Energy storage lithium ion battery capacity abnormity identification method, system and equipment

Abstract

The invention discloses an energy storage lithium ion battery capacity abnormity identification method, system and equipment, and relates to the technical field of lithium ion batteries. The method aims to overcome the defects that an existing battery capacity diagnosis method needs to depend on complete charging and discharging data, a complex battery model or large-scale historical operation data support, and it is difficult to meet capacity anomaly judgment requirements under the conditions of rapidness, disturbance and data sparseness in actual operation. Obtaining a standing voltage curve after charging and discharging of the energy storage lithium ion battery are stopped, and constructing a standing voltage time sequence; selecting a group of wavelet functions satisfying multi-scale orthogonality, performing wavelet decomposition on the standing voltage sequence, and extracting low-frequency approximate components; performing feature extraction on the obtained low-frequency approximate components to obtain standardized three-dimensional features of each battery monomer; and clustering the batteries with abnormal capacities by adopting a clustering algorithm to obtain a normal battery category and an abnormal battery category. The method is mainly used for identifying the capacity abnormity of the energy storage lithium ion battery.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery technology, and in particular to a method, system and device for identifying abnormal capacity of energy storage lithium-ion batteries. Background Technology

[0002] Lithium-ion batteries are widely used in energy storage systems due to their advantages such as high energy density, long cycle life, high output voltage, no memory effect, and environmental friendliness. Simultaneously, driven by both increasing demand for clean energy and policy support, the application scope of battery energy storage systems continues to expand, from new energy vehicles to large-scale energy storage projects, playing an increasingly important role in the energy transition. Energy storage battery systems consist of thousands of lithium-ion battery cells connected in series and parallel. Due to manufacturing differences, varying operating environments, and different aging rates, capacity deviations inevitably occur between cells during use. When capacities are inconsistent, lower-capacity cells will either fully charge or fully discharge first, triggering voltage protection limits and causing the entire battery pack to prematurely end the charging and discharging process. This not only limits the overall energy utilization rate of the system, but also, in long-term operation, lower-capacity cells are prone to accelerated aging due to overcharging and over-discharging, further widening the differences and creating a vicious cycle.

[0003] Chinese invention patent CN 120214621 A proposes a method for diagnosing short-circuit faults within lithium-ion battery packs based on relaxation time and an adaptive clustering algorithm. This method constructs relaxation time curves using an equivalent first-order circuit model and extracts their statistical features as fault criteria. It then combines this with an adaptive DBSCAN algorithm to perform cluster analysis on individual cells, quantifying outlier levels using critical neighborhood radii, and calculating fault scores to identify and locate short-circuit faults within the battery. However, this method relies on charging segments that meet certain charging duration and capacity conditions. It requires obtaining total current, total voltage, SOC, internal resistance, and individual cell voltage as prerequisites for extracting relaxation segments. Data selection is complex and subject to operational limitations, restricting the algorithm's applicability to incomplete data conditions in real-world scenarios. This method has certain advantages in fully utilizing the static relaxation behavior to reflect internal anomalies, but its feature extraction dimensions are relatively limited. It uses the time constant of the first-order equivalent circuit model as the analysis object and mainly relies on a small number of statistical features, making it difficult to characterize complex fault evolution processes. At the same time, the algorithm depends on the setting of diagnostic threshold parameters, resulting in poor robustness and generalization ability under different types of faults, and it cannot effectively identify non-short-circuit anomalies such as capacity degradation.

[0004] Chinese invention patent CN 113253142 B proposes a capacity anomaly diagnosis method based on wavelet transform and wavelet reconstruction. This method obtains a group of reconstructed signal curves by performing wavelet transform and wavelet reconstruction on the electrical signals generated during the operation of a lithium battery pack, and then uses this group of curves to determine the consistency of the lithium battery pack. However, this method lacks an effective signal filtering mechanism and processes all electrical signals generated during operation. This results in its diagnostic objective remaining at a generalized consistency assessment level, failing to further identify specific anomalies such as capacity decay, making it difficult to provide targeted guidance for battery maintenance. Furthermore, in terms of technical implementation, this method relies on manual observation of the distribution of the reconstructed curve group without providing methods for quantifying or identifying outlier curves. This leads to a highly subjective judgment process, making it difficult to achieve standardized and automated diagnosis.

