus ensuring the sensitivity of the risk score to the latest changes in the status quo.

[0126] Therefore, a global multidimensional composite temporal feature matrix is ​​constructed. Each row corresponds to a battery cell, and each column corresponds to a feature dimension, forming a complete battery cell-level timing feature representation.

[0127] Next, weight calculation and cell-level scoring are performed based on the adaptive empirical modified entropy weight method. First, the feature matrix X is normalized column-wise to obtain the probability distribution matrix. Its elements ,in Then, the information entropy of each time-series characteristic index is calculated. This leads to the coefficients of heterogeneity for each indicator. This coefficient reflects the degree of information diversity of each indicator in all battery cells. The higher the heterogeneity, the stronger the ability of the indicator to distinguish the risk level of different battery cells.

[0128] Based on the heterogeneity coefficient, calculate the entropy weight vector driven by pure data. ,in Then, an empirical fusion ratio coefficient was introduced. , will pre-set experience weights The weights are linearly combined with the data entropy weights and then normalized to obtain the final adaptive self-correcting weight vector. This dual-drive hybrid weight fusion mechanism retains the prior knowledge of key risk dimensions from expert experience, while adaptively reflecting the actual data distribution characteristics through data entropy weights, making the weight allocation more scientific and reasonable.

[0129] In one embodiment, step S144 described above may include steps S1441 to S1442: S1441. Calculate the information entropy of each feature in the multidimensional risk matrix.

[0130] In this embodiment, information entropy refers to a statistical measure used to assess the diversity and heterogeneity of information among various feature indicators in a multidimensional time-series feature set. Specifically, the probability distribution of the feature matrix is ​​first constructed, and then the information entropy of each feature indicator is calculated. The information entropy ranges from 0 to 1. When the information entropy is closer to 1, it indicates a more uniform distribution of the feature across individual battery cells, and a weaker discriminative ability. Conversely, when the information entropy is closer to 0, it indicates a more uneven distribution of the feature, and a stronger discriminative ability. By calculating the information entropy, the contribution of each dimension of features to risk identification can be objectively assessed, providing a data-driven basis for subsequent weight calculations.

[0131] S1442. Calculate the weight vector using the information entropy and empirical weights.

[0132] In this embodiment, the weight vector refers to the adaptive self-correcting weight vector obtained through a dual-drive hybrid weight fusion mechanism. This vector is used to balance the contribution ratio of data-driven and empirical knowledge in risk assessment. The specific calculation process is as follows: First, based on the information entropy calculated in step S1441... Calculate the heterogeneity coefficient (information diversity index) for each indicator. Then, the entropy weight vector driven purely by data is calculated. ,in Next, an empirical fusion matching coefficient ρ is introduced. The empirical weights are then... The weights are linearly combined with the data entropy weights and then normalized to obtain the final adaptive self-correcting weight vector. ,in This weight vector enables a dynamic fusion of expert experience and data adaptation, allowing risk assessment to be guided by domain knowledge while also adapting to the actual data characteristics of different battery packs.

[0133] Finally, the temporal feature matrix X from step S144 is combined with the aforementioned adaptive weight vector. Calculate the final comprehensive risk evolution score for each battery cell. : This risk score comprehensively reflects the performance of individual battery cells in three dimensions: abnormal proportion, continuous deterioration, and timeliness. After spatial coupling feature correction and multi-dimensional feature fusion, it effectively quantifies the degree of failure risk of individual battery cells, providing a reliable individual-level basis for subsequent battery pack-level risk assessment and operation and maintenance decisions.

