Standby power energy storage battery grading early warning method and system based on unsupervised learning algorithm

Abstract

The invention discloses a standby power energy storage battery grading early warning method and system based on an unsupervised learning algorithm, and belongs to the technical field of batteries, and the method comprises the steps: firstly, obtaining the data of a standby power energy storage battery; secondly, based on one or more data fields, segmenting charging and discharging data fields of the energy storage battery; then, four characteristics of voltage deviation difference, voltage distribution difference, voltage variability and voltage curve similarity are obtained, and a characteristic matrix is constructed; and then clustering processing is carried out on the characteristic matrix and grading early warning is carried out on the standby power energy storage battery. Based on the constructed feature matrix, a clustering method is adopted to carry out graded safety early warning on the battery system, data support is provided for reasonable configuration and fine management of batteries, the internal evolution trend of the standby power energy storage system can be identified through data driving, and the essential relation behind the data is analyzed; in addition, through grading early warning, problem monomers can be effectively positioned, and related problem parts can be checked and corrected in a targeted manner.

Description

Technical Field

[0001] This application belongs to the field of battery technology, and specifically relates to a graded early warning method and system for backup power storage batteries based on unsupervised learning algorithms. Background Technology

[0002] With the rapid development of technology and the ever-increasing demand for energy, backup power storage batteries are being used more and more widely in various fields. From new energy vehicles to smart homes, from data centers to communication base stations, backup power storage batteries play a crucial role, providing stable power support when the main power source fails or is insufficient, ensuring the normal operation of equipment. However, backup power storage batteries face many safety hazards and performance degradation issues during use. The internal chemical reactions of the battery are complex, and its state is affected by various factors such as the number of charge and discharge cycles, depth of charge and discharge, and ambient temperature. As the usage time increases, the battery will gradually age, with increased internal resistance, reduced capacity, and even serious failures such as thermal runaway and short circuits. These problems not only affect the battery's lifespan and performance but may also cause safety accidents such as fires and explosions, resulting in personal injury and property damage.

[0003] Currently, early warning methods for backup power storage batteries mainly focus on monitoring single or a few parameters such as battery voltage, current, and temperature, and setting fixed thresholds for judgment. When these parameters exceed the thresholds, corresponding warning signals are issued. However, this method has significant limitations. On the one hand, monitoring a single parameter cannot fully reflect the true state of the battery, because battery failures are often the result of multiple factors working together, and relying on a single parameter may lead to misjudgments or missed diagnoses. On the other hand, setting fixed thresholds lacks flexibility and cannot adapt to the differences in characteristics of different batteries and the complex and ever-changing usage environment. For example, under different temperature conditions, the normal operating parameter range of a battery will vary. Using fixed thresholds for early warning may issue false alarms in some cases, while failing to detect battery anomalies in a timely manner in others.

[0004] Furthermore, most existing early warning methods are based on static data analysis and lack the ability to dynamically track and predict battery status. Battery performance and status change continuously over time, and existing methods cannot capture these changes in real time, nor can they predict potential battery failures in advance, making it difficult to meet the requirements for battery safety and reliability in practical applications.

[0005] Furthermore, existing early warning methods also employ clustering algorithms. Clustering algorithms measure similarity based on the distance between data points, forming multiple clusters to provide early warnings. Currently, most clustering algorithms used in power battery safety early warning and fault diagnosis fall into two categories: DBSCAN and fast clustering. Chinese patent CN117648589A discloses a method, system, electronic device, and medium for early warning of thermal runaway in energy storage batteries, which uses the DBSCAN algorithm to achieve safety early warning of thermal runaway in energy storage batteries. This includes standardizing the correlation coefficient vector and autocorrelation coefficient vector respectively, constructing a two-dimensional array based on the standardized correlation coefficient vector and the corresponding standardized autocorrelation coefficient vector; and applying the DBSCAN clustering algorithm to cluster the two-dimensional array data to obtain analysis results. Based on the battery cells corresponding to the unclassified two-dimensional data, battery cells with thermal runaway risk are identified for early warning. Chinese patent CN108960321A discloses a battery fault prediction method for large-scale lithium battery energy storage power stations, which uses a fast clustering method to achieve battery fault prediction in lithium battery energy storage power stations. The specific method is as follows: ① The historical monitoring signals of the battery box cluster of the lithium battery energy storage power station are used as the original feature library; ② The main feature matrix of the battery box cluster at each sampling time is extracted from the original feature library through the sparse autoencoder algorithm; ③ The cluster center battery box at each sampling time is searched based on the fast clustering algorithm, and the cumulative eccentricity distance matrix of the battery box cluster is calculated; ④ The cumulative eccentricity distance matrix is ​​normalized and a warning threshold is set to finally realize the prediction of battery box failure.

