A method for rapid detection of health state of an electrochemical energy storage cell based on UMLC and ICA, a medium and a program product

Abstract

The application discloses a kind of based on UMLC and ICA's electrochemical energy storage cell health state rapid detection method, medium and program product, belong to electrochemical energy storage power station monitoring and battery management technical field.This method is by BMS real-time acquisition battery cluster in each cell monomer voltage, monomer temperature and battery cluster current and other operating data, constructs the characteristic vector containing voltage and temperature mean, standard deviation and deviation, and is combined with the unsupervised machine learning clustering (UMLC) and incremental capacity analysis (ICA) and other technologies not dependent on historical training data, identify and extract the key features of abnormal cell, to quickly diagnose cell health state.This method does not need historical data, with higher detection efficiency and interpretability, can be directly applied to on-site energy storage power station, realize to cell fast and accurate consistency analysis screening and positioning abnormal cause.

Description

Technical Field

[0001] This invention belongs to the field of monitoring and battery management technology for electrochemical energy storage power stations, and relates to battery consistency detection and fault diagnosis. In particular, it relates to a rapid detection method, medium and program product for the health of electrochemical energy storage cells based on unsupervised machine learning clustering (UMLC) and incremental capacity analysis (ICA), which can be used for rapid screening and analysis of battery health status in large-scale energy storage power stations. Background Technology

[0002] In large-scale electrochemical energy storage power stations, due to the large number of battery cells, the cells will gradually differ in terms of capacity, internal resistance, polarization, and voltage response due to manufacturing discreteness, assembly differences, and aging mechanism differentiation. This difference will accumulate and manifest as inconsistency degradation. This problem will directly affect the state estimation accuracy and available capacity release of the battery management system (BMS), leading to increased SOC (State of Charge) deviation and decreased energy utilization. In severe cases, it may induce safety risks such as local overcharging, over-discharging, and abnormal heating. Therefore, the consistency of each cell is crucial to the overall performance and safety of the power station.

[0003] Existing methods for testing and diagnosing battery consistency in energy storage systems often rely on time-consuming full-charge-discharge experiments or long-term historical data analysis. For example, conventional methods require periodically performing full charge-discharge tests on the battery pack to measure the capacity and internal resistance differences of each cell, thereby determining the deviation in cell performance. These methods are not only time-consuming (full-site testing often takes several hours to several days) but also highly dependent on human experience, requiring professionals to perform offline analysis of large amounts of data. Furthermore, traditional methods lack on-site operability, typically requiring the battery modules to be disassembled and tested using specialized equipment, making it impossible to directly perform consistency checks at the power station operation site. These shortcomings of existing technologies make it difficult to detect and diagnose battery consistency problems in energy storage power stations in a timely and accurate manner, seriously affecting the operation and maintenance efficiency of the power station and battery safety management.

[0004] In terms of health assessment and fault diagnosis, traditional methods such as capacity testing, pulse testing, and static calibration can obtain relatively direct SOH (State of Health) related characteristics, but they often require specific operating conditions or long testing cycles, making them difficult to coordinate with continuous power plant operation. Diagnostic methods based on physical models or mechanistic constraints are sensitive to parameter accuracy and boundary conditions; model mismatch, temperature coupling, and changes in aging stages may lead to unstable conclusions. In recent years, data-driven methods have been gradually applied to battery state assessment and anomaly identification, including supervised learning and unsupervised learning approaches, but they generally face engineering challenges such as high sample acquisition costs, difficulties in transferring operating conditions across sites, insufficient consistency in feature selection and labeling, sensitivity to missing data and sensor drift, and limited interpretability of results.

[0005] In summary, existing technologies for battery consistency testing in electrochemical energy storage power stations suffer from low testing efficiency, poor accuracy, poor interpretability, over-reliance on historical data, and difficulty in on-site implementation. Therefore, there is an urgent need to propose a rapid detection and analysis method for battery health in electrochemical energy storage power stations that does not rely on historical data, is available on-site, can quickly diagnose problems, and possesses a certain degree of mechanistic explanation capability, in order to solve the current technical challenges in the operation of electrochemical energy storage power stations. Summary of the Invention

[0006] (a) Purpose of the invention

[0007] To address the aforementioned deficiencies and shortcomings in existing technologies, this invention aims to provide a rapid detection method, medium, and program product for the health status of electrochemical energy storage cells based on unsupervised machine learning clustering (UMLC) and incremental capacity analysis (ICA). This method integrates an unsupervised clustering algorithm for initial screening of cell operating parameters to ensure consistency, and utilizes incremental capacity analysis for electrochemical mechanism diagnosis of abnormal cells. It can complete the screening of battery consistency anomalies and mechanism-assisted diagnosis within minutes using real-time data collected during power station operation, without interrupting normal power station operation or requiring battery pack removal. Through this invention, abnormal cells with performance deviations in large-scale energy storage power stations can be quickly identified. Furthermore, without disassembling the battery pack, the combination of unsupervised clustering algorithms and incremental capacity analysis accurately pinpoints the causes of health problems, achieving an organic combination of initial consistency screening and fault cause localization. The method is simple, easy to implement, and readily applicable.