[0134] Please see Figure 6In the graph, one cell on the horizontal axis represents one day, with a time span of 90 days. The y-axis represents the cell IDs within the battery pack. A yellow cell indicates that the cell was detected as an abnormal cell on that day. The DBSCAN clustering algorithm performs calculations once a day, with each input voltage data spanning two days. For example, if clustering analysis is performed at midnight on January 10th, the input data would be the voltage data of all cells in the battery pack during their charging and resting periods from midnight on January 8th to midnight on January 10th. Specifically, this graph shows the time sequence of cells within a battery pack being identified as potentially abnormal over 90 days. One cell on the horizontal axis represents one day, with a time span of 90 days. The y-axis represents the cell IDs within the battery pack. A yellow cell indicates that the cell was detected as an abnormal cell on that day. The DBSCAN clustering algorithm performs calculations daily, with each calculation using voltage data spanning two days. For example, if cluster analysis is performed at midnight on January 10th, the input data consists of the voltage data of all cells in the battery pack during the "charging and resting" period from midnight on January 8th to midnight on January 10th. Each horizontal row represents the number of times a cell was identified as an abnormal cell within 90 days; the more yellow cells in a row, the more days it was identified as an abnormal cell. Each vertical row represents the number of cells in the battery pack identified as abnormal on that day. The graph shows that the battery pack has 16 cells, with cell number 14 repeatedly identified as a potential abnormal cell, while the other cells were identified as normal cells within the 90-day period. This data serves as the basis for calculating risk scores at the individual cell and battery pack levels.

[0135] Each horizontal row represents the number of times the cell was identified as abnormal within 90 days. The more yellow cells in a horizontal row, the more days the cell was identified as abnormal. Each vertical row represents the number of cells in the battery pack identified as abnormal on that day. This data serves as the basis for calculating the scores at the individual cell level and the battery pack level.

[0136] S150. Based on the risk scores of all individual battery cells, construct a multi-dimensional risk matrix including key risk values, diffusion risk values, and concentration risk values, and calculate the overall risk score of the battery pack.

[0137] In this embodiment, the multidimensional risk matrix refers to a risk assessment framework composed of three complementary monitoring dimensions: critical risk value, diffusion risk value, and concentration risk value. It describes the risk status of abnormal cells inside the battery pack from three perspectives: worst-case scenario monitoring, high-risk group monitoring, and internal risk distribution structure monitoring.

[0138] The overall risk score refers to the global battery pack-level comprehensive risk score generated by combining the internal evolution comprehensive risk with the anomaly coverage factor through a dynamic weight fusion mechanism. This score comprehensively reflects the severity, diffusion trend, spatial aggregation characteristics and lateral impact range of the abnormal state within the battery pack, realizing a quantitative leap from individual risk to system-level risk.

[0139] After calculating the comprehensive risk evolution score of each individual cell within the battery pack, this step performs a battery pack-level risk assessment based on an adaptive correlation between data divergence and physical evolution mechanisms. Specifically, the multidimensional risk matrix refers to a risk assessment framework composed of three complementary monitoring dimensions, each characterizing the risk status of abnormal cells within the battery pack from different perspectives: Responsible for monitoring worst-case scenarios, extracting the highest score from the anomaly list, and preventing the bottleneck effect of short circuits in a single cell at extreme points; Responsible for monitoring high-risk scales, paying attention to high-risk groups, and preventing multiple points from deteriorating simultaneously and spreading; The Gini coefficient is responsible for monitoring the internal risk distribution structure. It doesn't consider the score ranking, but only whether the score distribution among abnormal cells is uniform. If a few cells have scores far away from the rest of the abnormal cells (high Gini coefficient), it indicates that the risk has formed a localized polarization cluster in space and structure, thus adjusting the final internal risk score. The overall risk score refers to the calculation of the proportion of abnormal cells within the battery pack (the battery pack includes both normal and abnormal cells), combined with the internal risk score of the abnormal cells calculated by the above multi-dimensional risk matrix, and the final battery pack-level risk score obtained through a dynamic weight fusion mechanism. This score comprehensively reflects the severity, diffusion trend, and spatial clustering characteristics of the abnormal state within the battery pack, realizing a quantitative leap from individual cell risk to system-level risk.

[0140] In one embodiment, step S150 described above may include steps S151 to S152.

[0141] S151. Based on the risk scores of all battery cells, construct a multidimensional risk matrix that includes key risk values, diffusion risk values, and concentration risk values.