[0006] Chinese patent CN118671602A discloses an unsupervised learning-based method and system for early warning of battery system faults. First, it employs a multi-stage DBSCAN clustering algorithm to achieve unsupervised learning-based early warning of battery faults. Specifically, the method involves extracting features from the battery pack charging voltage data for each effective charging cycle to obtain voltage and voltage gradient feature vectors. This captures minute deviations in the voltage signal, suppresses interference caused by random factors, and adapts to the randomness of actual operating conditions. Then, for the feature vectors, a bisection method is used to perform multiple DBSCAN clustering operations to solve for dimensionless parameters, thereby measuring the consistency of battery voltage. Finally, the dimensionless parameter values ​​for each charging cycle are compared with a threshold to establish a fault warning indicator, thus achieving unsupervised learning-based early warning of battery faults.

[0007] However, the shortcomings of the clustering algorithms used in the above applications are mainly reflected in the following aspects: While K-means and DBSCAN algorithms are widely used due to their simple models and fast computation, their clustering performance is significantly affected by the value of K or hyperparameters. Given the complex operating environment of batteries in reality, where characteristic parameters vary unpredictably, preset parameters are insufficient to ensure accurate identification of abnormal cells, easily leading to delayed fault diagnosis or even false alarms.

[0008] Therefore, there is an urgent need for a more accurate, reliable, and flexible method for classifying and warning about backup power storage batteries, in order to improve battery safety and lifespan and ensure the stable operation of related equipment.

[0009] Summary of the Invention To address the aforementioned issues, this application provides a graded early warning method and system for backup power storage batteries based on unsupervised learning algorithms. This method offers high early warning reliability and improves battery safety and lifespan.

[0010] The purpose of this disclosure is to provide a graded early warning method for backup power storage batteries based on an unsupervised learning algorithm, including, Obtain backup power storage battery data; Based on one or more data fields, the charging and discharging data segments of the energy storage battery are segmented. Based on the segmented charge and discharge data segments of the energy storage battery, voltage deviation difference characteristics, voltage distribution difference characteristics, voltage variability characteristics, and voltage curve similarity are obtained to construct a feature matrix; Cluster the feature matrix; Based on the clustering results, a graded early warning system is implemented for backup power storage batteries.

[0011] Optionally, acquiring backup power storage battery data includes collecting static and dynamic data of the backup power storage battery at preset time frame intervals. The collected backup power storage battery data includes data acquisition time, total voltage, total current, and individual cell voltage.

[0012] Optionally, it also includes preprocessing the backup power storage battery data, wherein the preprocessing includes data noise reduction using filters, including... Within a sliding window of length 2m+1, where m refers to the number of data frames, the signal data points are... ,in, i It refers to the first i Data points, y This refers to the voltage of a single cell, corresponding to the relative offset of that cell. By fitting the signal data points within the window using an nth-order polynomial p(t), the expression of the polynomial is as follows:

[0013] The goal is to solve for the coefficients of the polynomial using the least squares method. , so that:

[0014] In the formula, k Indicates the first k One data frame; The coefficients are solved using the least squares method. By substituting the offset of each point within the window into the polynomial, the matrix form of the system of equations is obtained.

[0015] in, , y A matrix of voltage signal data points.

[0016] The polynomial basis matrix constructed for the offset, where e is the error vector; By minimizing the sum of squared errors The optimal coefficients are obtained:

[0017] Extract the filter coefficients, where the filtered output value is the fitted value of the fitting polynomial at the window center position x=0. That is, the constant term of the polynomial. According to the least squares solution, It can be represented as:

[0018] in, It is a matrix The first row yields the filter coefficient vector. .

[0019] Optionally, the energy storage battery's charge and discharge data segments can be segmented based on one or more data fields, including segmenting the energy storage battery's charge and discharge data segments based on the state of charge, current, and / or SOC data fields to obtain charge and discharge cycle data segments.

[0020] Optionally, based on the segmented charge and discharge data segments of the energy storage battery, the voltage deviation difference characteristics are obtained, including: A voltage matrix is ​​constructed based on the extracted charge-discharge cycle data segments. :

[0021] Where K is the number of frames in the charge / discharge cycle data segment, and N is the number of individual battery cells. It is the voltage value of the j-th battery cell at sampling time t, j∈N, t∈K; Based on voltage matrix The mean voltage matrix is ​​obtained by taking the voltage values ​​of all individual cells at each time point. :

[0022]

[0023] in, This represents the average voltage value of all individual battery cells at time t; Based on voltage matrix Voltage values ​​and average voltage matrix of all battery cells The average voltage values ​​of all individual battery cells are used to obtain the voltage deviation difference characteristic DA, where the voltage deviation difference DA of the j-th cell is... j for: .