[0008] (II) Technical Solution

[0009] To achieve the objective of this invention and solve its technical problems, the present invention adopts the following technical solution:

[0010] The first objective of this invention is to provide a rapid detection method for the health status of electrochemical energy storage cells based on UMLC and ICA, comprising the following steps:

[0011] S100. Data Acquisition: Real-time acquisition and preprocessing of the operating data of each cell in the battery cluster, including at least individual cell voltage, individual cell temperature and battery cluster current, to form a cell operating dataset;

[0012] S200. Feature Construction: Based on the cell operation dataset, construct the feature vector of each cell within the preset analysis time window, which includes at least the mean, standard deviation and deviation of the cell voltage and cell temperature. The deviation of the cell voltage is the deviation between the cell voltage of each cell and the mean of the cell voltage of all cells in the battery cluster. The deviation of the cell temperature is the deviation between the cell temperature of each cell and the mean of the cell temperature of all cells in the battery cluster.

[0013] S300. UMLC Initial Screening: An unsupervised machine learning clustering algorithm (UMLC) is used to perform cluster analysis on the feature vectors of each battery cell. Battery cells whose feature vectors are close in distance in the high-dimensional feature space are grouped into the same cluster. The cluster containing the most battery cells is identified as the main cluster. The Euclidean distance between the feature vector of each battery cell and the center of its respective cluster is calculated. Battery cells whose distance from the center of the main cluster exceeds a preset distance threshold are marked as abnormal battery cells, or battery cells that form small independent clusters are marked as abnormal battery cells.

[0014] S400. ICA Calculation: Within a continuous time period where the current direction of the battery cluster is charging, the capacity sequence is obtained based on the integration of current with time, and voltage-capacity relationship data is formed by corresponding it with the voltage of each individual cell; for each abnormal cell and each cell in the main cluster, the derivative of capacity with respect to voltage dQ / dV is calculated to generate the corresponding incremental capacity ICA curve; the arithmetic mean of the ICA curves of each cell in the main cluster is performed to form the baseline ICA curve;

[0015] S500. Difference Determination: For each abnormal cell, calculate and identify the characteristic differences of its ICA curve relative to the reference ICA curve in terms of the position of the main peak, the peak amplitude, and the curve shape. Perform diagnostic analysis on the abnormal cell, determine the abnormal type, and locate the cause of the abnormality.

[0016] S600. Output Results: Outputs the detection results for each abnormal cell, forming a consistent detection and analysis result of the cell health status of the battery cluster.

[0017] The second objective of this invention is to provide a computer program product, including computer instructions, for executing the steps of the above-described rapid detection method for the health status of electrochemical energy storage cells based on UMLC and ICA.

[0018] The third objective of this invention is to provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described rapid detection method for the health status of electrochemical energy storage cells based on UMLC and ICA.

[0019] (III) Technical Effects

[0020] Compared with existing technologies, the rapid detection method, medium, and program products for the health status of electrochemical energy storage cells based on UMLC and ICA of the present invention have at least the following significant advantages:

[0021] (1) The detection efficiency is greatly improved. The cell consistency is initially screened using an unsupervised machine learning clustering algorithm, which can complete the detection of consistency anomalies of thousands of cells in minutes. This significantly shortens the detection cycle compared to traditional methods of group-by-group measurement or offline analysis. At the same time, this method uses real-time online data for detection, which is ready to use immediately without waiting for a lengthy charging and discharging process.

[0022] (2) Improved diagnostic accuracy and interpretability: By introducing mechanistic analysis methods such as incremental capacity analysis (ICA), this method can not only identify abnormal cells, but also provide visual diagnostic evidence (such as the comparison between ICA curves and average curves) to help determine the cause of abnormalities. Compared with the method of simply relying on numerical thresholds, the results such as ICA curves are more interpretable and can intuitively assess the health status of cells.

[0023] (3) Independent of large data samples, the method of this invention does not rely on pre-trained large data samples and complex models, but can be implemented based solely on the data characteristics under the current operating conditions. This means that whether it is a newly commissioned power plant or a power plant that has been in operation for many years, this method can be directly applied to cell health testing, which is suitable for the consistency assessment needs of different sites and has good universality and promotion value.

[0024] (4) No disassembly required and easy to deploy. This method can complete the analysis based on the voltage, temperature and other operating data collected by the existing battery management system without disassembling the battery pack or adding additional hardware. The software algorithm can be integrated into the power station monitoring system or operation and maintenance terminal to realize online diagnosis of abnormal cells, avoid interference with the normal operation of the battery system, and reduce detection costs and difficulties. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the rapid detection method for the health status of electrochemical energy storage cells based on UMLC and ICA according to the present invention.

[0026] Figure 2This is a comparison chart of the original charging voltage curve and the average curve of some battery cells with consistency deviation in a certain energy storage power station in Example 2;

[0027] Figure 3 This is a comparison chart of the incremental capacity (ICA) curve and the average ICA curve of a certain energy storage power station with some consistency deviations in Example 2. Detailed Implementation

[0028] This invention aims to provide a method, medium, and program product for rapid detection of the health status of electrochemical energy storage cells based on unsupervised machine learning clustering (UMLC) and incremental capacity analysis (ICA). To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions in the embodiments of this invention will be described in more detail below with reference to the accompanying drawings. The described embodiments are some, but not all, embodiments of this invention, and are exemplary, intended to explain the invention, and should not be construed as limiting the invention.