[0005] Existing technologies include several methods for diagnosing capacity anomalies. Traditional methods primarily involve Coulomb counting and model-based approaches, such as the Equivalent Circuit Model (ECM). First-order RC models are simple but have limited accuracy and cannot accurately describe dynamic responses. While multi-order RC models offer high accuracy, they involve a large number of parameters and unstable identification. Machine learning methods based on full lifecycle data (such as SVR, RF, and LSTM) require a large number of real labels for model training. However, data labels from real-world operating scenarios are often difficult to obtain, and offline test data struggles to reproduce real-world conditions, leading to decreased generalization performance. Feature analysis methods based on impedance spectroscopy (EIS) or impulse response require specialized, expensive equipment for high-precision EIS testing, making integration into lightweight embedded systems difficult. Furthermore, they necessitate equivalent circuit fitting (e.g., ECM parameter extraction) or feature mapping (e.g., Nyquist graphical analysis, spectral analysis) of frequency domain data. Impulse response testing requires specific temperatures and sleep times, making continuous testing during real-world operation difficult, time-consuming, and time-consuming, hindering practical deployment. The above methods generally suffer from the following problems: they rely on complete charge and discharge data, complex battery models, or large-scale historical operating data, making it difficult to meet the needs of capacity anomaly detection under rapid, disturbed, and data-sparse conditions in actual operation.

[0006] Therefore, there is a need for a method, system, and device for identifying abnormal capacity in energy storage lithium-ion batteries that can accurately identify battery capacity abnormalities solely based on the static voltage curve after charging and discharging has stopped. Summary of the Invention

[0007] This invention addresses the shortcomings of existing battery capacity diagnosis methods, which rely on complete charge-discharge data, complex battery models, or large-scale historical operating data, making it difficult to meet the needs of capacity anomaly identification under rapid, disturbed, and data-sparse conditions in actual operation. It provides a method, system, and device for identifying capacity anomalies in energy storage lithium-ion batteries that can accurately identify battery capacity anomalies based solely on the static voltage curve after charging and discharging has stopped.

[0008] The present invention provides a method for identifying abnormal capacity in energy storage lithium-ion batteries, comprising the following steps: S1. Obtain the static voltage curve of the energy storage lithium-ion battery after charging and discharging has stopped, and construct the static voltage time series. S2. Select a set of wavelet functions ψ(t) that satisfy multi-scale orthogonality, and then apply this to the resting voltage sequence. Perform wavelet decomposition to extract low-frequency approximate components; S3. Extract features from the obtained low-frequency approximation components to obtain the standardized three-dimensional features of each battery cell. S4. Use a clustering algorithm to cluster the three-dimensional features of all batteries to obtain normal battery categories and abnormal battery categories.

[0009] Further: In S4, the density clustering DBSCAN algorithm is used. The standardized three-dimensional features of each battery cell are used as input to the density clustering DBSCAN algorithm to obtain the clustering results. It is determined whether there are three-dimensional features in the clustering results that cannot be assigned to any cluster. If so, the battery cell to which the three-dimensional feature belongs is a cell with abnormal capacity.

[0010] Further: In S2, the static voltage sequence The formula is: ; In the formula, Let be the voltage value at time t, and J be the preset maximum number of decomposition layers. The minimum number of decomposition layers is given by k, where k is the translation parameter. Let be the wavelet coefficients of the j-th layer, representing the high-frequency detail components of that layer; Let be the wavelet function of the j-th layer, used to extract high-frequency details of that layer; For the first The scale factor of a layer represents the low-frequency approximate component of that layer; For the first The scaling function at the layer scale is used to extract the low-frequency approximation components of that layer.