[0025] Optionally, based on the segmented charge and discharge data segments of the energy storage battery, the voltage distribution difference characteristics are obtained, including: Based on voltage matrix For each cell in the data segment, calculate the mean matrix of each cell in the charge-discharge cycle data segment by taking the voltage value of each cell at all times. :

[0026]

[0027] Where K is the number of frames in the charge / discharge cycle data segment, and N is the number of individual battery cells. p This represents the average voltage of the p-th cell in the backup power storage battery within the charge / discharge cycle data segment. p ∈N, This represents the voltage value of the j-th battery cell at sampling time t. This represents the average voltage of the j-th battery cell in the charge-discharge cycle data segment of the backup power storage battery. Based on voltage matrix The voltage and mean matrix of each cell at all times. The average voltage of all battery cells in the charge-discharge cycle data segment is used to obtain the voltage distribution difference feature DT. Then, the voltage distribution difference of the j-th cell is... DT j for: .

[0029] Optionally, voltage variability characteristics can be obtained based on the segmented charge and discharge data segments of the energy storage battery. DV, Voltage variability characteristics of the j-th individual DV j for:

[0030] Where K is the number of frames in the charge / discharge cycle data segment.

[0031] Optionally, based on the charge and discharge data segments of the segmented energy storage batteries, the voltage curve similarity DF is obtained, then the voltage variability characteristics of the j-th cell are... DF j for:

[0032] in, Represents any value of the j-th monomer within the fragment. This represents the average voltage of all individual cells in the backup power storage battery at any given time within the segment.

[0033] Optionally, construct a feature matrix. include, Obtain the voltage deviation difference characteristics, voltage distribution difference characteristics, voltage variability characteristics, and voltage curve similarity of all individual battery cells; Based on the voltage deviation differences, voltage distribution differences, voltage variability, and voltage curve similarity of all individual battery cells, a feature matrix is ​​constructed. :

[0034] In the formula, K is the number of frames of valid data segments, and N is the number of individual battery cells. DA N Voltage deviation difference of the Nth battery cell; DV N The voltage variability characteristic of the Nth battery cell; DF N The voltage variability characteristic of the Nth battery cell; DT N The voltage distribution difference of the Nth battery cell.

[0035] Optionally, clustering the feature matrix includes, Determine the distance radius Minimum number of clusters Among them, the distance radius is determined by trial and error. For an infinite number of nodes, the minimum number of nodes required for clustering is [value missing]. Value , where N is the number of individual battery cells; By inputting the feature matrix into the clustering algorithm, the reachable distance of each individual cell voltage is calculated. .

[0036] Optionally, based on the clustering results, a tiered early warning system for backup power storage batteries is implemented, including: Set fault classification thresholds ; Reachable distance based on individual cell voltages Calculate the average distance achievable by the voltage of all individual units. and standard deviation ; Based on fault classification threshold and the calculated mean and standard deviation The thresholds for the three-level fault classification are determined; among them, the thresholds for the three-level fault classification include the minor fault safety threshold. Moderate fault safety threshold Severe fault safety threshold ; Based on reachable distance Based on the determined three-level fault classification thresholds, obtain the state classification of the backup power storage battery:

[0037] In the formula, This represents the clustering label value of the j-th battery cell. The larger the label value, the higher the level of anomaly. This represents the reachable distance of the j-th battery cell; If the extracted charge / discharge cycle data segment identifies battery cell information with diagnostic abnormalities, a safety warning will be issued based on the highest fault level observed in the battery cell, and the cell numbers of all warning cells will be output.

[0038] Another objective of this disclosure is to provide a graded early warning system for backup power storage batteries based on an unsupervised learning algorithm, including, The acquisition module is used to acquire data from backup power storage batteries; The segmentation module is used to segment the charge and discharge data segments of the energy storage battery based on one or more data fields. The feature extraction module is used to obtain voltage deviation difference features, voltage distribution difference features, voltage variability features, and voltage curve similarity features based on the segmented charge and discharge data segments of the energy storage battery, and construct a feature matrix. The clustering module is used to cluster the feature matrix. The early warning module is used to provide graded early warnings for backup power storage batteries based on the results of clustering.