[0142] Specifically, let the score of the set of cells in the current battery pack that actually have abnormalities be a vector. Where K is the number of battery cells that have experienced anomalies within the current time window; let the legally mandated total design capacity of the battery pack be... First, calculate the arithmetic mean of the scores of all abnormal cells in the current battery pack. with standard deviation ,in Introducing a dynamic boundary adjustment coefficient n, the dynamic high-risk judgment threshold is then... Defined as This dynamic threshold is used to filter out a subset of outlier, high-risk battery cells that exceed the boundary. Based on this subset, the system's adaptive diffusion risk factor... The solution formula is: This diffusion risk value is responsible for monitoring high-risk scales, focusing on high-risk groups, and preventing simultaneous deterioration at multiple points from forming a diffusion trend. It represents the maximum risk score within the abnormal cell set. for This key risk value is responsible for worst-case monitoring, extracting the highest score from the anomaly list to prevent the bottleneck effect of a single cell experiencing an extreme internal short circuit. The anomaly cell score vector S is sorted in ascending order to obtain an ordered vector. satisfy .like ,but Otherwise, construct the cumulative sum sequence. (where m = 1, 2, ..., K), the expression for the concentration factor is: The concentration risk value (Gini coefficient) is responsible for monitoring the internal risk distribution structure and aims to characterize the degree of clustering of abnormal scores on a few outlier cells. The higher the clustering, the greater the risk of local thermal runaway. Through the construction of the above three dimensions, a complete multi-dimensional risk matrix is ​​formed, realizing a multi-angle and multi-level characterization of the internal risk state of the battery pack.

[0143] Specifically, considering that battery system risks originate not only from extreme failures of local abnormal cells but also from the group propagation effect of abnormal diffusion scale, a dual-channel coupling mechanism for local extreme risk and group diffusion risk is constructed to quantify the synergistic contribution of different risk sources to the overall risk evolution. Based on this, a local risk contribution factor α and a diffusion risk contribution factor are introduced. (satisfy The two types of risks are then merged to obtain the initial internal risk. Subsequently, a physical triggering operator based on an exponential transition mechanism is introduced to nonlinearly modulate the internal risk, constructing the following mapping relationship: , where ψ is a nonlinear adjustment factor used to control the system’s response sensitivity to the coupling strength between local extreme risks and overall diffusion trends; δ is a concentration relaxation displacement constant used to define the limiting physical boundary of spatial heat accumulation in the system and to trigger the response transition of risks from linear superposition to exponential avalanche.

[0144] In one embodiment, step S151 described above may include steps S1511 to S1512.

[0145] S1511. The anomaly detection results are screened based on the dynamic boundary adjustment mechanism, and the internal risk is assessed by integrating local extreme value risk and population diffusion risk.

[0146] In this embodiment, internal risk refers to a comprehensive risk measure obtained by fusing local extreme value risk and group diffusion risk through a dual-channel coupling mechanism, followed by nonlinear modulation using a concentration factor and an exponential transition mechanism. Specifically, let the score of the set of cells that actually exhibit anomalies within the current battery pack be a vector. Where K is the number of battery cells that have experienced anomalies within the current time window. First, the arithmetic mean μ and standard deviation σ of the scores of all abnormal battery cells in the current battery pack are calculated. Then, a dynamic boundary adjustment coefficient n is introduced to define the dynamic high-risk judgment threshold. A subset of outlier, high-risk battery cells that exceed the dynamic boundary is selected. The system's adaptive diffusion risk factor The solution formula is: The highest risk score in the abnormal battery cell set. Considering that battery system risks originate not only from extreme failures of local abnormal cells but also from the group propagation effect of abnormal diffusion scale, a dual-channel coupling mechanism for local extreme risk and group diffusion risk is constructed. This mechanism introduces a local risk contribution factor α and a diffusion risk contribution factor β (satisfying α+β=1) to fuse the two types of risks, yielding the initial internal risk. This preliminary internal risk assessment integrates information from two dimensions: worst-case scenario monitoring and high-risk group monitoring, laying the foundation for subsequent nonlinear modulation and final scoring.

[0147] S1512. Calculate the concentration factor based on the distribution of abnormal battery cell scores, and use the exponential transition mechanism to nonlinearly modulate the internal risk.