[0029] Example 1: Rapid Detection and Analysis Method for Battery Cell Health

[0030] like Figure 1 As shown, the present invention provides a rapid detection method for the health status of electrochemical energy storage cells based on unsupervised machine learning clustering (UMLC) and incremental capacity analysis (ICA). This method is used to identify abnormal cells and output results without interrupting the operation of the energy storage power station and without disassembling the battery pack. Its implementation mainly includes the following steps:

[0031] S100. Data Acquisition:

[0032] Real-time acquisition and preprocessing of operational data for each cell within the battery cluster, including at least individual cell voltage, individual cell temperature, and battery cluster current, forms a cell operational dataset. This operational data is directly acquired by the Battery Management System (BMS), and the acquisition targets include the individual cell voltage, individual cell temperature, and battery cluster current of each cell within the battery cluster. The preprocessing includes at least timestamp alignment of the individual cell voltage, individual cell temperature, and battery cluster current data; interpolation completion of missing sampling points; filtering and noise reduction; and outlier removal. Timestamp alignment unifies data from different sensors to the same time base; filtering and noise reduction uses sliding window averaging or Kalman filtering to eliminate measurement noise; and outlier removal uses 3D filtering. σ The criteria or interquartile range-based method identifies and removes abnormal data points that deviate significantly from the normal range, ensuring the temporal consistency and quality reliability of the data used in subsequent analyses; and the resulting cell operation dataset contains at least the voltage distribution data and temperature distribution data of each cell at the same time or within the same short period of time, to support rapid initial screening analysis of the cell group status.

[0033] S200. Feature Construction:

[0034] Based on the battery cell operation dataset, a multidimensional feature vector is constructed for each battery cell within a preset analysis time window. The preset time window is a continuous time period during the charging or discharging process. The multidimensional feature vector includes at least the mean, standard deviation, and deviation of the individual cell voltage and temperature. The deviation of the individual cell voltage is the deviation between the individual cell voltage and the mean of the individual cell voltages of all cells in the battery cluster. The deviation of the individual cell temperature is the deviation between the individual cell temperature and the mean of the individual cell temperatures of all cells in the battery cluster. In addition to the mean, standard deviation, and deviation of the individual cell voltage and temperature, the feature vector also includes two dimensions: inter-cell voltage difference and inter-cell temperature difference. The inter-cell voltage difference is the difference between the maximum and minimum values ​​of the individual cell voltages within the battery cluster within the preset analysis time window. The inter-cell temperature difference, along with the statistical values ​​of individual cell voltage and temperature, is used for consistency characterization.

[0035] Furthermore, the feature vector also includes two dimensions: single-cell voltage change rate and single-cell temperature change rate. The single-cell voltage change rate is the ratio of the change in single-cell voltage within the analysis time window to the time span, and the single-cell temperature change rate is the ratio of the change in single-cell temperature within the analysis time window to the time span. By constructing a comprehensive feature vector that includes both statistical and dynamic features, the sensitivity of cluster analysis in identifying differences in cell status is improved.

[0036] S300. UMLC initial screening:

[0037] An unsupervised machine learning clustering algorithm (UMLC) is used to cluster the feature vectors of each battery cell. Cells whose feature vectors are close in distance in the high-dimensional feature space are grouped into the same cluster. The cluster containing the most cells is identified as the main cluster. The Euclidean distance between each cell and the center of its respective cluster is calculated. Cells whose distance from the center of the main cluster exceeds a preset distance threshold are marked as abnormal cells, or cells that form small independent clusters are marked as abnormal cells. The clustering analysis is completed based on the features of the currently collected data and does not rely on historical training data.

[0038] In this embodiment of the invention, when performing UMLC clustering analysis on the feature vector of each battery cell, the following steps are included:

[0039] S301. Feature standardization processing: The feature vectors of each cell are scaled uniformly, and standardization is completed based on the statistics of the corresponding dimensional features within the battery cluster, so that the features of each dimension meet the comparability and form a set of standardized feature vectors. The standardization processing adopts the Z-score standardization method to convert each feature parameter into a standard normal distribution with a mean of 0 and a standard deviation of 1, or adopts the minimum-maximum normalization method to linearly map each feature parameter to the interval [0,1].

[0040] S302. Clustering parameter determination: Perform clustering within the preset range of candidate cluster numbers and calculate the clustering effectiveness evaluation index. Determine the target cluster number based on the principle of optimal evaluation index, and determine the initialization rules and iteration termination conditions for cluster centers.

[0041] S303. Clustering Iteration and Cluster Center Update: In the high-dimensional feature space, the distance between standardized feature vectors is used as a similarity criterion to group cells that are close in distance into the same cluster, and the cluster center is updated according to the statistical center of the feature vectors within the cluster; the sample assignment and cluster center update are repeated until the iteration termination condition is met, and the cluster label of each cell and the cluster center of each cell are obtained.

[0042] S304. Identification of the main cluster: Count the number of battery cells contained in each cluster, determine the cluster with the most battery cells as the main cluster, and determine the center of the main cluster as the group consistency benchmark center;

[0043] S305. Distance Calculation and Consistency Index Generation: Calculate the distance between the standardized feature vector of each cell and the center of its respective cluster, and calculate the distance between each cell and the center of the main cluster. Generate a consistency deviation index based on the distance between the cell and the center of the main cluster to quantify the degree of consistency deviation of the cell relative to the group benchmark.

[0044] S306. Anomaly Detection: Anomaly identification is performed based on a preset distance threshold and a preset cluster size threshold: when the distance between the battery cell and the center of the main cluster exceeds the preset distance threshold, it is marked as an abnormal battery cell; when the size of the cluster to which the battery cell belongs is less than the size threshold, the battery cells within the cluster are marked as abnormal battery cells.

[0045] S307. Output Anomaly Set: Output the set of abnormal cells and their corresponding consistency deviation indicators and cluster labels, which serve as input objects for subsequent ICA calculation and difference determination.