[0039] Compared with the prior art, this application has the following advantages: The method disclosed herein constructs a feature matrix by acquiring voltage deviation difference characteristics, voltage distribution difference characteristics, voltage variability characteristics, and voltage curve similarity. Based on clustering, it performs graded safety early warning for battery systems, enabling data-driven identification of the internal evolution trend of backup power storage systems and analysis of the essential relationships behind the data. Furthermore, by issuing graded early warnings at different levels and locating problematic individual units, targeted inspections and corrections of relevant problem parts can be carried out. In addition, cluster analysis can help determine the differences between different batteries, which can assist in further analysis and evaluation of the health status of power batteries, providing data support for the rational configuration and refined management of batteries.

[0040] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 A schematic diagram of a graded early warning method for backup power storage batteries based on an unsupervised learning algorithm is shown in an embodiment of this disclosure. Figure 2 A schematic diagram of another method for graded early warning of backup power storage batteries based on an unsupervised learning algorithm is shown in an embodiment of this disclosure. Figure 3 A schematic diagram of a graded early warning system for backup power storage batteries based on an unsupervised learning algorithm is shown in an embodiment of this disclosure. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0044] like Figure 1 As shown in the embodiments of this disclosure, a hierarchical early warning method for backup energy storage batteries based on an unsupervised learning algorithm is introduced. The method includes: first, acquiring backup energy storage battery data; second, segmenting the charge / discharge data segments of the energy storage battery based on one or more data fields; then, based on the segmented charge / discharge data segments of the energy storage battery, acquiring voltage deviation difference features, voltage distribution difference features, voltage variability features, and voltage curve similarity, and constructing a feature matrix; then, performing clustering processing on the feature matrix; finally, based on the clustering results, performing hierarchical early warning for the backup energy storage battery. The above method extracts features from the battery voltage list data, acquires voltage deviation difference features, voltage distribution difference features, voltage variability features, and voltage curve similarity, constructs a feature matrix, and performs hierarchical early warning for backup energy storage battery safety through data-driven methods. It uses fewer features and preset parameters, has wider applicability, and better early warning capabilities.

[0045] Specifically, such as Figure 2 As shown, the graded early warning method for backup power storage batteries based on unsupervised learning algorithms specifically includes the following steps: Step S1, Data Acquisition: Acquire backup power storage battery data; wherein, the acquired data is raw operational data, including static and dynamic data of the backup power storage battery, collected at preset time frame intervals. A preset time frame can be 10 seconds, but is not limited to this; other preset times, such as 15 seconds, are also applicable to this disclosure. The acquired backup power storage battery data includes, but is not limited to, data acquisition time, total voltage, total current, and individual cell voltage.

[0046] Step S2: Preprocess the backup power storage battery data. Due to factors such as network transmission signals and system hardware failures, the raw data from the enterprise platform may contain outliers, missing values, duplicate records, and noisy data. These errors can affect the accuracy of subsequent clustering model diagnostics; therefore, data preprocessing is necessary. The data preprocessing methods used in this disclosure include: S21: Outlier Handling: This includes deleting obviously abnormal values ​​in data such as battery voltage, current, and temperature. For example, a sudden drop in single-cell voltage from 3.5V to 0V, or a sudden drop in temperature from the normal 25℃ to -40℃, could be due to sensor malfunction causing abnormal individual measurements.

[0047] S22: Duplicate Record Handling: If two adjacent rows of data have the same collection time and other data, retain the earliest record. This is most likely due to data collection errors causing duplicate records.

[0048] S23: Missing Value Handling: Based on the number of missing data points, the following two cases should be handled: S231: Data discontinuity and missing data: If a sample data is missing within 1 minute of sampling time, or if the missing data does not exceed 6 sampling points, the missing value is filled by the average of the non-missing data in the adjacent cells before and after it; S232: Continuous missing data: If data with a sampling time of more than 1 minute is missing, all data within the missing time period will be deleted directly.

[0049] S24: Data Denoising: For the voltage signal, a filter is used for noise reduction, specifically the Savitzky-Golay filter (a digital smoothing filtering algorithm) to remove high-frequency noise from the original voltage signal. However, this is not the only applicable filter; other filters are also suitable for this disclosure. The Savitzky-Golay filter achieves filtering by performing polynomial fitting within a sliding window. The Savitzky-Golay filter filtering process is as follows: S241: Polynomial Fitting Formula: Within a sliding window of length 2m+1 (m is a non-negative integer, the window contains 2m+1 data points), where m refers to the number of data frames, and the voltage signal data points are... ,in, i It refers to the first i Data points, y This refers to the individual unit voltage, and the corresponding relative offset (i.e., the offset relative to the median of the individual unit voltages in each frame) is... By fitting the signal data points within the window using an nth-order polynomial p(t), the expression of the polynomial is as follows:

[0050] The goal is to solve for the coefficients of the polynomial using the least squares method. , so that:

[0051] In the formula, k It refers to a specific data frame, that is, the first... k One data frame.