[0011] Further: In S2, the wavelet decomposition process is represented as follows: ; In the formula, This is the low-frequency approximation component of the j-th layer; In the high-frequency detail component sequence of the j-th layer; For the low-frequency approximation component of layer j-1, These are the coefficients of a wavelet low-pass filter. denoted as wavelet high-pass filter coefficients; n represents the sampling points, and b represents the sampling points of the filter coefficients.

[0012] Furthermore: In S3, the feature extraction is calculated as follows: ; ; ; In the formula, E represents the energy characteristic. Here, H represents the low-frequency approximation component of the J-th layer obtained after wavelet decomposition, and H is the entropy feature. The standard deviation is the feature, and N is the number of samples. Let be the mean of the low-frequency approximate components of the Jth layer, n=1,2,...,N.

[0013] Furthermore: In S4, the clustering is based on "density reachability," dividing the samples into three categories: Key point: It contains at least M points in its ε-neighborhood; Boundary point: less than M in its neighborhood, but within the neighborhood of a certain core point; Noise points: These are neither core points nor boundary points and are considered outliers.

[0014] The identification system for implementing the method for identifying abnormal capacity of energy storage lithium-ion batteries according to the present invention includes a battery static data acquisition module, a wavelet decomposition module, a feature extraction module, and a clustering diagnosis module. The battery static data acquisition module is used to acquire the static voltage curve of the energy storage lithium-ion battery after charging and discharging has stopped, and to construct a static voltage time series. The wavelet decomposition module is used to perform multi-scale wavelet decomposition on the static voltage time series and extract its low-frequency approximate components. The feature extraction module is used to obtain the standardized three-dimensional features of each battery cell; The clustering diagnosis module uses a clustering algorithm to diagnose abnormalities in battery capacity status based on the extracted three-dimensional features.

[0015] The present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor is used to execute the computer program to implement the method for identifying abnormal capacity of an energy storage lithium-ion battery.

[0016] The beneficial effects of this invention are: This invention provides a method for identifying abnormal capacity in energy storage lithium-ion batteries. Without requiring complete and continuous charge-discharge curves, it utilizes the static voltage curve after charging and discharging has stopped to accurately identify abnormal battery capacity.

[0017] This invention discloses a method for identifying capacity anomalies in energy storage lithium-ion batteries. It utilizes wavelet transform to extract three-dimensional features such as energy, entropy, and standard deviation from the low-frequency approximate components of the resting voltage curve at different scales. This comprehensively characterizes the capacity state changes inherent in the cell's resting behavior, enabling rapid identification of batteries with significant capacity degradation, abrupt changes, or deviations from the group average. Compared to traditional characterization methods using charge-discharge data, the resting voltage signal is easily obtained, less susceptible to operating condition disturbances, and possesses higher engineering practicality. Furthermore, this invention does not require preconditions such as total current, total voltage, SOC, internal resistance, or individual cell voltage data, avoiding the introduction of preceding errors, simplifying the diagnostic process, enhancing adaptability to non-standard operating data, and making it suitable for capacity anomaly screening tasks under complex operating conditions. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the practical application of a method for identifying abnormal capacity in energy storage lithium-ion batteries. Figure 2 This is a schematic diagram of the voltage curve of an energy storage battery module during operation; where the horizontal axis represents time and the vertical axis represents the voltage value. Figure 3 This is a schematic diagram of the process of extracting low-frequency components using discrete wavelet transform; Figure 4 This is a schematic diagram of the wavelet transform result of the static voltage curve of a single cell in a battery pack. Figure 5 This is a schematic diagram illustrating the results of identifying abnormal capacity in an energy storage lithium-ion battery. Detailed Implementation

[0019] The following are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. The embodiments described below are only for explaining the present invention and should not be construed as limiting the present invention. The scope of protection of the present invention should be determined by the scope of the claims. The embodiments of the present invention are described in detail below. In order to facilitate the description of the present invention and simplify the description, the technical terms used in the specification of the present invention should be interpreted broadly, including but not limited to conventional alternatives not mentioned in this application, and including both direct and indirect implementation methods.