[0046] In this embodiment of the invention, the main cluster is determined by the cluster with the largest proportion of battery cells after clustering and the smallest dispersion of feature vectors within the cluster. The dispersion within the cluster is characterized by the mean or variance of the distance between the feature vectors of each battery cell and the cluster center. After normalizing the distance between each battery cell and the center of its respective cluster, a consistency deviation index is formed. Battery cells with a deviation index greater than a preset threshold are marked as abnormal battery cells. This achieves quantitative screening of abnormal battery cells without relying on historical training data.

[0047] S400. ICA Calculation:

[0048] Within a continuous time interval where the battery cluster current is in the charging direction, the capacity sequence is obtained by integrating the battery cluster current with time. This sequence is then compared with the voltage sequence of each individual cell to form voltage-capacity relationship data. For each abnormal cell and each cell in the main cluster, the derivative of capacity with respect to voltage, dQ / dV, is calculated to generate the corresponding incremental capacity ICA curve. The ICA curves of each cell in the main cluster are then arithmetically averaged to form a baseline ICA curve.

[0049] In this embodiment of the invention, performing incremental capacity analysis (ICA) on abnormal cells includes at least the following:

[0050] S401. Charging segment determination: Based on the battery cluster current data, under the condition that the current direction is the charging direction, identify the time segment where the current direction is consistent and continuous, and use it as the ICA calculation segment;

[0051] S402. Data Synchronization and Segment Extraction: Time-align the battery cluster current sequence and the individual cell voltage sequence of the abnormal cell within the ICA calculation segment to form segment current-voltage synchronization data for the abnormal cell; and form corresponding segment current-voltage synchronization data for each cell within the main cluster.

[0052] S403. Capacity sequence calculation: Taking the start time of the ICA calculation segment as the integration starting point, the battery cluster current sequence is integrated with the charging current in the positive direction to obtain the capacity sequence. The sequence is then matched with each sampling time in the segment, and the capacity increment is used to characterize the change in charging capacity in the segment.

[0053] S404. Voltage-Capacity Relationship Construction: The capacity sequence is matched with the individual voltage sequence of the abnormal cell according to the sampling time to form voltage-capacity relationship data; the capacity sequence is matched with the individual voltage sequence of each cell in the main cluster according to the sampling time to form voltage-capacity relationship data for each cell respectively.

[0054] S405. ICA curve generation: The voltage-capacity relationship data is resampled with a uniform step size on the voltage axis, and the capacity-voltage relationship is smoothed after resampling to reduce the amplification of noise by the differential operation; on this basis, the derivative of the resampled capacity with respect to voltage is obtained to obtain dQ / dV, and the ICA curve of the abnormal cell and the ICA curve of each cell in the main cluster are generated.

[0055] S406. Construction of the benchmark ICA curve: The ICA curves of each cell in the main cluster are arithmetically averaged under a unified voltage axis coordinate to obtain the benchmark ICA curve that characterizes the average characteristics of the main cluster.

[0056] S407. ICA Result Output: Output the ICA curve of the abnormal cell and its corresponding reference ICA curve, and use them as input for step S500 difference judgment and mechanism diagnosis analysis.

[0057] S500. Difference Determination:

[0058] For each abnormal cell, calculate and identify the characteristic differences of its ICA curve relative to the reference ICA curve in terms of the position of the main peak, the peak amplitude, and the curve shape. Perform diagnostic analysis on the abnormal cell to determine the abnormal type and locate the cause of the abnormality.

[0059] In this embodiment of the invention, the characteristic differences include at least the curve crossover misalignment criterion and the peak amplitude reduction criterion in the charging platform region: when the abnormal cell ICA curve crosses and misaligns with the reference ICA curve in the charging platform region or the peak amplitude is lower than a preset proportional threshold of the reference peak amplitude, the abnormal cell is determined to be a capacity decay type abnormality; when the main peak position of the abnormal cell ICA curve shifts and the change in peak amplitude is less than a preset amplitude threshold, the abnormal cell is determined to be a characteristic performance deviation or polarization increase type abnormality.

[0060] Anomalies are categorized into capacity decay anomalies, internal resistance increase anomalies, and material difference anomalies. When the peak value of the main peak of the abnormal cell's ICA curve is significantly lower than that of the reference ICA curve, but the peak positions are basically the same, it is identified as a capacity decay anomaly, caused by reversible capacity reduction due to loss of active material or lithium ions. When the main peak position of the abnormal cell's ICA curve shifts significantly towards higher voltage while the peak amplitude remains basically the same, it is identified as an internal resistance increase anomaly, caused by intensified polarization due to electrolyte degradation or increased contact resistance. When the overall morphology of the abnormal cell's ICA curve shows a systematic difference from the reference curve, but the peak characteristics do not change significantly, it is identified as a material difference anomaly, caused by initial differences in electrode material ratio, particle size distribution, or manufacturing process parameters.

[0061] In this embodiment of the invention, the difference determination and diagnostic analysis of abnormal battery cells includes at least the following:

[0062] S501. Curve Alignment and Effective Range Determination: Obtain the ICA curve of the abnormal cell and the reference ICA curve, align the two curves under a unified voltage coordinate system, and determine the effective voltage range for difference judgment; the effective voltage range is limited by the range of single-cell voltage variation corresponding to the charging voltage-capacity relationship data to ensure that the two curves can be compared within the same voltage range.

[0063] S502. Main Peak Identification: Within the effective voltage range, perform peak search on the ICA curve of the abnormal cell and the reference ICA curve, calculate the peak amplitude, peak position and peak width parameters of each candidate peak, and determine the candidate peak with the largest amplitude that meets the preset peak significance condition as the main peak. The peak significance condition is limited by the difference between the peak amplitude of the candidate peak and the amplitude of its adjacent valley value not being less than the preset significance threshold.