[0052] In this embodiment of the disclosure, m takes the value 5 and n takes the value 3, but is not limited to this, and other values ​​are also applicable to this disclosure.

[0053] S242: The coefficients are solved using the least squares method. Substituting the offset of each point within the window into the polynomial, the matrix form of the system of equations is obtained:

[0054] in, ,y It is a matrix of voltage signal data points.

[0055]

[0056] The polynomial basis matrix constructed for the offset, where e is the error vector.

[0057] By minimizing the sum of squared errors The optimal coefficients are obtained:

[0058] S343: Extracting Filter Coefficients: The filtered output value is the fitted value of the polynomial at the center position x=0 of the window. That is, the constant term of the polynomial. Solve using the least squares method. It can be represented as:

[0059] in, It is a matrix The first row yields the filter coefficient vector. Finally, the original data is subjected to a sliding convolution using the filter coefficient vector to obtain the filtered data.

[0060] Step S3, Data Segmentation: Based on one or more data fields, the charge and discharge data segments of the energy storage battery are segmented. Specifically, the charge and discharge data segments are segmented based on data fields such as state of charge, current, and SOC (State of Charge) to obtain charge and discharge cycle data fragments. Optionally, charge and discharge cycle data fragments can also be extracted from the charge and discharge data based solely on the state of charge. For example, fragments that meet the SOC (State of Charge) range requirements are the data fragments to be extracted.

[0061] S4. Data Feature Extraction: Based on the segmented charge-discharge cycle data fragments of the energy storage battery, obtain voltage deviation difference features, voltage distribution difference features, voltage variability features, and voltage curve similarity features, and construct a feature matrix. S41. Based on the segmented charge-discharge cycle data fragments of the energy storage battery, obtain the voltage deviation difference feature DA. The voltage deviation difference feature DA is an indicator characterizing the voltage fluctuation of each individual cell in the backup energy storage battery. Typically, the voltage fluctuation of a cell containing a fault signal will be greater than that of a normal cell. The calculation method for this feature is as follows: S411: Based on the extracted charge-discharge cycle data segments, a voltage matrix is ​​constructed. As shown below:

[0062] Where K is the number of frames in the charge / discharge cycle data segment, and N is the number of individual battery cells. It is the voltage value of the j-th battery cell at sampling time t (i.e., the t-th frame), where j∈N, t∈K, and N and K are both integers.

[0063] S412: Calculate the mean voltage matrix :

[0064]

[0065] in, This represents the average voltage of all individual cells in the backup power storage battery at time t.

[0066] S413: Obtain the voltage deviation difference characteristic DA, then the voltage deviation difference of the j-th battery cell DA j for: .

[0068] Step S42: Based on the segmented charge-discharge cycle data fragments of the energy storage battery, obtain the voltage distribution difference feature (DT). The voltage distribution difference feature (DT) is an indicator characterizing the fluctuation of a single backup energy storage battery cell over time. Typically, the voltage signal fluctuation will be greater when fault signals are present than the normal voltage signal. The calculation method for this feature is as follows: S421: Based on voltage matrix Calculate the mean matrix of each cell in the charge-discharge cycle data segment. :

[0069]

[0070] Where K is the number of frames in the charge / discharge cycle data segment, and N is the number of individual battery cells. p This represents the average voltage of the p-th cell in the backup power storage battery within the charge / discharge cycle data segment. p ∈N, This represents the average voltage of the j-th cell in the segment of the backup power storage battery. This represents the voltage value of the j-th battery cell at sampling time t.

[0071] S422: Obtain the voltage distribution difference feature (DT), then the voltage distribution difference of the j-th cell... DT j for:

[0072] Step S43: Based on the segmented charge and discharge data segments of the energy storage battery, obtain the voltage variability characteristics. DV, Voltage variability characteristics of the j-th cell DV j for:

[0073] Where K is the number of frames in the charge / discharge cycle data segment.

[0074] Step S44: Based on the segmented charge-discharge cycle data fragments of the energy storage battery, obtain the voltage curve similarity DF. The voltage curve similarity is measured by the Frechet distance, which is an efficient method for evaluating curve similarity. Then, the voltage variability characteristics of the j-th battery cell are... DF j for:

[0075] in, This represents any value of the j-th battery cell within the segment. The above formula represents the average voltage of all individual cells in the backup power storage battery at any time within the segment. Through the "robust optimization" approach, it minimizes the maximum difference between the j-th cell and other cells (or the target voltage) under the worst-case scenarios such as maximum fluctuation and extreme deviation of the cell voltage. That is, it obtains the minimum difference of the j-th cell voltage under the worst-case scenario by constructing a min-max objective function and solving it in combination with constraints.