[0020] Example 1 Combination Figures 1-5 This embodiment describes a method for identifying abnormal capacity in energy storage lithium-ion batteries, comprising the following steps: S1. Obtain the static voltage curve of the energy storage lithium-ion battery after charging and discharging has stopped, and construct the static voltage time series.

[0021] Figure 2 The diagram shows the voltage curve of the energy storage battery module during operation, including the charging / discharging process and the resting process. Under complex charging / discharging conditions, the voltage characteristics of low-capacity batteries are difficult to distinguish directly from normal batteries. Furthermore, the charging / discharging conditions of batteries are complex and variable, with disturbances and sparse data, making complete data acquisition difficult. This embodiment starts with the relatively easily obtainable resting period. First, it obtains the resting voltage curve of the energy storage lithium-ion battery after charging and discharging stops when there are no scheduling instructions, and constructs a resting voltage sequence. w is the length of the voltage sequence.

[0022] Discrete wavelet transform (DWT) is a decomposition technique that breaks down a time-domain signal into multiple frequency bands. It possesses both time-domain and frequency-domain localization analysis capabilities, making it highly suitable for processing non-stationary signals with local abrupt changes. Wavelet transform achieves good time resolution in the high-frequency band and good frequency resolution in the low-frequency band, thus making it particularly suitable for slowly changing signals such as voltage that contain local fluctuations.

[0023] S2. Using discrete wavelet transform on the stationary voltage sequence Decompose it.

[0024] Discrete wavelet transform (DWT) is a decomposition technique that breaks down a time-domain signal into multiple frequency bands. It possesses both time-domain and frequency-domain localization analysis capabilities, making it highly suitable for processing non-stationary signals with local abrupt changes. Wavelet transform achieves good time resolution in the high-frequency band and good frequency resolution in the low-frequency band, thus making it particularly applicable to slowly changing signals with local fluctuations, such as stationary voltage.

[0025] Specifically, a set of wavelet functions ψ(t) that satisfy multi-scale orthogonality are selected, and the static voltage sequence is... Perform layer-by-layer decomposition.

[0026] ; In the formula, Let be the voltage value at time t, and J be the preset maximum number of decomposition layers. The minimum number of decomposition layers is given by k, where k is the translation parameter (lateral offset, representing the position in the time domain). Let be the wavelet coefficients of the j-th layer, representing the high-frequency detail components of that layer. Let be the wavelet function of the j-th layer, used to extract high-frequency details of that layer. For the first The scale factor of a layer represents the low-frequency approximate component of that layer. For the first The scaling function at the layer scale is used to extract the low-frequency approximation components of that layer.

[0027] The multi-scale orthogonality refers to the fact that the selected wavelet function and its scaled and translated family of functions are orthogonal to each other in different frequency ranges and at different time positions. That is, the basis functions in the same layer are uncorrelated with each other, and the basis functions in different layers are also uncorrelated with each other, thus forming a complete orthogonal basis in the entire signal space.

[0028] In discrete implementations, wavelet transform decomposes and reconstructs the signal using a filter bank, namely a pair of low-pass and high-pass filters. The wavelet decomposition process is represented as: ; In the formula, This is the low-frequency approximation component of the j-th layer; In the high-frequency detail component sequence of the j-th layer; For the low-frequency approximation component of layer j-1, These are the coefficients of a wavelet low-pass filter. denoted as wavelet high-pass filter coefficients; n represents the sampling points, and b represents the sampling points of the filter coefficients.