[0064] S503. Peak Difference Calculation: Based on the main peak of the abnormal cell and the reference main peak, extract the peak position, peak amplitude and peak width parameters respectively, and calculate the peak position difference, peak amplitude difference and peak width difference. The peak position difference is the absolute value of the difference between the two peak positions, the peak amplitude difference is the absolute value of the difference between the two peak amplitudes, and the peak width difference is the absolute value of the difference between the two peak width parameters, which are used to characterize the difference in the degree of peak shape broadening.

[0065] S504. Quantification of curve morphology differences: Within the effective voltage range, calculate a set of morphology difference indices for the abnormal cell ICA curve and the reference ICA curve, including at least curve correlation, morphology distance and area difference indices. The curve correlation index is the correlation coefficient between the two curves at a unified voltage sampling point, the morphology distance index is the distance measurement result between the two curves at a unified voltage sampling point, and the area difference index is the absolute value of the difference between the integrated areas of the two curves within the effective voltage range.

[0066] S505. Regularized Judgment and Anomaly Type Determination: The set of peak position difference, peak amplitude difference, peak width difference, and morphology difference indicators are input into a preset judgment rule set. The preset judgment rule set includes at least peak position difference threshold, peak amplitude difference threshold, peak width difference threshold, and morphology distance threshold rules. Anomaly type is output according to the preset judgment rule set. Preferably, the judgment rule set includes at least the following: when the peak amplitude difference and area difference indicators are both greater than their respective thresholds and the curve correlation indicator is lower than the correlation threshold, it is judged as a capacity-related anomaly; when the peak position difference and peak width difference are both greater than their respective thresholds and the morphology distance indicator is greater than the distance threshold, it is judged as a polarization-related anomaly; when the abnormal cell curve shows a discontinuous abrupt change within the effective voltage range and causes a significant increase in the morphology distance indicator, it is judged as a data anomaly-related anomaly.

[0067] S506. Anomaly Cause Localization and Confidence Assignment: Output anomaly cause labels based on the corresponding cause mapping relationship of anomaly type in the judgment rule set, and calculate confidence parameters based on the exceedance range of each difference index relative to its threshold, which are used to characterize the confidence level of anomaly type and anomaly cause localization results.

[0068] S507. Output Diagnostic Feature Set: Output the abnormal type, abnormal cause label, confidence parameters, and peak position difference, peak amplitude difference, peak width difference and morphology difference index set of the abnormal cell as the components of the output of step S600.

[0069] S600. Output:

[0070] Output the detection results for each abnormal cell to form a consistent detection and analysis result of the cell health status of the battery cluster, including: abnormal cell identification information, abnormal type determination result, abnormal cause location result, and comparison output of the abnormal cell ICA curve with the benchmark ICA curve; and generate maintenance and handling suggestions based on the abnormal type determination result. When the abnormality is determined to be capacity decay type, output equalization maintenance or replacement handling suggestions; when the abnormality is determined to be polarization rise type, output connection and heat dissipation status verification suggestions.

[0071] In this embodiment of the invention, steps S300 to S600 are repeatedly executed on multiple consecutive time segments corresponding to multiple consecutive charging processes to verify the consistency of the abnormal determination results of the same cell in multiple execution cycles. Only when the same cell is determined to be an abnormal cell in a preset number of execution cycles is the cell output as the final abnormal cell in step S600, thereby suppressing occasional false alarms caused by instantaneous operating condition disturbances.

[0072] Through the above steps, the method of this invention enables rapid detection and analysis of the health status of battery cells in electrochemical energy storage power stations. The entire process is completed based on existing real-time measurement data from the battery management system, without the need to disassemble the battery pack and without affecting the normal operation of the power station. From initial screening of consistency anomalies to output of mechanism diagnosis results, everything can be completed within minutes, significantly improving the efficiency and practicality of battery consistency testing in large-scale energy storage systems.

[0073] Example 2: Rapid Testing Application of an Energy Storage Power Station

[0074] Based on Example 1 above, Example 2 uses an electrochemical energy storage power station as an example to introduce the specific application process and practical application effect of the method of the present invention. This energy storage power station uses a lithium iron phosphate battery system with a rated capacity of 100MWh. The battery system consists of multiple battery clusters, each containing 238 individual cells. This example selects a complete charging process during normal operation of the energy storage power station as the test condition. Without interrupting the operation of the power station or disassembling the battery pack, the health status of the cells within the battery cluster is rapidly detected and analyzed.

[0075] First, the battery management system (BMS) of the power station acquires real-time operating data such as individual cell voltage, individual cell temperature, and corresponding battery cluster current for all cells within each battery cluster. Then, following step S100 in Example 1, timestamp alignment, missing point completion, filtering and noise reduction, and outlier removal are performed to form a cell operation dataset. Next, following step S200, a multi-dimensional feature vector containing the mean, standard deviation, voltage deviation, and temperature deviation is constructed within a preset analysis period.

[0076] Then, the multidimensional feature vectors formed by these data are input into an unsupervised machine learning clustering analysis model to perform fast unsupervised machine learning clustering of the cell states, identifying cells exhibiting abnormal behavior for further analysis. In the testing of this energy storage power station, all batteries were divided into several clusters for management. After applying the method of this invention to perform clustering analysis on the data of this station at a certain operating moment, the distribution of abnormal cells in each battery cluster can be obtained, as shown in Table 1 below.