[0076] Step S45: Construct the feature matrix include, Obtain the voltage deviation difference characteristics, voltage distribution difference characteristics, voltage variability characteristics, and voltage curve similarity of all individual battery cells; Based on the voltage deviation differences, voltage distribution differences, voltage variability, and voltage curve similarity of all individual battery cells, a feature matrix is ​​constructed. : .

[0077] In the formula, K is the number of frames of valid data segments, and N is the number of individual battery cells. DA N Voltage deviation difference of the Nth battery cell; DV N The voltage variability characteristic of the Nth battery cell; DF N The voltage variability characteristic of the Nth battery cell; DTN The voltage distribution difference of the Nth battery cell.

[0078] Step S5: Perform clustering processing on the feature matrix. Specifically, cluster the feature data using the OPTICS clustering algorithm (Ordering Points To Identify the Clustering Structure, a density-based hierarchical clustering algorithm), including: Step S51: Determine the distance radius using the OPTICS clustering method. Minimum number of clusters Among them, the distance radius is determined by trial and error. For an infinite number of nodes, the minimum number of nodes required for clustering is [value missing]. Value Where N is the number of battery cells; the core steps of determining MinPts (the core parameter of the OPTICS density clustering algorithm, i.e., the minimum number of points) using the trial-and-error method include initial value setting. Clustering Validation Iterative adjustment Convergence termination, its core function is ( In density clustering, it is the neighborhood radius. Given that the neighborhood radius is infinitely large (meaning the neighborhood includes all data points), the goal is to find the optimal system that can accurately distinguish between normal battery cell clusters and abnormal cells. Without the need for complex mathematical modeling, the parameter adaptation problem of "minimum core cluster size" in OPTICS clustering is solved solely through feedback from actual clustering results.

[0079] Step S52: By using the feature matrix The input is fed into the OPTICS clustering algorithm to calculate the reachability distance of each cell voltage. .

[0080] (1) Calculate the pairwise distances between all battery cells: For any two battery cells i and j, calculate the absolute value of the difference in the voltage deviation index of the battery cells d(i, j) = |v i -v j |, thus obtaining the distance matrix D, where elements D ij =d(i,j).

[0081] (2) Identify the core point and calculate the core distance: for each battery cell Extract the first element from the distance matrix. Sort all elements of the row in ascending order of these distances to obtain a sorted distance sequence. If the first MinPts distances exist after sorting (because... (If it must exist), then As the core point, the core distance is the MinPts-th distance. If the battery cell... Insufficient number of nearest neighbors (in practice) (If this situation does not hold true), then For non-core points, the core distance is undefined.

[0082] (3) Calculate the reachable distance of each battery cell: ① For each battery cell p, iterate through all core points. Calculate p for each battery cell to each reachable distance ; ②The final reachable distance of a single battery cell p, taking all The minimum value in the range (i.e., the reachable distance to the nearest core point): ; ③ Special cases: If there is no core point Then the reachable distance of all individual units is .

[0083] Step S6, Graded Early Warning: Based on the clustering results and combined with the sigma rule, graded early warning is performed on backup power storage batteries, including... Step S61: Set fault classification threshold And calculate the average distance achievable by the voltage of all individual units. and standard deviation ;

[0084]

[0085] In the formula, i represents the i-th battery cell. Let N represent the reachable distance of the i-th battery cell, and N be the number of battery cells.

[0086] Step S62: Based on fault classification threshold and the calculated mean and standard deviation The thresholds for the three-level fault classification are determined; among them, the thresholds for the three-level fault classification include the minor fault safety threshold. Moderate fault safety threshold Severe fault safety threshold ; Step S63: Based on reachable distance Based on the determined three-level fault classification thresholds, obtain the state classification of the backup power storage battery:

[0087] In the formula, This represents the clustering label value of the j-th battery cell. The larger the label value, the higher the level of anomaly. This represents the reachable distance of the j-th battery cell; Step S64: If abnormal battery cell information is identified in the extracted charge / discharge cycle data segment, a safety warning is issued based on the highest fault level occurring in the battery cell, and the cell numbers of all warning cells are output. For example, the identified... An alarm will be triggered if the value equals 3.

[0088] By acquiring characteristics of voltage deviation differences, voltage distribution differences, voltage variability, and voltage curve similarity, a feature matrix is ​​constructed. Based on clustering methods, a graded safety early warning system for the battery system can be implemented. This enables data-driven identification of the internal evolution trends of the backup power storage system and analysis of the underlying relationships behind the data. Furthermore, by issuing different warnings at different levels and locating problematic individual units, targeted inspections and corrections of relevant issues can be carried out. In addition, cluster analysis can help determine the differences between different batteries, which can assist in further analysis and evaluation of the health of the power battery, providing data support for the rational configuration and refined management of batteries.