[0029] The "low frequency" and "high frequency" are frequency ranges automatically defined through wavelet decomposition. Wavelet decomposition separates the signal at each level using low-pass and high-pass filters: the low-pass portion forms low-frequency approximate components, primarily reflecting the slow variation trend of the stationary voltage signal; the high-pass portion forms high-frequency detail components, primarily reflecting the rapid fluctuations of the voltage signal. Taking a stationary voltage signal with a sampling frequency of f as an example, the frequency range of the approximate components at the j-th level after wavelet decomposition is... The frequency range of the detail components is .

[0030] The Daubechies wavelet series of functions ensures that the components at each decomposition level are independent and do not interfere with each other, thus enabling the low-frequency approximation components to accurately represent the slow-changing trend of the voltage. In this embodiment, the Daubechies-6 wavelet function is used to perform discrete wavelet decomposition on the stationary voltage signal V(t), with a sampling frequency of 1Hz for V(t), and the maximum decomposition level J=6 is taken as an example.

[0031] Figure 3This diagram illustrates the discrete wavelet decomposition process. The original stationary voltage sequence V(t) is decomposed layer by layer. The first layer yields low-frequency approximate components and high-frequency detail components. The low-frequency approximate components are then decomposed layer by layer until the maximum number of decomposition layers is reached. A1-A6 represent the low-frequency approximate components of each layer, used to characterize the slow voltage change of the battery during the stationary process. D1-D6 represent the high-frequency detail components of each layer, mainly containing noise and short-term disturbances. The frequency band corresponding to the low-frequency approximate components of the first layer is... The frequency band of the high-frequency detail components is The frequency band corresponding to the low-frequency approximation component of the second-level decomposition is: The frequency band of the high-frequency detail components is And so on, up to the frequency range of the low-frequency approximation component in the 6th layer. It belongs to the extremely low frequency range, and the high frequency components correspond to The frequency range.

[0032] Figure 4 The figure shows the low-frequency approximate component curves of the static voltage sequence of different battery cells in a battery module in this embodiment. Cell 8 has a lower capacity than other battery cells in the same group, V(t) sampling frequency is 1Hz, and the maximum decomposition layer is 6. Starting from the 4th decomposition layer, the low-frequency approximate component curve of cell 8 separates from other normal cells. At this time, the frequency range corresponding to the low-frequency approximate component curve is... As the number of decomposition layers increases, the low-frequency approximation curve of the abnormal capacity cells deviates more significantly. During the resting process, the electrochemical reaction of the battery does not completely stop, but rather slow side reactions occur (such as SEI film repair, electrolyte decomposition, and ion concentration gradient dispersion), whose time constants are mainly in the low-frequency range (<0.1Hz). These processes are closely related to capacity decay, causing the resting voltage to show a slow trend of gradually approaching equilibrium. Therefore, extracting the low-frequency approximation component of the resting voltage can effectively preserve the long-term trend reflecting the actual electrochemical evolution and capacity change of the battery, while filtering out irrelevant interference, thereby improving the stability of capacity feature extraction. It should be noted that in practical applications, the maximum number of decomposition layers in this invention is not the only limitation; the number of decomposition layers needs to be determined based on the sampling frequency of the resting voltage and computational resources.

[0033] S3. Extract the three-dimensional features of the low-frequency approximation components of the J-th layer.

[0034] Specifically, the three-dimensional features include energy features, entropy features, and standard deviation features, and the specific calculation methods are as follows: ; ; ; In the formula, E represents the energy characteristic. Here, H represents the low-frequency approximation component of the J-th layer obtained after wavelet decomposition, and H is the entropy feature. The standard deviation is the feature, and N is the number of samples. Let be the mean of the low-frequency approximate components of the Jth layer, n=1,2,...,N.

[0035] S4. After extracting the above three-dimensional features from the static voltage curve through wavelet transform, a clustering algorithm is used to cluster the three-dimensional features of all batteries to obtain normal battery categories and abnormal battery categories.