[0077] Table 1. Battery cluster numbers and list of abnormal cells obtained from cluster analysis.

[0078]

[0079] Note: In Table 1, [] indicates that no abnormal cells were detected in the corresponding battery cluster during this cluster consistency screening; [n(x)] indicates that an abnormal cell with number n was detected in the battery cluster, and x in parentheses is the consistency index or consistency deviation index of the cell, which is calculated from the normalized distance between the cell feature vector and the center of the main cluster. The smaller the value, the greater the deviation from the group benchmark. The average benchmark value is 1.0.

[0080] As shown in Table 1, the clustering results revealed abnormal cell consistency in some battery clusters. For example, cluster 4 was marked with multiple abnormal cells (numbered 32, 2, 87, 90, 10, and 112), with cell #32 having the lowest consistency index at only 0.748 (relative to the average baseline of 1.0), indicating poor initial consistency screening results. Cluster 3 detected abnormal cells #108 and #53; cluster 5 detected abnormal cells #93, #161, #49, and #57; cluster 6 detected abnormal cells #145, #196, #57, and #205; and cluster 7 detected abnormal cells #149, #95, and #31. The clustering process took approximately 20 seconds to complete the initial consistency screening of 238 battery cells, successfully identifying the range of abnormal cells.

[0081] After identifying abnormal cells through clustering and screening, this embodiment selects the initially screened abnormal cells (#32, #2, #87, #90, #10, #112) from cluster 4 as the objects of further diagnostic analysis. For these abnormal cells, their individual cell voltage data during a single standard constant current charging process are extracted and compared with the average charging curve of other normal cells in cluster 4. For example... Figure 2 As shown, to clearly illustrate the differences, the figure plots the voltage curves (solid colored lines) of the six cells with poor consistency in cluster 4 during charging, as well as the average voltage curve (dashed black line) of the normal cells. It can be seen that: compared to the average curve, cell #32's voltage rises faster in the later stages of charging, reaching the charging cutoff voltage earlier than the average curve, indicating its capacity is below average; the voltage curves of cells #2 and #87 are also slightly lower than the average curve, and gradually widen the gap with the average curve towards the end of charging; the curves of other abnormal cells (#90, #10, #112) also deviate to varying degrees. These differences are particularly evident in the charging plateau stage (the plateau region where the voltage is close to saturation), suggesting that the characteristics of these cells deviate from the group's characteristics.

[0082] Furthermore, incremental capacity analysis (ICA) was performed on the charging data of the aforementioned abnormal cells to obtain the dQ / dV (incremental capacity) curve for each abnormal cell, and compared with the average ICA curve of normal cells. Figure 3As shown, the black dashed line is the average ICA curve of the normal cells in cluster 4, and the curve mainly contains the characteristic peaks of the charging platform of the corresponding battery materials; the colored solid line is the ICA curve of each abnormal cell (#32, #2, #87, #90, #10, #112). The graph shows that the peak height of the ICA curve for cell #32 is significantly lower than the average peak, and the overall curve area is also significantly reduced. This indicates that the cell can provide less capacity at the corresponding voltage platform, verifying its capacity decay anomaly—that is, due to repeated aging cycles, its usable capacity is significantly reduced compared to normal cells. The peak position of the ICA curve for cell #2 has shifted somewhat relative to the average curve, with a slight decrease in peak value, but the magnitude is not as significant as that of #32. At the same time, the dQ / dV curve corresponding to its charging platform shifts towards higher voltage. This phenomenon is usually related to increased cell polarization (increased internal resistance)—cells with higher internal resistance reach their upper limit of termination voltage earlier under the action of charging current, resulting in a slight loss of equivalent capacity. For cells #90, #10, and #112, the deviation of their ICA curves is relatively small, indicating that although they are identified as anomalies by clustering, the problem may not be as serious as that of cells like #32. They may be within the range of minor anomalies or detection errors, requiring further observation.

[0083] Through the above mechanistic diagnostic analysis, this invention not only identifies the list of abnormal battery cells but also provides the possible abnormality type and cause for each abnormal cell. In this embodiment, cell #32 is diagnosed as having severely degraded capacity, and it is recommended that maintenance personnel perform equalization maintenance or replace the module containing this cell as soon as possible to avoid dragging down the overall usable capacity of the battery cluster; cell #2 is diagnosed as having increased polarization resistance, and it is recommended to check the connection and heat dissipation of this cell, and consider replacement if necessary; for other detected abnormal cells, maintenance plans should also be developed according to their respective situations. These measures all benefit from the rapid and intuitive diagnostic results of this method, enabling maintenance personnel to take targeted actions to prevent potential faults from occurring.

[0084] This embodiment verifies the effectiveness of the detection method described in this invention in a real energy storage station. The entire detection process utilizes an existing normal charging process of the energy storage station, without interrupting the station's operation or disassembling the battery system. From the start of data acquisition to the output of the diagnostic report, the entire process is completed within approximately 5 minutes, with cluster screening taking about 1 minute, ICA analysis and result discrimination taking about 2 minutes, and the remaining time used for data transmission and processing. Compared with traditional methods, the detection efficiency is improved by tens of times, while the diagnostic results are more comprehensive and reliable. This indicates that the method of this invention can meet the need for rapid health checks of battery cells in large-scale energy storage stations, and has significant value in ensuring the safe operation of the station and optimizing operation and maintenance strategies.