[0089] like Figure 3 As shown in the embodiments of this disclosure, a graded early warning system for backup power storage batteries based on an unsupervised learning algorithm, capable of executing the above-described method, is also introduced. The system includes an acquisition module, a segmentation module, a feature extraction module, a clustering module, and an early warning module. The acquisition module acquires backup power storage battery data; the segmentation module segments the charge / discharge data segments of the storage battery based on one or more data fields; the feature extraction module extracts voltage deviation differences, voltage distribution differences, voltage variability, and voltage curve similarity features from the segmented charge / discharge data segments, constructing a feature matrix; the clustering module performs clustering processing on the feature matrix; and the early warning module provides graded early warnings for backup power storage batteries based on the clustering results. This system extracts features from the voltage list data in the collected data. The system uses fewer features and preset parameters, and through data-driven monitoring, it provides graded early warnings for backup power storage battery safety, exhibiting wider applicability and better early warning capabilities. The system described above provides graded early warnings for batteries. Cluster analysis can help identify the differences between different batteries and assist in further analyzing and evaluating the health of power batteries. This provides data support for the rational configuration and refined management of batteries. In the field of electric vehicles, it solves the problem of comprehensively and objectively scoring the health of power batteries based on a multi-indicator model that is inherently related.

[0090] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A graded early warning method for backup power storage batteries based on an unsupervised learning algorithm, characterized in that, include, Obtain backup power storage battery data; Based on one or more data fields, the charging and discharging data segments of the energy storage battery are segmented. Based on the segmented charge and discharge data segments of the energy storage battery, voltage deviation difference characteristics, voltage distribution difference characteristics, voltage variability characteristics, and voltage curve similarity are obtained to construct a feature matrix; Cluster the feature matrix; Based on the clustering results, a graded early warning system is implemented for backup power storage batteries.

2. The method for graded early warning of backup power storage batteries based on unsupervised learning algorithm according to claim 1, characterized in that, Acquiring backup power storage battery data includes collecting static and dynamic data of the backup power storage battery at preset time frame intervals. The collected backup power storage battery data includes data collection time, total voltage, total current, and individual cell voltage.

3. The method for graded early warning of backup power storage batteries based on unsupervised learning algorithm according to claim 2, characterized in that, It also includes preprocessing of backup power storage battery data, which includes data noise reduction using filters, including... Within a sliding window of length 2m+1, where m refers to the number of data frames, the signal data points are... ,in, i It refers to the first i Data points, y This refers to the voltage of a single cell, corresponding to the relative offset of that cell. By fitting the signal data points within the window using an nth-order polynomial p(t), the expression of the polynomial is as follows: The goal is to solve for the coefficients of the polynomial using the least squares method. , so that: In the formula, k Indicates the first k One data frame; The coefficients are solved using the least squares method. By substituting the offset of each point within the window into the polynomial, the matrix form of the system of equations is obtained. in, , y A matrix of voltage signal data points. The polynomial basis matrix constructed for the offset, where e is the error vector; By minimizing the sum of squared errors The optimal coefficients are obtained: Extract the filter coefficients, where the filtered output value is the fitted value of the fitting polynomial at the window center position x=0. That is, the constant term of the polynomial. According to the least squares solution, for: in, It is a matrix The first row yields the filter coefficient vector. .

4. The method for graded early warning of backup power storage batteries based on unsupervised learning algorithm according to claim 3, characterized in that, Based on one or more data fields, the charging and discharging data segments of the energy storage battery are segmented, including based on the state of charge, current and / or SOC data fields, to obtain charge and discharge cycle data segments.