[0148] Specifically, the abnormal cell rating vector S is arranged in ascending order to obtain an ordered vector. ,satisfy .like ,but Otherwise, construct a cumulative sum sequence. (in The expression for the concentration factor is: The concentration factor aims to characterize the degree of aggregation of anomaly scores on a few outlier cells; the higher the aggregation, the greater the risk of local thermal runaway. Subsequently, a physical triggering operator based on an exponential transition mechanism is introduced to nonlinearly modulate the internal risk, constructing the following mapping relationship: Here, ψ is a nonlinear adjustment factor used to control the system's sensitivity to the coupling strength between local extreme risks and the overall diffusion trend; δ is a concentration relaxation displacement constant used to define the limiting physical boundary of spatial heat accumulation in the system and trigger a response transition from linear superposition of risks to exponential avalanche. Through this exponential transition mechanism, when the concentration factor is high (i.e., the risk is highly concentrated in a few cells) and there are extremely high-risk cells, the internal risk will increase exponentially, thus more sensitively capturing the potential risk of local thermal runaway.

[0149] S152. Combining the dynamic boundary adjustment mechanism with the integration of internal evolution comprehensive risk and abnormal coverage factor, calculate the overall risk score of the battery pack.

[0150] After constructing the multidimensional risk matrix, this step introduces a physical triggering operator based on an exponential transition mechanism to nonlinearly modulate the internal risks and integrates the anomaly coverage factor to finally generate a global battery pack-level comprehensive risk score.

[0151] Anomaly Coverage Factor The lateral spread of a cell failure within the current battery pack is characterized by the ratio of the actual number of cells experiencing anomalies to the total designed number of cells in the battery pack. Set the base coverage weight. Maximum gain weight magnitude and convergence rate coefficient The dynamic weight relationship is constructed as follows: , Finally, by integrating the dynamically generated weights, a global battery pack-level comprehensive risk score is generated. Through the above steps, a complete mapping from individual unit risk scores to the overall risk score of the battery pack is achieved. This score comprehensively reflects the extreme risks, diffusion trends, spatial clustering characteristics, and anomaly coverage of abnormal states within the battery pack, providing a quantitative basis for battery pack safety status assessment and operation and maintenance decisions.

[0152] In one embodiment, step S152 described above may include steps S1521 to S1522.

[0153] S1521. Calculate the ratio of the actual number of abnormal battery cells to the total number of battery cells in the battery pack as the abnormality coverage factor.

[0154] In this embodiment, the anomaly coverage factor is a quantitative indicator characterizing the lateral spread of cell faults within the current battery pack. It is defined as the ratio of the actual number of cells with anomalies to the total designed number of cells in the battery pack. This factor reflects the extent to which abnormal states spread within the battery pack. A high anomaly coverage rate indicates a large area of ​​abnormal states within the battery pack, significantly increasing system-level risk. Conversely, a low anomaly coverage rate indicates that abnormal states are limited to a few cells, and the overall system risk is relatively controllable. By introducing the anomaly coverage rate factor, a horizontal dimension assessment of battery pack risk is achieved, complementing the internal risk assessment (vertical depth dimension) to form a complete battery pack-level risk assessment system.

[0155] S1522. The weight vector is dynamically adjusted based on the abnormal coverage factor and combined with the internal risk to calculate the overall risk score of the battery pack.

[0156] Specifically, setting the base coverage weight Maximum gain weight magnitude and convergence rate coefficient The dynamic weight relationship is constructed as follows: , The design logic of this dynamic weighting mechanism is that when the abnormal coverage is low, the coverage weight is close to the base value. Internal risks dominate; as the anomaly coverage increases, the coverage weight gradually increases, up to a maximum of [missing information]. This indicates that the impact of abnormal lateral spread on the overall system risk is becoming increasingly significant. This dynamic adjustment mechanism enables an adaptive response to the battery pack's risk status. Finally, by fusing dynamically generated weights, a global battery pack-level comprehensive risk score is generated. This overall risk score integrates the internal evolutionary comprehensive risk (reflecting the depth and clustering characteristics of anomalies) with the anomaly coverage factor (reflecting the lateral impact range of anomalies) through dynamic weighting, and sets an upper limit of 1 to ensure the boundedness and comparability of the score. Through this comprehensive score, maintenance personnel can intuitively understand the overall risk status of the battery pack and formulate differentiated maintenance strategies and repair plans accordingly.