[0036] To avoid manually setting the number of categories and improve sensitivity to isolated outliers, this embodiment employs the density-based clustering algorithm DBSCAN. DBSCAN is a density-based clustering algorithm capable of automatically discovering clusters of arbitrary shapes and identifying outliers. DBSCAN does not rely on a preset number of categories, can identify irregularly shaped clusters, and can automatically mark noise points, making it particularly suitable for battery status monitoring in cases with a small number of outliers, non-spherical structures, or uneven density distribution within clusters. Based on "density accessibility," DBSCAN divides samples into three categories: Key point: It contains at least M points in its ε-neighborhood; Boundary point: less than M in its neighborhood, but within the neighborhood of a certain core point; Noise points: These are neither core points nor boundary points and are considered outliers. The clustering process starts from a core point and recursively expands densely connected regions to form a cluster until it can no longer be expanded.

[0037] "Density reachability" means that for a battery sample p, if sample q falls within the ε-neighborhood of sample p, then q is said to be density reachable from p. Furthermore, if there exists a series of samples... Each sample point Previous sample point If the density can be reached, then it is called... from Density reachability. DBSCAN clusters samples by recursively expanding the region based on this density reachability.

[0038] Specifically: First, the three-dimensional features are standardized using the z-transform method: ; in, These are the standardized feature values.

[0039] The standardized three-dimensional features of each battery cell are used as input to the density clustering DBSCAN algorithm to obtain the clustering results. It is then determined whether there are any three-dimensional features in the clustering results that cannot be assigned to any cluster. If so, the battery cell to which the three-dimensional feature belongs is a cell with abnormal capacity.

[0040] Taking a 16-cell series-connected battery module as an example, the capacity anomaly diagnosis execution steps are S1-S4.

[0041] First, the static voltage sequence of 16 cells is obtained according to step S1, and the sampling frequency is 1Hz. According to step S2, discrete wavelet transforms are performed on the 16 stationary voltage sequences, with the maximum decomposition level set to 6, to obtain the following results. Figure 4 The sixth layer low-frequency approximation component of the 16 cells shown; According to step S3, the three-dimensional features of the sixth layer low-frequency approximation component of each of the 16 cells are extracted respectively; Finally, according to step S4, the 16 sets of three-dimensional features are standardized and used as input for density clustering algorithm for diagnosis and identification. The ε-neighborhood is set to 2, and the minimum number of samples M within a cluster is set to 10. Cells that cannot be assigned to any cluster are considered to have abnormal capacity. The judgment results are as follows: Figure 5 As shown, Figure 5 This is the result diagram of a capacity anomaly diagnosis for a battery module consisting of 16 cells connected in series. Cell number 8 has a lower capacity than the other cells in the module. Cell 8 is far from the normal cluster and cannot be assigned to any cluster, thus it is marked as a capacity anomaly. Cells numbered 10 and 11 are located in the same position. Although cells numbered 3, 4, 5, and 7 are also some distance from the other cells, they are considered normal cells because the distance does not exceed the threshold set for the neighboring area.

[0042] Example 2 A lithium-ion battery capacity anomaly identification system includes: Battery resting data acquisition module: used to acquire the voltage of the battery during the resting process in the energy storage system and construct a voltage time series; Wavelet decomposition module: used to perform multi-scale wavelet decomposition on the static voltage time series and extract its low-frequency approximate components; Feature extraction module: used to obtain the standardized three-dimensional features of each battery cell; Clustering Diagnosis Module: This module uses a clustering algorithm to diagnose abnormalities in battery capacity status based on the extracted 3D features.

[0043] Example 3 An electronic device includes a memory and a processor for storing a computer program on the memory and running on the processor, wherein the processor executes the computer program to implement the aforementioned method for identifying abnormal capacity of an energy storage lithium-ion battery.