5. The method for graded early warning of backup power storage batteries based on unsupervised learning algorithm according to claim 4, characterized in that, Based on the segmented charge and discharge data segments of the energy storage battery, the voltage deviation difference characteristics are obtained, including: A voltage matrix is ​​constructed based on the extracted charge-discharge cycle data segments. : Where K is the number of frames in the charge / discharge cycle data segment, and N is the number of individual battery cells. It is the voltage value of the j-th battery cell at sampling time t, j∈N, t∈K; Based on voltage matrix The mean voltage matrix is ​​obtained by taking the voltage values ​​of all individual cells at each time point. : in, This represents the average voltage value of all individual battery cells at time t; Based on voltage matrix Voltage values ​​and average voltage matrix of all battery cells The average voltage values ​​of all individual battery cells are used to obtain the voltage deviation difference characteristic DA, where the voltage deviation difference DA of the j-th cell is... j for: 。 6. The method for graded early warning of backup power storage batteries based on unsupervised learning algorithm according to claim 5, characterized in that, Based on the segmented charge and discharge data segments of the energy storage battery, the voltage distribution difference characteristics are obtained, including: Based on voltage matrix For each cell in the data segment, calculate the mean matrix of each cell in the charge-discharge cycle data segment by taking the voltage value of each cell at all times. : Where K is the number of frames in the charge / discharge cycle data segment, and N is the number of individual battery cells. p This represents the average voltage of the p-th cell in the backup power storage battery within the charge / discharge cycle data segment. p ∈N, This represents the voltage value of the j-th battery cell at sampling time t. This represents the average voltage of the j-th battery cell in the charge-discharge cycle data segment of the backup power storage battery. Based on voltage matrix The voltage and mean matrix of each cell at all times. The average voltage of all battery cells in the charge-discharge cycle data segment is used to obtain the voltage distribution difference feature DT. Then, the voltage distribution difference of the j-th cell is... DT j for: 。 7. The method for graded early warning of backup power storage batteries based on unsupervised learning algorithm according to claim 6, characterized in that, Based on the segmented charge and discharge data segments of the energy storage battery, voltage variability characteristics are obtained. DV, Voltage variability characteristics of the j-th individual DV j for: Where K is the number of frames in the charge / discharge cycle data segment.

8. The method for graded early warning of backup power storage batteries based on unsupervised learning algorithm according to claim 7, characterized in that, Based on the segmented charge and discharge data of the energy storage battery, the voltage curve similarity DF is obtained, and then the voltage variability characteristics of the j-th cell are determined. DF j for: in, Represents any value of the j-th monomer within the fragment. This represents the average voltage of all individual cells in the backup power storage battery at any given time within the segment.

9. The method for graded early warning of backup power storage batteries based on unsupervised learning algorithm according to claim 8, characterized in that, Constructing the feature matrix include, Obtain the voltage deviation difference characteristics, voltage distribution difference characteristics, voltage variability characteristics, and voltage curve similarity of all individual battery cells; Based on the voltage deviation differences, voltage distribution differences, voltage variability, and voltage curve similarity of all individual battery cells, a feature matrix is ​​constructed. : In the formula, K is the number of frames of valid data segments, and N is the number of individual battery cells. DA N Voltage deviation difference of the Nth battery cell; DV N The voltage variability characteristic of the Nth battery cell; DF N The voltage variability characteristic of the Nth battery cell; DT N The voltage distribution difference of the Nth battery cell.

10. The method for graded early warning of backup power storage batteries based on unsupervised learning algorithms according to any one of claims 1-9, characterized in that, Clustering of the feature matrix includes, Determine the distance radius Minimum number of clusters Among them, the distance radius is determined by trial and error. For an infinite number of nodes, the minimum number of nodes required for clustering is [value missing]. Value , where N is the number of individual battery cells; By inputting the feature matrix into the clustering algorithm, the reachable distance of each individual cell voltage is calculated. .

11. The method for graded early warning of backup power storage batteries based on unsupervised learning algorithm according to claim 10, characterized in that, Based on the clustering results, a tiered early warning system for backup power storage batteries is implemented, including: Set fault classification thresholds ; Reachable distance based on individual cell voltages Calculate the average distance achievable by the voltage of all individual units. and standard deviation ; Based on fault classification threshold and the calculated mean and standard deviation The thresholds for the three-level fault classification are determined; among them, the thresholds for the three-level fault classification include the minor fault safety threshold. Moderate fault safety threshold Severe fault safety threshold ; Based on reachable distance Based on the determined three-level fault classification thresholds, obtain the state classification of the backup power storage battery: In the formula, This represents the clustering label value of the j-th battery cell. The larger the label value, the higher the level of anomaly. This represents the reachable distance of the j-th battery cell; If the extracted charge / discharge cycle data segment identifies battery cell information with diagnostic abnormalities, a safety warning is issued based on the highest fault level observed in the battery cell, and the cell number of all warning cells is output.

12. A graded early warning system for backup power storage batteries based on an unsupervised learning algorithm, characterized in that, include, The acquisition module is used to acquire data from backup power storage batteries; The segmentation module is used to segment the charge and discharge data segments of the energy storage battery based on one or more data fields. The feature extraction module is used to obtain voltage deviation difference features, voltage distribution difference features, voltage variability features, and voltage curve similarity features based on the segmented charge and discharge data segments of the energy storage battery, and construct a feature matrix. The clustering module is used to cluster the feature matrix. The early warning module is used to provide graded early warnings for backup power storage batteries based on the results of clustering.

Images