[0157] By introducing multidimensional time-series composite feature extraction technology, this method transforms the previously scattered single-point fault identification into continuous tracking and analysis of abnormal battery cells over a long period. This method not only achieves a shift from "qualitative judgment of whether an anomaly exists" to "numerical quantification of risk severity assessment," but also solves several key problems in existing technologies: first, it can only identify and locate abnormal battery cells, but cannot combine long-term operational data for continuous tracking and analysis; second, it is difficult to effectively quantify the risk level of battery cell failures; and third, it is difficult to achieve a holistic assessment of the risk status of the entire battery pack.

[0158] Specifically, by employing multi-dimensional time-series composite feature extraction, the operational characteristics of each battery cell can be comprehensively captured over time, thereby more accurately identifying potentially risky behavioral patterns. Analyzing data over long periods allows for a deeper understanding of the health status and trends of each battery cell, enabling predictions of its future development path. This continuous tracking and analysis based on long-term data can precisely quantify the degree of failure risk based on the actual performance of each battery cell, rather than relying solely on simple threshold judgments or surface phenomena.

[0159] Furthermore, this method supports comprehensive risk assessment of the entire battery pack. Through detailed analysis of each individual battery cell and their interactions, a more complete and accurate overall health model of the battery pack can be constructed. This not only helps to promptly identify and address individual high-risk battery cells but also prevents chain reactions caused by the performance degradation of certain cells, ensuring the stability and safety of the entire energy storage system. Therefore, compared to traditional single-fault detection methods, multi-dimensional time-series composite feature extraction provides a more comprehensive, in-depth, and quantitative risk management strategy.

[0160] The aforementioned battery risk assessment method based on long-term anomaly evolution collects voltage data of each cell at different operating stages from the battery management system, calculates the voltage variation coefficient for each stage and converts it into a consistency score, normalizes the voltage data during the charging and resting stages of the low-scoring battery pack, and uses the DBSCAN algorithm to identify abnormal cells. Then, combining the anomaly detection results within a set number of days with spatial proximity characteristics, common-mode penalty and thermal propagation gain operators are used to adjust and calculate the risk score of each cell. Finally, a multi-dimensional risk matrix including critical risk values, diffusion risk values, and concentration risk values ​​is constructed to calculate the overall risk score of the battery pack. This method not only achieves continuous tracking and analysis of abnormal cells over a long period but also solves the problem of existing technologies that can only qualitatively identify and locate anomalies. It achieves effective quantification of fault risk levels and overall risk assessment of the battery pack, thereby significantly improving the accuracy and reliability of battery system fault early warning.

[0161] Figure 7 This is a schematic block diagram of a battery risk assessment system 300 based on long-term time-series anomalous evolution provided in an embodiment of the present invention. Figure 7 As shown, corresponding to the above-described battery risk assessment method based on long-term anomaly evolution, the present invention also provides a battery risk assessment system 300 based on long-term anomaly evolution. This battery risk assessment system 300 includes a unit for executing the above-described battery risk assessment method based on long-term anomaly evolution, and the system can be configured in a server. Specifically, please refer to... Figure 7The battery risk assessment system 300 based on long-term time-series abnormal evolution includes an acquisition unit 301, a consistency score calculation unit 302, an anomaly detection unit 303, a risk score calculation unit 304, and an overall risk scoring unit 305.

[0162] The acquisition unit 301 is used to collect voltage data of each battery cell from the battery management system during the four stages of charging, charging rest, discharging, and discharging rest. The consistency score calculation unit 302 is used to calculate the voltage variation coefficient of each stage based on the voltage data and convert it into a consistency score. The anomaly detection unit 303 is used to normalize the voltage data of the charging rest stage for battery packs with consistency scores lower than a set threshold, and use the DBSCAN algorithm to identify abnormal battery cells to obtain anomaly detection results. The risk score calculation unit 304 is used to analyze the anomaly detection results within a set number of days, combine spatial proximity characteristics, and use common mode penalty and thermal spread gain operators to adjust and calculate the risk score of each battery cell. The overall risk scoring unit 305 is used to construct a multi-dimensional risk matrix including key risk value, diffusion risk value, and concentration risk value based on the risk scores of all battery cells, and calculate the overall risk score of the battery pack.