Claims

1. A method for identifying abnormal capacity in energy storage lithium-ion batteries, characterized in that, Includes the following steps: S1. Obtain the static voltage curve of the energy storage lithium-ion battery after charging and discharging has stopped, and construct the static voltage time series. S2. Select a set of wavelet functions ψ(t) that satisfy multi-scale orthogonality, and then apply this to the resting voltage sequence. Perform wavelet decomposition to extract low-frequency approximate components; S3. Extract features from the obtained low-frequency approximation components to obtain the standardized three-dimensional features of each battery cell. S4. Use a clustering algorithm to cluster the three-dimensional features of all batteries to obtain normal battery categories and abnormal battery categories.

2. The method for identifying abnormal capacity of energy storage lithium-ion batteries according to claim 1, characterized in that, In S4, the density clustering DBSCAN algorithm is used. The standardized three-dimensional features of each battery cell are used as input to the density clustering DBSCAN algorithm to obtain the clustering results. It is determined whether there are three-dimensional features in the clustering results that cannot be assigned to any cluster. If so, the battery cell to which the three-dimensional feature belongs is a cell with abnormal capacity.

3. The method for identifying abnormal capacity of energy storage lithium-ion batteries according to claim 1, characterized in that, In S2, the static voltage sequence The formula is: ； In the formula, Let be the voltage value at time t, and J be the preset maximum number of decomposition layers. The minimum number of decomposition layers is given by k, where k is the translation parameter. Let be the wavelet coefficients of the j-th layer, representing the high-frequency detail components of that layer; Let be the wavelet function of the j-th layer, used to extract high-frequency details of that layer; For the first The scale factor of a layer represents the low-frequency approximate component of that layer; For the first The scaling function at the layer scale is used to extract the low-frequency approximation components of that layer.

4. The method for identifying abnormal capacity of energy storage lithium-ion batteries according to claim 1, characterized in that, In S2, the wavelet decomposition process is represented as follows: ； In the formula, This is the low-frequency approximation component of the j-th layer; In the high-frequency detail component sequence of the j-th layer; For the low-frequency approximation component of layer j-1, These are the coefficients of the wavelet low-pass filter. denoted as wavelet high-pass filter coefficients; n represents the sampling points, and b represents the sampling points of the filter coefficients.

5. The method for identifying abnormal capacity of energy storage lithium-ion batteries according to claim 1, characterized in that, In S3, the feature extraction is calculated as follows: ； ； ； In the formula, E represents the energy characteristic. Here, H represents the low-frequency approximation component of the J-th layer obtained after wavelet decomposition, and H is the entropy feature. The standard deviation is the feature, and N is the number of samples. Let be the mean of the low-frequency approximate components of the Jth layer, n=1,2,...,N.

6. The method for identifying abnormal capacity of energy storage lithium-ion batteries according to claim 1, characterized in that, In S4, the clustering is based on "density reachability" and divides the samples into three categories: Key point: It contains at least M points in its ε-neighborhood; Boundary point: less than M in its neighborhood, but within the neighborhood of a certain core point; Noise points: These are neither core points nor boundary points and are considered outliers.

7. An identification system for implementing the method for identifying abnormal capacity of an energy storage lithium-ion battery as described in any one of claims 1-6, characterized in that, It includes a battery static data acquisition module, a wavelet decomposition module, a feature extraction module, and a clustering diagnosis module; The battery static data acquisition module is used to acquire the static voltage curve of the energy storage lithium-ion battery after charging and discharging has stopped, and to construct a static voltage time series. The wavelet decomposition module is used to perform multi-scale wavelet decomposition on the static voltage time series and extract its low-frequency approximate components. The feature extraction module is used to obtain the standardized three-dimensional features of each battery cell; The clustering diagnosis module uses a clustering algorithm to diagnose abnormalities in battery capacity status based on the extracted three-dimensional features.

8. An electronic device, characterized in that, The device includes a memory and a processor for storing a computer program on the memory and running on the processor, the processor being used to execute the computer program to implement the method for identifying abnormal capacity of an energy storage lithium-ion battery as described in any one of claims 1-6.

Images