[0085] The objectives of this invention have been fully and effectively achieved through the above embodiments. Those skilled in the art will understand that this invention includes, but is not limited to, the contents described in the accompanying drawings and the specific embodiments described above. Although the invention has been described with reference to what is currently considered the most practical and preferred embodiments, it should be understood that the invention is not limited to the disclosed embodiments, and any modifications that do not depart from the functional and structural principles of the invention will be included within the scope of the claims.

Claims

1. A rapid detection method for the health status of electrochemical energy storage cells based on UMLC and ICA, characterized in that, Includes the following steps: S100. Real-time acquisition and preprocessing of the operating data of each cell in the battery cluster, including at least the individual cell voltage, individual cell temperature and battery cluster current, to form a cell operating dataset, wherein the cell operating dataset contains at least the voltage distribution data and temperature distribution data of each cell at the same moment or the same short period of time. S200. Construct a feature vector for each cell within a preset analysis time window based on the cell operation dataset. The feature vector includes at least the mean, standard deviation, and deviation of the cell voltage and cell temperature. The deviation of the cell voltage is the deviation between the cell voltage of each cell and the mean of the cell voltages of all cells in the battery cluster. The deviation of the cell temperature is the deviation between the cell temperature of each cell and the mean of the cell temperature of all cells in the battery cluster. S300. Perform UMLC clustering analysis on the feature vector of each cell, group cells whose feature vectors are close in distance in the high-dimensional feature space into the same cluster, identify the cluster with the largest proportion of cells and the smallest intra-cluster dispersion as the main cluster, the intra-cluster dispersion is characterized by the mean or variance of the distance between the feature vector of each cell and the cluster center, calculate the Euclidean distance between the feature vector of each cell and the center of its cluster, and mark cells whose distance from the center of the main cluster exceeds a preset distance threshold as abnormal cells, or mark cells that form small-scale clusters independently as abnormal cells; S400. In a continuous time period where the current direction of the battery cluster is charging, the capacity sequence is obtained based on the integration of current with time, and voltage-capacity relationship data is formed by corresponding with the voltage of each individual cell. After uniform step-size resampling and smoothing of the voltage-capacity relationship data of each abnormal cell and each cell in the main cluster on the voltage axis, the derivative of capacity with respect to voltage is calculated to generate the ICA curves corresponding to the abnormal cells and each cell in the main cluster. The ICA curves of each cell in the main cluster are then arithmetically averaged to form the baseline ICA curve. S500. Calculate and identify the characteristic differences of the ICA curve of each abnormal cell relative to the reference ICA curve in terms of the position of the main peak, the peak amplitude, and the curve shape, perform diagnostic analysis on the abnormal cells, determine the abnormal type, and locate the cause of the abnormality. S600. Outputs the detection results of each abnormal cell, forming a consistent detection and analysis result of the health status of the battery cluster cells.

2. The method according to claim 1, characterized in that, In step S100, the operating data is directly collected by the BMS, and the collected data includes the individual cell voltage, individual cell temperature and battery cluster current of each cell in the battery cluster; the data preprocessing includes at least the timestamp alignment processing of the individual cell voltage, individual cell temperature and battery cluster current data, interpolation completion of missing sampling points, filtering and noise reduction processing and abnormal data removal processing.

3. The method according to claim 1, characterized in that, In step S200, the feature vector also includes two dimensions: voltage difference between cells and temperature difference between cells. The voltage difference between cells is the difference between the maximum and minimum values ​​of the voltage of a single cell in the battery cluster within a preset analysis time window, and the temperature difference between cells is the difference between the maximum and minimum values ​​of the temperature of a single cell in the battery cluster within a preset analysis time window.

4. The method according to claim 1, characterized in that, In step S300, when performing UMLC clustering analysis on the feature vector of each cell, at least the following sub-steps are included: S301. Feature standardization processing: The feature vectors of each cell are scaled and standardized based on the statistics of the corresponding dimensional features within the battery cluster, forming a set of standardized feature vectors; S302. Clustering parameter determination: Perform clustering within the preset range of candidate cluster numbers and calculate the clustering effectiveness evaluation index. Determine the target cluster number based on the principle of optimal evaluation index, and determine the initialization rules and iteration termination conditions for cluster centers. S303. Clustering Iteration and Cluster Center Update: Using the distance between standardized feature vectors as the similarity criterion, cells with similar distances are grouped into the same cluster. The cluster center is updated based on the statistical center of the feature vectors within the cluster until the iteration termination condition is met, and the cluster label of each cell and the cluster center of each cluster are obtained. S304. Identification of the main cluster: Count the number of battery cells contained in each cluster, determine the cluster with the most battery cells as the main cluster, and determine the center of the main cluster as the group consistency benchmark center; S305. Distance Calculation and Consistency Index Generation: Calculate the distance between the standardized feature vector of each cell and the center of its respective cluster, and calculate the distance between each cell and the center of the main cluster. Generate a consistency deviation index based on the distance between the cell and the center of the main cluster. S306. Anomaly Detection: Anomaly identification is performed based on a preset distance threshold and a preset cluster size threshold: when the distance between the battery cell and the center of the main cluster exceeds the preset distance threshold, it is marked as an abnormal battery cell; when the size of the cluster to which the battery cell belongs is less than the size threshold, the battery cells within the cluster are marked as abnormal battery cells. S307. Output Anomaly Set: Output the set of abnormal cells and their corresponding consistency deviation indicators and cluster labels, which serve as input objects for subsequent ICA calculation and difference determination.

5. The method according to claim 1 or 4, characterized in that, In step S300, the distance between each cell and the center of its respective cluster is normalized to form a consistency deviation index. Cells with consistency deviation indices greater than a preset threshold are also marked as abnormal cells.