[0163] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the battery risk assessment system 300 based on long-term time-series abnormal evolution and each unit can be referred to the corresponding description in the foregoing method embodiments. For the sake of convenience and brevity, it will not be repeated here.

[0164] The aforementioned battery risk assessment system 300 based on long-term time-series anomalous evolution can be implemented as a computer program, which can be used in, for example... Figure 8 It runs on the computer device shown.

[0165] Please see Figure 8 , Figure 8 This is a schematic block diagram of a computer device provided in an embodiment of this application. The computer device 500 can be a server, wherein the server can be a standalone server or a server cluster composed of multiple servers.

[0166] See Figure 8 The computer device 500 includes a processor 502, a memory, and a network interface 505 connected via a system bus 501. The memory may include a non-volatile storage medium 503 and internal memory 504.

[0167] The non-volatile storage medium 503 may store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, cause the processor 502 to perform a battery risk assessment method based on long-term time-series abnormal evolution.

[0168] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0169] The internal memory 504 provides an environment for the execution of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a battery risk assessment method based on long-term time-series abnormal evolution.

[0170] This network interface 505 is used for network communication with other devices. Those skilled in the art will understand that... Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 500 to which the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0171] The processor 502 is used to run a computer program 5032 stored in the memory to implement all the steps of the battery risk assessment method based on long-term time-series abnormal evolution.

[0172] It should be understood that in the embodiments of this application, the processor 502 may be a central processing unit (CPU), or it may be 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. The general-purpose processor may be a microprocessor or any conventional processor.

[0173] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of each unit is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0174] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A battery risk assessment method based on long-term time-series anomalous evolution, characterized in that, include: Voltage data of each battery cell is collected from the battery management system during four stages: charging, charging rest, discharging, and discharging rest. Based on the voltage data, the voltage variation coefficient for each stage is calculated and converted into a consistency score; For battery packs with consistency scores below a set threshold, the voltage data during the charging and resting phase is normalized, and the DBSCAN algorithm is used to identify abnormal battery cells to obtain anomaly detection results. The anomaly detection results within a set number of days are analyzed, and combined with spatial proximity characteristics, the risk score of each battery cell is adjusted and calculated using common-mode penalty and thermal spread gain operators. Based on the risk scores of all individual battery cells, a multi-dimensional risk matrix is ​​constructed, including critical risk values, diffusion risk values, and concentration risk values, and the overall risk score of the battery pack is calculated.

2. The battery risk assessment method based on long-term time-series anomalous evolution according to claim 1, characterized in that, The step of calculating the voltage variation coefficient for each stage based on the voltage data and converting it into a consistency score includes: The battery pack's timeline is divided into four stages: charging, charging rest, discharging, and discharging rest. For each stage, calculate the maximum and minimum voltage values ​​of all individual cells in the battery pack at each moment; The root mean square error is calculated using the maximum and minimum voltage values. The voltage variation coefficient is obtained by dividing the root mean square error by the average voltage value. Based on the coefficient of variation, a segmented scoring model is used to map voltage differences into consistency score intervals.

3. The battery risk assessment method based on long-term time-series anomalous evolution according to claim 2, characterized in that, The process of normalizing voltage data during the charging and resting phase for battery packs with consistency scores below a set threshold, and using the DBSCAN algorithm to identify abnormal battery cells to obtain anomaly detection results, includes: Battery packs with a consistency score below a set threshold during the charging and resting phase were selected as the analysis objects to obtain voltage curve data of all individual battery cells of the analysis objects during the charging and resting phase. The voltage curve data is subjected to global normalization and transpose to obtain the processed voltage sequence. The DBSCAN algorithm is applied to cluster the processed voltage sequence to identify potential abnormal battery cells whose changing trends do not meet the requirements, so as to obtain the anomaly detection results.