6. The method according to claim 1, characterized in that, In step S400, when performing incremental capacity analysis (ICA) on the screened abnormal cells, at least the following sub-steps are included: S401. Charging segment determination: Based on the battery cluster current data, under the condition that the current direction is the charging direction, identify the time segment where the current direction is consistent and continuous, and use it as the ICA calculation segment; S402. Data Synchronization and Segment Extraction: Time-align the battery cluster current sequence and the individual cell voltage sequence of the abnormal cell within the ICA calculation segment to form segment current-voltage synchronization data for the abnormal cell, and form corresponding segment current-voltage synchronization data for each cell within the main cluster. S403. Capacity sequence calculation: Taking the start time of the ICA calculation segment as the integration starting point, the battery cluster current sequence is integrated with the charging current in the positive direction to obtain the capacity sequence. The sequence is then matched with each sampling time in the segment, and the capacity increment is used to characterize the change in charging capacity in the segment. S404. Voltage-Capacity Relationship Construction: The capacity sequence is matched with the individual voltage sequence of the abnormal cell according to the sampling time to form voltage-capacity relationship data; the capacity sequence is matched with the individual voltage sequence of each cell in the main cluster according to the sampling time to form voltage-capacity relationship data for each cell respectively. S405. ICA curve generation: The voltage-capacity relationship data is resampled with a uniform step size on the voltage axis, and the capacity-voltage relationship is smoothed after resampling. The capacity after resampling is differentiated with respect to voltage to generate the ICA curves of abnormal cells and cells in the main cluster. S406. Construction of the baseline ICA curve: The ICA curves of each cell in the main cluster are arithmetically averaged to obtain the baseline ICA curve that characterizes the average characteristics of the main cluster. S407. ICA Result Output: Outputs the ICA curve of the abnormal cell and the reference ICA curve.

7. The method according to claim 1, characterized in that, In step S500, the characteristic difference includes at least the curve crossover misalignment criterion and the peak amplitude reduction criterion in the charging platform region: when the abnormal cell ICA curve crosses and misaligns with the reference ICA curve in the charging platform region or the peak amplitude is lower than a preset proportional threshold of the reference peak amplitude, the abnormal cell is determined to be a capacity decay type abnormality; when the main peak position of the abnormal cell ICA curve shifts and the peak amplitude change is less than a preset amplitude threshold, the abnormal cell is determined to be a characteristic performance deviation or polarization increase type abnormality.

8. The method according to claim 1, characterized in that, In step S500, when performing difference determination and diagnostic analysis on abnormal cells, at least the following sub-steps are included: S501. Curve Alignment and Effective Range Determination: Obtain the ICA curve of the abnormal cell and the reference ICA curve, align the two curves, and determine the effective voltage range used for difference judgment; S502. Main Peak Identification: Perform peak search on the ICA curve of the abnormal cell and the reference ICA curve respectively, calculate the peak amplitude, peak position and peak width parameters of each candidate peak, and determine the candidate peak with the largest amplitude that meets the preset peak significance condition as the main peak. S503. Peak Difference Calculation: Based on the main peak of the abnormal cell and the reference main peak, extract the peak position, peak amplitude and peak width parameters respectively, and calculate the peak position difference, peak amplitude difference and peak width difference; S504. Quantification of curve morphology differences: Calculate a set of morphology difference indicators for the ICA curve of abnormal cells and the reference ICA curve, including at least curve correlation, morphology distance and area difference indicators; S505. Regularized Judgment and Anomaly Type Determination: Input the set of peak position difference, peak amplitude difference, peak width difference and morphology difference indicators into a preset judgment rule set. The preset judgment rule set includes at least peak position difference threshold, peak amplitude difference threshold, peak width difference threshold and morphology distance threshold rules. Output the anomaly type according to the preset judgment rule set. S506. Anomaly Cause Location and Confidence Assignment: Output anomaly cause labels based on the corresponding cause mapping relationship of anomaly type in the judgment rule set, and calculate confidence parameters based on the exceedance range of each difference index relative to its threshold. S507. Output Diagnostic Feature Set: The set of abnormality types, abnormality cause labels, confidence parameters, and peak position difference, peak amplitude difference, peak width difference, and morphological difference indicators used for judgment of abnormal cells.

9. The method according to claim 1, characterized in that, In step S600, the consistency detection and analysis results include: abnormal cell identification information, abnormal type determination results, abnormal cause location results, and comparison output of the abnormal cell ICA curve and the reference ICA curve; and maintenance handling suggestions are generated based on the abnormal type determination results. When the abnormality is determined to be a capacity decay type abnormality, a balanced maintenance or replacement handling suggestion is output. When the abnormality is determined to be a polarization rise type abnormality, a connection and heat dissipation status verification suggestion is output.

10. The method according to claim 1, characterized in that, Steps S300 to S600 are repeatedly executed on multiple consecutive time segments corresponding to multiple consecutive charging processes to verify the consistency of the abnormal determination results of the same cell in multiple execution cycles; only when the same cell is determined to be an abnormal cell in a preset number of execution cycles will the cell be output as the final abnormal cell in step S600.

11. A computer program product comprising computer instructions, characterized in that, The computer instructions are used to execute the steps of the rapid detection method for the health status of electrochemical energy storage cells based on UMLC and ICA as described in any one of claims 1 to 10.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the rapid detection method for the health status of electrochemical energy storage cells based on UMLC and ICA as described in any one of claims 1 to 10.

Images