4. The battery risk assessment method based on long-term time-series anomalous evolution according to claim 3, characterized in that, The DBSCAN algorithm is applied to cluster the processed voltage sequence to identify potential abnormal battery cells whose changing trends do not meet the requirements, in order to obtain anomaly detection results, including: Set the neighborhood radius and minimum number of samples; For each battery cell sample, calculate the number of neighboring samples in its neighborhood based on the set neighborhood radius and minimum sample number. If the number of samples in the neighborhood reaches or exceeds the minimum sample number, then mark the battery cell as a core sample. For each core sample, a cluster is gradually formed by finding other samples that satisfy the density reachability relationship with it; Each battery cell is assigned a corresponding cluster label. Battery cells belonging to a certain density cluster are assigned a corresponding cluster number. Battery cells that cannot establish an effective density connection with any core sample and do not meet the conditions for forming an independent cluster are marked as noise points and form potential abnormal battery cells. When the cluster label corresponding to a battery cell is a noise point, the battery cell is identified as a potentially abnormal battery cell, and the corresponding information is recorded to obtain the anomaly detection result.

5. The battery risk assessment method based on long-term time-series anomalous evolution according to claim 1, characterized in that, The analysis of the anomaly detection results within the set number of days, combined with spatial proximity features, uses common-mode penalty and thermal spread gain operators to adjust and calculate the risk score of each battery cell, including: The positional information of individual battery cells inside the battery pack is converted into grid points in the physical coordinate system; For the anomaly detection results within a set number of days, calculate the spatial distance and proximity between the abnormal battery cells; Based on the spatial distance and proximity, the effective number of abnormal days for each abnormal battery cell is adjusted using a common-mode penalty operator and a thermal spread gain operator. Based on the adjusted effective number of abnormal days and multidimensional time series characteristics, the risk score of each battery cell is calculated.

6. The battery risk assessment method based on long-term time-series anomalous evolution according to claim 5, characterized in that, The multidimensional time series features include anomaly ratio indicators, continuous deterioration time series indicators, and timeliness decay indicators.

7. The battery risk assessment method based on long-term time-series anomalous evolution according to claim 6, characterized in that, The risk score for each battery cell is calculated based on the adjusted effective number of abnormal days and multidimensional time-series characteristics, including: Calculate the information entropy of each feature in the multidimensional risk matrix; The weight vector is calculated using the information entropy and empirical weights.

8. The battery risk assessment method based on long-term time-series anomalous evolution according to claim 7, characterized in that, Based on the risk scores of all individual battery cells, a multi-dimensional risk matrix is ​​constructed, including critical risk values, diffusion risk values, and concentration risk values. The overall risk score of the battery pack is then calculated, including: Based on the risk scores of all individual battery cells, a multidimensional risk matrix is ​​constructed, which includes key risk values, diffusion risk values, and concentration risk values. By combining a dynamic boundary adjustment mechanism with the internal evolution comprehensive risk and anomaly coverage factor, the overall risk score of the battery pack is calculated.

9. A battery risk assessment system based on long-term time-series anomalous evolution, characterized in that, include: The acquisition unit is used to collect voltage data of each battery cell from the battery management system during the four stages of charging, charging rest, discharging, and discharging rest. The consistency score calculation unit is used to calculate the voltage variation coefficient of each stage based on the voltage data and convert it into a consistency score; The anomaly detection unit is used to normalize the voltage data of battery packs with consistency scores below a set threshold during the charging and resting phase, and to use the DBSCAN algorithm to identify abnormal battery cells in order to obtain anomaly detection results. The risk score calculation unit is used to analyze the anomaly detection results within a set number of days, combine spatial proximity characteristics, and adjust and calculate the risk score of each battery cell using common mode penalty and thermal spread gain operators. The overall risk scoring unit is used to construct a multi-dimensional risk matrix, including key risk values, diffusion risk values, and concentration risk values, based on the risk scores of all individual battery cells, and to calculate the overall risk score of the battery pack.

10. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 8.

Images